50
Foreign Direct Investment in Services and Manufacturing Productivity: Evidence for Chile Ana M. Fernandes a Caroline Paunov b The World Bank OECD March 2011 Journal of Development Economics forthcoming Abstract This paper examines the impact of substantial foreign direct investment (FDI) inflows in producer service sectors on the total factor productivity (TFP) of Chilean manufacturing firms. Positive effects are obtained in firm fixed effects instrumental variables regressions and show that forward linkages from FDI in services explain 7% of the observed increase in Chile’s manufacturing users’ TFP. Our findings also suggest that service FDI fosters innovation activities in manufacturing. Moreover, we show that service FDI offers opportunities for laggard firms to catch up with industry leaders. Keywords: Total Factor Productivity, Service Liberalization, Foreign Direct Investment, Chile, Firm Heterogeneity. JEL Classification codes: D24, L8, L9, F21, F23. a Ana Margarida Fernandes (corresponding author). The World Bank. Development Research Group. 1818 H Street NW, Washington DC, 20433, U.S.A. Email: [email protected]. b Caroline Paunov. OECD. Directorate for Science, Technology and Industry. 2, rue André Pascal, 75 775 Paris Cedex 16, France. Email: [email protected] and [email protected]. This paper is a modified version of the World Bank Policy Research Working Paper 4730. The authors would like to thank Eric Verhoogen (the co-editor) and two anonymous referees as well as Richard Disney, Ana Paula Fernandes, Jonathan Haskel, Beata Javorcik, Raimundo Soto, Peter-Paul Walsh, and seminar participants at Indiana University, the Chilean Central Bank, the University of Chile, Queen Mary University of London, the 6 th International Industrial Organization Conference, the OECD Development Centre, the 2008 Empirical Investigations in International Economics Conference in Slovenia, the 2008 North American Summer Meetings of the Econometric Society, the 2008 EEA ESEM Meetings, the 2008 European Trade Study Group Conference, 2008 LACEA-LAMES Meetings, the 4 th MEIDE Conference in Estonia for valuable comments. Support from the governments of Norway, Sweden and the United Kingdom through the Multi-Donor Trust Fund for Trade and Development is gratefully acknowledged. The findings expressed in this paper are those of the authors and do not necessarily represent the views of the World Bank or the OECD.

Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

  • Upload
    hakhanh

  • View
    216

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

Foreign Direct Investment in Services and Manufacturing Productivity

Evidence for Chile

Ana M Fernandes a Caroline Paunov

b

The World Bank

OECD

March 2011

Journal of Development Economics forthcoming

Abstract

This paper examines the impact of substantial foreign direct investment (FDI) inflows in

producer service sectors on the total factor productivity (TFP) of Chilean manufacturing

firms Positive effects are obtained in firm fixed effects instrumental variables regressions

and show that forward linkages from FDI in services explain 7 of the observed increase

in Chilersquos manufacturing usersrsquo TFP Our findings also suggest that service FDI fosters

innovation activities in manufacturing Moreover we show that service FDI offers

opportunities for laggard firms to catch up with industry leaders

Keywords Total Factor Productivity Service Liberalization Foreign Direct Investment

Chile Firm Heterogeneity

JEL Classification codes D24 L8 L9 F21 F23

a Ana Margarida Fernandes (corresponding author) The World Bank Development Research Group 1818 H Street

NW Washington DC 20433 USA Email afernandesworldbankorg b Caroline Paunov OECD Directorate for Science Technology and Industry 2 rue Andreacute Pascal 75 775 Paris Cedex

16 France Email carolinepaunovoecdorg and carolinepaunovgmailcom

This paper is a modified version of the World Bank Policy Research Working Paper 4730 The authors would like to

thank Eric Verhoogen (the co-editor) and two anonymous referees as well as Richard Disney Ana Paula Fernandes

Jonathan Haskel Beata Javorcik Raimundo Soto Peter-Paul Walsh and seminar participants at Indiana University the

Chilean Central Bank the University of Chile Queen Mary University of London the 6th International Industrial

Organization Conference the OECD Development Centre the 2008 Empirical Investigations in International

Economics Conference in Slovenia the 2008 North American Summer Meetings of the Econometric Society the 2008

EEA ESEM Meetings the 2008 European Trade Study Group Conference 2008 LACEA-LAMES Meetings the 4th

MEIDE Conference in Estonia for valuable comments Support from the governments of Norway Sweden and the

United Kingdom through the Multi-Donor Trust Fund for Trade and Development is gratefully acknowledged The

findings expressed in this paper are those of the authors and do not necessarily represent the views of the World Bank

or the OECD

1

1 Introduction

Foreign direct investment (FDI) inflows into the service sector experienced a boom

during the 1990s By 2002 services accounted for 60 of the world stock of FDI a four-

fold increase since 1990 (UNCTAD 2004) In developed and developing countries alike

the main recipients of FDI have been profit-seeking producer services which range from

network-intensive services such as electricity telecommunications and transport to

finance and business services These sectors are characterized by the facilitating and

intermediating role which they play for downstream user firms (Francois 1990) Thus

producer service sectors are an intricate component of a countryrsquos business environment

In emerging economies where manufacturing firms are constrained by cumbersome

business environments it is particularly relevant to understand how the performance of

service sectors can be improved and how that supports business development and thus

overall economic growth FDI is a potentially powerful means to achieve such

improvements as it might increase the quality and variety of services available as well as

lower their cost Manufacturing firms may benefit from their interaction with foreign

services suppliers through spillovers of management organizational marketing or

technological knowledge (Markusen 1989 Rivera-Batiz and Rivera-Batiz 1992)

Despite the relevance of this topic the effects of vertical linkages resulting from the

openness of producer services to FDI on downstream manufacturing firms have not been

widely documented (Hoekman 2006) This paper attempts to fill this gap by addressing

the following question did the increased penetration of FDI into producer service sectors

in Chile benefit total factor productivity (TFP) of manufacturing firms between 1995 and

2004 Chile is an interesting economy to study as its service sector received large FDI

2

inflows during the 1990s We show that foreign-owned service firms perform better in

terms of labor productivity and innovation than their domestic counterparts The fact that

this finding refers also to greenfield FDI - ie it is not simply foreign investors cherry-

picking the best-performing domestic service firms - suggests these firms introduced

superior standards in the Chilean services sector thus potentially offering improved

services for manufacturing firm users

Evaluating the causal impact of service FDI on manufacturing firm TFP is

challenging due to several valid endogeneity concerns discussed below Our strategy to

identify that impact is to estimate by instrumental variables (IV) a regression of firm TFP

on a service FDI linkage measure controlling for firm fixed effects as well as industry-

year and region-year fixed effects Our study uses TFP measures obtained as residuals

from production functions estimated following the methodologies of Levinsohn and

Petrin (2003) Olley and Pakes (1996) and Ackerberg et al (2006) to correct for the

endogeneity of input choices - including the choice of service inputs - with respect to

productivity Our service FDI linkage measure is defined as service FDI penetration

weighted by firm-level intensity of service usage The rationale underlying this novel

measure is the expectation that firms that use services more intensively benefit more from

any positive effects of service FDI Our measure of service usage intensity is based on

historic values for each firm following Blalock and Gertler (2009) to avoid the potential

endogeneity between firm TFP and service intensity Our choice of instrumental variables

estimation addresses another potential endogeneity concern related to fact that large FDI

inflows into Chilean service sectors may have responded to the strong performance of the

downstream manufacturing users Service FDI penetration weighted by the historic firm-

3

level intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile - Spain and the US - weighted

similarly The identification assumption is that while increases in total FDI outflows into

service sectors by those two countries affect service FDI penetration in Chilean service

sectors they are not correlated with the TFP of manufacturing firms in the country

through any other channel

We find evidence of a positive and significant effect of service FDI on Chilean

manufacturing firmsrsquo TFP This effect is robust to the control for differential productivity

trends across plants with different services intensity and to allowing the effects of service

FDI to operate with lags Alternative measures of service usage weights and service FDI

penetration including the use of industry-level weights from an input-output table

confirm our main findings The positive and significant effect of service FDI is also

obtained when considering a one-stage regression where output depends on inputs as well

as the service FDI linkage measure We also provide suggestive evidence supporting the

hypothesis that service FDI stimulates innovation by Chilean manufacturing firms

Moreover interestingly we find that service FDI offers an opportunity for laggard firms

to catch up in terms of TFP with industry leaders From a policy perspective this finding

is particularly relevant since it suggests that the benefits from service liberalization do not

accrue mostly to leading firms but seem to offer opportunities for firms that are further

behind

Our estimates suggest that a one-standard deviation increase in service FDI would

lead to an increase in TFP of Chilean firms by 3 all else constant The economic

magnitude of this impact is that forward linkages from FDI in services accounted for at

4

least 7 of the observed increase in manufacturing usersrsquo TFP in Chile during the sample

period This economic impact is quite meaningful in light of the finding by Haskel et al

(2007) that horizontal spillovers from manufacturing FDI explain a roughly similar share

of manufacturing TFP growth in the UK during the 1973-1992 period The positive

effects of service FDI on manufacturing firmsrsquo TFP may capture to some extent an

unmeasured decline in quality-adjusted services prices but also the spillover of

managerial and organizational knowledge from service providers to manufacturing users

The microeconomic evidence provided by our study contributes to the emerging

literature on the impact of services liberalization on growth and the performance of

services users At the macro level Mattoo et al (2006) and Eschenbach and Hoekman

(2006) show that countries with liberalized service sectors grow faster once all standard

growth correlates are controlled for Based on computable general equilibrium models

Konan and Maskus (2006) and Jensen et al (2007) argue that business services

liberalization could bring large GDP gains to Tunisia and Russia respectively1 The main

mechanism for these gains is the increase in the number of services available for

manufacturing users as a result of FDI2 At the industry level Francois and Woerz (2008)

show that the increased openness of business services through exports and FDI has strong

positive effects on exports value added and employment of manufacturing industries in

the OECD while Fernandes (2009) estimates positive and significant effects of

liberalization of finance and infrastructure on labor productivity of downstream

manufacturing industries in Eastern European countries At the firm-level Arnold et al

1 Markusen et al (2005) also show important GDP gains from services liberalization based on general equilibrium

simulations for a hypothetical country In their model the presence of foreign-owned service providers allows final

goods producers to rely on more specialized expertise 2 This increase in the number of services increases the TFP of manufacturing firms through a Dixit-Stiglitz-Ethier

framework (Dixit and Stiglitz 1977 Ethier 1982)

5

(2007) show significant positive effects of services liberalization in the Czech Republic

on manufacturing firmsrsquo TFP while Arnold et al (2010) find significant positive effects

of banking telecommunications and transport reforms on Indian manufacturing firmsrsquo

TFP Finally Javorcik and Li (2008) estimate a positive effect of FDI in Romaniarsquos retail

sector on the TFP of manufacturing suppliers to that sector3

Relative to the existing literature the contribution of our study is three-fold First

we exploit the heterogeneity in service usage by considering firm-level measures of the

intensity of service usage Evidence on the usage of services shows considerable

heterogeneity across Chilean firms which suggests that the practice followed in all the

aforementioned studies of using industry-level usage measures based on input-output

tables may be inappropriate The advantage of our measures is that they enable us to

identify the intensive service users within each industry Second we follow a rigorous

empirical approach by relying on firm fixed effects IV estimation to identify the causal

effects of services FDI on TFP Hence our specifications exploit the within-firm variation

in TFP in response to instrumented changes in the service FDI linkage measure Third

we go beyond previous studies by exploring the nature of the effects of service FDI

allowing for heterogeneity across industries relating to their potential for innovation We

also focus on heterogeneous effects across firms relating to their distance to

technologically advanced firms

The remainder of the paper proceeds as follows Section 2 describes recent trends in

FDI in services in Chile and the effects of FDI in services Section 3 describes the data

3 By considering the potential role of knowledge spillovers from service providers to manufacturing users our study

also relates to the literature on vertical spillovers from manufacturing FDI which are shown to be more important than

horizontal spillovers by Javorcik (2004) Kugler (2006) Blalock and Gertler (2008) Damijan et al (2008) and Marcin

(2008) A rationale provided in this literature for vertical forward linkages is that foreign suppliers provide assistance

and complementary services to local buyers

6

Section 4 describes our empirical specification Section 5 discusses our main results and

robustness checks Section 6 discusses extensions to our main results Section 7

concludes

2 FDI in Services Trends and Effects

21 Trends in FDI in Services in Chile

Over the last three decades liberalization privatization and deregulation reforms in

Chile opened its economy to trade and investment more than any other country in Latin

America (Moreira and Blyde 2006)4 In the 1980s most FDI inflows were related to the

extraction and processing of natural resources while in the 1990s inflows into service

sectors took on a leading role with electricity water transport telecommunications and

business services representing about 60 of net FDI inflows during the 1996-2001

period5 Figure 1 shows that these substantial inflows resulted in a growing FDI stock in

the main service sectors in Chile Also the ratio of FDI to output increased substantially

in most Chilean service sectors over the 1990s as shown in Figure 26

The large FDI inflows in Chile during the 1990s reflect first and foremost a

worldwide increase in FDI in services mainly motivated by the interest of multinationals

(MNCs) to become global service providers by gaining access to domestic and regional

markets particularly in the developing world (UNCTAD 2004) In sectors such as

electricity Chilean firms were privatized before 1990 and later acquired by foreign

4 FDI in Chile is governed by Decree Law 600 in place since 1974 which regulates conditions for market entry

capitalization and foreign capital remittances (ECLAC 2000) The decree law grants equal treatment to foreign and

domestic investments in mining manufacturing and most service sectors the exceptions being professional services

such as engineering or legal services (Moreira and Blyde 2006) 5 FDI inflows reached a peak in 1999 in the electricity and water sector due to the purchase of Enersis and Endesa-

Chile by the Spanish electricity firm Endesa-Spain (ECLAC 1999) 6 The computation of the variables shown in Figures 1 and 2 is described in Section 4 and in the Appendix

7

players Global MNCs identified Chilersquos largely privately-owned firms as an attractive

investment opportunity to consolidate their positions in Latin America (ECLAC 2000)

22 Effects of FDI in Services

FDI in the services sector can have four effects within the sector price reductions

quality improvements increased variety and knowledge spillovers7 First FDI is likely

to increase competition in local markets resulting in price reductions as incumbent firms -

eg in electricity and telecommunications sectors - no longer retain the rents they

obtained from being previously monopoly providers The available evidence for banking

electricity and telecommunications confirms price decreases for Chile (Stehmann 1995

Claessens et al 2001 Pollitt 2004) Second FDI may lead to service quality

improvements due to competition and the superior technological organizational and

managerial know-how of foreign-owned service providers FDI can also provide the

necessary finance for major upgrades and the expansion of existing electricity and

telecommunications networks improving the reliability of provision UNCTAD (2004)

and World Bank (2004) report evidence of such developments for Latin America Third

FDI may increase the variety of services provided including new technologically

advanced services or services provided to new regions or new types of clients FDI had

such effects in the Chilean telecommunications sector (ECLAC 2000) Fourth FDI may

result in leaking of managerial marketing and organizational know-how and best

practices (eg linked to the environment or labor codes) from foreign-owned to

domestic-owned services providers (Miroudout 2006)

7 FDI can also entail costs (1) foreign ownership in inherently monopolistic sectors such as electricity or

telecommunications may result in higher prices unless the regulatory system is well defined and managed by the

government and (2) foreign ownership may crowd out domestic firms eg in the banking sector (UNCTAD 2004)

8

The aforementioned positive effects of service FDI are based on the premise that

foreign-owned firms are better performers offering superior services and being more

productive than their domestic counterparts However due to data limitations evidence of

the superiority of foreign-owned service firms is rare The Fourth Chilean Innovation

Survey is a notable exception covering 612 service firms in the electricity generation real

estate financial intermediation business activities and transport storage and

communication sectors8 The survey collected data for 2003-2004 on firm innovation

outcomes accounting variables and basic characteristics Using this data we examine

whether foreign ownership is associated with better performance for Chilean service

firms If foreign investors acquire the best performing domestic service firms then a

positive effect of foreign ownership on firm performance would simply indicate the

endogeneity of the ownership status rather than the intrinsic advantages - eg better

technology - of the foreign parent company However this problem would not arise for

greenfield FDI Hence for each Chilean service firm we obtained information on

whether foreign ownership is due to greenfield FDI or foreign acquisition Table 1 shows

the results from regressions of three firm performance variables - labor productivity and

indicators for product and for process innovations - on dummy variables for greenfield

FDI and for foreign acquisition along with several controls The results show that

foreign-owned firms particularly those resulting from greenfield FDI exhibit better

productivity and innovation outcomes than their domestic counterparts While the

evidence in Table 1 is not causal due to the cross-sectional nature of the data it is

8 The survey includes a census of firms generating more than 2 megawatts of electricity per hour (category E of the

ISIC Rev 3 classification) and representative samples of the real estate renting and business activities (ISIC Rev 3

category K) financial intermediation services (ISIC Rev 3 category J) and transport storage and communication

services (ISIC Rev 3 category I)

9

suggestive of better performance by foreign-owned service firms in Chile and thus

provides support for the potential for FDI spillovers onto the TFP of manufacturing firms

in the country

The crucial hypothesis tested in this paper is whether the aforementioned FDI-

induced improvements in service sectors benefit TFP of downstream manufacturing

users If present these gains could be classified as pecuniary (rent) spillovers a by-

product of market transactions (Griliches 1992) Manufacturing firms benefit from

pecuniary spillovers if increases in the quality or variety of the services they use due to

FDI are not fully appropriated by service providers In imperfectly competitive service

sectors providers may not appropriate the full surplus from better and more diversified

services because of their inability to perfectly price discriminate whereas in sectors where

FDI increases competition competitive pressures may prevent providers from

appropriating the surplus FDI in services can also benefit manufacturing firms through

spillovers of lsquosoft technologyrsquo linked to eg managerial know-how or technical skills

Learning by manufacturing firms could result from demonstration effects personal

contacts manager or worker turnover9 Griliches (1992) distinguishes knowledge

spillovers from pecuniary spillovers since in principle only the former allow

manufacturing firms to improve their innovation capabilities But in practice pecuniary

spillovers may become knowledge spillovers if downstream users of better services apply

the embodied knowledge to improve their TFP (Branstetter 2001) First for knowledge-

intensive business services such as information technology (IT) the actual service

provided is a knowledge-intensive input upon which firms rely to improve their

9 Kugler (2006) notes a larger potential for knowledge spillovers from vertical relative to horizontal FDI as for the

former foreign-owned service suppliers are not concerned about knowledge leakages since they operate in different

sectors than their domestic manufacturing clients

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 2: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

1

1 Introduction

Foreign direct investment (FDI) inflows into the service sector experienced a boom

during the 1990s By 2002 services accounted for 60 of the world stock of FDI a four-

fold increase since 1990 (UNCTAD 2004) In developed and developing countries alike

the main recipients of FDI have been profit-seeking producer services which range from

network-intensive services such as electricity telecommunications and transport to

finance and business services These sectors are characterized by the facilitating and

intermediating role which they play for downstream user firms (Francois 1990) Thus

producer service sectors are an intricate component of a countryrsquos business environment

In emerging economies where manufacturing firms are constrained by cumbersome

business environments it is particularly relevant to understand how the performance of

service sectors can be improved and how that supports business development and thus

overall economic growth FDI is a potentially powerful means to achieve such

improvements as it might increase the quality and variety of services available as well as

lower their cost Manufacturing firms may benefit from their interaction with foreign

services suppliers through spillovers of management organizational marketing or

technological knowledge (Markusen 1989 Rivera-Batiz and Rivera-Batiz 1992)

Despite the relevance of this topic the effects of vertical linkages resulting from the

openness of producer services to FDI on downstream manufacturing firms have not been

widely documented (Hoekman 2006) This paper attempts to fill this gap by addressing

the following question did the increased penetration of FDI into producer service sectors

in Chile benefit total factor productivity (TFP) of manufacturing firms between 1995 and

2004 Chile is an interesting economy to study as its service sector received large FDI

2

inflows during the 1990s We show that foreign-owned service firms perform better in

terms of labor productivity and innovation than their domestic counterparts The fact that

this finding refers also to greenfield FDI - ie it is not simply foreign investors cherry-

picking the best-performing domestic service firms - suggests these firms introduced

superior standards in the Chilean services sector thus potentially offering improved

services for manufacturing firm users

Evaluating the causal impact of service FDI on manufacturing firm TFP is

challenging due to several valid endogeneity concerns discussed below Our strategy to

identify that impact is to estimate by instrumental variables (IV) a regression of firm TFP

on a service FDI linkage measure controlling for firm fixed effects as well as industry-

year and region-year fixed effects Our study uses TFP measures obtained as residuals

from production functions estimated following the methodologies of Levinsohn and

Petrin (2003) Olley and Pakes (1996) and Ackerberg et al (2006) to correct for the

endogeneity of input choices - including the choice of service inputs - with respect to

productivity Our service FDI linkage measure is defined as service FDI penetration

weighted by firm-level intensity of service usage The rationale underlying this novel

measure is the expectation that firms that use services more intensively benefit more from

any positive effects of service FDI Our measure of service usage intensity is based on

historic values for each firm following Blalock and Gertler (2009) to avoid the potential

endogeneity between firm TFP and service intensity Our choice of instrumental variables

estimation addresses another potential endogeneity concern related to fact that large FDI

inflows into Chilean service sectors may have responded to the strong performance of the

downstream manufacturing users Service FDI penetration weighted by the historic firm-

3

level intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile - Spain and the US - weighted

similarly The identification assumption is that while increases in total FDI outflows into

service sectors by those two countries affect service FDI penetration in Chilean service

sectors they are not correlated with the TFP of manufacturing firms in the country

through any other channel

We find evidence of a positive and significant effect of service FDI on Chilean

manufacturing firmsrsquo TFP This effect is robust to the control for differential productivity

trends across plants with different services intensity and to allowing the effects of service

FDI to operate with lags Alternative measures of service usage weights and service FDI

penetration including the use of industry-level weights from an input-output table

confirm our main findings The positive and significant effect of service FDI is also

obtained when considering a one-stage regression where output depends on inputs as well

as the service FDI linkage measure We also provide suggestive evidence supporting the

hypothesis that service FDI stimulates innovation by Chilean manufacturing firms

Moreover interestingly we find that service FDI offers an opportunity for laggard firms

to catch up in terms of TFP with industry leaders From a policy perspective this finding

is particularly relevant since it suggests that the benefits from service liberalization do not

accrue mostly to leading firms but seem to offer opportunities for firms that are further

behind

Our estimates suggest that a one-standard deviation increase in service FDI would

lead to an increase in TFP of Chilean firms by 3 all else constant The economic

magnitude of this impact is that forward linkages from FDI in services accounted for at

4

least 7 of the observed increase in manufacturing usersrsquo TFP in Chile during the sample

period This economic impact is quite meaningful in light of the finding by Haskel et al

(2007) that horizontal spillovers from manufacturing FDI explain a roughly similar share

of manufacturing TFP growth in the UK during the 1973-1992 period The positive

effects of service FDI on manufacturing firmsrsquo TFP may capture to some extent an

unmeasured decline in quality-adjusted services prices but also the spillover of

managerial and organizational knowledge from service providers to manufacturing users

The microeconomic evidence provided by our study contributes to the emerging

literature on the impact of services liberalization on growth and the performance of

services users At the macro level Mattoo et al (2006) and Eschenbach and Hoekman

(2006) show that countries with liberalized service sectors grow faster once all standard

growth correlates are controlled for Based on computable general equilibrium models

Konan and Maskus (2006) and Jensen et al (2007) argue that business services

liberalization could bring large GDP gains to Tunisia and Russia respectively1 The main

mechanism for these gains is the increase in the number of services available for

manufacturing users as a result of FDI2 At the industry level Francois and Woerz (2008)

show that the increased openness of business services through exports and FDI has strong

positive effects on exports value added and employment of manufacturing industries in

the OECD while Fernandes (2009) estimates positive and significant effects of

liberalization of finance and infrastructure on labor productivity of downstream

manufacturing industries in Eastern European countries At the firm-level Arnold et al

1 Markusen et al (2005) also show important GDP gains from services liberalization based on general equilibrium

simulations for a hypothetical country In their model the presence of foreign-owned service providers allows final

goods producers to rely on more specialized expertise 2 This increase in the number of services increases the TFP of manufacturing firms through a Dixit-Stiglitz-Ethier

framework (Dixit and Stiglitz 1977 Ethier 1982)

5

(2007) show significant positive effects of services liberalization in the Czech Republic

on manufacturing firmsrsquo TFP while Arnold et al (2010) find significant positive effects

of banking telecommunications and transport reforms on Indian manufacturing firmsrsquo

TFP Finally Javorcik and Li (2008) estimate a positive effect of FDI in Romaniarsquos retail

sector on the TFP of manufacturing suppliers to that sector3

Relative to the existing literature the contribution of our study is three-fold First

we exploit the heterogeneity in service usage by considering firm-level measures of the

intensity of service usage Evidence on the usage of services shows considerable

heterogeneity across Chilean firms which suggests that the practice followed in all the

aforementioned studies of using industry-level usage measures based on input-output

tables may be inappropriate The advantage of our measures is that they enable us to

identify the intensive service users within each industry Second we follow a rigorous

empirical approach by relying on firm fixed effects IV estimation to identify the causal

effects of services FDI on TFP Hence our specifications exploit the within-firm variation

in TFP in response to instrumented changes in the service FDI linkage measure Third

we go beyond previous studies by exploring the nature of the effects of service FDI

allowing for heterogeneity across industries relating to their potential for innovation We

also focus on heterogeneous effects across firms relating to their distance to

technologically advanced firms

The remainder of the paper proceeds as follows Section 2 describes recent trends in

FDI in services in Chile and the effects of FDI in services Section 3 describes the data

3 By considering the potential role of knowledge spillovers from service providers to manufacturing users our study

also relates to the literature on vertical spillovers from manufacturing FDI which are shown to be more important than

horizontal spillovers by Javorcik (2004) Kugler (2006) Blalock and Gertler (2008) Damijan et al (2008) and Marcin

(2008) A rationale provided in this literature for vertical forward linkages is that foreign suppliers provide assistance

and complementary services to local buyers

6

Section 4 describes our empirical specification Section 5 discusses our main results and

robustness checks Section 6 discusses extensions to our main results Section 7

concludes

2 FDI in Services Trends and Effects

21 Trends in FDI in Services in Chile

Over the last three decades liberalization privatization and deregulation reforms in

Chile opened its economy to trade and investment more than any other country in Latin

America (Moreira and Blyde 2006)4 In the 1980s most FDI inflows were related to the

extraction and processing of natural resources while in the 1990s inflows into service

sectors took on a leading role with electricity water transport telecommunications and

business services representing about 60 of net FDI inflows during the 1996-2001

period5 Figure 1 shows that these substantial inflows resulted in a growing FDI stock in

the main service sectors in Chile Also the ratio of FDI to output increased substantially

in most Chilean service sectors over the 1990s as shown in Figure 26

The large FDI inflows in Chile during the 1990s reflect first and foremost a

worldwide increase in FDI in services mainly motivated by the interest of multinationals

(MNCs) to become global service providers by gaining access to domestic and regional

markets particularly in the developing world (UNCTAD 2004) In sectors such as

electricity Chilean firms were privatized before 1990 and later acquired by foreign

4 FDI in Chile is governed by Decree Law 600 in place since 1974 which regulates conditions for market entry

capitalization and foreign capital remittances (ECLAC 2000) The decree law grants equal treatment to foreign and

domestic investments in mining manufacturing and most service sectors the exceptions being professional services

such as engineering or legal services (Moreira and Blyde 2006) 5 FDI inflows reached a peak in 1999 in the electricity and water sector due to the purchase of Enersis and Endesa-

Chile by the Spanish electricity firm Endesa-Spain (ECLAC 1999) 6 The computation of the variables shown in Figures 1 and 2 is described in Section 4 and in the Appendix

7

players Global MNCs identified Chilersquos largely privately-owned firms as an attractive

investment opportunity to consolidate their positions in Latin America (ECLAC 2000)

22 Effects of FDI in Services

FDI in the services sector can have four effects within the sector price reductions

quality improvements increased variety and knowledge spillovers7 First FDI is likely

to increase competition in local markets resulting in price reductions as incumbent firms -

eg in electricity and telecommunications sectors - no longer retain the rents they

obtained from being previously monopoly providers The available evidence for banking

electricity and telecommunications confirms price decreases for Chile (Stehmann 1995

Claessens et al 2001 Pollitt 2004) Second FDI may lead to service quality

improvements due to competition and the superior technological organizational and

managerial know-how of foreign-owned service providers FDI can also provide the

necessary finance for major upgrades and the expansion of existing electricity and

telecommunications networks improving the reliability of provision UNCTAD (2004)

and World Bank (2004) report evidence of such developments for Latin America Third

FDI may increase the variety of services provided including new technologically

advanced services or services provided to new regions or new types of clients FDI had

such effects in the Chilean telecommunications sector (ECLAC 2000) Fourth FDI may

result in leaking of managerial marketing and organizational know-how and best

practices (eg linked to the environment or labor codes) from foreign-owned to

domestic-owned services providers (Miroudout 2006)

7 FDI can also entail costs (1) foreign ownership in inherently monopolistic sectors such as electricity or

telecommunications may result in higher prices unless the regulatory system is well defined and managed by the

government and (2) foreign ownership may crowd out domestic firms eg in the banking sector (UNCTAD 2004)

8

The aforementioned positive effects of service FDI are based on the premise that

foreign-owned firms are better performers offering superior services and being more

productive than their domestic counterparts However due to data limitations evidence of

the superiority of foreign-owned service firms is rare The Fourth Chilean Innovation

Survey is a notable exception covering 612 service firms in the electricity generation real

estate financial intermediation business activities and transport storage and

communication sectors8 The survey collected data for 2003-2004 on firm innovation

outcomes accounting variables and basic characteristics Using this data we examine

whether foreign ownership is associated with better performance for Chilean service

firms If foreign investors acquire the best performing domestic service firms then a

positive effect of foreign ownership on firm performance would simply indicate the

endogeneity of the ownership status rather than the intrinsic advantages - eg better

technology - of the foreign parent company However this problem would not arise for

greenfield FDI Hence for each Chilean service firm we obtained information on

whether foreign ownership is due to greenfield FDI or foreign acquisition Table 1 shows

the results from regressions of three firm performance variables - labor productivity and

indicators for product and for process innovations - on dummy variables for greenfield

FDI and for foreign acquisition along with several controls The results show that

foreign-owned firms particularly those resulting from greenfield FDI exhibit better

productivity and innovation outcomes than their domestic counterparts While the

evidence in Table 1 is not causal due to the cross-sectional nature of the data it is

8 The survey includes a census of firms generating more than 2 megawatts of electricity per hour (category E of the

ISIC Rev 3 classification) and representative samples of the real estate renting and business activities (ISIC Rev 3

category K) financial intermediation services (ISIC Rev 3 category J) and transport storage and communication

services (ISIC Rev 3 category I)

9

suggestive of better performance by foreign-owned service firms in Chile and thus

provides support for the potential for FDI spillovers onto the TFP of manufacturing firms

in the country

The crucial hypothesis tested in this paper is whether the aforementioned FDI-

induced improvements in service sectors benefit TFP of downstream manufacturing

users If present these gains could be classified as pecuniary (rent) spillovers a by-

product of market transactions (Griliches 1992) Manufacturing firms benefit from

pecuniary spillovers if increases in the quality or variety of the services they use due to

FDI are not fully appropriated by service providers In imperfectly competitive service

sectors providers may not appropriate the full surplus from better and more diversified

services because of their inability to perfectly price discriminate whereas in sectors where

FDI increases competition competitive pressures may prevent providers from

appropriating the surplus FDI in services can also benefit manufacturing firms through

spillovers of lsquosoft technologyrsquo linked to eg managerial know-how or technical skills

Learning by manufacturing firms could result from demonstration effects personal

contacts manager or worker turnover9 Griliches (1992) distinguishes knowledge

spillovers from pecuniary spillovers since in principle only the former allow

manufacturing firms to improve their innovation capabilities But in practice pecuniary

spillovers may become knowledge spillovers if downstream users of better services apply

the embodied knowledge to improve their TFP (Branstetter 2001) First for knowledge-

intensive business services such as information technology (IT) the actual service

provided is a knowledge-intensive input upon which firms rely to improve their

9 Kugler (2006) notes a larger potential for knowledge spillovers from vertical relative to horizontal FDI as for the

former foreign-owned service suppliers are not concerned about knowledge leakages since they operate in different

sectors than their domestic manufacturing clients

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 3: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

2

inflows during the 1990s We show that foreign-owned service firms perform better in

terms of labor productivity and innovation than their domestic counterparts The fact that

this finding refers also to greenfield FDI - ie it is not simply foreign investors cherry-

picking the best-performing domestic service firms - suggests these firms introduced

superior standards in the Chilean services sector thus potentially offering improved

services for manufacturing firm users

Evaluating the causal impact of service FDI on manufacturing firm TFP is

challenging due to several valid endogeneity concerns discussed below Our strategy to

identify that impact is to estimate by instrumental variables (IV) a regression of firm TFP

on a service FDI linkage measure controlling for firm fixed effects as well as industry-

year and region-year fixed effects Our study uses TFP measures obtained as residuals

from production functions estimated following the methodologies of Levinsohn and

Petrin (2003) Olley and Pakes (1996) and Ackerberg et al (2006) to correct for the

endogeneity of input choices - including the choice of service inputs - with respect to

productivity Our service FDI linkage measure is defined as service FDI penetration

weighted by firm-level intensity of service usage The rationale underlying this novel

measure is the expectation that firms that use services more intensively benefit more from

any positive effects of service FDI Our measure of service usage intensity is based on

historic values for each firm following Blalock and Gertler (2009) to avoid the potential

endogeneity between firm TFP and service intensity Our choice of instrumental variables

estimation addresses another potential endogeneity concern related to fact that large FDI

inflows into Chilean service sectors may have responded to the strong performance of the

downstream manufacturing users Service FDI penetration weighted by the historic firm-

3

level intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile - Spain and the US - weighted

similarly The identification assumption is that while increases in total FDI outflows into

service sectors by those two countries affect service FDI penetration in Chilean service

sectors they are not correlated with the TFP of manufacturing firms in the country

through any other channel

We find evidence of a positive and significant effect of service FDI on Chilean

manufacturing firmsrsquo TFP This effect is robust to the control for differential productivity

trends across plants with different services intensity and to allowing the effects of service

FDI to operate with lags Alternative measures of service usage weights and service FDI

penetration including the use of industry-level weights from an input-output table

confirm our main findings The positive and significant effect of service FDI is also

obtained when considering a one-stage regression where output depends on inputs as well

as the service FDI linkage measure We also provide suggestive evidence supporting the

hypothesis that service FDI stimulates innovation by Chilean manufacturing firms

Moreover interestingly we find that service FDI offers an opportunity for laggard firms

to catch up in terms of TFP with industry leaders From a policy perspective this finding

is particularly relevant since it suggests that the benefits from service liberalization do not

accrue mostly to leading firms but seem to offer opportunities for firms that are further

behind

Our estimates suggest that a one-standard deviation increase in service FDI would

lead to an increase in TFP of Chilean firms by 3 all else constant The economic

magnitude of this impact is that forward linkages from FDI in services accounted for at

4

least 7 of the observed increase in manufacturing usersrsquo TFP in Chile during the sample

period This economic impact is quite meaningful in light of the finding by Haskel et al

(2007) that horizontal spillovers from manufacturing FDI explain a roughly similar share

of manufacturing TFP growth in the UK during the 1973-1992 period The positive

effects of service FDI on manufacturing firmsrsquo TFP may capture to some extent an

unmeasured decline in quality-adjusted services prices but also the spillover of

managerial and organizational knowledge from service providers to manufacturing users

The microeconomic evidence provided by our study contributes to the emerging

literature on the impact of services liberalization on growth and the performance of

services users At the macro level Mattoo et al (2006) and Eschenbach and Hoekman

(2006) show that countries with liberalized service sectors grow faster once all standard

growth correlates are controlled for Based on computable general equilibrium models

Konan and Maskus (2006) and Jensen et al (2007) argue that business services

liberalization could bring large GDP gains to Tunisia and Russia respectively1 The main

mechanism for these gains is the increase in the number of services available for

manufacturing users as a result of FDI2 At the industry level Francois and Woerz (2008)

show that the increased openness of business services through exports and FDI has strong

positive effects on exports value added and employment of manufacturing industries in

the OECD while Fernandes (2009) estimates positive and significant effects of

liberalization of finance and infrastructure on labor productivity of downstream

manufacturing industries in Eastern European countries At the firm-level Arnold et al

1 Markusen et al (2005) also show important GDP gains from services liberalization based on general equilibrium

simulations for a hypothetical country In their model the presence of foreign-owned service providers allows final

goods producers to rely on more specialized expertise 2 This increase in the number of services increases the TFP of manufacturing firms through a Dixit-Stiglitz-Ethier

framework (Dixit and Stiglitz 1977 Ethier 1982)

5

(2007) show significant positive effects of services liberalization in the Czech Republic

on manufacturing firmsrsquo TFP while Arnold et al (2010) find significant positive effects

of banking telecommunications and transport reforms on Indian manufacturing firmsrsquo

TFP Finally Javorcik and Li (2008) estimate a positive effect of FDI in Romaniarsquos retail

sector on the TFP of manufacturing suppliers to that sector3

Relative to the existing literature the contribution of our study is three-fold First

we exploit the heterogeneity in service usage by considering firm-level measures of the

intensity of service usage Evidence on the usage of services shows considerable

heterogeneity across Chilean firms which suggests that the practice followed in all the

aforementioned studies of using industry-level usage measures based on input-output

tables may be inappropriate The advantage of our measures is that they enable us to

identify the intensive service users within each industry Second we follow a rigorous

empirical approach by relying on firm fixed effects IV estimation to identify the causal

effects of services FDI on TFP Hence our specifications exploit the within-firm variation

in TFP in response to instrumented changes in the service FDI linkage measure Third

we go beyond previous studies by exploring the nature of the effects of service FDI

allowing for heterogeneity across industries relating to their potential for innovation We

also focus on heterogeneous effects across firms relating to their distance to

technologically advanced firms

The remainder of the paper proceeds as follows Section 2 describes recent trends in

FDI in services in Chile and the effects of FDI in services Section 3 describes the data

3 By considering the potential role of knowledge spillovers from service providers to manufacturing users our study

also relates to the literature on vertical spillovers from manufacturing FDI which are shown to be more important than

horizontal spillovers by Javorcik (2004) Kugler (2006) Blalock and Gertler (2008) Damijan et al (2008) and Marcin

(2008) A rationale provided in this literature for vertical forward linkages is that foreign suppliers provide assistance

and complementary services to local buyers

6

Section 4 describes our empirical specification Section 5 discusses our main results and

robustness checks Section 6 discusses extensions to our main results Section 7

concludes

2 FDI in Services Trends and Effects

21 Trends in FDI in Services in Chile

Over the last three decades liberalization privatization and deregulation reforms in

Chile opened its economy to trade and investment more than any other country in Latin

America (Moreira and Blyde 2006)4 In the 1980s most FDI inflows were related to the

extraction and processing of natural resources while in the 1990s inflows into service

sectors took on a leading role with electricity water transport telecommunications and

business services representing about 60 of net FDI inflows during the 1996-2001

period5 Figure 1 shows that these substantial inflows resulted in a growing FDI stock in

the main service sectors in Chile Also the ratio of FDI to output increased substantially

in most Chilean service sectors over the 1990s as shown in Figure 26

The large FDI inflows in Chile during the 1990s reflect first and foremost a

worldwide increase in FDI in services mainly motivated by the interest of multinationals

(MNCs) to become global service providers by gaining access to domestic and regional

markets particularly in the developing world (UNCTAD 2004) In sectors such as

electricity Chilean firms were privatized before 1990 and later acquired by foreign

4 FDI in Chile is governed by Decree Law 600 in place since 1974 which regulates conditions for market entry

capitalization and foreign capital remittances (ECLAC 2000) The decree law grants equal treatment to foreign and

domestic investments in mining manufacturing and most service sectors the exceptions being professional services

such as engineering or legal services (Moreira and Blyde 2006) 5 FDI inflows reached a peak in 1999 in the electricity and water sector due to the purchase of Enersis and Endesa-

Chile by the Spanish electricity firm Endesa-Spain (ECLAC 1999) 6 The computation of the variables shown in Figures 1 and 2 is described in Section 4 and in the Appendix

7

players Global MNCs identified Chilersquos largely privately-owned firms as an attractive

investment opportunity to consolidate their positions in Latin America (ECLAC 2000)

22 Effects of FDI in Services

FDI in the services sector can have four effects within the sector price reductions

quality improvements increased variety and knowledge spillovers7 First FDI is likely

to increase competition in local markets resulting in price reductions as incumbent firms -

eg in electricity and telecommunications sectors - no longer retain the rents they

obtained from being previously monopoly providers The available evidence for banking

electricity and telecommunications confirms price decreases for Chile (Stehmann 1995

Claessens et al 2001 Pollitt 2004) Second FDI may lead to service quality

improvements due to competition and the superior technological organizational and

managerial know-how of foreign-owned service providers FDI can also provide the

necessary finance for major upgrades and the expansion of existing electricity and

telecommunications networks improving the reliability of provision UNCTAD (2004)

and World Bank (2004) report evidence of such developments for Latin America Third

FDI may increase the variety of services provided including new technologically

advanced services or services provided to new regions or new types of clients FDI had

such effects in the Chilean telecommunications sector (ECLAC 2000) Fourth FDI may

result in leaking of managerial marketing and organizational know-how and best

practices (eg linked to the environment or labor codes) from foreign-owned to

domestic-owned services providers (Miroudout 2006)

7 FDI can also entail costs (1) foreign ownership in inherently monopolistic sectors such as electricity or

telecommunications may result in higher prices unless the regulatory system is well defined and managed by the

government and (2) foreign ownership may crowd out domestic firms eg in the banking sector (UNCTAD 2004)

8

The aforementioned positive effects of service FDI are based on the premise that

foreign-owned firms are better performers offering superior services and being more

productive than their domestic counterparts However due to data limitations evidence of

the superiority of foreign-owned service firms is rare The Fourth Chilean Innovation

Survey is a notable exception covering 612 service firms in the electricity generation real

estate financial intermediation business activities and transport storage and

communication sectors8 The survey collected data for 2003-2004 on firm innovation

outcomes accounting variables and basic characteristics Using this data we examine

whether foreign ownership is associated with better performance for Chilean service

firms If foreign investors acquire the best performing domestic service firms then a

positive effect of foreign ownership on firm performance would simply indicate the

endogeneity of the ownership status rather than the intrinsic advantages - eg better

technology - of the foreign parent company However this problem would not arise for

greenfield FDI Hence for each Chilean service firm we obtained information on

whether foreign ownership is due to greenfield FDI or foreign acquisition Table 1 shows

the results from regressions of three firm performance variables - labor productivity and

indicators for product and for process innovations - on dummy variables for greenfield

FDI and for foreign acquisition along with several controls The results show that

foreign-owned firms particularly those resulting from greenfield FDI exhibit better

productivity and innovation outcomes than their domestic counterparts While the

evidence in Table 1 is not causal due to the cross-sectional nature of the data it is

8 The survey includes a census of firms generating more than 2 megawatts of electricity per hour (category E of the

ISIC Rev 3 classification) and representative samples of the real estate renting and business activities (ISIC Rev 3

category K) financial intermediation services (ISIC Rev 3 category J) and transport storage and communication

services (ISIC Rev 3 category I)

9

suggestive of better performance by foreign-owned service firms in Chile and thus

provides support for the potential for FDI spillovers onto the TFP of manufacturing firms

in the country

The crucial hypothesis tested in this paper is whether the aforementioned FDI-

induced improvements in service sectors benefit TFP of downstream manufacturing

users If present these gains could be classified as pecuniary (rent) spillovers a by-

product of market transactions (Griliches 1992) Manufacturing firms benefit from

pecuniary spillovers if increases in the quality or variety of the services they use due to

FDI are not fully appropriated by service providers In imperfectly competitive service

sectors providers may not appropriate the full surplus from better and more diversified

services because of their inability to perfectly price discriminate whereas in sectors where

FDI increases competition competitive pressures may prevent providers from

appropriating the surplus FDI in services can also benefit manufacturing firms through

spillovers of lsquosoft technologyrsquo linked to eg managerial know-how or technical skills

Learning by manufacturing firms could result from demonstration effects personal

contacts manager or worker turnover9 Griliches (1992) distinguishes knowledge

spillovers from pecuniary spillovers since in principle only the former allow

manufacturing firms to improve their innovation capabilities But in practice pecuniary

spillovers may become knowledge spillovers if downstream users of better services apply

the embodied knowledge to improve their TFP (Branstetter 2001) First for knowledge-

intensive business services such as information technology (IT) the actual service

provided is a knowledge-intensive input upon which firms rely to improve their

9 Kugler (2006) notes a larger potential for knowledge spillovers from vertical relative to horizontal FDI as for the

former foreign-owned service suppliers are not concerned about knowledge leakages since they operate in different

sectors than their domestic manufacturing clients

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 4: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

3

level intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile - Spain and the US - weighted

similarly The identification assumption is that while increases in total FDI outflows into

service sectors by those two countries affect service FDI penetration in Chilean service

sectors they are not correlated with the TFP of manufacturing firms in the country

through any other channel

We find evidence of a positive and significant effect of service FDI on Chilean

manufacturing firmsrsquo TFP This effect is robust to the control for differential productivity

trends across plants with different services intensity and to allowing the effects of service

FDI to operate with lags Alternative measures of service usage weights and service FDI

penetration including the use of industry-level weights from an input-output table

confirm our main findings The positive and significant effect of service FDI is also

obtained when considering a one-stage regression where output depends on inputs as well

as the service FDI linkage measure We also provide suggestive evidence supporting the

hypothesis that service FDI stimulates innovation by Chilean manufacturing firms

Moreover interestingly we find that service FDI offers an opportunity for laggard firms

to catch up in terms of TFP with industry leaders From a policy perspective this finding

is particularly relevant since it suggests that the benefits from service liberalization do not

accrue mostly to leading firms but seem to offer opportunities for firms that are further

behind

Our estimates suggest that a one-standard deviation increase in service FDI would

lead to an increase in TFP of Chilean firms by 3 all else constant The economic

magnitude of this impact is that forward linkages from FDI in services accounted for at

4

least 7 of the observed increase in manufacturing usersrsquo TFP in Chile during the sample

period This economic impact is quite meaningful in light of the finding by Haskel et al

(2007) that horizontal spillovers from manufacturing FDI explain a roughly similar share

of manufacturing TFP growth in the UK during the 1973-1992 period The positive

effects of service FDI on manufacturing firmsrsquo TFP may capture to some extent an

unmeasured decline in quality-adjusted services prices but also the spillover of

managerial and organizational knowledge from service providers to manufacturing users

The microeconomic evidence provided by our study contributes to the emerging

literature on the impact of services liberalization on growth and the performance of

services users At the macro level Mattoo et al (2006) and Eschenbach and Hoekman

(2006) show that countries with liberalized service sectors grow faster once all standard

growth correlates are controlled for Based on computable general equilibrium models

Konan and Maskus (2006) and Jensen et al (2007) argue that business services

liberalization could bring large GDP gains to Tunisia and Russia respectively1 The main

mechanism for these gains is the increase in the number of services available for

manufacturing users as a result of FDI2 At the industry level Francois and Woerz (2008)

show that the increased openness of business services through exports and FDI has strong

positive effects on exports value added and employment of manufacturing industries in

the OECD while Fernandes (2009) estimates positive and significant effects of

liberalization of finance and infrastructure on labor productivity of downstream

manufacturing industries in Eastern European countries At the firm-level Arnold et al

1 Markusen et al (2005) also show important GDP gains from services liberalization based on general equilibrium

simulations for a hypothetical country In their model the presence of foreign-owned service providers allows final

goods producers to rely on more specialized expertise 2 This increase in the number of services increases the TFP of manufacturing firms through a Dixit-Stiglitz-Ethier

framework (Dixit and Stiglitz 1977 Ethier 1982)

5

(2007) show significant positive effects of services liberalization in the Czech Republic

on manufacturing firmsrsquo TFP while Arnold et al (2010) find significant positive effects

of banking telecommunications and transport reforms on Indian manufacturing firmsrsquo

TFP Finally Javorcik and Li (2008) estimate a positive effect of FDI in Romaniarsquos retail

sector on the TFP of manufacturing suppliers to that sector3

Relative to the existing literature the contribution of our study is three-fold First

we exploit the heterogeneity in service usage by considering firm-level measures of the

intensity of service usage Evidence on the usage of services shows considerable

heterogeneity across Chilean firms which suggests that the practice followed in all the

aforementioned studies of using industry-level usage measures based on input-output

tables may be inappropriate The advantage of our measures is that they enable us to

identify the intensive service users within each industry Second we follow a rigorous

empirical approach by relying on firm fixed effects IV estimation to identify the causal

effects of services FDI on TFP Hence our specifications exploit the within-firm variation

in TFP in response to instrumented changes in the service FDI linkage measure Third

we go beyond previous studies by exploring the nature of the effects of service FDI

allowing for heterogeneity across industries relating to their potential for innovation We

also focus on heterogeneous effects across firms relating to their distance to

technologically advanced firms

The remainder of the paper proceeds as follows Section 2 describes recent trends in

FDI in services in Chile and the effects of FDI in services Section 3 describes the data

3 By considering the potential role of knowledge spillovers from service providers to manufacturing users our study

also relates to the literature on vertical spillovers from manufacturing FDI which are shown to be more important than

horizontal spillovers by Javorcik (2004) Kugler (2006) Blalock and Gertler (2008) Damijan et al (2008) and Marcin

(2008) A rationale provided in this literature for vertical forward linkages is that foreign suppliers provide assistance

and complementary services to local buyers

6

Section 4 describes our empirical specification Section 5 discusses our main results and

robustness checks Section 6 discusses extensions to our main results Section 7

concludes

2 FDI in Services Trends and Effects

21 Trends in FDI in Services in Chile

Over the last three decades liberalization privatization and deregulation reforms in

Chile opened its economy to trade and investment more than any other country in Latin

America (Moreira and Blyde 2006)4 In the 1980s most FDI inflows were related to the

extraction and processing of natural resources while in the 1990s inflows into service

sectors took on a leading role with electricity water transport telecommunications and

business services representing about 60 of net FDI inflows during the 1996-2001

period5 Figure 1 shows that these substantial inflows resulted in a growing FDI stock in

the main service sectors in Chile Also the ratio of FDI to output increased substantially

in most Chilean service sectors over the 1990s as shown in Figure 26

The large FDI inflows in Chile during the 1990s reflect first and foremost a

worldwide increase in FDI in services mainly motivated by the interest of multinationals

(MNCs) to become global service providers by gaining access to domestic and regional

markets particularly in the developing world (UNCTAD 2004) In sectors such as

electricity Chilean firms were privatized before 1990 and later acquired by foreign

4 FDI in Chile is governed by Decree Law 600 in place since 1974 which regulates conditions for market entry

capitalization and foreign capital remittances (ECLAC 2000) The decree law grants equal treatment to foreign and

domestic investments in mining manufacturing and most service sectors the exceptions being professional services

such as engineering or legal services (Moreira and Blyde 2006) 5 FDI inflows reached a peak in 1999 in the electricity and water sector due to the purchase of Enersis and Endesa-

Chile by the Spanish electricity firm Endesa-Spain (ECLAC 1999) 6 The computation of the variables shown in Figures 1 and 2 is described in Section 4 and in the Appendix

7

players Global MNCs identified Chilersquos largely privately-owned firms as an attractive

investment opportunity to consolidate their positions in Latin America (ECLAC 2000)

22 Effects of FDI in Services

FDI in the services sector can have four effects within the sector price reductions

quality improvements increased variety and knowledge spillovers7 First FDI is likely

to increase competition in local markets resulting in price reductions as incumbent firms -

eg in electricity and telecommunications sectors - no longer retain the rents they

obtained from being previously monopoly providers The available evidence for banking

electricity and telecommunications confirms price decreases for Chile (Stehmann 1995

Claessens et al 2001 Pollitt 2004) Second FDI may lead to service quality

improvements due to competition and the superior technological organizational and

managerial know-how of foreign-owned service providers FDI can also provide the

necessary finance for major upgrades and the expansion of existing electricity and

telecommunications networks improving the reliability of provision UNCTAD (2004)

and World Bank (2004) report evidence of such developments for Latin America Third

FDI may increase the variety of services provided including new technologically

advanced services or services provided to new regions or new types of clients FDI had

such effects in the Chilean telecommunications sector (ECLAC 2000) Fourth FDI may

result in leaking of managerial marketing and organizational know-how and best

practices (eg linked to the environment or labor codes) from foreign-owned to

domestic-owned services providers (Miroudout 2006)

7 FDI can also entail costs (1) foreign ownership in inherently monopolistic sectors such as electricity or

telecommunications may result in higher prices unless the regulatory system is well defined and managed by the

government and (2) foreign ownership may crowd out domestic firms eg in the banking sector (UNCTAD 2004)

8

The aforementioned positive effects of service FDI are based on the premise that

foreign-owned firms are better performers offering superior services and being more

productive than their domestic counterparts However due to data limitations evidence of

the superiority of foreign-owned service firms is rare The Fourth Chilean Innovation

Survey is a notable exception covering 612 service firms in the electricity generation real

estate financial intermediation business activities and transport storage and

communication sectors8 The survey collected data for 2003-2004 on firm innovation

outcomes accounting variables and basic characteristics Using this data we examine

whether foreign ownership is associated with better performance for Chilean service

firms If foreign investors acquire the best performing domestic service firms then a

positive effect of foreign ownership on firm performance would simply indicate the

endogeneity of the ownership status rather than the intrinsic advantages - eg better

technology - of the foreign parent company However this problem would not arise for

greenfield FDI Hence for each Chilean service firm we obtained information on

whether foreign ownership is due to greenfield FDI or foreign acquisition Table 1 shows

the results from regressions of three firm performance variables - labor productivity and

indicators for product and for process innovations - on dummy variables for greenfield

FDI and for foreign acquisition along with several controls The results show that

foreign-owned firms particularly those resulting from greenfield FDI exhibit better

productivity and innovation outcomes than their domestic counterparts While the

evidence in Table 1 is not causal due to the cross-sectional nature of the data it is

8 The survey includes a census of firms generating more than 2 megawatts of electricity per hour (category E of the

ISIC Rev 3 classification) and representative samples of the real estate renting and business activities (ISIC Rev 3

category K) financial intermediation services (ISIC Rev 3 category J) and transport storage and communication

services (ISIC Rev 3 category I)

9

suggestive of better performance by foreign-owned service firms in Chile and thus

provides support for the potential for FDI spillovers onto the TFP of manufacturing firms

in the country

The crucial hypothesis tested in this paper is whether the aforementioned FDI-

induced improvements in service sectors benefit TFP of downstream manufacturing

users If present these gains could be classified as pecuniary (rent) spillovers a by-

product of market transactions (Griliches 1992) Manufacturing firms benefit from

pecuniary spillovers if increases in the quality or variety of the services they use due to

FDI are not fully appropriated by service providers In imperfectly competitive service

sectors providers may not appropriate the full surplus from better and more diversified

services because of their inability to perfectly price discriminate whereas in sectors where

FDI increases competition competitive pressures may prevent providers from

appropriating the surplus FDI in services can also benefit manufacturing firms through

spillovers of lsquosoft technologyrsquo linked to eg managerial know-how or technical skills

Learning by manufacturing firms could result from demonstration effects personal

contacts manager or worker turnover9 Griliches (1992) distinguishes knowledge

spillovers from pecuniary spillovers since in principle only the former allow

manufacturing firms to improve their innovation capabilities But in practice pecuniary

spillovers may become knowledge spillovers if downstream users of better services apply

the embodied knowledge to improve their TFP (Branstetter 2001) First for knowledge-

intensive business services such as information technology (IT) the actual service

provided is a knowledge-intensive input upon which firms rely to improve their

9 Kugler (2006) notes a larger potential for knowledge spillovers from vertical relative to horizontal FDI as for the

former foreign-owned service suppliers are not concerned about knowledge leakages since they operate in different

sectors than their domestic manufacturing clients

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 5: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

4

least 7 of the observed increase in manufacturing usersrsquo TFP in Chile during the sample

period This economic impact is quite meaningful in light of the finding by Haskel et al

(2007) that horizontal spillovers from manufacturing FDI explain a roughly similar share

of manufacturing TFP growth in the UK during the 1973-1992 period The positive

effects of service FDI on manufacturing firmsrsquo TFP may capture to some extent an

unmeasured decline in quality-adjusted services prices but also the spillover of

managerial and organizational knowledge from service providers to manufacturing users

The microeconomic evidence provided by our study contributes to the emerging

literature on the impact of services liberalization on growth and the performance of

services users At the macro level Mattoo et al (2006) and Eschenbach and Hoekman

(2006) show that countries with liberalized service sectors grow faster once all standard

growth correlates are controlled for Based on computable general equilibrium models

Konan and Maskus (2006) and Jensen et al (2007) argue that business services

liberalization could bring large GDP gains to Tunisia and Russia respectively1 The main

mechanism for these gains is the increase in the number of services available for

manufacturing users as a result of FDI2 At the industry level Francois and Woerz (2008)

show that the increased openness of business services through exports and FDI has strong

positive effects on exports value added and employment of manufacturing industries in

the OECD while Fernandes (2009) estimates positive and significant effects of

liberalization of finance and infrastructure on labor productivity of downstream

manufacturing industries in Eastern European countries At the firm-level Arnold et al

1 Markusen et al (2005) also show important GDP gains from services liberalization based on general equilibrium

simulations for a hypothetical country In their model the presence of foreign-owned service providers allows final

goods producers to rely on more specialized expertise 2 This increase in the number of services increases the TFP of manufacturing firms through a Dixit-Stiglitz-Ethier

framework (Dixit and Stiglitz 1977 Ethier 1982)

5

(2007) show significant positive effects of services liberalization in the Czech Republic

on manufacturing firmsrsquo TFP while Arnold et al (2010) find significant positive effects

of banking telecommunications and transport reforms on Indian manufacturing firmsrsquo

TFP Finally Javorcik and Li (2008) estimate a positive effect of FDI in Romaniarsquos retail

sector on the TFP of manufacturing suppliers to that sector3

Relative to the existing literature the contribution of our study is three-fold First

we exploit the heterogeneity in service usage by considering firm-level measures of the

intensity of service usage Evidence on the usage of services shows considerable

heterogeneity across Chilean firms which suggests that the practice followed in all the

aforementioned studies of using industry-level usage measures based on input-output

tables may be inappropriate The advantage of our measures is that they enable us to

identify the intensive service users within each industry Second we follow a rigorous

empirical approach by relying on firm fixed effects IV estimation to identify the causal

effects of services FDI on TFP Hence our specifications exploit the within-firm variation

in TFP in response to instrumented changes in the service FDI linkage measure Third

we go beyond previous studies by exploring the nature of the effects of service FDI

allowing for heterogeneity across industries relating to their potential for innovation We

also focus on heterogeneous effects across firms relating to their distance to

technologically advanced firms

The remainder of the paper proceeds as follows Section 2 describes recent trends in

FDI in services in Chile and the effects of FDI in services Section 3 describes the data

3 By considering the potential role of knowledge spillovers from service providers to manufacturing users our study

also relates to the literature on vertical spillovers from manufacturing FDI which are shown to be more important than

horizontal spillovers by Javorcik (2004) Kugler (2006) Blalock and Gertler (2008) Damijan et al (2008) and Marcin

(2008) A rationale provided in this literature for vertical forward linkages is that foreign suppliers provide assistance

and complementary services to local buyers

6

Section 4 describes our empirical specification Section 5 discusses our main results and

robustness checks Section 6 discusses extensions to our main results Section 7

concludes

2 FDI in Services Trends and Effects

21 Trends in FDI in Services in Chile

Over the last three decades liberalization privatization and deregulation reforms in

Chile opened its economy to trade and investment more than any other country in Latin

America (Moreira and Blyde 2006)4 In the 1980s most FDI inflows were related to the

extraction and processing of natural resources while in the 1990s inflows into service

sectors took on a leading role with electricity water transport telecommunications and

business services representing about 60 of net FDI inflows during the 1996-2001

period5 Figure 1 shows that these substantial inflows resulted in a growing FDI stock in

the main service sectors in Chile Also the ratio of FDI to output increased substantially

in most Chilean service sectors over the 1990s as shown in Figure 26

The large FDI inflows in Chile during the 1990s reflect first and foremost a

worldwide increase in FDI in services mainly motivated by the interest of multinationals

(MNCs) to become global service providers by gaining access to domestic and regional

markets particularly in the developing world (UNCTAD 2004) In sectors such as

electricity Chilean firms were privatized before 1990 and later acquired by foreign

4 FDI in Chile is governed by Decree Law 600 in place since 1974 which regulates conditions for market entry

capitalization and foreign capital remittances (ECLAC 2000) The decree law grants equal treatment to foreign and

domestic investments in mining manufacturing and most service sectors the exceptions being professional services

such as engineering or legal services (Moreira and Blyde 2006) 5 FDI inflows reached a peak in 1999 in the electricity and water sector due to the purchase of Enersis and Endesa-

Chile by the Spanish electricity firm Endesa-Spain (ECLAC 1999) 6 The computation of the variables shown in Figures 1 and 2 is described in Section 4 and in the Appendix

7

players Global MNCs identified Chilersquos largely privately-owned firms as an attractive

investment opportunity to consolidate their positions in Latin America (ECLAC 2000)

22 Effects of FDI in Services

FDI in the services sector can have four effects within the sector price reductions

quality improvements increased variety and knowledge spillovers7 First FDI is likely

to increase competition in local markets resulting in price reductions as incumbent firms -

eg in electricity and telecommunications sectors - no longer retain the rents they

obtained from being previously monopoly providers The available evidence for banking

electricity and telecommunications confirms price decreases for Chile (Stehmann 1995

Claessens et al 2001 Pollitt 2004) Second FDI may lead to service quality

improvements due to competition and the superior technological organizational and

managerial know-how of foreign-owned service providers FDI can also provide the

necessary finance for major upgrades and the expansion of existing electricity and

telecommunications networks improving the reliability of provision UNCTAD (2004)

and World Bank (2004) report evidence of such developments for Latin America Third

FDI may increase the variety of services provided including new technologically

advanced services or services provided to new regions or new types of clients FDI had

such effects in the Chilean telecommunications sector (ECLAC 2000) Fourth FDI may

result in leaking of managerial marketing and organizational know-how and best

practices (eg linked to the environment or labor codes) from foreign-owned to

domestic-owned services providers (Miroudout 2006)

7 FDI can also entail costs (1) foreign ownership in inherently monopolistic sectors such as electricity or

telecommunications may result in higher prices unless the regulatory system is well defined and managed by the

government and (2) foreign ownership may crowd out domestic firms eg in the banking sector (UNCTAD 2004)

8

The aforementioned positive effects of service FDI are based on the premise that

foreign-owned firms are better performers offering superior services and being more

productive than their domestic counterparts However due to data limitations evidence of

the superiority of foreign-owned service firms is rare The Fourth Chilean Innovation

Survey is a notable exception covering 612 service firms in the electricity generation real

estate financial intermediation business activities and transport storage and

communication sectors8 The survey collected data for 2003-2004 on firm innovation

outcomes accounting variables and basic characteristics Using this data we examine

whether foreign ownership is associated with better performance for Chilean service

firms If foreign investors acquire the best performing domestic service firms then a

positive effect of foreign ownership on firm performance would simply indicate the

endogeneity of the ownership status rather than the intrinsic advantages - eg better

technology - of the foreign parent company However this problem would not arise for

greenfield FDI Hence for each Chilean service firm we obtained information on

whether foreign ownership is due to greenfield FDI or foreign acquisition Table 1 shows

the results from regressions of three firm performance variables - labor productivity and

indicators for product and for process innovations - on dummy variables for greenfield

FDI and for foreign acquisition along with several controls The results show that

foreign-owned firms particularly those resulting from greenfield FDI exhibit better

productivity and innovation outcomes than their domestic counterparts While the

evidence in Table 1 is not causal due to the cross-sectional nature of the data it is

8 The survey includes a census of firms generating more than 2 megawatts of electricity per hour (category E of the

ISIC Rev 3 classification) and representative samples of the real estate renting and business activities (ISIC Rev 3

category K) financial intermediation services (ISIC Rev 3 category J) and transport storage and communication

services (ISIC Rev 3 category I)

9

suggestive of better performance by foreign-owned service firms in Chile and thus

provides support for the potential for FDI spillovers onto the TFP of manufacturing firms

in the country

The crucial hypothesis tested in this paper is whether the aforementioned FDI-

induced improvements in service sectors benefit TFP of downstream manufacturing

users If present these gains could be classified as pecuniary (rent) spillovers a by-

product of market transactions (Griliches 1992) Manufacturing firms benefit from

pecuniary spillovers if increases in the quality or variety of the services they use due to

FDI are not fully appropriated by service providers In imperfectly competitive service

sectors providers may not appropriate the full surplus from better and more diversified

services because of their inability to perfectly price discriminate whereas in sectors where

FDI increases competition competitive pressures may prevent providers from

appropriating the surplus FDI in services can also benefit manufacturing firms through

spillovers of lsquosoft technologyrsquo linked to eg managerial know-how or technical skills

Learning by manufacturing firms could result from demonstration effects personal

contacts manager or worker turnover9 Griliches (1992) distinguishes knowledge

spillovers from pecuniary spillovers since in principle only the former allow

manufacturing firms to improve their innovation capabilities But in practice pecuniary

spillovers may become knowledge spillovers if downstream users of better services apply

the embodied knowledge to improve their TFP (Branstetter 2001) First for knowledge-

intensive business services such as information technology (IT) the actual service

provided is a knowledge-intensive input upon which firms rely to improve their

9 Kugler (2006) notes a larger potential for knowledge spillovers from vertical relative to horizontal FDI as for the

former foreign-owned service suppliers are not concerned about knowledge leakages since they operate in different

sectors than their domestic manufacturing clients

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 6: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

5

(2007) show significant positive effects of services liberalization in the Czech Republic

on manufacturing firmsrsquo TFP while Arnold et al (2010) find significant positive effects

of banking telecommunications and transport reforms on Indian manufacturing firmsrsquo

TFP Finally Javorcik and Li (2008) estimate a positive effect of FDI in Romaniarsquos retail

sector on the TFP of manufacturing suppliers to that sector3

Relative to the existing literature the contribution of our study is three-fold First

we exploit the heterogeneity in service usage by considering firm-level measures of the

intensity of service usage Evidence on the usage of services shows considerable

heterogeneity across Chilean firms which suggests that the practice followed in all the

aforementioned studies of using industry-level usage measures based on input-output

tables may be inappropriate The advantage of our measures is that they enable us to

identify the intensive service users within each industry Second we follow a rigorous

empirical approach by relying on firm fixed effects IV estimation to identify the causal

effects of services FDI on TFP Hence our specifications exploit the within-firm variation

in TFP in response to instrumented changes in the service FDI linkage measure Third

we go beyond previous studies by exploring the nature of the effects of service FDI

allowing for heterogeneity across industries relating to their potential for innovation We

also focus on heterogeneous effects across firms relating to their distance to

technologically advanced firms

The remainder of the paper proceeds as follows Section 2 describes recent trends in

FDI in services in Chile and the effects of FDI in services Section 3 describes the data

3 By considering the potential role of knowledge spillovers from service providers to manufacturing users our study

also relates to the literature on vertical spillovers from manufacturing FDI which are shown to be more important than

horizontal spillovers by Javorcik (2004) Kugler (2006) Blalock and Gertler (2008) Damijan et al (2008) and Marcin

(2008) A rationale provided in this literature for vertical forward linkages is that foreign suppliers provide assistance

and complementary services to local buyers

6

Section 4 describes our empirical specification Section 5 discusses our main results and

robustness checks Section 6 discusses extensions to our main results Section 7

concludes

2 FDI in Services Trends and Effects

21 Trends in FDI in Services in Chile

Over the last three decades liberalization privatization and deregulation reforms in

Chile opened its economy to trade and investment more than any other country in Latin

America (Moreira and Blyde 2006)4 In the 1980s most FDI inflows were related to the

extraction and processing of natural resources while in the 1990s inflows into service

sectors took on a leading role with electricity water transport telecommunications and

business services representing about 60 of net FDI inflows during the 1996-2001

period5 Figure 1 shows that these substantial inflows resulted in a growing FDI stock in

the main service sectors in Chile Also the ratio of FDI to output increased substantially

in most Chilean service sectors over the 1990s as shown in Figure 26

The large FDI inflows in Chile during the 1990s reflect first and foremost a

worldwide increase in FDI in services mainly motivated by the interest of multinationals

(MNCs) to become global service providers by gaining access to domestic and regional

markets particularly in the developing world (UNCTAD 2004) In sectors such as

electricity Chilean firms were privatized before 1990 and later acquired by foreign

4 FDI in Chile is governed by Decree Law 600 in place since 1974 which regulates conditions for market entry

capitalization and foreign capital remittances (ECLAC 2000) The decree law grants equal treatment to foreign and

domestic investments in mining manufacturing and most service sectors the exceptions being professional services

such as engineering or legal services (Moreira and Blyde 2006) 5 FDI inflows reached a peak in 1999 in the electricity and water sector due to the purchase of Enersis and Endesa-

Chile by the Spanish electricity firm Endesa-Spain (ECLAC 1999) 6 The computation of the variables shown in Figures 1 and 2 is described in Section 4 and in the Appendix

7

players Global MNCs identified Chilersquos largely privately-owned firms as an attractive

investment opportunity to consolidate their positions in Latin America (ECLAC 2000)

22 Effects of FDI in Services

FDI in the services sector can have four effects within the sector price reductions

quality improvements increased variety and knowledge spillovers7 First FDI is likely

to increase competition in local markets resulting in price reductions as incumbent firms -

eg in electricity and telecommunications sectors - no longer retain the rents they

obtained from being previously monopoly providers The available evidence for banking

electricity and telecommunications confirms price decreases for Chile (Stehmann 1995

Claessens et al 2001 Pollitt 2004) Second FDI may lead to service quality

improvements due to competition and the superior technological organizational and

managerial know-how of foreign-owned service providers FDI can also provide the

necessary finance for major upgrades and the expansion of existing electricity and

telecommunications networks improving the reliability of provision UNCTAD (2004)

and World Bank (2004) report evidence of such developments for Latin America Third

FDI may increase the variety of services provided including new technologically

advanced services or services provided to new regions or new types of clients FDI had

such effects in the Chilean telecommunications sector (ECLAC 2000) Fourth FDI may

result in leaking of managerial marketing and organizational know-how and best

practices (eg linked to the environment or labor codes) from foreign-owned to

domestic-owned services providers (Miroudout 2006)

7 FDI can also entail costs (1) foreign ownership in inherently monopolistic sectors such as electricity or

telecommunications may result in higher prices unless the regulatory system is well defined and managed by the

government and (2) foreign ownership may crowd out domestic firms eg in the banking sector (UNCTAD 2004)

8

The aforementioned positive effects of service FDI are based on the premise that

foreign-owned firms are better performers offering superior services and being more

productive than their domestic counterparts However due to data limitations evidence of

the superiority of foreign-owned service firms is rare The Fourth Chilean Innovation

Survey is a notable exception covering 612 service firms in the electricity generation real

estate financial intermediation business activities and transport storage and

communication sectors8 The survey collected data for 2003-2004 on firm innovation

outcomes accounting variables and basic characteristics Using this data we examine

whether foreign ownership is associated with better performance for Chilean service

firms If foreign investors acquire the best performing domestic service firms then a

positive effect of foreign ownership on firm performance would simply indicate the

endogeneity of the ownership status rather than the intrinsic advantages - eg better

technology - of the foreign parent company However this problem would not arise for

greenfield FDI Hence for each Chilean service firm we obtained information on

whether foreign ownership is due to greenfield FDI or foreign acquisition Table 1 shows

the results from regressions of three firm performance variables - labor productivity and

indicators for product and for process innovations - on dummy variables for greenfield

FDI and for foreign acquisition along with several controls The results show that

foreign-owned firms particularly those resulting from greenfield FDI exhibit better

productivity and innovation outcomes than their domestic counterparts While the

evidence in Table 1 is not causal due to the cross-sectional nature of the data it is

8 The survey includes a census of firms generating more than 2 megawatts of electricity per hour (category E of the

ISIC Rev 3 classification) and representative samples of the real estate renting and business activities (ISIC Rev 3

category K) financial intermediation services (ISIC Rev 3 category J) and transport storage and communication

services (ISIC Rev 3 category I)

9

suggestive of better performance by foreign-owned service firms in Chile and thus

provides support for the potential for FDI spillovers onto the TFP of manufacturing firms

in the country

The crucial hypothesis tested in this paper is whether the aforementioned FDI-

induced improvements in service sectors benefit TFP of downstream manufacturing

users If present these gains could be classified as pecuniary (rent) spillovers a by-

product of market transactions (Griliches 1992) Manufacturing firms benefit from

pecuniary spillovers if increases in the quality or variety of the services they use due to

FDI are not fully appropriated by service providers In imperfectly competitive service

sectors providers may not appropriate the full surplus from better and more diversified

services because of their inability to perfectly price discriminate whereas in sectors where

FDI increases competition competitive pressures may prevent providers from

appropriating the surplus FDI in services can also benefit manufacturing firms through

spillovers of lsquosoft technologyrsquo linked to eg managerial know-how or technical skills

Learning by manufacturing firms could result from demonstration effects personal

contacts manager or worker turnover9 Griliches (1992) distinguishes knowledge

spillovers from pecuniary spillovers since in principle only the former allow

manufacturing firms to improve their innovation capabilities But in practice pecuniary

spillovers may become knowledge spillovers if downstream users of better services apply

the embodied knowledge to improve their TFP (Branstetter 2001) First for knowledge-

intensive business services such as information technology (IT) the actual service

provided is a knowledge-intensive input upon which firms rely to improve their

9 Kugler (2006) notes a larger potential for knowledge spillovers from vertical relative to horizontal FDI as for the

former foreign-owned service suppliers are not concerned about knowledge leakages since they operate in different

sectors than their domestic manufacturing clients

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 7: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

6

Section 4 describes our empirical specification Section 5 discusses our main results and

robustness checks Section 6 discusses extensions to our main results Section 7

concludes

2 FDI in Services Trends and Effects

21 Trends in FDI in Services in Chile

Over the last three decades liberalization privatization and deregulation reforms in

Chile opened its economy to trade and investment more than any other country in Latin

America (Moreira and Blyde 2006)4 In the 1980s most FDI inflows were related to the

extraction and processing of natural resources while in the 1990s inflows into service

sectors took on a leading role with electricity water transport telecommunications and

business services representing about 60 of net FDI inflows during the 1996-2001

period5 Figure 1 shows that these substantial inflows resulted in a growing FDI stock in

the main service sectors in Chile Also the ratio of FDI to output increased substantially

in most Chilean service sectors over the 1990s as shown in Figure 26

The large FDI inflows in Chile during the 1990s reflect first and foremost a

worldwide increase in FDI in services mainly motivated by the interest of multinationals

(MNCs) to become global service providers by gaining access to domestic and regional

markets particularly in the developing world (UNCTAD 2004) In sectors such as

electricity Chilean firms were privatized before 1990 and later acquired by foreign

4 FDI in Chile is governed by Decree Law 600 in place since 1974 which regulates conditions for market entry

capitalization and foreign capital remittances (ECLAC 2000) The decree law grants equal treatment to foreign and

domestic investments in mining manufacturing and most service sectors the exceptions being professional services

such as engineering or legal services (Moreira and Blyde 2006) 5 FDI inflows reached a peak in 1999 in the electricity and water sector due to the purchase of Enersis and Endesa-

Chile by the Spanish electricity firm Endesa-Spain (ECLAC 1999) 6 The computation of the variables shown in Figures 1 and 2 is described in Section 4 and in the Appendix

7

players Global MNCs identified Chilersquos largely privately-owned firms as an attractive

investment opportunity to consolidate their positions in Latin America (ECLAC 2000)

22 Effects of FDI in Services

FDI in the services sector can have four effects within the sector price reductions

quality improvements increased variety and knowledge spillovers7 First FDI is likely

to increase competition in local markets resulting in price reductions as incumbent firms -

eg in electricity and telecommunications sectors - no longer retain the rents they

obtained from being previously monopoly providers The available evidence for banking

electricity and telecommunications confirms price decreases for Chile (Stehmann 1995

Claessens et al 2001 Pollitt 2004) Second FDI may lead to service quality

improvements due to competition and the superior technological organizational and

managerial know-how of foreign-owned service providers FDI can also provide the

necessary finance for major upgrades and the expansion of existing electricity and

telecommunications networks improving the reliability of provision UNCTAD (2004)

and World Bank (2004) report evidence of such developments for Latin America Third

FDI may increase the variety of services provided including new technologically

advanced services or services provided to new regions or new types of clients FDI had

such effects in the Chilean telecommunications sector (ECLAC 2000) Fourth FDI may

result in leaking of managerial marketing and organizational know-how and best

practices (eg linked to the environment or labor codes) from foreign-owned to

domestic-owned services providers (Miroudout 2006)

7 FDI can also entail costs (1) foreign ownership in inherently monopolistic sectors such as electricity or

telecommunications may result in higher prices unless the regulatory system is well defined and managed by the

government and (2) foreign ownership may crowd out domestic firms eg in the banking sector (UNCTAD 2004)

8

The aforementioned positive effects of service FDI are based on the premise that

foreign-owned firms are better performers offering superior services and being more

productive than their domestic counterparts However due to data limitations evidence of

the superiority of foreign-owned service firms is rare The Fourth Chilean Innovation

Survey is a notable exception covering 612 service firms in the electricity generation real

estate financial intermediation business activities and transport storage and

communication sectors8 The survey collected data for 2003-2004 on firm innovation

outcomes accounting variables and basic characteristics Using this data we examine

whether foreign ownership is associated with better performance for Chilean service

firms If foreign investors acquire the best performing domestic service firms then a

positive effect of foreign ownership on firm performance would simply indicate the

endogeneity of the ownership status rather than the intrinsic advantages - eg better

technology - of the foreign parent company However this problem would not arise for

greenfield FDI Hence for each Chilean service firm we obtained information on

whether foreign ownership is due to greenfield FDI or foreign acquisition Table 1 shows

the results from regressions of three firm performance variables - labor productivity and

indicators for product and for process innovations - on dummy variables for greenfield

FDI and for foreign acquisition along with several controls The results show that

foreign-owned firms particularly those resulting from greenfield FDI exhibit better

productivity and innovation outcomes than their domestic counterparts While the

evidence in Table 1 is not causal due to the cross-sectional nature of the data it is

8 The survey includes a census of firms generating more than 2 megawatts of electricity per hour (category E of the

ISIC Rev 3 classification) and representative samples of the real estate renting and business activities (ISIC Rev 3

category K) financial intermediation services (ISIC Rev 3 category J) and transport storage and communication

services (ISIC Rev 3 category I)

9

suggestive of better performance by foreign-owned service firms in Chile and thus

provides support for the potential for FDI spillovers onto the TFP of manufacturing firms

in the country

The crucial hypothesis tested in this paper is whether the aforementioned FDI-

induced improvements in service sectors benefit TFP of downstream manufacturing

users If present these gains could be classified as pecuniary (rent) spillovers a by-

product of market transactions (Griliches 1992) Manufacturing firms benefit from

pecuniary spillovers if increases in the quality or variety of the services they use due to

FDI are not fully appropriated by service providers In imperfectly competitive service

sectors providers may not appropriate the full surplus from better and more diversified

services because of their inability to perfectly price discriminate whereas in sectors where

FDI increases competition competitive pressures may prevent providers from

appropriating the surplus FDI in services can also benefit manufacturing firms through

spillovers of lsquosoft technologyrsquo linked to eg managerial know-how or technical skills

Learning by manufacturing firms could result from demonstration effects personal

contacts manager or worker turnover9 Griliches (1992) distinguishes knowledge

spillovers from pecuniary spillovers since in principle only the former allow

manufacturing firms to improve their innovation capabilities But in practice pecuniary

spillovers may become knowledge spillovers if downstream users of better services apply

the embodied knowledge to improve their TFP (Branstetter 2001) First for knowledge-

intensive business services such as information technology (IT) the actual service

provided is a knowledge-intensive input upon which firms rely to improve their

9 Kugler (2006) notes a larger potential for knowledge spillovers from vertical relative to horizontal FDI as for the

former foreign-owned service suppliers are not concerned about knowledge leakages since they operate in different

sectors than their domestic manufacturing clients

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 8: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

7

players Global MNCs identified Chilersquos largely privately-owned firms as an attractive

investment opportunity to consolidate their positions in Latin America (ECLAC 2000)

22 Effects of FDI in Services

FDI in the services sector can have four effects within the sector price reductions

quality improvements increased variety and knowledge spillovers7 First FDI is likely

to increase competition in local markets resulting in price reductions as incumbent firms -

eg in electricity and telecommunications sectors - no longer retain the rents they

obtained from being previously monopoly providers The available evidence for banking

electricity and telecommunications confirms price decreases for Chile (Stehmann 1995

Claessens et al 2001 Pollitt 2004) Second FDI may lead to service quality

improvements due to competition and the superior technological organizational and

managerial know-how of foreign-owned service providers FDI can also provide the

necessary finance for major upgrades and the expansion of existing electricity and

telecommunications networks improving the reliability of provision UNCTAD (2004)

and World Bank (2004) report evidence of such developments for Latin America Third

FDI may increase the variety of services provided including new technologically

advanced services or services provided to new regions or new types of clients FDI had

such effects in the Chilean telecommunications sector (ECLAC 2000) Fourth FDI may

result in leaking of managerial marketing and organizational know-how and best

practices (eg linked to the environment or labor codes) from foreign-owned to

domestic-owned services providers (Miroudout 2006)

7 FDI can also entail costs (1) foreign ownership in inherently monopolistic sectors such as electricity or

telecommunications may result in higher prices unless the regulatory system is well defined and managed by the

government and (2) foreign ownership may crowd out domestic firms eg in the banking sector (UNCTAD 2004)

8

The aforementioned positive effects of service FDI are based on the premise that

foreign-owned firms are better performers offering superior services and being more

productive than their domestic counterparts However due to data limitations evidence of

the superiority of foreign-owned service firms is rare The Fourth Chilean Innovation

Survey is a notable exception covering 612 service firms in the electricity generation real

estate financial intermediation business activities and transport storage and

communication sectors8 The survey collected data for 2003-2004 on firm innovation

outcomes accounting variables and basic characteristics Using this data we examine

whether foreign ownership is associated with better performance for Chilean service

firms If foreign investors acquire the best performing domestic service firms then a

positive effect of foreign ownership on firm performance would simply indicate the

endogeneity of the ownership status rather than the intrinsic advantages - eg better

technology - of the foreign parent company However this problem would not arise for

greenfield FDI Hence for each Chilean service firm we obtained information on

whether foreign ownership is due to greenfield FDI or foreign acquisition Table 1 shows

the results from regressions of three firm performance variables - labor productivity and

indicators for product and for process innovations - on dummy variables for greenfield

FDI and for foreign acquisition along with several controls The results show that

foreign-owned firms particularly those resulting from greenfield FDI exhibit better

productivity and innovation outcomes than their domestic counterparts While the

evidence in Table 1 is not causal due to the cross-sectional nature of the data it is

8 The survey includes a census of firms generating more than 2 megawatts of electricity per hour (category E of the

ISIC Rev 3 classification) and representative samples of the real estate renting and business activities (ISIC Rev 3

category K) financial intermediation services (ISIC Rev 3 category J) and transport storage and communication

services (ISIC Rev 3 category I)

9

suggestive of better performance by foreign-owned service firms in Chile and thus

provides support for the potential for FDI spillovers onto the TFP of manufacturing firms

in the country

The crucial hypothesis tested in this paper is whether the aforementioned FDI-

induced improvements in service sectors benefit TFP of downstream manufacturing

users If present these gains could be classified as pecuniary (rent) spillovers a by-

product of market transactions (Griliches 1992) Manufacturing firms benefit from

pecuniary spillovers if increases in the quality or variety of the services they use due to

FDI are not fully appropriated by service providers In imperfectly competitive service

sectors providers may not appropriate the full surplus from better and more diversified

services because of their inability to perfectly price discriminate whereas in sectors where

FDI increases competition competitive pressures may prevent providers from

appropriating the surplus FDI in services can also benefit manufacturing firms through

spillovers of lsquosoft technologyrsquo linked to eg managerial know-how or technical skills

Learning by manufacturing firms could result from demonstration effects personal

contacts manager or worker turnover9 Griliches (1992) distinguishes knowledge

spillovers from pecuniary spillovers since in principle only the former allow

manufacturing firms to improve their innovation capabilities But in practice pecuniary

spillovers may become knowledge spillovers if downstream users of better services apply

the embodied knowledge to improve their TFP (Branstetter 2001) First for knowledge-

intensive business services such as information technology (IT) the actual service

provided is a knowledge-intensive input upon which firms rely to improve their

9 Kugler (2006) notes a larger potential for knowledge spillovers from vertical relative to horizontal FDI as for the

former foreign-owned service suppliers are not concerned about knowledge leakages since they operate in different

sectors than their domestic manufacturing clients

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 9: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

8

The aforementioned positive effects of service FDI are based on the premise that

foreign-owned firms are better performers offering superior services and being more

productive than their domestic counterparts However due to data limitations evidence of

the superiority of foreign-owned service firms is rare The Fourth Chilean Innovation

Survey is a notable exception covering 612 service firms in the electricity generation real

estate financial intermediation business activities and transport storage and

communication sectors8 The survey collected data for 2003-2004 on firm innovation

outcomes accounting variables and basic characteristics Using this data we examine

whether foreign ownership is associated with better performance for Chilean service

firms If foreign investors acquire the best performing domestic service firms then a

positive effect of foreign ownership on firm performance would simply indicate the

endogeneity of the ownership status rather than the intrinsic advantages - eg better

technology - of the foreign parent company However this problem would not arise for

greenfield FDI Hence for each Chilean service firm we obtained information on

whether foreign ownership is due to greenfield FDI or foreign acquisition Table 1 shows

the results from regressions of three firm performance variables - labor productivity and

indicators for product and for process innovations - on dummy variables for greenfield

FDI and for foreign acquisition along with several controls The results show that

foreign-owned firms particularly those resulting from greenfield FDI exhibit better

productivity and innovation outcomes than their domestic counterparts While the

evidence in Table 1 is not causal due to the cross-sectional nature of the data it is

8 The survey includes a census of firms generating more than 2 megawatts of electricity per hour (category E of the

ISIC Rev 3 classification) and representative samples of the real estate renting and business activities (ISIC Rev 3

category K) financial intermediation services (ISIC Rev 3 category J) and transport storage and communication

services (ISIC Rev 3 category I)

9

suggestive of better performance by foreign-owned service firms in Chile and thus

provides support for the potential for FDI spillovers onto the TFP of manufacturing firms

in the country

The crucial hypothesis tested in this paper is whether the aforementioned FDI-

induced improvements in service sectors benefit TFP of downstream manufacturing

users If present these gains could be classified as pecuniary (rent) spillovers a by-

product of market transactions (Griliches 1992) Manufacturing firms benefit from

pecuniary spillovers if increases in the quality or variety of the services they use due to

FDI are not fully appropriated by service providers In imperfectly competitive service

sectors providers may not appropriate the full surplus from better and more diversified

services because of their inability to perfectly price discriminate whereas in sectors where

FDI increases competition competitive pressures may prevent providers from

appropriating the surplus FDI in services can also benefit manufacturing firms through

spillovers of lsquosoft technologyrsquo linked to eg managerial know-how or technical skills

Learning by manufacturing firms could result from demonstration effects personal

contacts manager or worker turnover9 Griliches (1992) distinguishes knowledge

spillovers from pecuniary spillovers since in principle only the former allow

manufacturing firms to improve their innovation capabilities But in practice pecuniary

spillovers may become knowledge spillovers if downstream users of better services apply

the embodied knowledge to improve their TFP (Branstetter 2001) First for knowledge-

intensive business services such as information technology (IT) the actual service

provided is a knowledge-intensive input upon which firms rely to improve their

9 Kugler (2006) notes a larger potential for knowledge spillovers from vertical relative to horizontal FDI as for the

former foreign-owned service suppliers are not concerned about knowledge leakages since they operate in different

sectors than their domestic manufacturing clients

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 10: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

9

suggestive of better performance by foreign-owned service firms in Chile and thus

provides support for the potential for FDI spillovers onto the TFP of manufacturing firms

in the country

The crucial hypothesis tested in this paper is whether the aforementioned FDI-

induced improvements in service sectors benefit TFP of downstream manufacturing

users If present these gains could be classified as pecuniary (rent) spillovers a by-

product of market transactions (Griliches 1992) Manufacturing firms benefit from

pecuniary spillovers if increases in the quality or variety of the services they use due to

FDI are not fully appropriated by service providers In imperfectly competitive service

sectors providers may not appropriate the full surplus from better and more diversified

services because of their inability to perfectly price discriminate whereas in sectors where

FDI increases competition competitive pressures may prevent providers from

appropriating the surplus FDI in services can also benefit manufacturing firms through

spillovers of lsquosoft technologyrsquo linked to eg managerial know-how or technical skills

Learning by manufacturing firms could result from demonstration effects personal

contacts manager or worker turnover9 Griliches (1992) distinguishes knowledge

spillovers from pecuniary spillovers since in principle only the former allow

manufacturing firms to improve their innovation capabilities But in practice pecuniary

spillovers may become knowledge spillovers if downstream users of better services apply

the embodied knowledge to improve their TFP (Branstetter 2001) First for knowledge-

intensive business services such as information technology (IT) the actual service

provided is a knowledge-intensive input upon which firms rely to improve their

9 Kugler (2006) notes a larger potential for knowledge spillovers from vertical relative to horizontal FDI as for the

former foreign-owned service suppliers are not concerned about knowledge leakages since they operate in different

sectors than their domestic manufacturing clients

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 11: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

10

innovation capabilities and TFP (Kox and Rubalcaba 2007) Second the usage of newer

services (eg internet banking) may embody technological knowledge allowing

manufacturing firms to improve their production and operations (eg by increasing their

IT investmentsrsquo efficacy) Third the FDI-induced increased reliability of service

provision may allow firms to optimize their machinery usage (eg production processes

are less disrupted due to electricity outages) and encourage firms to use technologically

more advanced production processes which depend on telecom or internetdata

connection These possibilities capture multiple dimensions of technological change thus

motivating a positive effect of FDI in services on firm TFP and epitomize the overlap

between pecuniary and knowledge spillovers which will characterize our main results

3 Manufacturing Firm-Level Data

The main dataset used in our analysis is the Encuesta Nacional Industrial Annual

(ENIA) which is the annual manufacturing survey of Chilean firms with more than 10

employees10

The dataset is an unbalanced panel capturing firm entry and exit that

includes an average of 4913 firms per year for the 1992-2004 period classified into 4-

digit ISIC revision 2 industries The Appendix provides details on how the final sample

of 57025 observations is obtained The ENIA survey collects firm-level information on

sales employment raw materials investments (buildings machinery and equipment

transportation and land) which are used to construct output and inputs for the production

function discussed in Section 4 All nominal variables are expressed in real terms using

10 The Chilean Statistical Institute (INE) collects information on which plants

in the ENIA are part of a multi-plant firm ie a firm with at least two plants responding to the survey The information

was kindly provided by INE to the authors for the purposes of a related research project During the 1997-2003 period

on average 917 of plants are single-plant firms Thus plant data corresponds to a large extent to firm data and we

designate the units of observation as firms throughout the paper The composition of our sample across years and

industries as well as summary statistics for the variables used in our econometric analysis are provided in the working

paper Fernandes and Paunov (2008)

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 12: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

11

appropriate deflators and capital is constructed applying the perpetual inventory method

as described in the Appendix

A particularly interesting and novel feature of the ENIA survey is that it collects

information on firm-level expenditures on a variety of services advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services electricity and water This information allows us to include a bundle of

services (excluding electricity) appropriately deflated as inputs in the production function

discussed in Section 4 For electricity the quantity consumed is the input included This

information also enables us to construct firm-specific weights representing the intensity

of service usage as detailed in Section 42

4 Empirical Specification

41 Basic Framework

In this section we present the reduced form framework used to estimate the impact

of FDI in services on Chilean manufacturing firmsrsquo TFP We consider a Cobb-Douglas

production function in logarithms for firm i in industry j at time t as in

j

it

j

K

j

it

j

s

j

it

j

E

j

it

j

M

j

it

j

UL

j

it

j

SL

j

it

j

it KSEMULSLAY lnlnlnlnlnlnlnln (1)

where Y is output SL is skilled labor UL is unskilled labor M is materials E is

electricity S is services K is capital and A is a firm-specific index of Hicks-neutral TFP

measuring the firmrsquos efficiency in transforming inputs into output The hypothesis tested

in this paper is that FDI in services affects firm TFP This effect could result from

pecuniary spillovers showing up in measured TFP through unaccounted for increases in

services quality and variety Equally important is the possibility that FDI in services

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 13: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

12

generates knowledge spillovers for manufacturing users and pecuniary spillovers can

result in knowledge spillovers Thus we allow firm TFP Aln to depend on the service

FDI linkage measure sFDI _ as in (ignoring the industry subscript j)

ititZitsfdiit ZsFDIA _ln _ (2)

where Z is a vector of control variables discussed in Section 43 and is a stochastic

residual A positive sfdi_ indicates a beneficial impact of FDI in services on firm TFP

Next we present our service FDI linkage measure and discuss the econometric issues

associated with the estimation of equation (2)

42 Service FDI Linkage Measure and Endogeneity Issues

To estimate the effects of service FDI on manufacturing TFP we obtain a

composite measure interacting FDI penetration in services and a firm-level measure of

the intensity of services usage The measure reflects the rationale that Chilean firms that

are relatively heavy users of services should (ceteris paribus) benefit disproportionately

more from increases in service FDI than firms that are less heavy users of services11

To

capture the intensity of service usage by firms we compute the ratio of firm expenditures

on four categories of services (henceforth designated as lsquofour service sectorsrsquo) - (1)

electricity and water (2) transport and communications (3) financial insurance and

business services and (4) real estate - to firm sales12

Our final measures for each firmrsquos

intensity of usage of each of the four services are obtained taking the average of the

corresponding historic service expenditure to sales ratios over the first three years of data

11 This assumption is inspired by that made by Rajan and Zingales (1998) in the estimation of the benefits of access to

external finance for industry growth Our assumption implies that there are no restrictions (other than cost) preventing

access to certain services by certain users 12 Business services encompass advertising legal technical and accounting services licenses and foreign technical

assistance and other services The four categories of services are dictated by the availability of sectoral GDP data from

the Chilean Central Bank used below

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 14: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

13

for each firm13

Our estimating sample for the effect of service FDI on TFP includes the

remaining years of data for each firm covering a total of 33390 observations The

separation of our panel dataset into these two groups addresses the potential endogeneity

of the services intensity measure with respect to firm TFP This approach follows the

study by Blalock and Gertler (2009) on how the effects of horizontal FDI on Indonesian

manufacturing firmsrsquo TFP are mediated by firm capabilities14

To capture the presence of FDI we compute for each service sector net FDI

inflows based on data from the Chilean Foreign Investment Committee by subtracting

from annual FDI inflows the corresponding annual FDI outflows (ie foreign investorsrsquo

repatriation of capital profits and dividends) Net FDI inflows do not adequately capture

the importance of FDI in a sector and year because they neither account for past

investments nor for the sectorrsquos size Thus we cumulate net FDI inflows using the

perpetual inventory method to construct an FDI stock for each sector Our measure of

FDI penetration in a service sector is given by the ratio of the sectorrsquos FDI stock to the

sectorrsquos output (GDP) obtained from the Chilean Central Bank15

Our final firm-level time-varying service FDI linkage measure that captures both

the presence of FDI in services and firm usage of those services is computed as

K

k

kt

k

iit FDIpensFDI1

_ where ktFDIpen is the FDI penetration ratio in service

13

Ideally we would like to measure a plantrsquos usage of foreign-provided services as separate from domestic-provided

services but such information is not available However to the extent that domestic service providers increase their

quality and variety and lower their prices due to the presence of FDI in their sector - through increased competition or

knowledge spillovers - measuring total service usage (whether services are purchased from foreign or domestic

suppliers) is adequate to capture the overall benefits from service FDI 14 The authors use the first three years of each firm in the sample to measure its capabilities (ie its technology gap

relative to the most productive firms) and use the remaining years of each firm for their main regression that includes

the interaction between horizontal FDI and firm capabilities 15 One caveat with this measure of FDI penetration is that it differs from that used in most studies on FDI spillovers in

manufacturing that relies on the ratio of foreign affiliatesrsquo employment relative to the total employment in the sector

Unfortunately we are unable to compute such measure for Chilean services firms since no firm-level census of services

firms is available for our sample period

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 15: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

14

sector k in year t which is weighted by k

i the historic intensity of usage of services from

sector k by firm i The sum is computed over the four aforementioned service sectors

More details on the construction of the measure are provided in the Appendix

Our service FDI linkage measure is inspired by the measures used by Javorcik

(2004) Blalock and Gertler (2008) and Arnold et al (2007a) but differs from those by

relying on a firm-level measure of service usage instead of service usage measures based

on input-output coefficients at the 4-digit industry level The latter measures provide

information on average industry usage which does not identify the heavy users of

services within an industry Figure 3 shows a significant degree of heterogeneity in the

average intensity of service usage across 2-digit industries in Chile However unreported

variance decompositions suggest that almost all the variation in the average intensity of

service usage is due to variation across firms within industries rather than across

industries This suggests that industry-level measures may be strongly misleading about

the service usage of firms Hence we choose to measure the intensity of service usage by

our average service expenditures to sales ratios based on historic values for each firm

Two issues could raise the possibility of endogeneity in the FDI penetration

component of the service FDI linkage measure with respect to TFP A first issue is that

manufacturing industries with higher TFP may lobby the government for services

liberalization However since Chilersquos FDI regime was liberal since the 1980s lobbying

by manufacturing industries for services liberalization would have occurred well before

our sample period and is thus unlikely to bias our estimates16

A second issue is that

manufacturing TFP in Chile in the 1990s may have been a driving force for service

16 In fact one may even question whether such lobbying played any role given that the privatization of service firms

starting in the late 1980s was partly motivated by the need to solve a public deficit problem (Bitran and Saez 1994)

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 16: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

15

MNCs to invest in the country in expectation of strong demand for their services17

ECLAC (2004) argues that foreign investors were attracted by the sound performance of

recently privatized service firms in Chile and the strategic global positioning of MNCs

played a crucial role in their FDI decisions However it may also be the case that the

intensity of FDI inflows into Chilean service sectors was related to manufacturing thus a

positive coefficient on the service FDI linkage measure in firm TFP regressions could

reflect the choice of foreign investors rather than positive effects of service FDI on TFP

As a first approach to this endogeneity problem we will experiment with using one- and

two-period lags of the service FDI linkage variable but we are fully aware that this only

attenuates the problem Thus our preferred approach is to use IV estimation We select

two related instruments for the intensity of FDI in service sectors in Chile the stocks of

FDI outflows of Spain and of the US in service sectors18

The instrumental service FDI

linkage measures used are the sum across the four service sectors of Spanish or US

stocks of FDI outflows in a service sector weighted by the historic firm intensity of usage

of services from that sector Spain and the US are the main sources of FDI inflows into

Chile Hence it is natural to expect their overall FDI outflows to have an impact on

services FDI penetration in Chile However it is highly unlikely that Spanish and US

overall FDI stocks abroad are correlated with Chilean manufacturing firmsrsquo TFP through

any other channel other than the Chilean service FDI penetration

43 Other Econometric Issues and Final Specification

17 Evidence of such mechanism is provided by Nefussi and Schwellnus (2010) who estimate a positive effect of

downstream demand for services by French manufacturing firms on the location choice business services firms 18 Details on the construction of the outward FDI stocks for these countries are provided in the Appendix

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 17: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

16

In order to identify the effect of service FDI on firm TFP we use unbiased TFP

estimates that correct for the potential endogeneity between input choices - particularly

the choice of service inputs - and firm unobserved productivity We estimate our

production function (equation (1)) following the methodologies of Levinsohn and Petrin

(2003) [below LP] Olley and Pakes (1996) and Ackerberg et al (2006)19

The main idea

behind these methodologies is that unobserved firm productivity shocks can be

approximated by a non-parametric function of observable firm characteristics ndash eg in

the LP case electricity and capital - and as a result unbiased estimates of the production

function coefficients are obtained To allow for differences in production technology

across industries a separate production function for each 2-digit industry is estimated by

each of the three methodologies The production function coefficient estimates are shown

in Appendix Table 1 The coefficients are generally significant and have magnitudes in

line with those in previous studies Based on the three unbiased sets of estimates we

obtain three firm-level time-varying logarithmic TFP measures as residuals from equation

(1)20

Simple summary statistics on our TFP estimates suggest that they are quite

sensible For example several patterns are consistent with the predictions from models of

industry dynamics where competitive pressures lead to market selection based on

productivity differences within-industry TFP dispersion is larger for more concentrated

industries and within industries firm TFP is negatively associated with firm exit

19 The latter addresses collinearity problems in the Olley and Pakes (1996) and Levinsohn and Petrin (2003)

methodologies that prevent the identification of the variable input coefficients The Appendix provides further details

on the production function estimation 20 Note that our TFP estimates are obtained from production functions where nominal sales revenues (input

expenditures) deflated with industry price indexes are used to measure output (inputs) and are therefore subject to the

criticism by Katayama et al (2009) that they are revenue-per-unit-input-bundle measures that combine true efficiency

and price-cost markups Eslava et al (2004) address this problem empirically using information on firm prices

Theoretically as long as mark-ups increase with efficiency revenue-based TFP estimates will be positively correlated

with firm efficiency The differentiated products model developed by Bernard et al (2003) predicts in equilibrium a

positive relationship between firm size (market shares) price-cost markups and efficiency Hence this model provides

a rationale to expect our revenue-based TFP estimates to be positively correlated with unobserved true firm efficiency

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 18: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

17

Moreover firm TFP is found to be positively correlated with firm export status and firm

size within industries21

It is important to note that our TFP estimates are purged from the

effects of services since service use is explicitly accounted for as an input in the

production function Correlations between firm TFP and service intensity suggest in fact

that firms that use services more intensively are allocated a lower value of TFP

Second we consider as controls some time-varying industry- or firm-level

observable factors that could be correlated with the service FDI linkage and firm TFP and

whose omission could bias our coefficient of interest We account for the potential

spillovers from FDI in manufacturing or mining by FDI linkage measures for

manufacturing and mining22

These variables are allowed to be endogenous in our IV

specifications Thus we add to the aforementioned service FDI linkage measure

constructed based on Spanish or US stocks of service FDI outflows manufacturing and

mining FDI linkage measures constructed based on Spanish or US stocks of

manufacturing and mining FDI outflows as instruments23

To allow for differences in

service usage and in TFP for entrants into the export market we include a control for

firm exporter status in our specification

Third to control for static firm unobservables such as managerial ability we allow

the residual in equation (2) to include a firm-specific component f such that

21 See Bartelsman and Doms (2000) for a discussion of empirical evidence related to models of industry dynamics The

finding on dispersion is obtained measuring for each 2-digit industry concentration by its normalized Herfindahl index

of firm market shares and TFP dispersion by the difference between the 10th and the 90th percentile of the TFP

distribution The evidence for firm exit firm export status and firm size is obtained in regressions that include either 2-

digit industry or 2-digit industry-year fixed effects that are necessary to account for the differences in technology across

industries (ie for the fact that TFP levels are not comparable across industries) 22

The mining and manufacturing FDI linkage variables are constructed using coefficients from the 1996 Chilean input-

output table as described in the Appendix Hence in equation (3) below they are indexed by the industry subscript j

instead of the firm subscript i 23 Specifically the stocks of Spanish or US outward FDI in manufacturing and mining are interacted with the 1996

Chilean input-output table weights

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 19: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

18

itiit uf where u is an independent and identically distributed (iid) disturbance

Hence our final specifications are estimated by firm fixed effects IV

Fourth industries differ in productivity levels technological progress experienced

and market structure dynamics Chilean regions may also exhibit differential performance

over time due to the evolving nature and importance of agglomeration economies To

account for these possibilities we add 2-digit industry-year interaction fixed effects and

region-year interaction fixed effects to equation (2)

The considerations above lead to our final empirical specification

itireg

inditzjtmnfdijtmfdiitsfdiit

ufyearreg

yearinddxmnFDImfFDIsFDIA

___ln ___

(3)

where mfFDI _ and mnFDI _ are manufacturing and mining FDI linkages dx is a

dummy for firm export status indyear and reg year are 2-digit industry-year and

region-year interaction fixed effects and u is iid

5 Results

51 Main Results

Our main results are shown in Tables 2 and 4 for TFP estimates obtained from

Levinsohn and Petrin (2003) (Panel A) Olley and Pakes (1996) (Panel B) and Ackerberg

et al (2006) (Panel C) In Table 2 column 1 presents the results from estimating by firm

fixed effects a variant of equation (3) with no controls while column 2 presents the results

from the exact specification in equation (3) The estimates show a positive and significant

effect of service FDI penetration on firm TFP In columns 3 and 4 of Table 2 a variant of

equation (3) is estimated by firm fixed effects including respectively the one- and two-

period lags of the service FDI linkage measure The significant positive effect of FDI in

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 20: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

19

services on firm TFP is maintained Recognizing that the use of lagged values only

attenuates any potential endogeneity concerns we proceed to our preferred method IV

estimation with firm fixed effects whose results are shown in Table 4 while the

corresponding first-stage results are shown in Table 3 The first-stage regression results

for the Chilean service FDI linkage measure show a strong correlation between that

endogenous variable and the instrument which is a service FDI linkage measure

constructed based on current stocks of Spanish outward FDI in services24

Note that to

ensure that the controls are not driving our IV results we show in columns 1-4 of Table 4

the estimates from intermediate specifications before showing the estimates from the full

specification in column 5 the specification in column 1 includes only the service linkage

those in columns 2 and 3 add only the manufacturing linkage and the manufacturing and

mining linkages and that in column 4 adds only the export dummy We find that the

service FDI linkage variable affects Chilean manufacturing firmsrsquo TFP positively The R-

squared from those regressions suggests that our instruments explain a substantial

fraction (38) of the variation in the service FDI linkage measure The standard errors in

Table 2 and all subsequent tables are robust and clustered at the firm-level to allow for

possible serial correlation across observations belonging to the same firm25

52 Magnitudes and Robustness Checks

Following standard practice in instrumental variables estimation we consider in

Table 5 specifications that use alternative instruments for the Chilean service FDI linkage

24 Note that in columns 2 3 and 5 of Table 4 FDI linkage measures based on the current stocks of Spanish outward

FDI in manufacturing and mining are also used as instruments since Chilean manufacturing and mining FDI linkages

are included in the second-stage regressions and also allowed to be endogenous Table 3 presents only one set of first-

stage regressions for the service FDI linkage given that the specifications and the sample used for IV estimation are the

same for the second-stage IV regressions shown in all panels of Table 4 (only the dependent variable TFP changes

across panels) 25 The use of non-clustered robust standard errors does not affect the significance of our results

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 21: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

20

measure and re-estimate its effects on our three TFP measures Column 1 presents the

results when the service FDI linkage instrument is constructed based on the current US

stocks of outward FDI in services column 2 combines that measure with the measure

used in Table 4 based on current Spanish FDI in services as instruments and column 3

shows the results from using both current and one-period lagged services FDI linkage

measures constructed based on both Spanish and US stocks of outward FDI in services

as instruments26

In all cases the estimated effects of the service FDI linkage measure on

TFP are positive and significant It is also important to point out that our specifications do

not suffer from weak instrument problems as reflected by the p-values for the

Kleibergen-Paap under-identification test Also our instruments appear to be adequate as

indicated by the p-values from the Hansen over-identification tests

Thus our findings indicate that increased FDI in services in Chile during the

sample period led to a significant increase in TFP for firms using the services more

intensively The coefficients in column 5 across all panels of Table 4 suggest that a one-

standard deviation increase in service FDI (ie a 0085 increase in the FDI linkage

measure) would bring a 3 increase in TFP of Chilean firms all else constant To

quantify further the economic magnitude of the impact of FDI in services we use the

lowest point estimate in columns 1-3 of Table 5 (ie 0224 in column 3 of Panel C) as a

lower bound on that impact According to our estimates TFP in the Chilean

manufacturing sector increased by about 12 over the sample period and the service FDI

linkage variable increase over the sample period was 003827

Thus the forward linkages

26

Manufacturing and mining FDI linkage measures constructed analogously are also used as instruments 27

In order to obtain TFP growth for the manufacturing sector over the sample period for each of our TFP estimates we

proceed in four steps First we compute for each 2-digit industry and year a weighted average TFP level which weighs

each firmrsquos TFP by the firmrsquos share in total sales of the 2-digit industry in that year Second across every two

consecutive years we calculate the growth rates in this 2-digit industry TFP measure Third for each year we compute

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 22: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

21

from FDI in services explain about 7 of the observed increase in Chilersquos manufacturing

usersrsquo TFP28

This economic impact is quite meaningful in light of the finding by Haskel

et al (2007) that horizontal spillovers from manufacturing FDI explain 5 of

manufacturing TFP growth in the UK between 1973 and 1992

To verify the robustness of our main results in Table 4 we conduct an extensive set

of tests including the control for differential TFP trends across plants with different

services usage changes in the dynamic pattern of the effects outlier exclusion variations

in the definition of the service FDI linkage measure and the consideration of a one-stage

regression IV estimation is used for all these robustness tests and the results are shown in

columns 4-11 of Table 5 in Panels A-C for each of the TFP measures and in column 12 of

Table 5 in Panel D for the one-stage regression

First column 4 considers the possibility that the service FDI linkage variable could

be picking up differential TFP trends across firms with different services intensity rather

than an effect of services FDI on firm TFP Within each industry we divide firms into

quartiles according to the distribution of firm initial services intensity Equation (3) is re-

estimated including four interaction terms corresponding to the interaction of a dummy

identifying each of those quartiles and a time trend29

The results show that the effect of

services FDI is maintained

TFP growth of the manufacturing sector as the weighted average of the 2-digit industry TFP growth rates where the

weights are given by the share of the 2-digit industry in manufacturing sector sales Fourth we sum this weighted

average TFP growth over all sample years to obtain a measure of the total TFP growth of the manufacturing sector over

the sample period 28 The 7 figure is obtained as the ratio of (i) the product of the regression coefficient (0224) and the increase in the

service FDI linkage variable over the sample period (0038) to (ii) the total TFP growth in the manufacturing sector

over the sample period (12) 29 Firm initial services intensity is computed as the ratio of the sum of expenditures on all services to sales averaged

across the first three years for each firm and a firm-specific time trend is used We thank Eric Verhoogen (the co-

editor) for suggesting this specification

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 23: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

22

Second while some effects of service FDI are likely to be immediate such as

pecuniary price spillovers knowledge spillovers may take time to materialize Thus

columns 5 and 6 show the results from a variant of equation (3) including our services

FDI linkage measure in period t-1 or t-2 and adjusting the instruments accordingly to be

also lagged The results are qualitatively maintained

Third columns 7 and 8 investigate whether our results are driven by outliers in our

TFP estimates We re-estimate equation (3) for a sample where the observations whose

TFP values are above or below 15 times the inter-quartile range (the difference between

the 75th

and the 25th

percentile) of the TFP distribution in each 2-digit industry and year

are removed and for a sample where the top and bottom 1 of the TFP distribution in

each 2-digit industry and year are eliminated In both cases our results are maintained

Fourth while we believe that our service FDI linkage measure captures the

importance of FDI in services in Chile we verify in columns 9-10 whether our results are

robust to modifications in the two components of that measure the weights and the

service FDI penetration Column 9 reports the results from using 4-digit industry service

usage coefficients from the 1996 Chilean Input-Output table as weights for the service

FDI penetration The effect of service FDI on firm TFP is positive and significant at the

5 confidence level As noted in Section 1 relying on 4-digit industry service usage

weights has the caveat of assuming that all firms within an industry use services in the

same proportion even though that is highly disproved by the Chilean data However that

very assumption can be used as a check to our main results in the following sense If

there was a systematic relationship between a firm characteristic (say firm size) and TFP

and between that characteristic and the intensity of usage of services then our findings in

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 24: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

23

Table 4 could be conflating the firm size effect with the effect of service FDI However

that does not seem to be the case given our finding of a positive effect of service FDI on

firm TFP relying on the FDI linkage measure based on 4-digit industry service usage

weights that are uncorrelated with any firm characteristics In column 10 we use dummy

variables that identify firms whose service usage intensity (in the first 3 sample years) is

above the sample median to multiply service FDI penetration in the four service sectors

The magnitude of the IV estimate is naturally very different from those in Table 4 but

still shows a positive and significant effect of service FDI on TFP of higher-than-median

service users Column 11 reports results using the logarithm of service FDI penetration

stocks without output denominator with the same firm-level weights as in Table 4 The

positive effect of the service FDI linkage measure on firm TFP is maintained 30

Fifth equations (1) and (3) can be viewed as two steps in a two-step estimation

procedure for the effect of services FDI on TFP which assumes that the covariates from

the first step are not correlated with the covariates in the second step To address the

possibility of correlation we show in column 12 of Panel D of Table 5 the results from a

one-stage regression which is an augmented version of equation (1) that includes all the

regressors in equation (3) in addition to the production inputs The IV fixed effects

estimates show that the positive and significant effect of services FDI is maintained31

30 While the industry-year interaction effects included in our specifications account in general terms for industry-

specific technological progress and competition we also experimented with replacing those by observable measures of

the degree of competition the normalized Herfindahl index of firm market shares and the market share of the top 4

firms both at the 3-digit industry level The unreported effects of service FDI on firm TFP are maintained 31 We conducted another unreported robustness check changing the dependent variable in equation (3) to be a firm TFP

index computed following Aw et al (2001) The estimates show a positive effect of service FDI on the firm TFP index

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 25: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

24

Sixth the coefficients on services shown in Appendix Table 1 are generally lower

that the observed input shares shown in Figure 332

Thus firms with higher service input

usage may have been assigned a higher TFP level than they would if the estimated

coefficients were more in line with the observed input shares If service input usage was

correlated with service sector FDI then this would raise the concern that the effect of FDI

in services was an artifact of estimation rather than a true productivity effect To address

this concern we re-estimated our IV regressions using TFP index measures computed

following Aw et al (2001) using the average service input shares in Figure 3 The results

which are available from the authors upon request are qualitatively maintained 33

6 Characterizing the Effects of FDI in Services

Our findings so far concern the average impact of FDI in services on firm TFP

across the Chilean manufacturing sector However the importance of service FDI for

firm TFP may differ across industries One reason for potential differences is that

services namely knowledge-based services - may play a particularly important role for

innovation Indeed for OECD countries Francois and Woerz (2008) show that the

openness of service sectors (in particular to FDI) enhances the performance of skill-

intensive and technology-intensive industries To address the possibility that FDI in

services is a contributor to innovation we follow two approaches The first approach is to

32 This statement refers to the bundle of business real estate and transport and communications services only The

coefficient on electricity shown in Appendix Table 1 is not comparable to the electricity input share shown in Figure 3

given that the electricity quantity rather than deflated expenditure (as is the case for other services) is the input entering

the production function 33 We thank a referee for pointing to this issue The bundle of electricity and services is obtained as the sum of

electricity expenditures and services expenditures deflated by a weighted average of the price deflators for these two

categories of services (using 3-digit industry input-output weights) Since TFP index measures impose the restrictive

assumption of constant returns to scale and perfect competition we examine the robustness of those results by

obtaining TFP measures as residuals from production functions that include this bundle of electricity and services by

OLS and by Levinsohn and Petrin (2003) and using them to re-estimate Equation (3) The results are again qualitatively

maintained

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 26: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

25

estimate a variant of equation (3) where the effect of service FDI on firm TFP is allowed

to vary with an industry variable related to its potential for innovation We consider the

definition of differentiated product industries proposed by Rauch (1999) and the RampD

intensity of industries used by Kugler and Verhoogen (2008)34

Columns 1 and 2 of

Panels A-C of Table 6 show the corresponding IV estimates The effects of service FDI

on firm TFP are stronger in differentiated product industries and in RampD intensive

industries The F-tests shows that the difference in the effects of FDI in services across

the different types of industries is statistically significant The coefficients in column 1 of

Panel B imply that a one-standard deviation increase in the service FDI linkage would

lead to a TFP increase of 3 for firms in differentiated product industries and of 2 for

firms in other industries all else constant The economic magnitude of these effects is

that the forward linkages from FDI in services explain about 16 of the increase in

manufacturing usersrsquo TFP in differentiated product industries but only 3 of the increase

in TFP in other industries

The second approach is to consider an indirect measure of firm innovation its

investment-capital ratio Huggett and Ospina (2001) argue that substantial increases in

investment-capital ratios represent the adoption of new technology thus process

innovation by a firm Column 3 of Panel D shows the results from estimating equation

(3) using firm-level machinery and vehicle investment-capital ratios as the dependent

variable The estimated effect of service FDI on investment-output ratios is positive and

34 Differentiated products are defined to be those that are neither (i) homogenous products traded in organized

exchanges (eg steel) nor (ii) reference-priced products which have listed prices in trade publications (eg some

chemical products) and require a more important degree of buyer-seller interaction To apply Rauchrsquos definition we

establish a correspondence between his classification of products based on 4-digit SITC rev 2 codes and our 4-digit

ISIC rev 2 codes For the printing industry (ISIC 342) we are unable to establish an unambiguous correspondence and

thus drop it from the regressions in Table 5 Kugler and Verhoogen (2008) use data from the US 1976 Federal Trade

Commission Line of Business Survey to compute the ratio of RampD expenditures to total sales for large firms in US

industries and we use those ratios as a benchmark for our Chilean 4-digit industries RampD-intensive industries are

defined as those with higher than median RampD intensity in the sample

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 27: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

26

significant The findings in Table 6 are suggestive of a role of services FDI for innovation

outcomes in manufacturing This reinforces the importance of services liberalization in

view of their providing essential knowledge services required for firms to engage in

innovative activities

The differences across industries just described are stark but another important

heterogeneity dimension concerns the variability in effects across firms within a given

industry It is important to identify which manufacturing firms benefit more from FDI in

services in Chile In Table 7 we allow the effect of the service FDI linkage measure on

firm TFP to vary with the firmrsquos technology gap relative to the lsquotechnological leadersrsquo in

its industry In column 1 the closeness to lsquotechnological leadersrsquo is defined as the average

distance between a firmrsquos TFP and the average TFP of firms in the top 10th

percentile of

the TFP distribution in the first three years of the firm whereas in column 2 it is defined

as the average distance between a firmrsquos TFP and the average TFP of the top 10th

percentile of the TFP distribution of foreign-owned firms in the first three years of the

firm Both results suggest that firms that are more distant from the technological frontier

experience stronger TFP benefits from service FDI The economic magnitude of the

effects in column 1 of Panel A of Table 7 is that an increase in the service FDI linkage by

one standard deviation would lead to an increase of 7 in TFP for firms in the 10th

percentile of the closeness to frontier distribution to no change in TFP for firms in the

50th

percentile of the closeness to frontier distribution and to a decline of 3 in TFP for

firms in the 90th

percentile of the closeness to frontier distribution35

The interpretation

35 The 7 figure is obtained by multiplying one standard deviation in the service FDI linkage (0066) by the total effect

of the linkage for the 10th percentile of the closeness to frontier distribution (10741) the latter being equal to the sum

of the coefficient on the linkage (1176) and the product of the coefficient on the interaction (-1657) and the 10th

percentile of the closeness to frontier distribution (0059)

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 28: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

27

for these findings is that the technologically less advanced Chilean firms have an

opportunity to catch-up by learning about advanced managerial and organizational

techniques optimizing their machinery usage and improving their production as a result

of the increased reliability of service provision and the knowledge embodied in new

services brought by FDI Technologically more advanced firms by contrast have less to

gain since they already use better lsquosoftrsquo as well as lsquohardrsquo technologies Our findings echo

those by Blalock and Simon (2010) who show that Indonesian firms with better

capabilities gain less from supplying foreign multinationals and that the least productive

firms have most room for improvement and thus to gain from new technology adoption

In face of the large degree of heterogeneity across firms it is interesting to see that one of

the contributions of services FDI in Chile was to provide opportunities for lsquolaggardrsquo firms

to catch up

7 Conclusion

This paper finds that FDI in services had a positive effect on Chilean manufacturing

firms TFP In the estimation particular care was taken to measure the intensity of service

usage by firms given the large degree of heterogeneity of usage found within industries

Also we employ three different methodologies to obtain unbiased TFP estimates

Moreover endogeneity concerns that inevitably arise when attempting to estimate causal

effects of service FDI on manufacturing firmsrsquo TFP are addressed using firm fixed effects

instrumental variable estimation whereby service FDI penetration weighted by historic

firm intensity of service usage is instrumented using values of the outward FDI stocks in

service sectors of the two major foreign investors in Chile Spain and the US weighted

by historic firm intensity of service usage

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 29: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

28

While governments spend large sums to attract FDI inflows in expectation of

spillovers the literature focusing the manufacturing sector has provided mixed evidence

relatively weak for horizontal spillovers and strong for vertical spillovers This is not

altogether surprising as foreign-owned firms have strong incentives to avoid technology

spillovers to local competitors in their industry whereas the same does not apply for

downstream providers and upstream clients of their products and services Our study

suggests that the service sector might turn out to be a particularly valuable source of

positive spillover effects from FDI Reducing the barriers that still protect FDI in services

in many emerging and developing economies may help improve TFP in their

manufacturing sectors Our evidence also hints at a role of services FDI in stimulating

innovation by manufacturing firms It is not surprising that we find evidence of such

effects as innovative activities require access to leading knowledge services that foreign

affiliates are particularly specialized at providing Services FDI can thus be seen as a

vehicle to stimulate innovative activities of manufacturing firms in countries behind the

technology frontier complementing catch-up opportunities provided by trade relations

More direct evidence on the impacts of service FDI on innovation would be valuable

Interestingly we also find that services FDI offers opportunities for lagging firms to

catch up with industry leaders This contradicts the frequently held idea that only the

most advanced firms in emerging economies can benefit from increased relations with

foreign-owned firms via FDI or trade Our results suggest that in some cases learning

opportunities can benefit more those further behind Concerns about opening emerging

economies further to FDI and trade based on alleged negative effects on less advanced

firms do not appear to be therefore a valid general objection

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 30: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

29

References

Ackerberg D Caves K Frazer G 2006 Structural Identification of Production

Functions UCLA mimeo

Arnold J Javorcik B Mattoo A 2007 Does Services Liberalization Benefit

Manufacturing Firms Evidence from the Czech Republic World Bank Policy Research

Working Paper No 4109

Arnold J Javorcik B Lipscomb M Mattoo A 2010 Services Reform and

Manufacturing Performance Evidence from India CEPR Discussion Paper 8011

Aw Y-B Chen X Roberts M 2001 Firm-Level Evidence on Productivity

Differentials and Turnover in Taiwanese Manufacturing Journal of Development

Economics 66 51-86

Bartelsman E Doms M 2000 Understanding Productivity Lessons from Longitudinal

Data Journal of Economic Literature 38 569-594

Bernard A Eaton J Jensen J Kortum S 2003 Plants and Productivity in

International Trade America Economic Review 93 1268-1290

Bitran E Saez R 1994 Privatization and Regulation in Chile in Bosworth B

Dornbush R Laban R (Eds) The Chilean Economy Policy Lessons and Challenges

The Brookings Institution Washington DC pp 329-367

Blalock G Gertler P 2008 Welfare Gains from Foreign Direct Investment through

Technology Transfer to Local Suppliers Journal of International Economics 38 402-421

Blalock G Gertler P 2009 How Firm Capabilities Affect Who Benefits from Foreign

Technology Journal of Development Economics 90 192-199

Blalock G Simon D 2009 Do All Firms Benefit Equally from Downstream FDI The

Moderating Effect of Local Suppliers Capabilities on Productivity Gains Journal of

International Business Studies 40 1095-1112

Branstetter L 2001 Are Knowledge Spillovers International or Intranational in Scope

Microeconometric Evidence from the US and Japan Journal of International

Economics 53 53-79

Claessens S Demirguc-Kunt A Huizinga H 2001 How Does Foreign Entry Affect

Domestic Banking Markets Journal of Banking and Finance 25 891-911

Coe D Helpman E 1995 International RampD Spillovers European Economic Review

39 859-887

Damijan J Rojec M Majcen B Knell M 2008 Impact of Firm Heterogeneity on

Direct and Spillover Effects of FDI Micro Evidence from Ten Transition Countries

Katholieke Universiteit Leuven LICOS Discussion Paper No 2182008

Dixit A Stiglitz J 1977 Monopolistic Competition and Optimum Product Diversity

American Economic Review 67 297-308

ECLAC 1999 Summary and Conclusions in ECLAC Foreign Investment in Latin

America and the Caribbean

ECLAC 2000 Chile Foreign Direct Investment and Corporate Strategies Chapter 2 in

ECLAC Foreign Investment in Latin America and the Caribbean

ECLAC 2004 Energiacutea eleacutectrica Inversiones y estrategias empresariales en los paiacuteses del

Cono Sur in ECLAC La Inversioacuten Extranjera en Ameacuterica Latina y el Caribe

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 31: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

30

Eschenbach F Hoekman B 2006 Services Policy Reform and Economic Growth in

Transition Economies 1990-2004 Review of World Economics 142 746-764

Eslava M Haltiwanger J Kugler A Kugler M 2004 The Effects of Structural

Reforms on Productivity and Profitability Enhancing Reallocation Evidence from

Colombia Journal of Development Economics 75 333-371

Ethier W 1982 National and International Returns to Scale in the Modern Theory of

International Trade American Economic Review 72 389-405

Fernandes A 2009 Structure and Performance of the Service Sector in Transition

Economies Economics of Transition 17 467-501

Fernandes A Paunov C 2008 Foreign Direct Investment in Services and

Manufacturing Productivity Growth Evidence for Chile World Bank Policy Research

Working Paper No 4730

Francois J 1990 Trade in Producer Services and Returns due to Specialization under

Monopolistic Competition Canadian Journal of Economics 23 109-124

Francois J Woerz J 2008 Producer Services Manufacturing Linkages and Trade

Journal of Industry Competition and Trade 8 199-229

Griliches Z 1992 The Search for RampD Spillovers Scandinavian Journal of Economics

94 29-47

Haskel J Pereira S Slaughter M 2007 Does Inward Foreign Direct Investment

Boost the Productivity of Domestic Firms Review of Economics and Statistics 89 482-

496

Hoekman B 2006 Liberalizing Trade in Services A Survey World Bank Policy

Research Working Paper No 4030

Hugget M Ospina S 2001 Does Productivity Growth Fall After the Adoption of New

Technology Journal of Monetary Economics 48 173-95

Javorcik B 2004 Does Foreign Direct Investment Increase the Productivity of

Domestic Firms In Search of Spillovers through Backward Linkages American

Economic Review 94 605-627

Javorcik B Li Y 2008 Do the Biggest Aisles Serve a Brighter Future Global Retail

Chains and Their Implications for Romania World Bank Policy Research Working Paper

No 4650

Jensen J Rutherford T Tarr D 2007 The Impact of Liberalizing Barriers to Foreign

Direct Investment in Services The Case of Russian Accession to the World Trade

Organization Review of Development Economics 11 482-506

Katayama H Lu S Tybout J 2009 Firm-level productivity studies Illusions and a

solution International Journal of Industrial Organization 27 403-413

Konan D Maskus K 2006 Quantifying the Impact of Services Liberalization in a

Developing Country Journal of Development Economics 81 142-162

Kox H Rubalcaba L 2007 Business Services and the Changing Structure of European

Economic Growth Munich Personal RePEc Archive Paper No 3750

Kugler M 2006 Spillovers from Foreign Direct Investment Within or Between

Industries Journal of Development Economics 80 444-477

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 32: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

31

Kugler M Verhoogen E 2008 The Quality-Complementarity Hypothesis Theory and

Evidence from Colombia NBER Working Paper 14418

Levinsohn J Petrin A 2003 Estimating Production Functions Using Inputs to Control

for Unobservables Review of Economic Studies 70 317-341

Marcin K 2008 How Does FDI Inflow Affect Productivity of Domestic Firms The

Role of Horizontal and Vertical Spillovers Absorptive Capacity and Competition

Journal of International Trade and Economic Development 17 155-173

Markusen J 1989 Trade in Producer Services and in Other Specialized Intermediate

Inputs American Economic Review 79 85-95

Markusen J Rutherford T Tarr D 2005 Trade and Direct Investment in Producer

Services and the Domestic Market for Expertise Canadian Journal of Economics 38

758-777

Mattoo A Rathindran R Subramanian A 2006 Measuring Services Trade

Liberalization and Its Impact on Economic Growth An Illustration Journal of Economic

Integration 21 64-98

Mirodout S 2006 The Linkages between Open Services Markets and Technology

Transfer OECD Trade Policy Working Paper No 29

Moreira M Blyde J 2006 Chilersquos Integration Strategy Is There Room for

Improvement INTAL-ITD Working Paper 21

Nefussi B and C Schwellnus 2010 Does FDI in Manufacturing Cause FDI in Business

Services Evidence from French Firm-Level Data Canadian Journal of Economics 43

180-203

Olley S Pakes A 1996 The Dynamics of Productivity in the Telecommunications

Equipment Industry Econometrica 64 1263-1297

Pollitt M 2004 Electricity Reform in Chile Lessons for Developing Countries

Massachusetts Institute of Technology Center for Energy and Environmental Policy

Research Working Paper 0416

Pombo C 1999 Productividad Industrial en Colombia Una Aplicacioacuten de Nuacutemeros

Indice Revista de Economiacutea del Rosario 2 107-139

Rajan R Zingales L 1998 Financial Dependence and Growth American Economic

Review 88 559-586

Rauch J 1999 Networks versus Markets in International Trade Journal of International

Economics 48 7-35

Rivera-Batiz F Rivera-Batiz L 1992 Europe 1992 and the Liberalization of Direct

Investment Flows Services Versus Manufacturing International Economic Journal 6 45-

57

Stehmann O 1995 Network Liberalization and Developing Countries The case of

Chile Telecommunications Policy 19 667-684

UNCTAD 2004 World Investment Report The Shift Towards Services New York and

Geneva

World Bank 2004 Reforming Infrastructure Privatization Regulation and Competition

Washington DC

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 33: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

32

Figure 1 Stocks of FDI in Chilean Service Sectors

Source Authorrsquos calculations based on data from the Chile Foreign Investment Committee

0

1000

2000

3000

4000

5000

6000

7000

8000

1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

Stocks of FDI in services in million 2000 US$

Electricity and Water Transport and Communications

Business and Real Estate Services

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 34: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

33

Figure 2 Average Ratio of Sectoral FDI Stocks to Sectoral GDP

Source Authorrsquos calculations based on data from the Chilersquos Foreign Investment Committee and the Central Bank

Note The figures show the average ratio of sectoral FDI stocks to sectoral GDP in each of the two time periods

003

221

000

050

100

150

200

250

1989-1996 1997-2004

Electricity and Water

015

044

000

010

020

030

040

050

1989-1996 1997-2004

Transport and Communication

016

027

000

010

020

030

1989-1996 1997-2004

Business and Real Estate

Services

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 35: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

34

Figure 3 Intensity of Service Usage across Industries

Source Authorsrsquo calculations based on ENIA survey data

Notes The figure shows the average ratio of service usage to sales computed across firms over the sample period in each

of the ISIC rev 2 2-digit industries

002 001003 001 002 002 003

001 001

002002

004002

003 004004

002 003

009014

014

011011

011010

012

015

000

010

020

Electricity and Water Transport and CommunicationsBusiness and Real Estate Services

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 36: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

35

Table 1 Foreign Ownership and Firm Performance in Service Sectors

Notes Robust standard errors in parentheses and indicate significance at 1 5 and 10 confidence

levels respectively Firm controls include an exporter dummy and size and age dummies Industry fixed effects are at

the 2-digit ISIC Rev 3 level

OLS Estimation

Firm-Level Log

of Labor

Productivity

Firm-Level

Indicator of

Product

Innovation

Firm-Level

Indicator of

Process

Innovation

(1) (2) (3)

Greenfield FDI Dummy 0919 0884 0767

(0481) (0409) (0414)

Foreign Acquisition Dummy 1912 0358 0440

(0297) (0351) (0325)

Firm Controls Yes Yes Yes

Industry Fixed Effects Yes Yes Yes

Region Fixed Effects Yes Yes Yes

Observations 492 487 498

R-Squared 032

Dependent Variable

Probit Estimation

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 37: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

36

Table 2 Effect of FDI in Services on Firm TFP - OLS Estimation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively

(1) (2) (3) (4)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0278 0277

(0095) (0094)

Service FDI Linkage t-1 0274

(0068)

Service FDI Linkage t-2 0180

(0047)

Manufacturing FDI Linkage t -0485 -0217 -0365

(0651) (0738) (0821)

Mining FDI Linkage t -0176 -0140 -0203

(0112) (0114) (0123)

Export Dummy t 0021 0027 0030

(0010) (0012) (0013)

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0211 0210

(0069) (0068)

Service FDI Linkage t-1 0236

(0053)

Service FDI Linkage t-2 0175

(0045)

Manufacturing FDI Linkage t -0715 -0225 -0312

(0616) (0696) (0772)

Mining FDI Linkage t -0239 -0189 -0237

(0107) (0108) (0115)

Export Dummy t 0034 0041 0045

(0010) (0011) (0013)

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0228 0227

(0086) (0085)

Service FDI Linkage t-1 0233

(0071)

Service FDI Linkage t-2 0163

(0043)

Manufacturing FDI Linkage t -0592 -0135 -0163

(0648) (0738) (0823)

Mining FDI Linkage t -0227 -0179 -0221

(0109) (0109) (0117)

Export Dummy t 0027 0034 0037

(0010) (0012) (0013)

Firm Fixed Effects Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes

Observations 33390 33390 26925 21502

Number of Firms 5817 5817 5103 4354

OLS Estimation

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 38: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

37

Table 3 First-Stage IV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses indicates significance at the 1 confidence

level

(1) (2) (3) (4) (5)

Service FDI Linkage Based on Service FDI Stocks of Spain 0239 0239 0239 0239 0239

(0065) (0065) (0065) (0065) (0065)

Manufacturing FDI Linkage Based on Manufacturing FDI

Stocks of Spain 0002 0002 0002

(0002) (0002) (0002)

Mining FDI Linkage Based on Mining FDI Stocks of Spain -0002 -0002

(0003) (0003)

Export Dummy 0001 0001

(0001) (0001)

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

R-Squared 038 038 038 038 038

Observations 32682 32682 32682 32682 32682

OLS Estimation

First Stage Regressions Corresponding to Table 4

Dependent Variable Service FDI Linkage

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 39: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

38

Table 4 Effect of FDI in Services on Firm TFP ndashIV Estimation Results

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively FDI linkage variables are instrumented using service manufacturing and mining FDI

linkage measures constructed based on service manufacturing and mining FDI stocks of Spain at t

(1) (2) (3) (4) (5)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0422 0422 0422 0423 0423

(0131) (0130) (0131) (0131) (0130)

Manufacturing FDI Linkage t -1517 -1597 -1629

(1419) (1460) (1462)

Mining FDI Linkage t -0122 -0125

(0163) (0163)

Export Dummy t 0021 0022

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0384 0384 0384 0385 0385

(0125) (0125) (0125) (0126) (0125)

Manufacturing FDI Linkage t -1394 -1444 -1494

(1417) (1461) (1461)

Mining FDI Linkage t -0077 -0082

(0160) (0160)

Export Dummy t 0033 0034

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0362 0362 0362 0363 0363

(0124) (0124) (0124) (0124) (0124)

Manufacturing FDI Linkage t 0321 0304 0265

(1520) (1566) (1570)

Mining FDI Linkage t -0026 -0030

(0167) (0167)

Export Dummy t 0026 0026

(0010) (0010)

R-Squared of Service FDI Linkage First Stage

Regression038 038 038 038 038

Kleinbergen LM Statistic (under-identif test) 002 000 000 002 000

Firm Fixed Effects Yes Yes Yes Yes Yes

IndYear Fixed Effects Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes

Observations 33390 33390 33390 33390 33390

Number of Firms 5817 5817 5817 5817 5817

IV Estimation

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 40: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

Table 5 Robustness Results ndash Part 1

Alternative

Instruments -

Stocks US

at t

Alternative

Instruments -

Stocks

Spain US

at t

Alternative

Instruments -

Stocks

Spain US

at t and t-1

Including

Service-

Intensity

Trends

Service FDI

Linkage

at t-1

Service FDI

Linkage

at t-2

(1) (2) (3) (4) (5) (6)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0465 0393 0299 0673

(0177) (0108) (0098) (0202)

Service FDI Linkage Lagged 0274 0211

(0109) (0091)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)080 049

Observations 33390 33390 26925 33390 26925 21502

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0445 0338 0277 0636

(0174) (0104) (0092) (0189)

Service FDI Linkage Lagged 0275 0235

(0108) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)032 014

Observations 33390 33390 26925 33390 26925 21502

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0427 0320 0224 0600

(0172) (0104) (0093) (0185)

Service FDI Linkage Lagged 0221 0176

(0105) (0093)

Alternative Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression021 046 052 039 045 047

Hansen J Statistic (over-identif test of all

instruments)000 000 000

Kleinbergen LM Statistic (under-identif test) 017 012 000 000 000

Observations 33390 33390 26925 33390 26925 21502

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t

R-Squared of Service FDI Linkage First Stage

Regression

Kleinbergen LM Statistic (under-identif test)

Observations

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 41: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

40

Table 5 Robustness Results ndash Part 2

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 5

and 10 confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and

mining linkage variables The latter two variables are endogenous in the IV estimation In column 1 the instruments for

the Chilean FDI linkage variables are service manufacturing and mining FDI linkage measures constructed based on

the corresponding FDI stocks of the US at t In column 2 the instruments for the Chilean linkage variables are service

manufacturing and mining FDI linkage measures constructed based on the corresponding FDI stocks of Spain and the

US at t In column 3 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain and the US at t and t-1 In columns 4 7

8 1011 the instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI linkage

measures constructed based on the corresponding FDI stocks of Spain at t while in columns 5 and 6 the instruments are

dated t-1 and t-2 respectively The service intensity trends and the alternative service FDI linkage measures are

described in the text

Drop TFP

Above Below

15Interq

Range by

Industry-Year

Drop

TopBottom

1 TFP by

Industry-

Year

Service FDI

Linkage

Using Input-

Output

Weights

Service FDI

Linkage

Using

Median

Values

Service FDI

Linkage

Using Log

FDI Stocks

One-Stage

Equation (1)

Adding

Regressors of

Equation (3)

(7) (8) (9) (10) (11) (12)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 0318 0351

(0142) (0156)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0668 0024 0196

(0304) (0008) (0057)

R-Squared of Service FDI Linkage First Stage

Regression026 026 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 32038 32889 52630 33390 33390

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 0194 0207

(0107) (0116)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0617 0017 0178

(0296) (0007) (0057)

R-Squared of Service FDI Linkage First Stage

Regression031 031 034 026 064

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Hansen J Statistic (over-identif test of all

instruments)

Observations 31701 32869 52630 33390 33390

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0292 0305

(0104) (0112)

Service FDI Linkage Lagged

Alternative Service FDI Linkage t 0786 0017 0167

(0329) (0008) (0056)

R-Squared of Service FDI Linkage First Stage

Regression032 032 034 026 064

Hansen J Statistic (over-identif test of all

instruments)

Kleinbergen LM Statistic (under-identif test) 000 000 000 000 000

Observations 32112 32854 52630 33390 33390

Panel D Dependent Variable - Firm-Level Output

Service FDI Linkage t 0386

(0139)

R-Squared of Service FDI Linkage First Stage

Regression038

Kleinbergen LM Statistic (under-identif test) 000

Observations 33390

Firm Fixed Effects Yes Yes Yes Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes Yes Yes Yes

IV Estimation

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 42: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

41

Table 6 Services FDI and Innovation

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The instruments for the Chilean FDI linkage variables are service manufacturing and mining FDI

linkage measures constructed based on the corresponding FDI stocks of Spain all at t and t-1 for the results in columns

(1)-(2) and based on the corresponding FDI stocks of Spain and the US at t for the results in column (3)

Differentiated products industries and RampD intensive industries are defined in the text Columns (1)-(2) show the R-

squares of two first stage regressions corresponding to the service FDI linkage variable interacted with each of the

groups

(1) (2) (3)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t Differentiated Products 0538

(0180)

Service FDI Linkage t Non-Differentiated Products 0232

(0118)

Service FDI Linkage t RampD - Intensive 0391

(0156)

Service FDI Linkage t Non RampD-Intensive -0037

(0198)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 009 001

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t Differentiated Products 0534

(0173)

Service FDI Linkage t Non-Differentiated Products 0240

(0114)

Service FDI Linkage t RampD - Intensive 0399

(0156)

Service FDI Linkage t Non RampD-Intensive -0047

(0191)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 008 001

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t Differentiated Products 0518

(0169)

Service FDI Linkage t Non-Differentiated Products 0146

(0113)

Service FDI Linkage t RampD - Intensive 0331

(0153)

Service FDI Linkage t Non RampD-Intensive -0118

(0205)

R-Squared of Service FDI Linkage First Stage Regressions 050 amp 043 046 amp 044

Kleinbergen LM Statistic (under-identif test) 000 000

P-value for F-Test of Difference in Coeff across Groups 003 002

Panel D Dependent Variable - Investment-Capital Ratio

Service FDI Linkage t 0070

(0032)

R-Squared of Service FDI Linkage First Stage Regression 046

Kleinbergen LM Statistic (under-identif test) 000

Firm Fixed Effects Yes Yes Yes

IndustryYear Fixed Effects Yes Yes Yes

RegionYear Fixed Effects Yes Yes Yes

Observations 25752 26925 34426

Number of Firms 4900 5103 6017

IV Estimation

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 43: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

42

Table 7 Firm Heterogeneity in the Effects of Service FDI on TFP

Notes Robust standard errors clustered at the firm level in parentheses and indicate significance at 1 and 5

confidence levels respectively All regressions include a firm export dummy and Chilean manufacturing and mining

linkage variables The latter two variables are endogenous in the IV estimation The instruments for the Chilean FDI

linkage variables are service manufacturing and mining FDI linkage measures constructed based on the corresponding

FDI stocks of Spain at t Closeness to leading performers and to top foreign firms is defined in the text The R-squared

shown are for two first stage regressions corresponding to the service FDI linkage variable by itself and interacted with

the measure of closeness to leading performers (column 1) and to top foreign firms (column 2)

(1) (2)

Panel A Dependent Variable - Firm-Level TFP Estimates Levinsohn and Petrin (2003)

Service FDI Linkage t 1172 1175

(0357) (0358)

Service FDI Linkage t Closeness to Leading Performers -1657

(0651)

Service FDI Linkage t Closeness to Top Foreign Firms -1699

(0671)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel B Dependent Variable - Firm-Level TFP Estimates Olley and Pakes (1996)

Service FDI Linkage t 1056 1059

(0342) (0343)

Service FDI Linkage t Closeness to Leading Performers -1531

(0608)

Service FDI Linkage t Closeness to Top Foreign Firms -1571

(0627)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Panel C Dependent Variable - Firm-Level TFP Estimates Ackerberg et al (2006)

Service FDI Linkage t 0905 0908

(0319) (0320)

Service FDI Linkage t Closeness to Leading Performers -1210

(0588)

Service FDI Linkage t Closeness to Top Foreign Firms -1244

(0608)

R-Squared of Service FDI Linkage First Stage Regressions 022 amp 032 022 amp 032

Kleinbergen LM Statistic (under-identification test) 000 000

Firm Fixed Effects Yes Yes

2-Digit IndustryYear Fixed Effects Yes Yes

RegionYear Fixed Effects Yes Yes

Observations 33017 33017

Number of Firms 5749 5749

IV Estimation

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 44: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

43

Appendix

A Sample Details

From 1992 to 2002 the ENIA survey gives each firm a unique identifier that allows

us to link firms over time to generate a panel dataset In 2003 the firm identifier changed

We established a correspondence between the old and the new firm identifier by merging

two versions of the 2001 dataset (one including the pre-2003 identifier and one including

the post-2003 identifier) according to more than 100 variables We confirm the

correspondence by merging two versions of the 2002 dataset (one including the pre-2003

identifier and one including the post-2003 identifier) Thus we are able to create a panel

of firms from 1992 to 2004 In cases where the correspondence between the old and the

new firm identifier was ambiguous we kept the firm with the old identifier and the firm

with the new identifier in the sample as separate firms The sample includes some firms

with discontinuous data over the sample period For those firms we consider only the

observations across consecutive years for which yearly growth rates can be computed

Since the ENIA survey data is judged to be of high quality and has been widely

used in research only minor data cleaning procedures are applied First we exclude from

the analysis firms with missing identifiers missing output or input variables or missing

industry affiliation Second we impute output and inputs to correct for non-reporting by a

firm in a single year (occurring in fewer than 30 firm-year observations) Third we

eliminate from the sample for production function estimation outliers in output and

inputs firms whose output growth is larger than (smaller than) 400 and those whose

output growth ranges between 100 and 300 (-300 and -100) but is not

accompanied by corresponding high (low) growth rates of inputs After applying these

data cleaning procedures the final sample for 1992-2004 consists of 57025 firm-year

observations However the estimating sample consists of 33390 observations since the

first three years of each firm are excluded to be used for the construction of firm-level

historic service usage intensities

B Production Function Variables

Output is measured by deflated sales The output price deflator is based on information

on indexes of total sales and indexes of physical production for each 3-digit industry from

the Chilean Statistical Institute Based on the equality total sales=physical production

price one obtains growth in total sales=growth in output + growth in prices Using this

formula we compute an industry output price deflator using 2002 as the base year For

years 1992-2002 the price deflator is obtained for 3-digit ISIC Rev 2 industries while for

2003-2004 it is obtained for 3-digit ISIC Rev 3 industries

Skilled and unskilled labor are measured by the number of workers in the following

occupational categories (a) skilled owners managers administrative personnel and

specialized production workers and (b) unskilled workers directly or indirectly involved

in the production process and home workers

Materials is measured by deflated materials expenditures The materials price deflator is

based on a weighted average of the aforementioned 3-digit output price deflators where

the weights are given by the share that each 3-digit industryrsquos output represents in total

manufacturing intermediates used by all 3-digit industries based on an input-output table

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 45: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

44

For years 1992-2002 [2003-2004] the weights are based on the 1986 [1996] Chilean

input-output table

Electricity is the quantity of electricity bought plus the quantity of electricity generated

minus the quantity of electricity sold in thousands of kilowatts

Services is measured by the deflated sum of expenditures on advertising banking

commissions and interest payments communications insurance legal technical and

accounting services licenses and foreign technical assistance rental payments transport

other services and water The services price deflator is based on GDP deflators for 4

groups of services from the Chilean Central Bank (i) electricity and water (ii) transport

and communications (iii) financial services insurance and business services and (iv)

real estate We calculate a weighted average of these GDP deflators where the weights

are given by the share that each of these 4 groups of services represents in total

intermediate expenditures (manufacturing plus services) for each 3-digit industry based

on the 1996 Chilean input-output table

Investment is computed as the sum of deflated net investment flows for each type of

capital in the ENIA survey where net investment flows are the sum of purchases of new

capital purchases of used capital and improvements to capital minus the sales of capital

and are deflated by an investment price deflator constructed as the ratio of current gross

capital formation to constant gross capital formation (in local currency units) from the

World Development Indicators with base year 2002 The ENIA survey provides

information on four types of capital buildings machinery and equipment transport

equipment and land

Capital is computed using the perpetual inventory method (PIM) for each of the four

types of capital in the ENIA survey and summed across the four types of capital For each

type of capital the PIM formula Kit+1 = (1 ndash δ) Kit + Iit is applied where Iit are real net

investment flows and δ is a depreciation rate Since detailed studies of depreciation rates

in Chile are unavailable we use the following rates proposed by Pombo (1999) who

studied the same type of capital goods in Colombia 3 for buildings 7 for machinery

and equipment and 119 for transport equipment Land is assumed not to depreciate

We also experimented with alternative rates of depreciation but did not find this to make

a substantial difference to the final capital stock values nor to our main results The initial

value of the capital stock needed to apply the PIM formula is given by the book value of

each of the four types of capital in the first year of firm presence in the sample Whenever

the book value is available only in a latter year we back out that value until the firmrsquos

first year in the sample taking into account the investment price deflator and the

corresponding depreciation rate

C FDI Linkage Measures

The main service FDI linkage measure is obtained based on the following five steps

1) For each service sector k net FDI inflows NI are given by FDI

kt

FDI

ktkt OINI where I

are sectoral inflows and O are outflows for each year t between 1974 and 2004 obtained

from the Chilean Foreign Investment Committee

2) Using the PIM formula we compute an FDI stock FDIS for each service sector k in

year t as FDI

kt

FDI

kt

FDI

kt SNIS )1( where δ is the depreciation rate assumed to be equal

to 565 which is the average of the depreciation rates for the capital goods machinery

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 46: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

45

buildings vehicles and land used in the construction of the capital stock for Chilean

manufacturing firms The initial value of the FDI stock needed to apply the PIM formula

is given by the net FDI inflows in 1974 for each service sector k Using these inflows as

initial value is reasonable given that FDI inflows into service sectors prior to 1974 were

minor While FDI stocks are calculated for the 1974-2004 period only the values for the

1995-2004 period are used in Steps 3 and 5 The construction of the FDI stock using the

PIM formula follows that used by Coe and Helpman (1995) to construct RampD stocks

3) For each service sector k we calculate a measure of FDI penetration in year t as

kt

FDI

ktkt GDPSFDIpen where GDP is total sectoral output obtained from the Chilean

Central Bank

4) For firm i the intensity of service usage in year t is given by it

k

it

k

it salesspending

ie firm expenditures on services from sector k as a ratio to sales We compute the

average of this intensity of service usage in the first three years of firm presence in the

sample to obtain a historic service usage intensity for each firm i k

i For the few firms

whose service usage intensity from sector k is unusually large (above 2) we replace the

firm intensity by the median service usage intensity in its 3-digit industry

5) We use FDI penetration from Step 3 and historic firm intensity of service usage from

Step 4 to construct the weighted sum which gives the main service FDI linkage measure

as

K

k

kt

k

iit FDIpensFDI1

_ where the K=4 services are (1) electricity and water (2)

transport and communications (3) financial insurance and business services and (4) real

estate

The service FDI linkage measure based on input-output weights used in column 9 of

Table 5 is obtained as follows We calculate the share that each service sector k

represents in total intermediate inputs (mining plus manufacturing plus services) kj used

by 4-digit manufacturing industry j based on the 1996 Chilean input-output table We use

the FDI penetration from Step 3 above and the formula for the weighted sum in Step 5

replacing k

it by kj to construct the service FDI linkage measure

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed as follows First we compute FDI penetration ratios for manufacturing

and mining sectors as described in Steps 1-3 above for service sectors Second for

manufacturing we calculate the share that each 4-digit manufacturing industry m

represents in total intermediate inputs (mining plus manufacturing plus services) mj used

by a 4-digit manufacturing industry j based on the 1996 Chilean input-output table We

interact each manufacturing industryrsquos share with the corresponding manufacturing FDI

penetration to obtain the FDI linkage variable as m

mtmjjt mfFDIpenmfFDI __

The mining FDI linkage measure is obtained analogously

To construct the instruments for the Chilean service manufacturing and mining FDI

linkages we follow the following 4 steps

1) For each service sector k of country c (ie Spain or US) we use data on overall real

service FDI outflows between 1993 and 2004 respectively provided by the OECD

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 47: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

46

2) Using the PIM formula we compute the stock of service FDI outflows FDI

cktO for each

service sector k in year t in country c as FDI

ckt

FDI

ckt

FDI

ckt ONIO )1( where δ is the

depreciation rate set equal to 10 Note that experimenting with alternative depreciation

rates produced no substantial differences The initial value of the FDI outflow stock

needed to apply the PIM formula is given by the FDI outflows in 1993

3) Is the same as Step 4 above

4) We use FDI penetration from Step 2 and historic firm intensity of service usage from

Step 3 to construct the weighted sum which gives the main service FDI instrumental

measure as

K

k

FDI

ckt

k

iit OcsFDI1

_ where the K=4 services are the same as above

The manufacturing and mining FDI linkage measures are based on input-output weights

and are computed analogously

D Production Function Estimation Methodologies

For production function estimation following Levinsohn and Petrin (2003) [below LP]

and Olley and Pakes (1996) [below OP] we point the reader to those two papers for

details In LP estimation we proxy for unobserved productivity using the variable input

electricity In OP estimation we proxy for unobserved productivity using investment and

thus only firms with positive investment are included in the estimating sample

For production function estimation following Ackerberg et al (2006) [below ACF] the

details are as follows We consider for expositional simplicity a three input logarithmic

Cobb-Douglas production function (instead of the six input production function in

equation (1))

itititkitmitlit kmly (D1)

where y is log output l is log labor m is log intermediates k is log capital for firm i at

time t ω are shocks observable to the firm but unobservable to the econometrician andis an iid residual Labor and intermediates are assumed to be perfectly variable inputs

therefore they are chosen by the firm after it observes ω Thus OLS estimates of the input

coefficients are biased To address this endogeneity LP assume that a firmrsquos profit-

maximizing demand for intermediates is a monotonically increasing function of its

productivity shock ω conditional on capital )( itittit kfm hence it can be inverted to

express the unobserved ω as a function of observable firm-level variables as in

)(1

itittit kmf Two other assumptions made by LP and retained by ACF are

(1) productivity ω follows a first-order Markov process (as in OP)

11 itititit pIp where 1itI is the information set available to the firm at t-1 so

one can write itititit E ][ 1 where it is the unexpected part of productivity

which is mean independent of all information known at t-1

(2) capital for use at t is decided at t-1 it is a state variable which can only be affected by

the expected value of productivity it conditional on 1it (but not by it )

LP propose that the function )(1

itittit kmf be inserted into equation (D1) to control

for the unobserved ω resulting in the first stage semi-parametric equation

ititittitlit kmly )( (D2)

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 48: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

47

where )()( 1

itittitlitmititt kmfkmkm

According to LP Equation (D2) is estimated by OLS approximating the unknown

function ()t by a third-degree polynomial on )( itit km This generates an unbiased

estimate for βl as well as for ()t since the error term is assumed to be uncorrelated with

the regressors

ACF identify a crucial identification problem with this first-stage estimate of βl They

argue that a consistent estimate of βl requires that labor varies independently (ie is not

collinear) with the non-parametric function )(1

ititt kmf

Since LP assume that intermediates and labor choices are made simultaneously and the

two inputs are perfectly variable it is natural to expect that they are determined in very

similar ways ie in the same vein of )( itittit kfm a function )( itittit kgl also

exists

Although ()tf and ()tg are likely to be different functions due to differences in input

prices both functions depend on the same two state variables )( itit k It follows that

)())(( 1

itittitititttit kmhkkmfgl This means that there is a serious collinearity

problem in the first stage of the estimation making it impossible to separately identify βl

from the non-parametric function ()t given that both depend on the same variables

To solve this collinearity problem ACF propose the following modification to LP labor

is assumed to be chosen at sub-period t-b (0ltblt1) after capital is known but before

intermediates are chosen at t Thus labor (like capital) can be affected by the expected

value of productivity it conditional on 1it ACF assume that it evolves according to

a first-order Markov process between the sub-periods t-1 t-b and t

bititbitit pIp and 11 itbititbit pIp A rationale for this timing

choice and for the fact that labor inputs are less variable than intermediates may be for

example restrictions in hiring or firing of workers

The fact that in ACF intermediates and labor are chosen with different information sets

generates independent variation But in this case the profit-maximizing firm demand for

intermediates will depend on labor )( ititittit lkfm

Assuming the conditions for invertibility ACF propose that the function

)(1

ititittit lkmf be inserted into equation (D1) to control for the unobserved ω

resulting in the ACF first stage semi-parametric equation

itititittit lkmy )( (D3)

where )()( 1

ititittitlitmitlitititt lkmfkmllkm

Equation (D3) is estimated by OLS approximating ()t by a third-degree polynomial

on )( ititit lkm ACF propose that the first stage is used to obtain unbiased estimates of

()t only not of βl The estimate for ()t is given by the polynomial expression

evaluated at the estimated polynomial coefficients and represents output net of the

untransmitted shock it By conditioning on a firmrsquos choice of intermediates this

procedure allows us to isolate the portion of output that is determined by unanticipated

shocks or measurement error at time t

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 49: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

48

In the second stage of the estimation ACF obtain the input coefficients using three

independent moment conditions Capital used at t is uncorrelated with the unexpected

part of productivity it so the moment condition that identifies the coefficient on capital

is given by 0][ itit kE A similar moment condition for labor does not identify its

coefficient given that labor is chosen at t-b and thus can be correlated with at least part of

it (in addition to being possibly correlated with the conditional expected value of it )

Lagged labor was chosen at t-b-1 and is part of 1itI and therefore is uncorrelated with

the unexpected part of productivity it Hence the moment condition that identifies the

coefficient on labor is 0][ 1 itit lE The moment condition that identifies the

coefficient on intermediates is similar to that in LP and is given by 0][ 1 itit mE

In order to operationalize these moment conditions we proceed as follows First for any

given set of candidate coefficients mll we use the first stage estimate for ()t to

obtain an estimate for kmlit as itkitmitltkmlit kml ()

Second we regress non-parametrically kmlit on kmlit 1 (obtained

similarly as the current value) and a constant The residuals from this non-parametric

regression are an estimate for kmlit Then we construct the sample analog to the

three moment conditions as

t i

it

it

it

kmlit

l

k

m

NT1

1

11

The minimization of this sample analog using an iterative procedure obtains consistent

estimates for the coefficients The standard errors for the coefficients are obtained by

block- bootstrap (ie if a firm is randomly selected to be part of the bootstrap sample

then all years of the firm are included in the bootstrap sample)

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation

Page 50: Foreign Direct Investment in Services and …siteresources.worldbank.org/INTTRADERESEARCH/Resources/...Foreign Direct Investment in Services and Manufacturing Productivity: Evidence

49

Appendix Table 1 Production Function Estimates

Notes Bootstrapped standard errors in parentheses and indicate significance at 1 5 and 10

confidence levels respectively

Food

(ISIC Rev 2

31)

Textiles

Apparel (ISIC

Rev 2 32)

Wood

Furniture

(ISIC Rev 2

33)

Paper

Printing (ISIC

Rev 2 34)

Chemicals

(ISIC Rev 2

35)

Nonmet

Minerals

(ISIC Rev 2

36)

Basic Metals

(ISIC Rev 2

37)

Machinery

(ISIC Rev 2

38)

Other

Manuf (ISIC

Rev 2 39)

Log of Skilled Labour 0077 0135 0085 0144 0142 0105 0061 0122 0142

(0007) (0013) (0011) (0017) (0021) (0024) (0039) (0014) (0052)

Log of Unskilled Labour 0038 0058 0044 0020 0018 0017 -0025 0052 0084

(0005) (0007) (0009) (0008) (0009) (0012) (0020) (0008) (0024)

Log of Materials 0772 0657 0724 0657 0680 0644 0576 0642 0612

(0011) (0012) (0016) (0020) (0018) (0023) (0039) (0013) (0058)

Log of Services 0009 0072 0040 0031 0014 -0006 -0006 0059 0011

(0003) (0009) (0008) (0011) (0007) (0003) (0010) (0009) (0020)

Log of Electricity 0160 0140 0080 0080 0220 0030 0290 0160 0030

(0022) (0016) (0044) (0050) (0095) (0086) (0136) (0016) (0066)

Log of Capital 0010 0010 0010 0090 0010 0010 0050 0010 0120

(0004) (0000) (0023) (0068) (0176) (0032) (0236) (0003) (0120)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Log of Skilled Labour 0077 0157 0100 0146 0140 0097 0113 0127 0197

(0009) (0014) (0014) (0019) (0034) (0034) (0063) (0014) (0074)

Log of Unskilled Labour 0049 0072 0071 0022 0027 0041 0002 0062 0106

(0007) (0010) (0014) (0010) (0013) (0018) (0031) (0009) (0036)

Log of Materials 0770 0642 0707 0629 0690 0574 0552 0634 0656

(0012) (0017) (0023) (0019) (0025) (0026) (0044) (0014) (0075)

Log of Services 0006 0075 0040 0044 0020 -0012 -0002 0056 0044

(0008) (0010) (0014) (0013) (0010) (0016) (0038) (0011) (0043)

Log of Electricity 0031 0051 0053 0058 0015 0169 0145 0074 0020

(0004) (0014) (0010) (0012) (0008) (0005) (0019) (0010) (0055)

Log of Capital 0045 0016 0121 0137 0140 0139 0211 0057 -0083

(0020) (0017) (0029) (0040) (0056) (0066) (0073) (0026) (0097)

Observations 9433 4620 3153 2126 4478 1268 552 5420 336

Log of Skilled Labour 0090 0160 0120 0110 0170 0160 0170 0150 0140

(0008) (0019) (0014) (0034) (0023) (0034) (0040) (0028) (0064)

Log of Unskilled Labour 0170 0090 0040 0080 0020 0080 0020 0100 0100

(0014) (0027) (0013) (0090) (0027) (0074) (0022) (0030) (0054)

Log of Materials 0771 0666 0709 0640 0695 0622 0614 0646 0610

(0012) (0018) (0015) (0016) (0020) (0026) (0042) (0031) (0052)

Log of Services 0010 0074 0040 0031 0015 -0004 -0002 0062 0028

(0003) (0011) (0009) (0009) (0007) (0008) (0014) (0017) (0015)

Log of Electricity 0160 0020 0080 0120 0060 0140 0160 0100 0070

(0009) (0024) (0012) (0058) (0016) (0033) (0035) (0033) (0041)

Log of Capital 0120 0042 0070 0050 0130 0052 0050 0040 0140

(0006) (0018) (0011) (0026) (0014) (0026) (0027) (0018) (0071)

Observations 16917 8676 5448 3503 6589 2293 841 9405 696

Panel C Ackerberg et al (2006) Estimation

Dependent Variable Log of Output

Panel A Levinsohn and Petrin (2003) Estimation

Panel B Olley and Pakes (1996) Estimation