The Impact of Trade Liberalization on Firm Size and Productivity Distributions:

Evidence from India’s Formal and Informal Manufacturing Sectors

(Job Market Paper)

Shanthi Nataraj∗

August 26, 2009

Abstract

Despite a large literature investigating the impacts of trade liberalization on productivity,

no study has tested whether trade liberalization changes the size and productivity distributions

of firms as predicted by recent trade theory. In this paper, I show that two recent models yield

contrasting predictions for the effects of a unilateral trade liberalization on the reallocation

of output among firms, and thus for the firm size and productivity distributions. I test these

predictions for the case of India’s trade reforms, using a unique, firm-level dataset that is repre-

sentative of the entire manufacturing industry, including small, informal firms as well as large,

formal firms. The results suggest that the fall in final goods tariffs increased overall produc-

tivity as the least productive informal firms exited. The firm size distribution was compressed,

consistent with the exit of the smallest firms and a reduction in output among surviving firms.

Although nearly all previous empirical work has focused on the formal sector, my results indi-

cate that small, informal firms were the driving factor behind aggregate productivity gains.

JEL Classification: F12, L11, O17, O24

∗Ph.D. candidate, Department of Agricultural and Resource Economics, University of California, Berkeley, 207 Gi-annini Hall #3310, Berkeley, CA 94720-3310. E-mail: shanthi@are.berkeley.edu. I would like to thank Ann Harrison,Pranab Bardhan, Sofia Villas-Boas, Paul Gertler, Larry Karp, Jill Luoto, Clair Null, and Elisabeth Sadoulet, as well asseminar participants at UC Berkeley for their valuable comments. I am indebted to Mr. M.L. Philip, Mr. P.C.Nirala,Dr. Praveen Shukla, and Mr. M.M.Hasija of the Ministry of Statistics and Programme Implementation for their help inanswering numerous questions about the data sets used in this paper. I am also grateful to Devashish Mitra for provid-ing me with the data on non-tariff barriers; to José Machado for his advice on estimating counterfactual densities; andto Gunjan Sharma and Jagadeesh Sivadasan for their assistance in interpreting the Annual Survey of Industries data.Any remaining errors are mine alone. This material is partly based upon work supported under a National ScienceFoundation Graduate Research Fellowship. Any opinions, findings, conclusions or recommendations expressed in thispublication are those of the author and do not necessarily reflect the views of the National Science Foundation.

1

1 Introduction

A key prediction of recent trade theory is that a fall in final goods tariffs reallocates output amongfirms of different productivity levels, thus changing aggregate productivity as well as the size (out-put) distribution of firms. Despite a large literature examining the impact of trade liberalizationon productivity, though, no study has tested whether trade liberalization changes the firm size andproductivity distributions as predicted by theory.

The effect of a unilateral trade liberalization on the productivity and size distributions of firmsis theoretically ambiguous. A recent paper by Demidova and Rodriguez-Clare (2009, hereafterreferred to as DR), modifies the seminal framework developed by Melitz (2003) to make it moretractable, and suggests that unilateral cuts in final goods tariffs will increase aggregate productivityand compress the firm size distribution. This occurs because the smallest, least productive firms areforced to exit, while surviving firms decrease their domestic production as consumers shift spendingtowards imported goods. In contrast, Melitz and Ottaviano (2008, hereafter referred to as MO)develop a model in which a unilateral fall in final goods tariffs decreases aggregate productivityand stretches out the firm size distribution. In this case, the fall in final goods tariffs decreasesentry in the liberalizing country, thus allowing less productive firms to survive; existing firms canincrease their domestic output because of the reduction in competition.1 Interestingly, both modelspredict that average firm size will increase; however, examining changes to the entire firm sizedistribution can help to distinguish between the two models. In addition, looking at changes inthe distribution of productivity, rather than simply average productivity, can shed light on whetherentry and exit are likely to be mechanisms through which productivity changes occur.

In this paper, I use the DR and MO models to derive predictions for how a unilateral trade liber-alization should change the firm size and productivity distributions, as output is reallocated amongfirms. I then test whether the predictions are supported by empirical evidence from India’s unilateraltrade liberalization, using a unique, firm-level database that is representative of all manufacturingfirms, including the smallest. India provides an excellent case study for several reasons. During the1990’s, it went through a large, unilateral trade liberalization that lowered average final goods tar-iffs from 95% to 35%. Although a number of other reforms occurred at the same time as the tariffreductions, the variation in tariffs across industries and over time allows me to isolate the impactsof tariff changes. I employ a difference-in-differences method that controls for macroeconomicchanges that affected all industries in the same way, as well as for time-invariant, industry-specificfactors. In addition, distinguishing between the impacts of trade liberalization on firms of differentsizes is particularly important in India, where over 80% of manufacturing employment is found in

1As discussed in more detail below, DR also predict that export output among existing exporters will rise, while MOpredict that export output among existing exporters will fall. I assume that the change in domestic output outweighs thechange in export output; however, the basic reallocation mechanisms are the same regardless of which effect dominates.

2

informal firms with fewer than 10 employees. India conducts periodic surveys of small, informalmanufacturing firms, as well as annual surveys of larger, formal manufacturing firms. I have con-structed a unique dataset that combines firm-level data from the formal and informal sectors. Thedataset provides three cross-sectional snapshots of the entire manufacturing industry in 1989, twoyears before the trade reforms began; in 1994, in the midst of the reforms; and in 1999, at the endof the major tariff cuts.

This study makes two main contributions to the literature. First, it estimates the impacts of aunilateral trade liberalization on productivity in the entire Indian manufacturing industry, includ-ing small and informal firms. There is a large literature examining the impact of trade reforms onproductivity (Tybout, de Melo and Corbo (1991) and Pavcnik (2002) for Chile, Levinsohn (1993)for Turkey, Harrison (1994) for Cote d’Ivoire, Tybout and Westbrook (1995) for Mexico, Trefler(2004) for Canada, Topalova (2007) for India, and Amiti and Konings (2007) for Indonesia, amongothers). However, nearly all of these studies exclude the smallest firms from their analyses, al-though theory suggests that the entry and exit of small, relatively unproductive firms plays a crucialrole in how trade affects aggregate productivity. The omission of small firms is even more impor-tant in developing countries, where a large share of manufacturing employment (often upwards of50%) is found in small firms (Tybout 2000). Understanding how small firms react to trade lib-eralization is therefore an important component of understanding post-reform growth. In India,over 80% of manufacturing employment is found in the informal sector, which consists largely forfirms with fewer than 10 employees; yet the analysis of India’s trade reforms has focused on large,formal firms. Recent studies show that the fall in final goods tariffs increase productivity amongamong large, listed firms (Topalova 2007) and among formal firms with five or more employees(Sivadasan 2008). However, there is very little evidence on how India’s trade reforms affect in-formal firms, and much of the existing literature is descriptive (Unni, Lalitha and Rani 2001, forexample).

The second contribution of this paper is to investigate whether changes in the firm size andproductivity distributions are consistent with the entry/exit and reallocation effects suggested byrecent trade models. As discussed above, I focus on two recent models - DR and MO - that explic-itly consider the effects of a unilateral tariff change. As I show in Section 2, a variety of differentmodeling assumptions lead to different predictions for the firm size and productivity distributions.In DR, the firm size distribution is compressed. The left tail of the size distribution shifts right, asthe smallest firms are forced to exit. The right tail of the size distribution shifts left, as survivingfirms reduce their domestic output.2 In MO, the firm size distribution is stretched out. The left tailof the size distribution does not change, while the right tail of the size distribution shifts right, as

2As discussed in more detail in Section 2, I assume that changes in domestic output outweigh changes in exportoutput, which seems to be a reasonable assumption for the Indian case.

3

firms increase their domestic output. Figure 1 presents a stylized plot of how we would expect thefirm size distribution to change when final goods tariffs fall, in each framework.

The two models also differ in their predictions for productivity. DR predict that average produc-tivity will rise. Since the increase in productivity occurs because of the exit of the least productivefirms, the rise in productivity will be most pronounced at the left tail of the productivity distribu-tion, and will taper off towards the right tail. In contrast, MO predict that average productivity willfall. In this case, since the decrease in productivity occurs because less productive firms are addedto the distribution, the fall in productivity will be most pronounced at the left tail of the productivitydistribution, and will taper off towards the right tail.

Despite the emphasis in recent theoretical work on how trade reallocates market share betweenfirms of different sizes, there are no existing studies that examine the impact of a unilateral tradeliberalization on firm size distribution. There is some evidence that trade affects firm size, butthe evidence is limited. In the US, Bernard, Jensen and Schott (2006) find that falling trade costsincrease exports among existing exporters; however, they find no economically significant relation-ship between trade and domestic sales. Their findings contrast to some extent with work by Headand Ries (1999) and Trefler (2004), who find that tariff cuts decrease overall output and employ-ment size, respectively, in Canada. The evidence from developing countries is also inconclusive.Tybout et al. (1991) find that a fall in import protection for Chilean firms increases output size butdecreases employment size. In contrast, Dutz (1996) shows that import competition in Moroccodecreases output size.

This paper is also related to studies that examine the reallocation of output among firms by fol-lowing a panel of firms over time. For example, Pavcnik (2002) follows firms in the years followingChile’s trade liberalization. She finds that that exiting firms are less productive than surviving firms,and that the reallocation of output towards more productive firms accounts for approximately two-thirds of the aggregate productivity gains. In contrast, Tybout and Westbrook (1995) find that theimpact of output reallocation is relatively small after Mexico’s trade liberalization. Like nearly allother studies on trade and productivity, both of these papers exclude the smallest firms.

In this paper, I add to this sparse literature on trade, firm size, and output reallocation by esti-mating the impact of India’s unilateral trade liberalization on the firm size and productivity distri-butions of the entire universe of manufacturing firms. I begin by examining the effect of the tradeliberalization on average firm size and productivity. I find that a 50 percentage point fall in finalgoods tariffs (the average reduction between 1989 and 1999) leads to a 15% fall in firm size amongformal firms, but a 24% increase in firm size among informal firms. Similarly, formal sector pro-ductivity falls by 5.5%, but informal sector productivity rises by 15%. The results for both sectorsare robust to controlling for other industrial policy changes that occurred at the same time and toa variety of ways of measuring productivity. Given the large size of the informal sector relative to

4

the formal sector, the overall result is a strong positive relationship between trade liberalization andboth average firm size and productivity. The rise in average firm size is consistent with both the DRand MO models, while the rise in average productivity supports the DR model.

Within the formal sector, I find that the effects of a fall in final goods tariffs are offset by theconcurrent fall in input tariffs. While the fall in final goods tariffs decreases average formal firmsize and productivity by 15% and 5.5%, respectively, the fall in input tariffs increases average firmsize and productivity by 49% and 29.5%, respectively. Furthermore, I find that the fall in formalfirm size and productivity is attenuated in more export-oriented industries, which is also consistentwith the DR model.

I then explore the effects of trade liberalization on the complete size and productivity distri-butions of firms, which sheds light on how changes at different percentiles of the distributionscontribute to the average increases in both firm size and productivity. I use a simulation techniquebased on conditional quantile regression analysis to compare the distributions of size and produc-tivity that prevailed prior to the reforms, to those that would have prevailed if final goods tariffs hadbeen distributed as they were after the reforms, ceteris paribus. I find that the left tail of the sizedistribution (composed of informal firms) shifts right, while the right tail of the size distribution(composed of formal firms) shifts left. The overall effect is to compress the size distribution offirms. Within the formal sector, the fall in size is less pronounced for the largest firms, which aremost likely to be exporters, and are therefore less affected by domestic competition. The produc-tivity distribution exhibits a large increase at the left tail of the informal sector, which is consistentwith the exit of the least productive firms.

These results highlight the importance of including small firms when investigating the impactsof trade liberalization on firm behavior. The sharp increase in productivity at the left tail of thedistribution, which is consistent with the forced exit of the least productive firms, would not havebeen identified if only formal firms were considered, as in nearly all previous studies. Furthermore,while neither the formal nor the informal sector, considered independently, conforms to the predic-tions of either the MO or DR models, the combination of changes in the productivity and firm sizedistributions in the two sectors supports the reallocation mechanisms predicted by DR.

Although we expect small and large firms to react differently to trade, the stark contrast be-tween the reactions of the formal and informal sectors deserves further examination. I investigatewhether the different responses of the two sectors are driven by regulations that make it difficult forformal firms to exit or downsize. India’s stringent labor regulations have long been criticized for de-creasing formal sector output and increasing informal sector output (Besley and Burgess 2004). Anumber of Indian politicians and commentators have argued that India’s trade reforms have shiftedmore output and employment from the formal sector to the informal sector, as firms seek to avoidregulations and lower costs (Hensman 2001, Jhabvala and Sinha 2002, Vajpayee 2003, among oth-

5

ers). In states where it is relatively difficult for formal firms to close or to adjust their employmentsize, I find that the average decline in formal firm size following the trade reforms is attenuated.However, there are also fewer formal firms in these states, suggesting that when formal firms findit difficult to adjust their size in response to a decrease in consumer demand, there will be less netentry in the long run. State regulations on firm size adjustment and closure have no statisticallysignificant impacts on informal firm size or output, as we would expect. The differential effects ofstate labor regulations on formal and informal firm size and net entry indicate that the stark differ-ences between the formal and informal sector reactions to trade may be caused, to some extent, bydiffering regulations that govern the two sectors.

Finally, I extend my analysis to examine whether the trade reforms affected India’s employ-ment size distribution. Like many other developing countries, India’s employment is concentratedamong tiny firms and very large firms, leaving a “missing middle” in the employment size distri-bution (see, for example, Mazumdar (1998)). Although there is concern that India’s trade reformsshifted more employment into small, informal firms, there is almost no quantitative evidence link-ing trade and the employment size distribution. Using the simulation technique discussed above,I find that the fall in final goods tariffs reduces average employment size in the formal sector; thisdecrease occurs because some employment is shifted away from mid-sized formal firms, with 10-50 employees, to small formal firms, with fewer than 10 employees. However, this fall in formalsector employment size is driven by capital-intensive industries. In labor-intensive industries (inwhich India would presumably have a comparative advantage), formal sector employment size ac-tually increases following the trade reforms. Furthermore, there is no evidence that the fall in finalgoods tariffs affects informal employment size. The combination of these results suggests that In-dia’s trade liberalization did not significantly increase its “missing middle” in the employment sizedistribution.

The rest of this paper is organized as follows. Section 2 discusses the relevant predictions ofthe two theoretical models in more detail. Section 3 presents a brief overview of the Indian tradereforms. Section 4 provides an overview of the combined formal and informal firm data. Section 5describes the empirical strategy and presents results. Section 6 extends the investigation to considerthe size distribution of employment, and Section 7 concludes.

2 Theoretical Motivation

In this section, I outline and compare the DR and MO models, and highlight the key implicationsof each model for the impact of a unilateral trade liberalization on productivity and firm size. Inboth of these models, firms pay a fixed entry cost in order to receive a productivity draw froma known distribution. The most productive firms become exporters (and also sell domestically);

6

less productive firms only sell domestically; and the least productive firms exit. All firms pro-duce differentiated varieties of the same good, so the models highlight intra-industry (rather thancomparative-advantage) trade. Firm-level productivity does not change, but aggregate productivitycan change due to entry and exit, as well as the reallocation of output between firms of differentproductivity levels.

DR base their model on the seminal work by Melitz (2003). They make Melitz’s (2003) frame-work more tractable for examining asymmetric countries and unilateral tariff changes by simplify-ing the model in several ways, most notably by assuming that the Home country is small, so thatHome firms that export to Foreign take the Foreign demand schedule as given. Consumers exhibitCES preferences. As in Melitz (2003), the wage is endogenously determined and is affected by afall in final goods tariffs, thereby changing firm-level output. DR show that when final goods tariffsat Home fall, consumers substitute towards imports and away from domestic varieties. All firmsreduce their domestic output, and the smallest (least productive) firms are forced to shut down.However, export output among existing exporters (which are the largest firms) actually increases.

In MO, the effect of trade on heterogeneous firms operates through a different channel. Thismodel assumes that the wage is set by production of an outside good in a competitive market withconstant returns to scale, so trade does not affect firm output through the factor market channel.Rather, consumer demand is assumed to be linear, so firm prices and output depend on the pro-ductivity level of the marginal firm (the firm that is indifferent between exiting and producing).When final goods tariffs at Home fall, MO predict that fewer firms choose to enter at Home, whilemore firms choose to enter in Foreign.3 Competition decreases at Home, so less productive firmsto survive; thus, the productivity of the marginal firm decreases, and existing firms can increasetheir domestic output. In contrast, since there is more competition in Foreign, the marginal firmin Foreign is more productive, and Home exporters are forced to decrease export output or stopexporting.

I use each of these frameworks to derive predictions for the size and productivity distributionsof firms when there is a unilateral reduction in final goods tariffs. I assume that domestic productionis sufficiently large, relative to export production, that the effects of final goods tariffs on domesticoutput are dominant. This assumption seems reasonable in the Indian context. Unfortunately, firm-level surveys during this time period did not ask for export status, or for the share of exports in totalproduction. However, in 1989, India’s share of exports in manufacturing output was only 10.1%,rising to 14.6% by the end of the reforms in 1999. In comparison, the export share in manufacturingoutput was 20.3% in China, 34.4% in Korea, 45.6% in Taiwan, and 39.2% in Thailand during thesame time period.4 In the US, which had a similar export share in manufacturing as India (15.9%

3MO note that this “de-location” effect - the movement of firms away from the liberalizing country - was present inthe older literature on trade with imperfect competition as well.

4Data on export shares in manufacturing are from the World Bank’s Trade and Production database.

7

in 1999), Bernard, Jensen, Redding and Schott (2007a) report that even among firms that export,the average share of exports in total shipments was only 14%.

Section 2.1 lays out and compares the basic setup of each model. Section 2.2 shows how the twomodels lead to differing predictions for productivity and firm size, and summarizes the predictionsthat I will test.

2.1 Free Entry Equilibrium with Costly Trade

Productivity Distributions

In each model, firms must pay a fixed entry cost fe. Upon entry, each firm receives a productivitydraw from a known distribution. Based on that productivity draw, the firm decides whether to exit,to produce and sell only to the domestic market (“domestic” firms), or to produce and sell to boththe domestic and export markets (“exporters”). Firms behave as monopolistic competitors, andentry or exit takes place until the expected profit from entering is equal to the fixed entry cost.There are two countries - Home and Foreign. I will focus on the case in which Home lowers itsfinal goods tariffs unilaterally.

To facilitate comparison between the two models, I have modified the notation of each slightly.In MO, productivity draws are given by 1/c where c is the firm’s marginal cost of production.All of MO’s key results are presented in terms of cost draws, rather than productivity draws. Foreasier comparison with DR, I have rewritten all relevant expressions in terms of productivity drawsφ ≡ 1/c. Furthermore, similar to DR, let x define the productivity level of the firm that is indifferentbetween exiting and producing only for the domestic market (the marginal firm), and let y define theproductivity level of the firm that is indifferent between producing only for the domestic market,and producing for both the domestic and export markets. In addition, to simplify the analysis, Iremove the consumption and export subsidies that play a role in DR’s model.

DR assume that productivity draws φ follow a Pareto distribution with lower bound b and shapeparameter β:

DR : G(φ) = 1− [b/φ]β, φ > b, β > σ (1)

where σ is the elasticity of substitution between varieties.Similarly, MO assume that productivity draws 1/c ≡ φ are distributed Pareto. Again, to make

the models more comparable, I use DR’s notation for the lower bound of productivity (b, equal toφM in MO) and the shape parameter (β, equal to k in MO). In terms of the revised notation, thisimplies that:

MO : G(φ) = 1− [b/φ]β, φ ∈ [b,∞), β ≥ 1 (2)

8

The assumptions on productivity draws are similar, though DR require that the shape parameter βbe larger than the elasticity of substitution σ (which itself is greater than 1) while MO only requireβ > 1.

Demand

In DR, Home consumers have CES preferences over domestic and foreign varieties indexed by νand ν ′, respectively:

DR : U =[ ∫

ν∈Ω

q(ν)ρdν +

∫ν′∈Ωm

qm(ν ′)ρdν ′]1/ρ

, 0 < ρ < 1 (3)

The sets of available domestic and imported varieties are given by Ω and Ωm, respectively; q(ν)

and qm(ν ′) are the quantities of domestic and imported varieties consumed; and the elasticity ofsubstitution between varieties is σ ≡ 1/[1− ρ]. This utility function implies the following demandfor each variety:

DR : q(ν) = RP σ−1[p(ν)]−σ, qm(ν ′) = RP σ−1[τHpm(ν ′)]−σ (4)

where R is aggregate expenditure at Home, P is the aggregate price level at Home, p and pm arethe prices of domestic and imported varieties, respectively, and τH is the Home tariff on imports.

In contrast, MO employ a linear demand structure:

MO : U = qc0 + α

∫i∈Ω

qcidi−1

2γ

∫i∈Ω

[qci ]2di− 1

2η[ ∫

qcidi]2 (5)

The set of available varieties is given by Ω; qci is the quantity of each variety consumed; γ representsthe elasticity of substitution between the differentiated varieties, and is similar to σ in DR; and α andη govern the substitution between the differentiated products and an outside good. This producesthe following demand for each variety:

MO : qci =α− pi − ηQc

γ(6)

where QC is total consumption of the differentiated varieties.The utility and demand functions highlight two key differences between DR and MO. The

first is the structure of demand. In DR, the CES preference structure means that firm prices area constant markup over marginal cost. In MO, the linear demand structure means that markupsdepend on the productivity of the marginal firm; a more competitive environment, indicated bya more productive marginal firm, leads to lower markups. The second key difference is the useof an outside, numeraire good in MO. In DR, consumers only substitute between differentiated

9

varieties of the same good. In MO, consumers substitute between differentiated varieties as wellas an outside good, which is produced in a competitive market under constant returns to scale withunit cost. This sets wages, so that changes in Home tariffs do not affect the differentiated sectorthrough the factor market. Rather, firm size changes because of the linear demand structure, whichallows each firm’s prices and quantities to depend on the marginal firm.

Production

In DR, firms produce their output with increasing returns to scale using only labor l, according tol = f + q/φ. Home firms behave as monopolistic competitors and choose optimal domestic outputqDH :

DR : qDH = RP σ−1[ wρφ

]−σ (7)

where w is the wage. As discussed above, the DR model is a simplified version of Melitz (2003).One of the key simplifications that the authors introduce is to make Home a “small” country in thesense that Home firms that export to Foreign take the Foreign demand schedule as given. Firms haveto pay a fixed cost wfexp in order to export. They optimize separately over exports and domesticproduction, so that optimal exports qXH are:

DR : qXH = A[ wρφ

]−σ (8)

where A is taken as given by Home firms.In MO, firms produce their output with constant returns to scale using only labor, according

to l = q/φ. Again, Home firms behave as monopolistic competitors and choose optimal domesticoutput qDH :

MO : qDH =LHγ

[pD(1

φ)− 1

φ] (9)

Foreign demand for Home exports is given by Foreign consumer preferences that are assumed tobe identical to Home preferences, though country size (LF in Foreign and LH at Home) is allowedto differ. In addition, there is a per-unit trade cost τF for Home exporters that sell to Foreignconsumers. Optimal exports for Home firms are then given by:

MO : qXH =LFγ

[pX(1

φ)− τF

φ] (10)

Equilibrium

In each model, firms enter until the expected profit from operating equals the entry cost. DRfollow Melitz (2003) and define φ(x) = [

∫∞xφσ−1µ(φ)dφ]1/σ−1 with µ(φ) = g(φ)/[1 − G(x)] =

10

βxβ/φβ+1. They show that the expected profit from entering is:

DR : π = πD(φ(x)) +mexpπexp(φ(y)) = wf [θ − 1] + wmexpfexp[θ − 1] (11)

where mexp is the mass of exporters and θ ≡ β/[β − σ + 1]. Therefore, the free entry conditiondefines the cutoff productivity level for survival at Home, xH , and can be written as:

DR : π[1−G(x)] = wfe =⇒ [θ − 1]x−βH [f +mexpfexp] =febβ

(12)

Similarly, only firms with sufficiently high productivity can make enough profits from exportingto cover the additional fixed cost of exporting, so we can write down a condition that defines theproductivity cutoff for the marginal exporter with productivity yH :

DR : Aw1−σ[ρyH ]σ−1 = σwfexp (13)

Equilibrium is determined by the combination of these two conditions, as well as three additionalconditions governing the trade balance between Home and Foreign, the cutoff productivity for For-eign firms exporting to Home, and the equivalence between the total amount of revenues obtainedby Home firms and the total wage bill. Interested readers are referred to the original paper by DRfor more details on the latter three conditions.

In MO, the cutoff productivity for domestic production at Home, xH , is similarly determinedby setting expected profits from entering equal to the fixed entry cost:

MO :

∫ 1/xl

0

πDl (1/φ)dG(1/φ) +

∫ 1/yl

0

πFl (1/φ)dG(1/φ) = fe

This expression holds for both countries l = H,F so that the free entry condition generates twoequilibrium expressions, which can be solved for the domestic productivity cutoff at Home xH :

MO :[ 1

xH

]k+2=[2γ[β + 1][β + 2][1/b]βfe

LH

1− τ−βF1− τ−βH τ−βF

](14)

where τH and τF are Home and Foreign final goods tariffs, respectively. The cutoff productivity forexporting is equal to the cutoff productivity for domestic production by Foreign firms, divided bythe per-unit trade costs. Intuitively, exporters must be slightly more productive than domestic firmsin order to cover trade costs. Let xF be the Foreign productivity cutoff for domestic production;then, the Home exporting cutoff is

MO : yH = τFxF (15)

11

Therefore, the Home exporting cutoff yH is defined by:

MO :[ τFyH

]k+2=[γ[β + 1][β + 2][1/b]βfe

LF

1− τ−βH1− τ−βH τ−βF

](16)

These cutoff productivities can be used to rewrite each firm’s domestic and export output in thetwo models:

DR : qDH =f [σ − 1]

xHσ−1φσ (17)

DR : qXH =fexp[σ − 1]

yHσ−1φσ (18)

MO : qDH =LH2γ

[1

xH− 1

φ] (19)

MO : qXH =LF2γτF [

1

yH− 1

φ] (20)

These expressions imply a positive and monotonic relationship between output and productivity inboth models.

I now turn to average productivity and firm size. Each model defines average productivity ina slightly different way. To facilitate comparison, I follow MO’s method of calculating averageproductivity. In DR, I define average productivity as φ = [

∫∞xHφdG(φ)]/[1 − G(xH)]. In keeping

more closely with the use of cost rather than productivity in MO, I define average cost as 1/φ =

[∫ 1/xH

0[1/φ]dG(1/φ)]/[G(1/xH)]. When average cost falls, average productivity rises, and vice

versa. These definitions can be evaluated to yield expressions for average productivity in DR, andfor average cost in MO:

DR : φH =

∫∞xHφdG(φ)

1−G(xH)=

βxHβ − 1

(21)

MO : 1/φH =

∫ 1/xH

0[1/φ]dG(1/φ)

G(1/xH)=

β

[β + 1]

1

xH(22)

Note that in both cases, average productivity rises (or average cost falls) when the domestic produc-tivity cutoff xH rises. Similarly, we can derive expressions for average domestic and export outputas functions of the domestic and exporting productivity cutoffs:

DR : qDH =

∫∞xHqDH(φ)dG(φ)

1−G(xH)=

xHβ

β − σf [σ − 1] (23)

12

MO : qDH =

∫ 1/xH

0qDH(1/φ)dG(1/φ)

G(1/xH)=LH

2γ

1

[β + 1]

1

xH(24)

DR : qXH =

∫∞yHqXH (φ)dG(φ)

1−G(yH)=

yHβ

β − σfexp[σ − 1] (25)

MO : qXH =

∫ 1/yH

0qXH (1/φ)dG(1/φ)

G(1/yH)=LF

2γ

1

[β + 1]τF

1

yH(26)

In DR, average domestic output increases with a rise in the domestic productivity cutoff xH ; in MO,average domestic output decreases with a rise in xH . Equations 17 and 19 imply that in both models,an individual firm’s domestic output falls when xH rises. However, the least productive (smallest)firms can no longer survive when xH rises. The latter effect outweighs the former in DR, but not inMO, resulting in the differing predictions. Similarly, in DR, average export output increases with arise in the exporting productivity cutoff yH ; in MO, average export output decreases with a rise inyH .

Finally, we can write down expressions for the share of exports in total output, for firms thatexport:

DR : S ≡ qXHqDH + qXH

=fexp[σ − 1]φσ/yσ−1

H

fexp[σ − 1]φσ/yσ−1H + f [σ − 1]φσ/xσ−1

H

(27)

MO : S ≡ qXHqDH + qXH

=[LF/2γ]τF [1/yH − 1/φ]

[LF/2γ]τF [1/yH − 1/φ] + [LH/2γ][1/xH − 1/φ](28)

It is easy to show that in both models, the share of exports in output is strictly increasing in pro-ductivity, dS/dφ > 0. This means that larger exporters are relatively less exposed to the domesticmarket, which will have a bearing on the reallocations that follow a unilateral trade liberalization.

2.2 Effects of a Unilateral Fall in Final Goods Tariffs

Both DR and MO explicitly consider the effects of a fall in final goods tariffs at Home on thedomestic and exporting productivity cutoffs at Home. In DR, the fall in Home final goods tariffscauses the domestic productivity cutoff xH to rise and the exporting productivity cutoff yH tofall. This occurs because consumers to shift spending away from domestic varieties and towardsimports. This reduction in spending on domestic varieties forces the least productive domestic firmsto exit, raising xH . At the same time, yH falls, which allows more Home firms to begin exporting.5

5The proof that a fall in final goods tariffs increases xH and decreases yH is not presented here. Interested readersare referred to the original paper for more details.

13

In contrast, MO show that the fall in final goods tariffs causes xH to fall and yH to rise6. Thisoccurs because fewer firms choose to enter at Home when there is less protection. Less entry yieldsless competition at Home, which allows less productive firms to survive. Conversely, MO showthat there is more entry in Foreign, so yH rises.

I use these predicted changes in the domestic and exporting productivity cutoffs to derivetestable predictions for average firm size and productivity, as well as for the firm size and pro-ductivity distributions. Interestingly, both models predict the same change in average firm size;however, examining changes to the left and right tails of the size distribution, as discussed below,can help to distinguish between the two models. In addition, looking at changes in the distributionof productivity, rather than simply average productivity, can shed light on whether entry/exit arelikely to be mechanisms through which productivity changes occur.

Table 1 summarizes the predictions that I will test:Average productivity. DR predict that xH rises, so from Equation 21, average productivity falls.

MO predict that xH falls, so from Equation 22, average productivity falls (average cost rises).Average firm size. In DR, xH rises and yH falls; therefore, from Equations 23 and 25, average

domestic output increases, while average export output decreases. As discussed previously, I as-sume that the effects of domestic output dominate in the Indian case; therefore, average firm sizeincreases.7 In MO, xH falls and yH rises; therefore, from Equations 24 and 26, average domesticoutput increases, while average export output decreases, so average firm size increases.

Left tail of productivity distribution. In DR, since the least productive firms exit, the left tail ofthe productivity distribution (at the lowest percentiles) should increase, with the increase taperingoff among higher percentiles of the distribution. In MO, since less productive firms can survive,the left tail of the productivity distribution (at the lowest percentiles) should decrease, with thedecrease tapering off among higher percentiles of the distribution.

Left tail of the firm size distribution. There is a positive and monotonic relationship betweensize and productivity in both models. The size of the smallest firm in DR therefore has productivitylevel xH . From Equation 17, then, we can show that the size of the smallest firm is given by:

DR : qDH(φ = xH) = f [σ − 1]xH (29)

A fall in final goods tariffs raises xH , which increases the size of the smallest firm and moves theleft tail of the size distribution to the right. In MO, the smallest firm, which has productivity xH ,always has zero output (see Equation 19). Therefore, the left tail of the size distribution does notmove.

6This can be seen from Equations 14 and 16.7A sufficient condition for this assumption to hold, in the DR framework, is that the fixed cost for Home firms to

export to Foreign, fexp, is small compared to the fixed cost of production, f . In the MO framework, I assume that thesize of the Foreign market, LF , is sufficiently small compared to the Home market LH .

14

Right tail of the firm size distribution. In DR, Equations 17 and 18 show that when xH rises andyH falls, the domestic output of existing firms falls, while the export output of existing exportersrises. Assuming that changes in domestic production dominate, the right tail of the size distributionmoves left. In MO, Equations 19 and 20 show that that when xH falls and yH rises, the domesticoutput of exiting firms rises, while the export output of existing exporters falls. Again assumingthat changes in domestic production dominate, the right tail of the size distribution moves right.

Overall size distribution. In DR, the overall size distribution is compressed as the left tail shiftsrightward, while the right tail shifts leftward. In MO, the overall size distribution is stretched out,as the left tail does not move, while the right tail shifts rightward.

Change in size of largest firms relative to mid-sized firms. In both models, Equations 27 and28 imply that among the largest firms, which are exporters, there is a positive correlation betweenfirm size and export share in production. Therefore, the relative impact of export output to domesticoutput will be larger for these firms. In DR, this means that the fall in output for existing firms willbe attenuated for the largest firms, relative to mid-sized firms. Similarly, in MO, the increase inoutput for existing firms will be attenuated for the largest firms, relative to mid-sized firms.

3 Indian Trade Reforms

I use India’s trade liberalization to test the impacts of a fall in final goods tariffs on the size andproductivity distributions of firms. Prior to 1991, import substitution was the cornerstone of India’strade regime. Just before the 1991 reforms, the average final goods tariff was approximately 95%.Tariffs constituted a large share of tax revenues, and Aksoy (1992) notes that “rather than adjust-ing the exchange rate...during every foreign exchange crisis, the average tariff collection rate wasratcheted up to a higher plateau.” India also had restrictive non-tariff barriers, which required firmsto apply for licenses in order to import most items, and banned many imports altogether.

Throughout the 1980’s, India’s budget deficit continued to grow, as did its balance of paymentsdeficit. In 1991, a combination of economic and political shocks - namely, a rise in oil prices,a decrease in remittances from Indians living abroad, and an unstable political climate - forcedadditional change by creating a balance of payments crisis. The IMF granted India a Stand-ByAgreement on the condition that it undertake several reforms (Topalova 2007). In July 1991, India’sgovernment announced a series of major reforms, including further delicensing, FDI liberalization,exchange rate liberalization, export liberalization, the removal of import licensing requirementsfor capital and intermediate goods, and a reduction and harmonization of tariffs across industries(Hasan, Mitra and Ramaswamy 2007).

Between 1989 and 1999, the average final goods tariff rate fell from 95% to 35%. I calculate fi-nal goods tariff rates for each industry based on the Government of India’s Customs Tariff Working

15

Schedules, in which rates are given for approximately 5,000 product lines. Using the concordanceof Debroy and Santhanam (1993), I match the product lines with 3-digit National Industrial Classi-fication (NIC-87) codes, and calculate average final goods tariff rates within each of approximately170 codes. Table 2 shows average final goods tariffs for broad manufacturing industry groups in1989, 1994, and 1999. Prior to the reforms, the average final goods tariff rate was approximately95% ad valorem and varied significantly across industries. Panel (a) of Figure 2, which shows var-ious percentiles of the final goods tariff distribution in each year, illustrates that final goods tariffswere both lowered and harmonized during the 1990’s.

India’s trade liberalization provides an excellent case study because of the manner in whichthe reforms took place. Tariffs were lowered and harmonized across all industries; therefore, theindustries with the highest pre-reform tariffs faced the highest tariff cuts (Panel (b) of Figure 2).This pattern provides rich variation in tariffs across industries and over time, allowing me to controlfor macroeconomic shocks that affected all industries in the same way, as well as for time-invariant,industry-specific characteristics. It also lowers the chance that industries were selected into tariffreforms based on political factors. It is still possible, though, that the pre-reform tariff levels arecorrelated with industry characteristics. I explore this possibility by looking at the correlationsbetween changes in final goods tariffs (1989-1999) and pre-reform industry characteristics in 1989.I consider the following firm-level characteristics that might influence political decisions abouttariff protection: employment size, output size, capital size, the capital-employee ratio, wages,and productivity. or each of these variables, I calculate the average values for each industry priorto the reforms, and regress the changes in tariffs between 1989 and 1999 on average pre-reformcharacteristics in each industry. I also look for correlations between tariff changes and pre-reformindustry size (measured in terms of total employment, total capital, total output, and the number offirms). For the formal sector, I consider the possibility that more concentrated industries (for whichI use the industry’s four-firm concentration ratio as a proxy) might have more political leverage.

I check for correlations between tariff changes and pre-reform firm and industry characteristics,in levels, in both sectors. I also check for correlations between tariff changes and pre-reformchanges in industry characteristics in the formal sector.8 Unfortunately, since I only have one yearof pre-reform data for the informal sector, I cannot check whether pre-reform trends in informalsector characteristics are associated with tariff changes. However, it is likely that any politicalinfluence would come from formal rather than informal firms.

Panels (a) and (b) of Table 3 present results for firm-level and industry-level characteristics,respectively. Consistent with earlier work by Topalova (2007), I find no significant correlationsbetween tariff changes and pre-reform characteristics in either the formal or informal sectors.9

8I use changes from 1987-1989, though the results are robust to using changes from 1988-1989 or 1986-1989instead).

9Topalova (2007) does find that tariff cuts after 1997 may have been related to productivity levels among large

16

Recent work by Amiti and Konings (2007) suggests that input tariffs may have a larger impacton firm productivity than final goods tariffs. To address this possibility, I calculate input tariffsusing India’s input-output transactions table (IOTT), following the method suggested by Amiti andKonings (2007). For example, if the footwear industry derives 80% of its inputs from the leatherindustry and 20% from the textile industry, then the input tariff for the footwear industry is 0.8times the final goods tariff for the leather industry plus 0.2 times the final goods tariff for thetextile industry. This measure of input tariffs is not perfect; the IOTT provides data at a relativelyaggregated level (it has 66 manufacturing industries, compared to 171 manufacturing industries forfinal goods tariffs). In addition, I cannot identify which individual firms import raw materials, andare therefore most likely to be affected by the fall in input tariffs.

One potential concern is that final goods and input tariffs may be highly correlated, thus leadingto multicollinearity problems in estimation. Panel (c) of Figure 2 shows the relationship betweenthe change in final goods tariffs and the change in input tariffs for a given industry. Though thetwo measures are related, there are a number of industries that received relatively large reductionsin final goods tariffs but relatively small reductions in input tariffs, and vice versa.

4 Data

A key contribution of this work is the creation of a unique dataset that links both formal andinformal Indian firms, thus providing three snapshots of the entire manufacturing industry over theperiod of the trade reforms. All manufacturing firms with more than 10 workers that use electricalpower, or 20 workers that do not use electrical power, are required to register under the FactoriesAct; these firms are considered organized, or formal. Firms below these employment thresholds arenot required to register under the Factories Act, and unregistered firms are considered unorganized,or informal. India’s formal manufacturing sector has approximately 100,000 firms and employs 8million people, while the informal sector includes 15 million firms that employ nearly 30 millionpeople. I combine informal sector data from 1989-90, 1994-95, and 2000-01 with formal sectordata from 1989-90, 1994-95, and 1999-2000.10

The formal sector is surveyed by the Central Statistical Organisation (CSO) every year throughthe Annual Survey of Industries (ASI). The sampling universe for the ASI is all firms that areregistered under Sections 2m(i) and 2m(ii) of the Factories Act, as well as firms registered under the

Indian firms; I address this issue in more detail in Section 5.4.10In 1989, informal firms with six or more employees were excluded from the survey I use; however, in the 1994 and

1999 surveys these firms represent fewer than 5% of the population of informal firms. The effect of this omission thatdoes not vary across industries will be addressed by the inclusion of year dummies in the empirical work. I also findno correlations between observed firm size in the informal sector in 1989 and subsequent final goods tariff changes, asshown in Table 3.

17

Bidi & Cigar Workers Act, and a number of utility providers.11 Large firms (those with 100 or moreemployees from 1987-1996, and with 200 or more employees or a certain output value between1997 and 2002) are surveyed every year, while approximately one-third of the smaller firms aresurveyed. In states designated as “industrially-backwards,” all firms are surveyed, regardless ofsize. The ASI provides multipliers for each firm, indicating the inverse sampling probability. SinceI am interested in estimating the effects of trade on the population of firms, I use the multipliers tore-weight firm-level observations in all of my analyses.

The informal sector is surveyed by the National Sample Survey Organisation (NSSO), with asampling universe of all manufacturing firms that are not registered under the Factories Act.12 TheNSSO surveyed informal enterprises in 1989-90 as a follow-up round to the all-India EconomicCensus of 1980; in 1994-95 as a follow-up round to the 1990 Economic Census; and in 2000-01 asa follow-up to the 1998 Economic Census. Given the long period of time between the EconomicCensus and the follow-up surveys, it is possible that there is a large amount of entry and exitbetween the two, particularly among small firms. However, this concern is mitigated by the NSSO’ssampling procedure. Information from the Economic and Population Censuses is used to selectfirst-stage sampling units (FSUs); however, once an NSSO investigator arrives at an FSU, s/hecreates an updated list of households and enterprises to be used in second- and third-stage sampling.Each sample consists of approximately 1% of informal firms. Large firms are oversampled toensure that enough of them are included; as with the ASI data, I use sampling weights provided bythe NSSO to re-weight firm-level observations.

Although firms with fewer than 10 employees are not required to register under the FactoriesAct, and therefore should not appear in the ASI data, between 15% and 20% of the ASI firmsin each year have fewer than 10 employees. Bedi and Banerjee (2007) argue that the process of“deregistration” is difficult for firms that have cut employment, and that many closed firms stillappear in the list of registered (formal) firms. For my analyses, I exclude firms that are reportedto be closed, but I retain firms in the ASI dataset that have fewer than 10 employees. As Bediand Banerjee (2007) point out, some firms may appear too small to be formal because they havetemporarily reduced employment. It is also possible that small firms may choose to register in theexpectation that they may grow in the future, or as a signal to creditors. Similarly, a few firms in theNSSO survey (less than 1% in each year) report having more than 20 employees, while another 2-3% report having more than 10 employees while spending money on electricity (which suggests thatthey should be registered, or formal). The number of firms in the NSSO survey that should likely beregistered is small when compared to the total number of informal firms. However, assuming thatthese firms are representative of the population of informal firms, this finding suggests that over

11I exclude all non-manufacturing firms from my analyses.12The NSSO surveys include non-manufacturing firms in some years; these firms are excluded from my analyses.

18

100,000 additional firms should be registered (a near doubling of the number of registered firms).For the analyses below, I classify firms as formal when they appear in the ASI and as informal whenthey appear in the NSSO’s survey, but the overlap between the two sectors provides an additionalreason to consider the entire universe of firms.

A number of firms in both sectors (though predominantly in the informal sector) report produc-ing no physical products; the source of their revenue is uncertain, but likely comes from services.I restrict my analysis to firms that use raw materials to produce physical products; this restrictionseems reasonable given that final goods and input tariffs are likely to affect these firms, rather thanthose that only provide services.13

Table 4 provides summary statistics for output, employment, and capital in the sample. Outputis the total value of manufactured products, deflated using industry-level price deflators.14 Laboris measured as the number of employees. Capital stock values are deflated using the perpetualinventory method of Harrison (1994), as modified by Sivadasan (2008).15

To combine the three years of data, I first make the industrial codes comparable over time. The1989-90 and 1994-95 firm-level data classify firms using India’s NIC-87 industrial code; however,the 1999-2001 data use a different (NIC-98) industrial code. At the level at which aggregate data arepublished (3-digit NIC-87 code, or 4-digit NIC-98 code), there are many instances in which mul-tiple NIC-98 codes map to multiple NIC-87 codes, thus confounding matching over these years.However, the firm-level data provide more detailed (5-digit) NIC-98 codes, which enables me tomap each firm in the 1999-2001 data to a unique 3-digit NIC-87 code. I then combine the firm-leveldata with tariff data (discussed in Section 3) at the 3-digit NIC-87 industry level using the concor-dance table provided by Debroy and Santhanam (1993). At this level, there are approximately 140unique industries in my dataset.

Figure 3, Panels (a)-(d), illustrates some features of formal and informal firms by showingkernel density plots of employment, capital, output, and total factor productivity (TFP) in 1989.16

The densities are weighted using the sampling multipliers, so the distributions are representativeof the population of firms. The modal value of employment is two in the informal sector andslightly over 10 in the formal sector, which is consistent with the requirement that firms registerwhen they have 10 employees. There does not appear to be any lumpiness around the 10- or 20-

13I have performed the baseline analyses presented below using all firms, and the results are similar in sign, signifi-cance, and magnitude; results are not shown here, but are available upon request.

14As a robustness check, in Section 5.4 I use the method suggested by Hsieh and Klenow (forthcoming) to accountfor differential firm pricing by assuming particular values for the elasticity of substitution between goods.

15I start with the initial value of capital stock in 1989, and assume a 10% depreciation rate. The value of real capitalstock in industry j and year t is Kjt = 0.9Kjt−1 + Ijt where Kjt−1 is real capital stock in the previous year and Ijt

is nominal investment in year t. Data on nominal investment and capital stocks are calculated based on industry-levelformal sector data only, since informal sector data are not available every year. Each firm’s nominal capital stock isdeflated by Kjt/Kj,1989.

16TFP measurement is discussed in Appendix A.

19

employee mark in the informal sector (even when the data are plotted using smaller bandwidthsor histograms), suggesting that the 10- or 20-employee constraint is not binding for most informalfirms. There is little overlap between the capital and output distributions of the formal and informalsectors. Consistent with previous studies (LaPorta and Shleifer 2008, among others), I find thatinformal firms are generally less productive than formal firms, but there is considerable overlapbetween the two sectors’ productivity distributions.

5 Empirical Strategy and Results

5.1 Baseline Results: Average Firm Size and Productivity

Before turning to the distributional analysis, I estimate the average effects of the trade reforms onfirm size and productivity. As shown in Table 1, both models predict that average firm size willincrease. However, DR predict that average productivity will increase, while MO predict that itwill decrease.

I employ a difference-in-differences approach that exploits the variation in tariffs across indus-tries and over time. I model the outcome of interest (firm size and TFP) for firm i in industry j attime t as:

yijt = β1τjt + β2τIjt + αj + αt + εijt (30)

whereyijst=log of firm size (output) or TFPτjt=final goods tariff, as a negative fraction of value (e.g., a 100% ad valorem tariff corresponds to-1)τ Ijt=input tariff, as a negative fraction of valueαj=industry fixed effectsαt=year fixed effects

Output is measured as discussed in Section 4. I construct a measure of TFP using a chain-linked index number method suggested by Aw, Chen and Roberts (2001); details are provided inAppendix A. The inclusion of year fixed effects allows me to control for macroeconomic changesthat affected all industries at the same time and in the same manner, while industry fixed effectscontrol for time-invariant industry characteristics, such as minimum efficient scale, that could affectfirm size and productivity. For ease of interpretation, I include final goods tariffs and input tariffvalues as negative numbers, so that the coefficients on these two variables can be interpreted as theeffect of a fall in tariffs on the outcome of interest.

Table 5 presents the baseline estimates for the entire sample, as well as for the formal andinformal sectors separately. All specifications include year and industry fixed effects, and standard

20

errors are clustered at the state-industry level.17 Column (1) includes only final goods tariffs (notinput tariffs) and shows a highly significant, positive relationship between the fall in final goodstariffs and average firm size. Since final goods tariffs and input tariffs are included as a negativefraction of value (e.g., a 100% ad valorem tariff is equivalent to τjt = −1), the average fall in finalgoods tariffs between 1989 and 1999 of 50 percentage points corresponds to a change of 0.5 inτjt. The magnitude of the coefficient (0.62) suggests that this fall in final goods tariffs is associatedwith an 31% increase in average firm size. In Column (2), I control for input tariffs as well, andfind that the effect of final goods tariffs is attenuated (to 0.46, or 23% given the average fall in finalgoods tariffs) but remains significant at the 5% level.

Columns (3) through (6) of Panel (a) show that the positive relationship between final goodstariffs and firm size is driven by informal firms. Once input tariffs are taken into account, a 50percentage point fall in final goods tariffs decreases average formal firm size by 15%, but increasesaverage informal firm size by 24%. Panel (b) shows similar results for TFP. When controllingfor input tariffs, a 50 percentage point fall in final goods tariffs increases average productivity byapproximately 15%. Informal sector productivity increases by 15.5%, while formal sector produc-tivity decreases by 5.5%. These initial results lend support to the DR framework, which predictsthat a fall in final goods tariffs will increase both average firm size and productivity.

Although both firm size and productivity in the formal sector fall when final goods tariffs arereduced, both effects are offset by a concurrent fall in input tariffs. The coefficients on input tariffssuggest that a 50 percentage point fall in input tariffs would increase formal firm size by 49% andTFP by 29.5%. The input tariff results should be cautiously interpreted, due to the aggregate natureof the input-output tables used to construct them. However, the pattern is consistent with recentwork by Amiti and Konings (2007), who find that a reduction in Indonesian input tariffs had agreater effect on productivity than the reduction in final goods tariffs. Similarly, a recent workingpaper by Goldberg, Khandelwal, Pavcnik and Topalova (2008) finds that imported intermediateinputs were an important driver of the increase in product scope among large, formal Indian firms.The finding that input tariffs increase firm size and productivity among formal firms, but do notsignificantly affect informal firm size or productivity, is also consistent with Amiti and Konings(2007), who show that a fall in input tariffs increases productivity more among firms that actuallyimport intermediate goods. It is logical that formal firms, which are presumably more able to importintermediates, are more likely to gain from a fall in input tariffs than informal firms.

17The standard errors are similar, and significance levels do not change, when I use the multi-way clustering methoddeveloped by Cameron, Gelbach and Miller (2006) to account for within-industry clustering across states, as well aswithin-state clustering across industries.

21

5.2 Size and Productivity Distributions

The initial results suggest that a fall in final goods tariffs increases average firm size and produc-tivity, as predicted by DR. I now investigate how changes in different percentiles of the size andproductivity distributions contributed to the average increases in size and productivity.

As noted in Table 1, DR predict that the firm size distribution will be compressed, while MOpredict that it will be stretched out. Furthermore, DR suggest that the left tail of the productivitydistribution will shift right, as the least productive firms exit, while MO suggest the opposite.

To test whether the Indian case is consistent with these predictions, I turn to a quantile regres-sion (QR) framework. In this section, I focus on final goods tariffs rather than input tariffs for tworeasons. First, as discussed above, my measure of input tariffs is available at a more aggregatelevel than my measure of final goods tariffs, and is based on input-output tables for relatively broadmanufacturing industries. Second, I want to investigate whether changes in the size distribution offirms are consistent with two recent trade models, which do not incorporate input tariffs. Of course,the results in Section 5.1 indicate that it is important to control for input tariffs when estimating theeffects of final goods tariffs on firm size and productivity. Therefore, in the QR framework below,I control for input tariffs as well as final goods tariffs, but input tariff results are not presented, andthe counterfactual simulations consider only a fall in final goods tariffs.

I begin by estimating the effects of final goods tariffs on various quantiles of the firm size andproductivity distributions. While ordinary least squares (OLS) estimates the mean of the outcomeof interest conditional on the covariates, QR estimates the locations of various percentiles of thedistribution (e.g., the median, 25th percentile, etc.). Let z′ijt = (τjt, τ

Ijt, α

′j, α

′t)′, and Equation 30

can be modified to allow the effects of the covariates to vary across percentiles:

yijt = z′ijtβθ + εθijt(31)

Let Fijt(yijt|zijt) be the distribution of yijt conditional on zijt. Then the θth conditional quantileof yijt (the location of the θth percentile of yijt, conditional on the covariate vector zijt) is:

Qθ(yijt|zijt) ≡ infy|Fijt(y|z) ≥ θ = z′ijtβ(θ) (32)

where we assume that Qθ(εθijt|zijt) = 0. As shown by Koenker and Bassett (1978), the estimator

for β(θ) is found by minimizing:

n−1

n∑i=1

ρθ(yijt − z′ijtβθ), ρθ(u) =

θu u > 0

(θ − 1)u u ≤ 0(33)

22

The estimated effect of a marginal change in the kth covariate zkijt is given by:

∂Qθ(yijt|zijt)∂zkijt

= βkθ (34)

How should βkθ be interpreted? The result does not imply that the individual firm at the θth per-centile of the distribution will change its size or productivity by βkθ , but rather that the θth percentileof the distribution will change by βkθ .

A key advantage of QR is that it allows estimation of the whole conditional distribution of firmsize, rather than just the mean. Both models discussed in Section 2 predict that trade liberalizationwill increase average firm size, but DR suggest that the distribution of output will be compressed,while MO suggest that it will be stretched out. The OLS results for all firms, presented in the pre-vious section, are consistent with either model, but the QR results can help to distinguish betweenthe two. Another advantage of QR is its robustness to outliers, which is particularly important inthis context since the size distribution (even in logs) is non-normal.

I conduct quantile regressions for firm size and TFP in the formal and informal sectors individ-ually, at every 5th percentile of the distributions.18 As before, the coefficients on final goods andinput tariffs are negative fractions of value, so that the coefficients can be interpreted as the effectof a fall in tariffs.

Figures 4 and 5 show the QR results for final goods tariffs graphically. The solid line plots theQR coefficient at every 5th percentile, while the two dotted lines show the 90% confidence intervalsfrom a block bootstrap estimate of the standard errors.19

Panel (a) of Figure 4 shows that the fall in final goods tariffs increases firm size across all quan-tiles in the informal sector. However, in the formal sector, the fall in final goods tariffs decreasesfirm size (Panel (b)). This decrease in formal firm size is attenuated in higher quantiles, and is indis-tinguishable from zero for the largest firms. Similarly, Panels (a) and (b) of Figure 5 present the QRcoefficients for TFP. The fall in final goods tariffs increases informal TFP across all quantiles, witha sharp increase among the lowest quantiles. This finding is consistent with recent work by Eslava,Haltiwanger, Kugler and Kugler (2009), who find that productivity among the bottom percentile offirms in Colombia increased by more than the median productivity after a trade liberalization. Theauthors consider this evidence of market selection at work, as the least productive firms are forcedout. Since the least productive firms are informal (see Panel (d) of Figure 3), this type of market

18I also conducted quantile regressions for the overall manufacturing industry, including both formal and informalfirms. Results are not shown because given the overwhelming size of the informal sector relative to the formal sector,the overall results are largely the same as the informal sector results.

19I re-sample over state-industry clusters in order to correct for possible serial correlation (Bertrand, Duflo andMullainathan 2004). In order to reduce computational time, I follow Abrevaya (2001) and sample (with replacement)m out of n clusters, where m = 0.1n. Standard errors are based on 100 bootstrap estimates, and are corrected by√m/n.

23

selection is also consistent with my QR results. It is interesting to note that I would not have foundthis result if I had only considered formal firms; Panel (b) of Figure 5 shows that final goods tariffsactually decrease the bottom quantiles of productivity in the formal sector, while having no effecton productivity among the top quantiles.

A more intuitive way to interpret these results is to compare the density of firm size in 1989, tothe density that would have prevailed in 1989 had final goods tariffs been distributed as in 1999,with all other covariates distributed as in 1989. Machado and Mata (2005) show that it is pos-sible to simulate such counterfactual densities using regression quantiles. I implement a slightlymodified version of their approach as follows.20 First, I take m=1,000 draws of θ from a uniformdistribution U ∼ [0, 1], and estimate βθ for each draw. I then draw a representative sample of 1,000observations of zijt from the 1989 data (with probability proportional to each observation’s sam-pling multiplier). Then yijt = z′ijtβ(θ)ml=1 is a sample of 1,000 observations from the estimatedmarginal density of the outcome yijt, if all of the covariates were distributed as in 1989. Next, Idraw 1,000 observations of τjt from the 1999 data and 1,000 observations of all covariates expectτjt from the 1989 data. I use this counterfactual sample of covariates zcijt to obtain 1,000 samplesfrom the estimated counterfactual density of outcome yijt, ycijt = (zcijt)

′β(θ)ml=1 if all covariatesexcept final goods tariffs were distributed as in 1989, and final goods tariffs were distributed as in1999.

Figure 6 shows the differences between the 1999 counterfactual distributions and the 1989distributions. Panel (a) shows that in the informal sector, the fall in final goods tariffs removesmass from the left half of the size distribution, and adds it to the right half of the size distribution.In other words, the fall in final goods tariffs shifts the informal firm size distribution to the right.In the formal sector, mass is removed from the middle of the size distribution and added to the lefttail of the size distribution, indicating that the bulk of the formal size distribution shifts left. Theupper tail of the formal size distribution is essentially unchanged, which is consistent with Figure4, showing that the top quantiles of the formal size distribution did not respond to the fall in finalgoods tariffs.

Panel (b) presents the results for the productivity distributions in each sector. The productivitydistribution for the informal sector shifts right, as mass is moved from the left side of the distributionto the right. In the formal sector, the productivity distribution shifts left, as mass is moved from theright side of the distribution to the left. These changes in productivity distributions are similar tothe changes in firm size distributions. However, in the case of the productivity distributions, thereis considerable overlap between the two sectors, so the results offset rather than complement eachother, as they did in the case of the size distribution.

Table 6 summarizes the results for the formal and informal sectors, and compares them to

20I am indebted to José Machado for his advice on using this technique.

24

the predictions of the DR and MO models. When considered individually, neither the informalnor the formal sector fit the predictions of either model. However, when we consider the overalleffects of the two sectors, the results are consistent with DR’s predictions. As discussed earlier,overall average firm size and productivity both increase. Informal firms constitute the bulk of theleast productive firms (see Figure 3); therefore, the relatively large increase in the left tail of theinformal sector productivity distribution, which tapers off towards the right, is consistent with DR’sprediction that the least productive firms are forced to exit. The left end of the firm size distribution(represented by informal firms) shifts right, while the right tail (represented by formal firms) shiftsleft. The overall distribution is therefore compressed, as predicted by DR. Finally, within the formalsector, the decrease in size is less pronounced at the right tail of the size distribution, which is whatwe would expect if the largest formal firms are relatively more export-oriented.

5.3 Firm Size Adjustment and Net Entry

Why are the results for the formal and informal sectors so different? On one hand, we wouldexpect exit to be more prevalent in the informal rather than the formal sector, since informal firmsare generally less productive; thus we would expect to see an increase in productivity and firm sizein this sector. In the formal sector, which is at the right tail of the size distribution, it is reasonablethat the contraction in output among existing firms would outweigh any exit effects, thus resultingin a fall in size among formal firms. On the other hand, it is surprising that there is such a starkcontrast between the two sectors, even when we compare the upper quantiles of the informal sectordistribution with the lower quantiles of the formal sector distribution.

One possibility is that there are barriers to the adjustment of firm size, as well as to exit, inthe formal sector. There is substantial evidence indicating that India’s stringent labor regulations,which make it difficult for large, formal firms to adjust their employment or to close down, causedthe two sectors to behave differently even before the trade reforms. Besley and Burgess (2004) clas-sify state amendments to India’s Industrial Disputes Act (IDA) as “pro-worker” or “pro-employer”;they find that “pro-worker” states have lower formal sector output, and higher informal sector out-put, relative to “pro-employer” states. Since the 1990’s, there has also been growing concern thatIndia’s trade liberalization further shifted resources from the formal to the informal sectors. In a2003 speech, former Indian Prime Minister A.B. Vajpayee argued that “the size and the scope ofthe informal sector have vastly expanded - not shrunk - because of the forces of liberalization andglobalisation,” and a number of commentators have expressed similar concerns (see, for example,Jhabvala and Sinha (2002), Hensman (2001)). While there is little quantitative evidence linkingIndia’s liberalization with informality, recent work by Hasan et al. (2007) finds that the elasticityof labor demand increased in the formal sector following the trade reforms, and that the increases

25

were more pronounced in states with more stringent labor laws.I explore the hypothesis that the difference between the formal and informal sectors’ reactions

to trade reform may be, at least partially, caused by the ease of adjusting firm size and of exiting, byexploiting the variation in labor laws across Indian states. First, I employ a measure of the difficultyof laying off workers or of closing down a large firm, from Ahsan and Pages (2007). Their measureis based on the work by Besley and Burgess (2004), but it isolates the labor regulations that pertainto ease of firm size adjustment and exit. I use pre-reform data in order to avoid the possibility thatstates changed their labor regulations in response to the trade reforms.21 I construct three additionalmeasures of the difficulty of adjusting firm size or closing based on the outcomes of court cases.India’s Ministry of Labour reports the number of cases in which large firms requested permissionto lay off or retrench workers, or to close down, between 1988 and 1992, by state; they also reportthe number of cases in which the request was granted or denied. My measures of the difficultyof laying off or retrenching workers, or closing, are the fraction of cases in which permission wasdenied.

I explore the impact of each of these measures on firm size using the following specification:

yijst = β1τjt + γτjtLs + β2τIjt + αj + αs + αt + εijst (35)

where yijst is firm size for firm i in industry j and state s at time t; τjt and τ Ijt are tariffs andinput tariffs, measured as negative fractions of value as discussed above; Ls is the measure of laborregulations in state s; and αj , αs, and αt are industry, state, and year dummy variables, respectively.

I also investigate whether labor regulations are responsible for differential net entry into eachsector by examining the (estimated) number of firms in each state-by-industry cluster:

njst = β1τjt + γτjtLs + β2τIjt + αj + αs + αt + εjst (36)

where njst is the number of firms in each state-by-industry cluster at time t. Each state-by-industryobservation is weighted by the number of firms in the cluster to make the overall estimate repre-sentative of the entire population.

Table 7 presents results from these two specifications. Panel (a) shows that the fall in formal firmsize is attenuated in states in which it is difficult to adjust size; the coefficients on all four measuresof labor regulations are positive, though only those on the Ahsan and Pages (2007) measure andon the fraction of layoffs denied are statistically significant. However, Panel (b) shows that thesestates also exhibit lower net entry of formal firms. Most of the results for the informal sector,though generally opposite in sign to those in the formal sector, are not statistically significant.These findings are consistent with the movement of resources from the formal to the informal

21However, as Ahsan and Pages (2007) note, major changes to the IDA are confined to 1989 and before.

26

sector when final goods tariffs fall. Formal firm size decreases, while informal firm size increases.The decrease in formal firm size is attenuated in states where it is difficult to adjust employment.However, these states also show increased exit from the formal sector, indicating that if formalfirms are not allowed to adjust their sizes to meet decreased demand, there will be less net entry inthe long run.

5.4 Robustness Checks

The baseline results suggest that India’s trade liberalization compressed the size distribution ofoutput and increased average productivity. In this section, I test the robustness of these resultsin a variety of ways. Due to the computational intensity of the QR process, I perform all of therobustness checks on the mean, rather than the quantiles, of the distribution.

Lagged Tariffs

In Section 3, I find no relationship between pre-reform industry characteristics and final goods tariffchanges. Topalova (2007) performs a similar exercise and finds no relationship between final goodstariff changes and pre-reform industry characteristics among formal firms; similarly, she finds norelationship between TFP among large firms in an industry in year t and final goods tariffs in yeart + 1, from 1989 to 1996. She does, however, find that industries with higher TFP levels amonglarge firms in year t faced lower final goods tariffs in year t + 1, from 1997 to 2001. She notesthat final goods tariffs were uniformly changed until 1997 (during India’s Eighth Five-Year Plan),but that they may have diverged somewhat after 1997, indicating that political factors may haveinfluenced changes after this point. To address the potentially endogenous selection of industriesinto tariff changes after 1997, I re-run the analysis using 1997 final goods and input tariffs in placeof 1999 tariffs. Table 8 shows that the coefficients are similar in magnitude and significance for thetwo different tariff measures, for both firm size and TFP.

Alternate TFP Measures

The main advantage of using an index number method to calculate TFP is that it is highly robust tovariation in technology across firms (VanBiesebroeck 2007). This is a key concern in my datasetbecause large and small firms within the same industry are likely to have very different productiontechniques. However, the index number method also has several limitations, including (1) theassumptions of perfect competition and constant returns to scale and (2) the failure to account formeasurement error in the variables. I therefore explore whether my findings are robust to alternatemethods of measuring TFP. First, I recalculate productivity with the index number method usingrevenue shares instead of cost shares, which allows me to avoid calculating rental rates for capital.

27

Second, I estimate TFP as the residuals from an OLS regression of output on labor, capital, andmaterial (in logs).

Third, I attempt to address the drawback of using industry rather than firm-level output pricedeflators. Heterogeneous firms models generally predict that more productive firms have higheroutput and revenue, but charge lower prices. Recent work by Foster, Haltiwanger and Syverson(2008) confirms this finding empirically for certain industries using US data. If this is the case inmy dataset, then deflating firm revenues by industry-level price deflators will systematically un-derestimate large firms’ productivities while overestimating small firms’ productivities. Hsieh andKlenow (forthcoming) suggest that if we assume the same Cobb-Douglas production technologyfor all firms, and if we model industry output as the CES aggregate of all products in the industry,then we can correct for differential pricing within an industry by raising each firm’s nominal out-put (sales) to the power σ/(σ − 1) where σ is the elasticity of substitution between differentiatedgoods. The authors note that typical values of σ range from 3 to 10 in the literature; I thereforemodify observed revenues using each of these bounds for σ, and use the modified values of outputto construct TFP.

Finally, I use labor productivity (the log of output per number of employees) as an alternateproductivity measure. While labor productivity is likely to overstate the productivity of larger,more capital-intensive firms, it avoids having to calculate real capital input values.

Table 9 presents the results from each of these alternate TFP measures in each sector. In bothpanels, Column (1) reproduces the original TFP estimates, and Columns (2) through (6) presentresults from each of these alternate measures. The coefficient on final goods tariffs ranges from-0.085 to -0.26 in the formal sector and is highly significant in all cases. In the informal sector, thecoefficient on final goods tariffs ranges from 0.30 to 0.55 and is significant at the 5% level in allspecifications.

Non-Tariff Barriers and Other Industrial Policy Changes

As discussed in Section 3, tariffs were harmonized beginning in 1991. However, India’s pre-reformtrade regime also included large non-tariff barriers (NTB), most of which were not immediatelyreduced. During the 1980’s, imports were categorized into numerous lists, which are summarizedin five broad categories by Aksoy (1992): “banned” (items that could not be imported), “restricted”and “limited permissible” (items that required licenses), “open general license” (items that didnot require licenses) and “canalized” (items that could only be imported by public sector compa-nies). The 1991 reforms eliminated these lists, and provided only one list of disallowed imports(Ahluwalia 1994). However, while NTBs were reduced sharply on capital and intermediate goods,high NTB barriers remained on consumer goods until the late 1990’s.

Measuring NTB coverage is not as straightforward as measuring tariffs, but Pandey (1999)

28

attempts to do so by calculating the fraction of product lines within each of 64 broad industrygroups that required an import license.22 Panel (a) of Figure 7 shows the changes in NTB protectionover time. I group industries into three classifications: basic and capital goods, intermediate goods,and consumer goods, based on Nouroz (2001). The first goods to have NTB restrictions liftedwere capital and basic goods; consumer durables and non-durables, such as apparel and tea, wereremoved from the negative list only during the latter half of the 1990’s. In contrast, Panel (b) ofFigure 7 shows that final goods tariffs were reduced across all types of industries similarly, whichis consistent with the results in Section 3 showing that tariff cuts are uncorrelated with pre-reformindustry characteristics.

The aggregate nature of the NTB data and the selection of industries into NTB liberalizationcreate potential estimation challenges. However, if NTBs, rather than tariffs, are the binding con-straint on imports, then I would expect the effect of tariff liberalization to be greater in industriesthat were also NTB liberalized. In Table 10, I explore this possibility by interacting the final goodstariff variable with a dummy variable for NTB liberalization. The median NTB coverage ratio fellby approximately 0.6 between 1989 and 1999. I construct a dummy variable for NTB liberaliza-tion that takes on a value of one if the industry’s NTB coverage ratio fell by more than 0.6 between1989 and 1999, zero otherwise. Panels (a) and (b) of Table 10 present results for firm size andproductivity in the formal and informal sectors separately. The coefficients on final goods tariffsdo not change substantially when the interaction between final goods tariffs and NTB liberalizationis included. In Panel (a), the coefficients on the interaction term are not statistically significant,though they indicate that NTB liberalization does amplify the effects of the fall in final goods tar-iffs on firm size in both sectors, as expected. In Panel (b), the coefficient on the interaction term forthe formal sector (0.059) is significant at the 5% level and indicates that the fall in formal sectorproductivity is mitigated in NTB-liberalized industries. A similar increase in productivity, thoughstatistically insignificant, is also found in the informal sector.

Table 10 also investigates potential confounding effects of two other industrial policy changesthat occurred during the 1990’s: the dismantling of the “license raj” and the allowance of foreigndirect investment (FDI) into most industries without case-by-case approval. Until the 1980’s, In-dia’s “license raj” required firms with more than 50 employees (100 employees without power)and a certain amount of assets were required to obtain a license in order to operate. The licensespecified, among other things, the amount of output a firm could produce, the types of goods itmade, and the firm’s location (Sharma 2008). In 1985, approximately one-third of industries were“delicensed” (the requirement for a license was dropped); in 1991, most industries were delicensedas part of the broader reforms package (Aghion, Burgess, Redding and Zilibotti 2008). Using ag-gregate industry-level data from 1980 to 1997, Aghion et al. (2008) find that delicensing increases

22I am grateful to Devashish Mitra for sharing the data from Pandey (1999) with me.

29

the number of formal firms (as well as output, to some extent) in the formal sector, and increasesformal sector growth among “pro-employer” states relative to “pro-worker” states. A second im-portant policy change that occurred in 1991 was the liberalization of FDI inflows. Prior to 1991,FDI was capped at 40% for most industries; beginning in 1991, FDI inflows of up to 51% wereallowed in selected industries with “automatic” approval (Sivadasan 2008). Using firm-level ASIdata, Sivadasan (2008) finds that the FDI liberalization increases productivity by approximately15% among formal firms with more than 5 employees.

Table 10 shows that the impacts of final goods tariffs are robust to controlling for these indus-trial policies.23. The coefficients on the fall in final goods tariffs in the formal sector are slightlyattenuated (from -0.30 to -0.26 for firm size, and from -0.11 to -0.09 for productivity) but remainhighly significant. In the informal sector, the coefficients on firm size range from 0.48 to 0.60,while the coefficients on productivity range from 0.21 to 0.32; all are significant at the 5% level.

I find that delicensing decreases average firm size and productivity in both the formal and infor-mal sectors. The negative coefficient on formal firm size is consistent with the findings of Aghionet al. (2008). These authors show that delicensing increases the number of formal firms by 6%,while increasing output by 3% (though the result is not statistically significant). The combinationof those two effects is to decrease average formal firm size, as I find. One possible reason forthe decreases in firm size and productivity in both sectors is the movement of the most productiveinformal firms to the formal sector, when an industry is delicensed; a recent working paper bySharma (2009) shows that the informal sector contracts when industries are delicensed. Finally, thepositive relationship between FDI liberalization and both firm size and productivity is consistentwith recent work by Sivadasan’s (2008), who shows that FDI liberalization increases productivityamong formal Indian firms with more than five employees.

Export Orientation

In Section 2, I argued that it was reasonable to assume that changes in domestic output wouldoverwhelm changes in export output, given India’s small share of exports in manufacturing output.Unfortunately, my dataset does not provide information on whether individual firms are exporters,so I cannot directly test whether exporters and non-exporters behave differently. However, if myassumption is reasonable, and the predictions in DR are correct, then I would expect the fall informal firm size to be attenuated for relatively export-oriented industries. To test this proposition,I calculate the share of exports in manufacturing output at the broad industry level in 1989 (beforethe trade reforms), using data from the World Bank’s Trade and Production database. I then includean interaction term between pre-reform export share and tariffs in the baseline specification.

23Data on delicensing and FDI reform are from Aghion et al. (2008). Since their data stop in 1997, I use the 1997delicensing and FDI reform variables for the 1999-2000 survey round.

30

Table 11 presents results. Consistent with the predictions of DR, I find that the decrease informal firm size due to the fall in final goods tariffs is attenuated in industries with higher exportshares. If export share were increased by one standard deviation (0.33), the fall in formal firmsize would be reduced by 9.6 percentage points, or 30%. The finding that informal firms are notsignificantly affected by export orientation is also consistent the DR’s implication that small firmsare unlikely to be exporters.

6 The Missing Middle

The previous sections have considered firm size in terms of output, and most of the trade theoryfocuses on output size. Yet the size distribution of employment in India has also been a subjectof interest among researchers. India, like many other developing countries, exhibits a “missingmiddle”: a concentration of employment among small and large firms, with little employmentin mid-sized firms, compared to the US or many other large Asian economies (Mazumdar 1998,Tybout 2000, among others). A number of reasons for the missing middle have been proposed,including regulations and credit constraints that prevent small firms from growing; Tybout (2000)provides an excellent summary of possible causes of this phenomenon. As he notes, however, thereis little empirical evidence on what causes the size distribution in developing countries to differfrom that of developed countries. A notable exception is VanBiesebroeck (2005), who documentsthe slow growth of small firms relative to large firms in South Africa, and shows that the allocationof workers and credit to high-productivity firms is an important factor.

In India, a number of commentators and politicians (Hensman 2001, Jhabvala and Sinha 2002,Vajpayee 2003, among others) have argued that globalization has pushed more employment intosmall, informal firms, but there is little quantitative evidence to substantiate these claims.24 In thissection, I address the links between India’s trade liberalization and employment size in two ways.First, I test whether the fall in final goods tariffs affect the employment size distribution, using theanalytical methods described in Section 5. Second, I consider whether the effects of a fall in finalgoods tariffs, and in particular the employment size effects, differ across labor-intensive versuscapital-intensive industries.

24There are a few studies that consider the links between trade and informal employment in other countries, but theresults are mixed. Goldberg and Pavcnik (2003) find no evidence that trade affects the share of informal employment inBrazil, while Menezes-Filho and Muendler (2007) find that trade liberalization is associated with lower formal sectoremployment and higher informal sector employment in Brazil. Goldberg and Pavcnik (2003) do find a link betweentrade and informal employment in Colombia, but only before Colombia’s labor market is liberalized as well.

31

6.1 The Employment Size Distribution

Figure 8 investigates the distributional effects of the fall in final goods tariffs using the quantileregression methods described above. Panel (a) shows that the fall in final goods tariffs decreasesemployment somewhat across all quantiles, though the effect is only statistically significant amongthe lower quantiles. Panel (b) shows that the fall in final goods tariffs shifts employment from firmswith 10-50 employees to firms with fewer than 10 employees.

I also performed quantile regressions for informal sector employment, and for the overall em-ployment distribution. Results are not shown here because the coefficients on final goods and inputtariffs were indistinguishable from zero (both statistically and economically speaking) across allquantiles. While there is some evidence that the fall in final goods tariffs shifted formal sector em-ployment towards smaller firms, the fact that informal sector employment (which accounts for 80%of total manufacturing employment) did not change at all suggests that India’s trade liberalizationdid not significantly impact the missing middle.

6.2 Labor-Intensive Industries

I further investigate the change in the employment size distribution by considering whether the de-crease in formal employment size is concentrated in certain industries. India is relatively abundantin labor; therefore, in a standard trade model, we would expect trade liberalization to reallocateresources away from capital-intensive (comparative disadvantage) industries and towards labor-intensive (comparative advantage) industries. The DR and MO models both assume that there isonly one factor of production, and therefore preclude the possibility of trade driven by comparativeadvantage. However, Bernard, Redding and Schott (2007b) use a similar framework in which thereis product differentiation within an industry, as well as comparative advantage between industries.They show that a fall in final goods tariffs increases average productivity and firm size, and does soby more in the comparative advantage industry.

I test this proposition by constructing a “comparative advantage” dummy variable. To do so,I calculate the average capital-to-employee ratio in 1989 (the pre-reform year) for each industry.Industries with capital-to-employee ratios less than the median are classified as comparative ad-vantage (labor intensive) industries, while those with capital-to-employee ratios greater than themedian are classified as comparative disadvantage (capital-intensive) industries. I modify the base-line specification to include an interaction term between the comparative advantage variable andthe fall in final goods tariffs.

The results in Panel (a) of Table 12 show that overall firm size is approximately the same be-tween labor-intensive and capital-intensive industries; however, within the formal sector, averagefirm size actually rises in labor-intensive industries, while it falls in capital-intensive industries.

32

Panel (b) indicates that informal sector productivity rises by more in labor-intensive (comparativeadvantage) industries, while formal sector productivity falls by slightly less in labor-intensive in-dustries (though the latter effect is not statistically significant). As shown in Panel (c), the fall infinal goods tariffs has no effect on informal employment in either labor- or capital-intensive indus-tries. However, consistent with the results on employment size distribution, Columns (3) and (4)of Panel (c) indicate that a 50 percentage point fall in final goods tariffs reduces formal employ-ment size in capital-intensive industries, but increases formal employment size in labor-intensiveindustries.

These findings are consistent with the predictions of Bernard et al. (2007b) and suggest that thefall in employment size among formal firms may reflect a shift of employment away from capital-intensive industries and towards labor-intensive industries. The fact that there was no significantchange in employment size in the informal sector is consistent with this hypothesis, since mostinformal firms (regardless of industry) are likely to use labor-intensive production techniques.

7 Discussion and Conclusion

India’s 1991 reforms spurred a large and growing literature on the effects of trade and other policieson firm behavior. This paper adds to that body of work by considering the effects of a major tariffliberalization on the entire manufacturing industry, including informal firms that comprise 80% ofmanufacturing employment. I construct a unique dataset that combines firm-level data on formaland informal firms from 1989, 1994, and 1999, a period of time that spans the major trade reforms.I use a differences-in-differences approach to isolate the impact of final goods tariffs and inputtariffs on firm size and productivity, exploiting the variation of tariffs across industries and overtime, and the fact that there does not appear to be a systematic correlation between tariff cutsand pre-reform industry characteristics. I find that average firm size and productivity increase inindustries that experience large final goods tariff cuts, relative to other industries. An importantcontribution of this paper is to show that the increases in average firm size and productivity aredriven by the informal sector; final goods tariff cuts lower firm size and productivity in the formalsector while raising it in the informal sector. The difference between the two sectors highlights theimportance of considering the manufacturing industry as a whole when evaluating the effects oftrade liberalization.

I also find that the decrease in formal sector firm size and productivity is driven by capital-intensive industries. In labor-intensive industries, in which India is likely to have a comparativeadvantage, formal firm size actually increases when final goods tariffs fall. Furthermore, even incapital-intensive industries, the decrease in formal firm size and productivity, due to the fall in finalgoods tariffs, is offset by the increase in both firm size and productivity, due to the concurrent fall

33

in input tariffs.I then employ quantile regression techniques to test whether the fall in final goods tariffs shifts

the size and productivity distributions of firms, as predicted by recent trade models. I modify a sim-ulation technique proposed by Machado and Mata (2005) to compare the distributions of size andproductivity that prevailed in 1989 to those that would have prevailed in 1989, if final goods tariffshad been distributed as they were in 1999. When considered individually, neither the informal northe formal sector fit the predictions of either model. However, when we consider the overall effectsof the two sectors, the results are consistent with Demidova and Rodriguez-Clare’s (2009) predic-tions. I find that the fall in tariffs compresses the overall output distribution of firms; the bulk of thedistribution (represented by the informal sector) shifts right, while the right tail of the distribution(represented by the formal sector) shifts left. This finding is consistent with the predictions put forthby Demidova and Rodriguez-Clare (2009). In their framework, a fall in Home final goods tariffsforces the least productive (smallest) firms to exit, decreases domestic output among existing firms,and increases exports among existing exporters. Assuming that domestic sales are more importantthan export sales for most firms, the net effect is an increase in the size of the smallest firms and adecrease in the size of the largest firms, which is consistent with my findings. The rise in averageproductivity, and the relatively large increase in productivity among the smallest quantiles of theinformal productivity distribution, also support Demidova and Rodriguez-Clare’s (2009) predictionthat a unilateral liberalization forces the least productive firms to exit.

Of course, the division between formal and informal firms is not the same as the division be-tween exporters and non-exporters in Demidova and Rodriguez-Clare’s (2009)’s framework. Nev-ertheless, the same basic logic applies. In their model, the reallocation of resources from domesticfirms to exporters occurs because there is an increase in competition as consumers spend moremoney on imports and less on domestic goods. The least productive firms - those that make thelowest profits - can no longer survive, while exporters - those least exposed to domestic markets -expand relative to domestic firms. We can think of an analogous change for the formal and infor-mal sectors. As competition from imports increases, the least productive, informal firms are forcedto exit. The size of most formal firms shrinks, but the drop in firm size is attenuated for largerfirms - those that are more likely to be exporters. This finding is supported by the fact that the fallin formal firm size is attenuated in industries that have higher export shares, while the change ininformal firm size is not affected by export orientation.

Finally, I investigate possible links between India’s trade liberalization and the “missing mid-dle” in employment. While the fall in final goods tariffs is, to some extent, correlated with a fall informal employment, the effect is offset by the rise in employment caused by the fall in input tariffs.Furthermore, formal employment actually rises in comparative advantage (labor-intensive) indus-tries in which final goods tariffs are liberalized, suggesting that labor may have been reallocated

34

from capital-intensive industries to labor-intensive industries in the formal sector. The “missingmiddle” remains an interesting phenomenon, with potentially important consequences for growth,but India’s trade reforms do not appear to have significantly affected its unusual employment distri-bution. Investigating how other policies affect the employment distribution of firms is an importantarea for future research.

References

Abrevaya, Jason, “The effects of demographics and maternal behavior on the distribution of birthoutcomes,” Empirical Economics, 2001, 26, 247–257.

Aghion, Philippe, Robin Burgess, Stephen Redding, and Fabrizio Zilibotti, “The Unequal Effectsof Liberalization: Evidence from Dismantling the License Raj in India,” The AmericanEconomic Review, September 2008, 98 (4), 1397–1412.

Ahluwalia, Montek S., “India’s Economic Reforms,” March 1994. Based on an address at theSeminar on India’s Economic Reforms at Merton College, Oxford.

Ahsan, Ahmad and Carmen Pages, “Are All Labor Regulations Equal? Assessing the Effects ofJob Security, Labor Dispute and Contract Labor Laws in India,” June 2007. World Bank PolicyResearch Working Paper 4259.

Aksoy, M. Ataman, “The Indian Trade Regime,” October 1992. Policy Research Working Papers,Country Department, World Bank WPS989.

Amiti, Mary and Jozef Konings, “Trade Liberalization, Intermediate Inputs, and Productivity:Evidence from Indonesia,” The American Economic Review, December 2007, 97 (5),1611–1638.

Aw, Bee Yan, Xiaomin Chen, and Mark J. Roberts, “Firm-level evidence on productivitydifferentials and turnover in Taiwanese manufacturing,” Journal of Development Economics,2001, 66, 51–86.

Bedi, Jatinder S. and Purnendu K. Banerjee, “Discrepancies and Validation of IndianManufacturing Data,” Economic and Political Weekly, March 10 2007.

Bernard, A. B., J. B. Jensen, and Peter K. Schott, “Trade Costs, Firms and Productivity,” Journalof Monetary Economics, 2006, 53, 917–937.

Bernard, Andrew B., J. Bradford Jensen, Stephen J. Redding, and Peter K. Schott, “Firms inInternational Trade,” April 2007. NBER working Paper 13054., Stephen J. Redding, and Peter K. Schott, “Comparative Advantage and Heterogeneous

Firms,” Review of Economic Studies, 2007, 74, 31–66.Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan, “How Much Should We Trust

Differences-in-Differences Estimates?,” Quarterly Journal of Economics, 2004, 119, 249–275.

35

Besley, Timothy and Robin Burgess, “Can Labor Regulation Hinder Economic Performance?Evidence from India,” Quarterly Journal of Economics, February 2004.

Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller, “Robust Inference with Multi-wayClustering,” September 2006. Working Paper.

Debroy, B. and A.T. Santhanam, “Matching Trade Codes with Industrial Codes,” Foreign TradeBulletin, July 1993, 24 (1), 5–27.

Demidova, Svetlana and Andres Rodriguez-Clare, “Trade Policy under firm-level heterogeneity ina small economy,” Journal of International Economics, June 2009, 78 (1), 100–112.

Dutz, Mark A., “Oligopolistic Firms’ Adjustments to Quota Liberalization: Theory andEvidence,” in Mark Roberts and James Tybout, eds., Industrial Evolution in DevelopingCountries, Oxford University Press, 1996.

Eslava, Marcela, John C. Haltiwanger, Adriana D. Kugler, and Maurice Kugler, “Trade Reformsand Market Selection: Evidence from Manufacturing Plants in Colombia,” April 2009. NBERWorking Paper 14935.

Foster, Lucia, John Haltiwanger, and Chad Syverson, “Reallocation, Firm Turnover, andEfficiency: Selection on Productivity or Profitability?,” The American Economic Review, 2008,98 (1), 394–425.

Goldberg, Pinelopi K., Amit Khandelwal, Nina Pavcnik, and Petia Topalova, “ImportedIntermediate Inputs and Domestic Product Growth: Evidence from India,” October 2008.NBER Working Paper 14416.

and Nina Pavcnik, “The Response of the Informal Sector to Trade Liberalization,” January2003. Working Paper.

Harrison, Ann, “Productivity, Imperfect Competition and Trade Reform,” Journal of InternationalEconomics, 1994, 36, 53–73.

Hasan, Rana, Devashish Mitra, and K. V. Ramaswamy, “Trade Reforms, Labor Regulations andLabor Demand Elasticities: Empirical Evidence from India,” Review of Economics andStatistics, August 2007, 89 (3), 466–481.

Head, Keith and John Ries, “Rationalization effects of tariff reductions,” Journal of InternationalEconomics, 1999, 47, 295–320.

Hensman, Rohini, “Globalisation and Informalisation,” Economic and Political Weekly, March 312001, pp. 1073–1074.

Hsieh, Chang-Tai and Peter J. Klenow, “Misallocation and Manufacturing TFP in China andIndia,” Quarterly Journal of Economics, forthcoming.

Jhabvala, Renana and Shalini Sinha, “Liberalisation and the Woman Worker,” Economic andPolitical Weekly, 2002, May 25, 2037–2044.

Koenker, Roger and Gilbert Bassett, “Regression Quantiles,” Econometrica, 1978, 46, 33–50.

36

LaPorta, Rafael and Andrei Shleifer, “The Unofficial Economy and Economic Development,”November 2008. Working Paper.

Levinsohn, James, “Testing the imports-as-market-discipline hypothesis,” Journal of InternationalEconomics, 1993, 35, 1–22.

Machado, Jose A. F. and Jose Mata, “Counterfactual Decomposition of Changes in WageDistributions Using Quantile Regression,” Journal of Applied Econometrics, 2005, 20,445–465.

Mazumdar, Dipak, “Size-Structure of Manufacturing Establishments and the ProductivityDifferentials between Large and Small firms: A Comparative Study of Asian Economies,”1998. CIS Working Paper 1998-7.

Melitz, Marc J., “The Impact of Trade on Intra-Industry Reallocations and Aggregate IndustryProductivity,” Econometrica, November 2003, 71 (6), 1695–1725.

and Giancarlo I. P. Ottaviano, “Market Size, Trade, and Productivity,” Review of EconomicStudies, 2008, 75, 295–316.

Menezes-Filho, Naercio Aquino and Marc-Andreas Muendler, “Labor Reallocation in Responseto Trade Reform,” November 2007. Working Paper.

Nouroz, Hasheem, Protection in Indian Manufacturing: An Empirical Study, MacMillan IndiaLtd., Delhi, 2001.

Pandey, Mihir, NCAER Report on Trade Protection in India, National Council of AppliedEconomic Research, New Delhi, India, 1999.

Pavcnik, Nina, “Trade Liberalization, Exit, and Productivity Improvements: Evidence fromChilean Plants,” Review of Economic Studies, 2002, 69, 245–276.

Sharma, Gunjan, “Competing or Collaborating Siblings? Industrial and Trade Policies in India,”February 2008. Working Paper.

Sharma, Siddharth, “Entry Regulation, Labor Laws and Informality,” April 2009. Working Paper.Sivadasan, Jagadeesh, “Barriers to Competition and Productivity: Evidence from India,”

November 2008. Working Paper.Topalova, Petia, “Trade Liberalization and Firm Productivity: The Case of India,” 2007. IMF

Working Paper WP/04/28.Trefler, Daniel, “The Long and Short of the Canada-U.S. Free Trade Agreement,” The American

Economic Review, September 2004, 94 (4), 870–895.Tybout, James, “Manufacturing Firms in Developing Countries: How Well Do They Do, and

Why?,” Journal of Economic Literature, 2000, 38 (1), 11–44.and M. Daniel Westbrook, “Trade liberalization and the dimensions of efficiency change in

Mexican manufacturing industries,” Journal of International Economics, 1995, 39, 53–78., Jaime de Melo, and Vittorio Corbo, “The Effects of Trade Reforms on Scale and Technical

37

Efficiency,” Journal of International Economics, 1991, pp. 231–250.Unni, Jeemol, N. Lalitha, and Uma Rani, “Economic Reforms and Productivity Trends in Indian

ManufacturingEconomic Reforms and Productivity Trends in Indian Manufacturing,”Economic and Political Weekly, October 13 2001, pp. 3914–3922.

Vajpayee, Atal Bihari, “Prime Minister’s Inaugural Speech at the "Gender Poverty Summit",”November 2003.

VanBiesebroeck, Johannes, “Firm Size Matters: Growth and Productivity Growth in AfricanManufacturing,” Economic Development and Cultural Change, March 2005, 53 (3), 545–584., “Robustness of Productivity Estimates,” Journal of Industrial Economics, September 2007,

LV (3), 529–569.

Figure 1: Predicted Changes in Firm Size Distribution (Stylized)

!"#$"#%

$&'%

()'*+)%#+,&)%-./)+,-.0,1*2%

34)+%#+,&)%-./)+,-.0,1*25%67%

34)+%#+,&)%-./)+,-.0,1*25%8!%

38

Figure 2: Tariff Reforms0

50

10

01

50

Fin

al G

oo

ds T

ariff

s,

% a

d v

alo

rem

1989

1994

1999

Year

5th Percentile 25th Percentile

Median 75th Percentile

95th Percentile

Change in Final Goods Tariffs over Time

(a) Change in Tariffs

−2

50

−2

00

−1

50

−1

00

−5

00

Ch

an

ge

in

Fin

al G

oo

ds T

ariff

s,

19

89

−1

99

9

0 100 200 300 400Final Goods Tariffs, 1989

Change in Final Goods Tariffs vs. Pre−Reform Tariffs

(b) Change in Tariffs vs. Pre-Reform Tariffs

−1

00

−8

0−

60

−4

0C

ha

ng

e in

In

pu

t T

ariff

s (

% a

d v

alo

rem

), 1

98

9−

19

99

−250 −200 −150 −100 −50 0Change in Final Goods Tariffs (% ad valorem), 1989−1999

Change in Final Goods vs. Input Tariffs

(c) Change in Final Goods vs. Input TariffsPanel (a) shows the 5th, 25th, 50th, 75th, and 90th percentiles of tariffs by 3-digit National In-dustrial Code (NIC) in each year. Panel (b) shows the correlation between the change in tariffsfrom 1989-1999 and 1989 tariffs. Panel (c) shows the relationship between the changes in finalgoods and input tariffs between 1989 and 1999. Source: Author’s calculations based on variouspublications of the Government of India.

39

Figure 3: Employment, Capital, Output, and TFP in Formal and Informal Sectors0

.1.2

.3.4

.5D

en

sity

1 10 100

1000

10,0

00

Employment (log scale)

Informal Formal

Employment Distribution in 1989

(a) Employment

0.1

.2.3

De

nsity

1 10 100

1000

10,0

00

100,

000

1 m

illion

10 m

illion

100

milli

on

1 billio

n

Capital (log scale)

Informal Formal

Capital Distribution in 1989

(b) Capital

0.1

.2.3

De

nsity

10 100

1,00

0

10,0

00

100,

000

1 m

illion

10 m

illion

100

milli

on

1 billio

n

Output (log scale)

Informal Formal

Output Distribution in 1989

(c) Output

0.5

11

.5D

en

sity

−2 −1 0 1 2Total Factor Productivity

Informal Formal

Productivity Distribution in 1989

(d) ProductivityKernel density plots of employment, capital, output, and productivity in the formal and informalsectors in 1989 (pre-reform). Output and capital are measured in 1989 Rupees, while employmentmeasures the number of employees. Plots use Epanechnikov kernel function with a bandwidthof 0.8. All observations are weighted using inverse sampling probabilities, so distributions arerepresentative of the population of firms. Source: Author’s calculations based on ASI and NSSOdata. Productivity is calculated using a chain-linked index method (Aw et al. 2001).

40

Figure 4: Quantile Regression Coefficients: Firm Size−

1−

.8−

.6−

.4−

.20

.2.4

.6.8

1Q

R E

stim

ate

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95

Quantile

Effect of Fall in Final Goods Tariffs on Informal Firm Size

(a) Informal Sector

−1

−.8

−.6

−.4

−.2

0.2

.4.6

.81

QR

Estim

ate

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95

Quantile

Effect of Fall in Final Goods Tariffs on Formal Firm Size

(b) Formal SectorEffect of a fall in final goods tariffs on various quantiles of the firm size distribution. Dependentvariable is log of firm size. Solid lines show quantile regression (QR) coefficients each of 19quantiles (5th, 10th,...,95th) of the firm size distribution. Dashed lines indicate 90% confidenceintervals, based on standard errors calculated using a block bootstrap.

Figure 5: Quantile Regression Coefficients: Productivity

(a) Informal Sector

−.3

−.1

.1.3

.5.7

QR

Estim

ate

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95

Quantile

Effect of Fall in Final Goods Tariffs on Informal TFP

(b) Formal Sector

−.3

−.1

.1.3

.5.7

QR

Estim

ate

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95

Quantile

Effect of Fall in Final Goods Tariffs on Formal TFP

Effect of a fall in final goods tariffs on various quantiles of the total factor productivity (TFP)distribution. Dependent variable is log of TFP. Solid lines show quantile regression (QR)coefficients each of 19 quantiles (5th, 10th,...,95th) of the TFP distribution. Dashed lines indicate90% confidence intervals, based on standard errors calculated using a block bootstrap.

41

Figure 6: Changes in Size and Productivity Distributions−

.05

0.0

5P

red

icte

d C

ha

ng

e in

De

nsity

10 100

1,00

0

10,0

00

100,

000

1 m

illion

10 m

illion

100

milli

on

1 billio

n

Output (log scale)

Informal Formal

Change in Marginal Density of Firm Size, 1989−1999

(a) Change in Marginal Density of Firm Size

−.2

−.1

0.1

.2P

red

icte

d C

ha

ng

e in

De

nsity

−4 −2 0 2Productivity

Informal Formal

Change in Marginal Density of Productivity, 1989−1999

(b) Change in Marginal Density of ProductivityDifference between the estimated marginal density of firm size (Panel (a)) and total factorproductivity (Panel (b)) in 1989 and marginal density that would have prevailed had tariffs beendistributed as in 1999, with all other covariates distributed as in 1989. Kernel density wasestimated at the same points for both densities, using an Epanechnikov kernel using Silverman’soptimal bandwidth.

Figure 7: Non-Tariff Barriers and Tariffs by Industry Type

0.2

.4.6

.81

Ch

an

ge

in

No

n−

Ta

riff

Ba

rrie

r C

ove

rag

e R

atio

1989 1994 1999

excludes outside values

Non−Tariff Barriers

Capital and Basic Goods Intermediate Goods

Consumer Goods

(a) Non-Tariff Barriers by Industry Type

.2.4

.6.8

11

.2C

ha

ng

e in

Fin

al G

oo

ds T

ariff

s

1989 1994 1999

excludes outside values

Final Goods Tariffs by Industry Type

Capital and Basic Goods Intermediate Goods

Consumer Goods

(b) Tariffs by Industry TypeNon-tariff barriers (NTB) and tariffs by year and industry type. Box plots show the 25th, 50th, and75th percentiles of NTBs, along with upper and lower adjacent lines. Industries are divided intothree categories using the classification in Nouroz (2001). The NTB coverage ratio is the fractionof product lines within an industry requiring import licenses. Source: Author’s calculations basedon Pandey (1999) and various publication of the Government of India.

42

Figure 8: Tariffs and the Formal Employment Size Distribution−

.2−

.15

−.1

−.0

50

QR

Estim

ate

5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95

Quantile

Effect of Fall in Final Goods Tariffs on Formal Employment

(a) Quantile Regression Coefficients

−.0

2−

.01

0.0

1.0

2.0

3P

red

icte

d C

ha

ng

e in

De

nsity

1 10 100

1000

10,0

00

Employment (log scale)

Change in Density of Formal Employment, 1989−1999

(b) Changes in Counterfactual DistributionsPanel (a) presents quantile regression (QR) coefficients for the effect of a fall in final goods tariffson each of 19 quantiles (5th, 10th,...,95th) of the employment size distribution. Dashed linesindicate 90% confidence intervals, based on standard errors calculated using a block bootstrap.Panel (b) shows the difference between the estimated marginal density of formal employment in1989 and marginal density of employment that would have prevailed had tariffs been distributed asin 1999, with all other covariates distributed as in 1989. Kernel density was estimated at the samepoints for both densities, using an Epanechnikov kernel using Silverman’s optimal bandwidth.

Table 1: Summary of Predictions for a Unilateral Reduction in Final Goods Tariffs

DR MOAverage productivity + -

Average firm size + +Left tail of productivity distribution + -

Left tail of firm size distribution + 0Right tail of firm size distribution - +

Firm size distribution Compressed Stretched outChange in size of largest firms relative to mid-sized firms Less negative Less positive

Summary of changes in the firm size and productivity distributions from Demidova and Rodriguez-Clare (2009) (DR) and Melitz and Ottaviano (2008) (MO). Predictions for the size distribution arebased on the assumption that changes in domestic output dominate changes in export output.

43

Table 2: Average Tariffs by 2-Digit Manufacturing Industry

1989 1994 1999Food Products 105.38 52.57 32.76Beverages and Tobacco 157.86 133.44 78.78Cotton Textiles 87.38 65.00 37.65Wool, Silk, and Synthetic Fibers 94.55 64.26 36.70Jute and Vegetable Fibers 61.43 65.00 39.00Textile Products 102.85 65.00 39.74Wood 71.71 65.00 35.47Paper 58.25 54.14 23.65Leather 94.29 65.00 33.57Basic Chemicals 114.00 64.09 34.43Rubber, Plastic, Petroleum, and Coal 89.82 54.25 33.15Non-Metallic Mineral Products 96.86 65.00 37.06Basic Metals, Alloys 102.32 49.26 32.35Metal Products and Parts 156.52 54.99 35.22Machinery and Equipment 84.93 64.87 27.32Electrical Machinery 102.27 64.98 31.57Transport Equipment 70.27 64.36 37.32Other 92.25 63.64 34.03

Average tariffs by 2-digit manufacturing sector in 1989, 1994, and 1999. Source: Author’s calculations, based onCustoms Tariff Working Schedules published by the Government of India.

44

Table 3: Tariff Changes and Pre-Reform Industry Characteristics

Panel A: Average Firm-Level Characteristics(1) (2) (3)

Informal Formal ∆ Formallog(Average Employment) 0.036 0.047 0.014

(0.14) (0.090) (0.11)

log(Average Output) 0.022 0.050 -0.031(0.050) (0.058) (0.11)

log(Average Capital) -0.029 -0.080 0.092(0.18) (0.074) (0.100)

log(Average Capital-Employee Ratio) 0.030 0.14 0.029(0.18) (0.085) (0.10)

log(Average Wage) -0.036 0.0073 0.017(0.087) (0.12) (0.25)

log(Average Productivity) 0.015 -0.022 0.29(0.16) (0.074) (0.90)

Observations 134 137 119R2 0.005 0.055 0.028

Panel B: Overall Industry Characteristics(1) (2) (3)

Informal Formal ∆ Formallog(Total Employment) 0.014 -0.057 0.0069

(0.12) (0.063) (0.10)

log(Total Capital) -0.0085 0.028 0.10(0.044) (0.046) (0.084)

log(Total Output) 0.018 0.058 -0.030(0.046) (0.057) (0.11)

log(No. Firms) -0.028 -0.015 -0.060(0.11) (0.047) (0.100)

C4 Ratio -0.18 0.088(0.23) (0.28)

Observations 134 137 119R2 0.004 0.055 0.027

Dependent variable is the absolute value of the change in final goods tariffs from 1989 to 1999. Column (1) in bothpanels shows the correlation between pre-reform, informal sector characteristics in 1989 and tariff changes from 1989-1999. Column (2) shows the correlation between pre-reform, formal sector characteristics in 1989 and tariff changesfrom 1989-1999. Column (3) shows the correlation between changes in pre-reform, formal sector characteristics from1987-1989 and tariff changes from 1989-1999. C4 is the sum of the market shares of the four largest firms in theformal sector. Standard errors are in parentheses. *, **, and *** represent significance at the 10%, 5% and 1% levels,respectively.

45

Table 4: Summary Statistics for Firm-Level Data

Mean St. Dev. Minimum MaximumFormal SectorEmployment 130.92 454.41 1 36,988Capital (thousands of Rs. 1989) 8,705 48,400 0.00 1,950,000Output (thousands of Rs. 1989) 32,800 107,000 0.00 5,160,000No. Observations 93,768Informal SectorEmployment 3.36 5.01 1 382Capital (thousands of Rs. 1989) 48.47 264.21 0.00 40,900Output (thousands of Rs. 1989) 138.46 1,046.54 0.00 199,000No. Observations 179,417

Summary statistics for firm-level data in the formal and informal sectors. Capital and output aredeflated as discussed in Section 4. Note that the summary statistics are for the sample data, notthe estimated values for the population of firms. Source: Author’s calculations, based on ASI andNSSO data.

46

Table 5: Effect of Tariffs on Average Firm Size and Productivity

Panel A: Firm Size(1) (2) (3) (4) (5) (6)All All Formal Formal Informal Informal

Fall in Final Goods Tariffs 0.62*** 0.46** -0.13** -0.30*** 0.65*** 0.48**(0.15) (0.19) (0.051) (0.061) (0.15) (0.19)

Fall in Input Tariffs 1.03 0.98*** 1.18(1.00) (0.21) (1.05)

Observations 273185 273185 93768 93768 179417 179417R2 0.408 0.408 0.193 0.194 0.399 0.400

Panel B: Total Factor Productivity(1) (2) (3) (4) (5) (6)All All Formal Formal Informal Informal

Fall in Final Goods Tariffs 0.39*** 0.30*** -0.0046 -0.11*** 0.40*** 0.31***(0.077) (0.093) (0.020) (0.024) (0.079) (0.097)

Fall in Input Tariffs 0.57 0.59*** 0.60(0.58) (0.095) (0.61)

Observations 271728 271728 93586 93586 178142 178142R2 0.086 0.086 0.292 0.293 0.088 0.088

Final goods tariffs and input tariffs are measured as a fraction of value, and all values are negative,so that the coefficients may be interpreted as the effects of a fall in final goods or input tariffs (e.g., a100% ad valorem tariff is represented by −1). All specifications include year and industry dummyvariables. Standard errors are in parentheses, and are clustered at the state-industry level. *, **,and *** represent significance at the 10%, 5% and 1% levels, respectively.

47

Tabl

e6:

Sum

mar

yof

Res

ults

fora

Uni

late

ralR

educ

tion

inFi

nalG

oods

Tari

ffs

Pred

icte

dA

ctua

lD

RM

OIn

form

alFo

rmal

Ove

rall

Ave

rage

prod

uctiv

ity+

-+

-+

Ave

rage

firm

size

++

+-

+L

eftt

ailo

fpro

duct

ivity

dist

ribu

tion

+-

+-

+L

eftt

ailo

ffirm

size

dist

ribu

tion

+0

+-

+R

ight

tail

offir

msi

zedi

stri

butio

n-

++

0-

Firm

size

dist

ribu

tion

Com

pres

sed

Stre

tche

dou

tSh

ifte

dri

ght

Shif

ted

left

Com

pres

sed

Cha

nge

insi

zeof

larg

estfi

rms

rela

tive

tom

id-s

ized

firm

sL

ess

nega

tive

Les

spo

sitiv

eSa

me

Les

sne

gativ

eL

ess

nega

tive

48

Tabl

e7:

Tari

ffs

and

Lab

orR

egul

atio

ns

Pane

lA:F

irm

Size

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

Form

alFo

rmal

Form

alFo

rmal

Form

alIn

form

alIn

form

alIn

form

alIn

form

alIn

form

alFa

llin

Fina

lGoo

dsTa

riff

s-0

.29*

**-0

.44*

**-0

.56*

**-0

.34*

**-0

.44*

**0.

46**

0.76

**0.

74*

0.55

**0.

61**

(0.0

60)

(0.1

1)(0

.12)

(0.0

82)

(0.1

7)(0

.20)

(0.3

0)(0

.41)

(0.2

5)(0

.30)

Fall

inTa

riff

sxA

djus

t0.

23*

-0.3

2(0

.12)

(0.2

2)

Fall

inTa

riff

sxL

ayof

f0.

43**

-0.4

6(0

.19)

(0.4

9)

Fall

inTa

riff

sxR

etre

nch

0.05

4-0

.19

(0.1

1)(0

.21)

Fall

inTa

riff

sxC

lose

0.18

-0.1

8(0

.21)

(0.3

7)O

bser

vatio

ns93

768

6236

868

836

6883

668

836

1794

1710

5740

9814

998

149

9814

9R

20.

207

0.20

50.

197

0.19

70.

197

0.45

40.

460

0.47

80.

478

0.47

8Pa

nelB

:Num

ber

ofFi

rms

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

Form

alFo

rmal

Form

alFo

rmal

Form

alIn

form

alIn

form

alIn

form

alIn

form

alIn

form

alFa

llin

Fina

lGoo

dsTa

riff

s-0

.085

-0.0

850.

200.

093

0.24

**-0

.17

0.03

2-0

.38

-0.4

6-0

.25

(0.0

54)

(0.0

95)

(0.1

3)(0

.082

)(0

.12)

(0.3

1)(0

.32)

(0.4

7)(0

.36)

(0.3

6)

Fall

inTa

riff

sxA

djus

t-0

.11

0.03

1(0

.10)

(0.2

0)

Fall

inTa

riff

sxL

ayof

f-0

.42*

*0.

13(0

.20)

(0.5

3)

Fall

inTa

riff

sxR

etre

nch

-0.2

7**

0.48

(0.1

1)(0

.30)

Fall

inTa

riff

sxC

lose

-0.4

1***

-0.0

76(0

.14)

(0.3

1)O

bser

vatio

ns66

6236

1438

8238

8238

8259

2028

6231

1831

1831

18R

20.

663

0.68

80.

648

0.64

80.

648

0.74

20.

768

0.76

50.

766

0.76

5Fi

nalg

oods

tari

ffs

and

inpu

ttar

iffs

are

mea

sure

das

afr

actio

nof

valu

e,an

dal

lval

ues

are

nega

tive,

soth

atth

eco

effic

ient

sm

aybe

inte

rpre

ted

asth

eef

fect

sof

afa

llin

final

good

sor

inpu

ttar

iffs

(e.g

.,a

100%

adva

lore

mta

riff

isre

pres

ente

dby−

1).

“Adj

ust”

isa

vari

able

that

indi

cate

sho

wdi

fficu

ltit

isfo

ra

larg

efir

mto

adju

stits

empl

oym

entc

apac

ity,b

ased

onA

hsan

and

Page

s(2

007)

;pos

itive

valu

esin

dica

tegr

eate

rdiffi

culty

inad

just

men

t.“L

ayof

f”,“

Ret

renc

h”,a

nd“C

lose

”sh

owth

efr

actio

nof

appe

als

byla

rge

firm

sto

layo

ffw

orke

rs,r

etre

nch

wor

kers

,and

clos

edo

wn,

that

wer

ere

ject

edbe

twee

n19

88an

d19

92,b

ased

onda

tafr

omIn

dia’

sM

inis

try

ofL

abou

r.A

llsp

ecifi

catio

nsin

clud

eye

ar,i

ndus

try,

and

stat

edu

mm

yva

riab

les.

Stan

dard

erro

rsar

ein

pare

nthe

ses,

and

are

clus

tere

dat

the

stat

e-in

dust

ryle

vel.

*,**

,and

***

repr

esen

tsig

nific

ance

atth

e10

%,5

%an

d1%

leve

ls,r

espe

ctiv

ely.

49

Table 8: 1997 Tariffs

Panel A: Firm Size(1) (2) (3) (4)

Formal Formal Informal InformalFall in Final Goods Tariffs (1999) -0.30*** 0.48**

(0.061) (0.19)

Fall in Final Goods Tariffs (1997) -0.30*** 0.45**(0.061) (0.19)

Fall in Input Tariffs 0.98*** 0.97*** 1.18 1.25(0.21) (0.21) (1.05) (1.04)

Observations 93768 93768 179417 179417R2 0.194 0.194 0.400 0.400

Panel B: Total Factor Productivity

(1) (2) (3) (4)Formal Formal Informal Informal

Fall in Final Goods Tariffs (1999) -0.11*** 0.31***(0.024) (0.097)

Fall in Final Goods Tariffs (1997) -0.11*** 0.30***(0.024) (0.094)

Fall in Input Tariffs 0.59*** 0.59*** 0.60 0.65(0.095) (0.094) (0.61) (0.60)

Observations 93586 93586 178142 178142R2 0.293 0.293 0.088 0.088

Columns (1) and (3) use 1999 tariffs, while Columns (2) and (4) use 1997 tariffs in place of 1999tariffs. Final goods tariffs and input tariffs are measured as a fraction of value, and all values arenegative, so that the coefficients may be interpreted as the effects of a fall in final goods or inputtariffs (e.g., a 100% ad valorem tariff is represented by −1). All specifications include year andindustry dummy variables. Standard errors are in parentheses, and are clustered at the state-industrylevel. *, **, and *** represent significance at the 10%, 5% and 1% levels, respectively.

50

Table 9: Alternate TFP Measures

Panel A: Formal Sector(1) (2) (3) (4) (5) (6)

Cost Shares Revenue Shares OLS σ Low σ High Labor Prod.Fall in Final Goods Tariffs -0.11*** -0.14*** -0.085*** -0.26*** -0.14*** -0.19***

(0.024) (0.027) (0.024) (0.050) (0.029) (0.049)

Fall in Input Tariffs 0.59*** 0.77*** 0.52*** 1.08*** 0.70*** 0.66***(0.095) (0.12) (0.095) (0.18) (0.11) (0.16)

Observations 93586 93586 93768 93586 93586 93768R2 0.293 0.259 0.126 0.432 0.351 0.219

Panel B: Informal Sector

(1) (2) (3) (4) (5) (6)Cost Shares Revenue Shares OLS σ Low σ High Labor Prod.

Fall in Final Goods Tariffs 0.31*** 0.34*** 0.30*** 0.55*** 0.37*** 0.41**(0.097) (0.11) (0.084) (0.19) (0.12) (0.18)

Fall in Input Tariffs 0.60 0.62 0.88*** 1.20 0.73 0.98(0.61) (0.62) (0.25) (1.12) (0.72) (0.97)

Observations 178142 178142 179417 178142 178142 179417R2 0.088 0.076 0.236 0.111 0.094 0.387

Dependent variable is log of total factor productivity (TFP) or log of labor productivity. In Columns (1) and (2), TFPis calculated using an index number method following Aw et al. (2001), using factor cost shares and revenue shares,respectively. In Column (3), TFP is calculated as the residuals from an OLS regression of output on inputs. In Columns(4) and (5), TFP is corrected using low and high values for the elasticities of substitution, as in Hsieh and Klenow(forthcoming). In Column (6), the dependent variable is the log of labor productivity (output per employee). Finalgoods tariffs and input tariffs are measured as a fraction of value, and all values are negative, so that the coefficientsmay be interpreted as the effects of a fall in final goods or input tariffs (e.g., a 100% ad valorem tariff is representedby −1). Standard errors are in parentheses, and are clustered at the state-industry level. *, **, and *** representsignificance at the 10%, 5% and 1% levels, respectively.

51

Tabl

e10

:Oth

erIn

dust

rial

Polic

ies

Pane

lA:F

irm

Size

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Form

alFo

rmal

Form

alFo

rmal

Info

rmal

Info

rmal

Info

rmal

Info

rmal

Fall

inFi

nalG

oods

Tari

ffs

-0.3

0***

-0.2

9***

-0.2

6***

-0.2

7***

0.48

**0.

49**

0.60

***

0.42

**(0

.061

)(0

.061

)(0

.069

)(0

.071

)(0

.19)

(0.1

9)(0

.18)

(0.1

8)

Fall

inFi

nalG

oods

Tari

ffs

xN

TB

-0.0

380.

18(0

.071

)(0

.18)

Del

ic.

-0.1

6***

-0.1

5**

(0.0

49)

(0.0

77)

FDI

0.25

***

0.69

***

(0.0

60)

(0.2

0)

Fall

inIn

putT

ariff

s0.

98**

*1.

00**

*0.

73**

*0.

55**

1.18

0.95

0.80

0.72

(0.2

1)(0

.21)

(0.2

6)(0

.27)

(1.0

5)(1

.07)

(0.8

2)(0

.81)

Obs

erva

tions

9376

893

768

7508

775

087

1794

1717

9417

1326

8213

2682

R2

0.19

40.

194

0.20

30.

203

0.40

00.

400

0.40

20.

403

Pane

lB:T

otal

Fact

orP

rodu

ctiv

ity

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Form

alFo

rmal

Form

alFo

rmal

Info

rmal

Info

rmal

Info

rmal

Info

rmal

Fall

inFi

nalG

oods

Tari

ffs

-0.1

1***

-0.1

3***

-0.0

89**

*-0

.089

***

0.31

***

0.32

***

0.32

***

0.21

**(0

.024

)(0

.024

)(0

.028

)(0

.028

)(0

.097

)(0

.097

)(0

.084

)(0

.087

)

Fall

inFi

nalG

oods

Tari

ffs

xN

TB

0.05

9**

0.06

5(0

.027

)(0

.11)

Del

ic.

-0.0

42**

-0.1

4***

(0.0

19)

(0.0

49)

FDI

0.03

30.

39**

*(0

.021

)(0

.089

)

Fall

inIn

putT

ariff

s0.

59**

*0.

56**

*0.

47**

*0.

43**

*0.

600.

520.

80*

0.78

*(0

.095

)(0

.092

)(0

.13)

(0.1

4)(0

.61)

(0.6

3)(0

.44)

(0.4

3)O

bser

vatio

ns93

586

9358

675

087

7508

717

8142

1781

4213

2682

1326

82R

20.

293

0.29

30.

272

0.27

20.

088

0.08

80.

086

0.08

6Fi

nalg

oods

tari

ffs

and

inpu

ttar

iffs

are

mea

sure

das

afr

actio

nof

valu

e,an

dal

lval

ues

are

nega

tive,

soth

atth

eco

effic

ient

sm

aybe

inte

rpre

ted

asth

eef

fect

sof

afa

llin

final

good

sor

inpu

ttar

iffs

(e.g

.,a

100%

adva

lore

mta

riff

isre

pres

ente

dby−

1).

“NT

BL

iber

aliz

ed”

isa

dum

my

vari

able

that

take

son

ava

lue

of1

ifth

ech

ange

inan

indu

stry

’sno

n-ta

riff

barr

ier(

NT

B)c

over

age

ratio

isgr

eate

rtha

nth

em

edia

nbe

twee

n19

89an

d19

99,0

othe

rwis

e.N

TB

data

are

from

Pand

ey(1

999)

.“D

elic

”is

adu

mm

yva

riab

leth

atta

kes

ona

valu

eof

1if

the

indu

stry

isde

licen

sed,

0ot

herw

ise.

“FD

I”is

the

frac

tion

ofpr

oduc

tlin

esin

anin

dust

ryth

atal

low

FDI

inflo

ws

upto

51%

with

outc

ase-

by-c

ase

appr

oval

.D

ata

onde

licen

sing

and

FDI

refo

rmar

efr

omA

ghio

net

al.(

2008

).A

llsp

ecifi

catio

nsin

clud

eye

aran

din

dust

rydu

mm

yva

riab

les.

Stan

dard

erro

rsar

ein

pare

nthe

ses,

and

are

clus

tere

dat

the

stat

e-in

dust

ryle

vel.

*,**

,and

***

repr

esen

tsig

nific

ance

atth

e10

%,5

%an

d1%

leve

ls,r

espe

ctiv

ely.

52

Table 11: Tariffs and Export Orientation

Panel A: Firm Size(1) (2) (3) (4) (5) (6)All All Formal Formal Informal Informal

Fall in Final Goods Tariffs 0.46** 0.40** -0.30*** -0.33*** 0.48** 0.44**(0.19) (0.20) (0.061) (0.063) (0.19) (0.20)

Fall in Final Goods Tariffs x Export Share -0.032 0.29** -0.085(0.14) (0.12) (0.14)

Fall in Input Tariffs 1.03 1.51** 0.98*** 1.01*** 1.18 1.70**(1.00) (0.69) (0.21) (0.21) (1.05) (0.71)

Observations 273185 261823 93768 91988 179417 169835R2 0.408 0.418 0.194 0.195 0.400 0.412

Panel B: Total Factor Productivity

(1) (2) (3) (4) (5) (6)All All Formal Formal Informal Informal

Fall in Final Goods Tariffs 0.30*** 0.20** -0.11*** -0.13*** 0.31*** 0.21**(0.093) (0.091) (0.024) (0.026) (0.097) (0.094)

Fall in Final Goods Tariffs x Export Share 0.067 0.15*** 0.055(0.059) (0.051) (0.060)

Fall in Input Tariffs 0.57 1.02*** 0.59*** 0.61*** 0.60 1.07***(0.58) (0.32) (0.095) (0.097) (0.61) (0.34)

Observations 271728 260366 93586 91806 178142 168560R2 0.086 0.086 0.293 0.282 0.088 0.088

Final goods tariffs and input tariffs are measured as a fraction of value, and all values are negative, so that the co-efficients may be interpreted as the effects of a fall in final goods or input tariffs (e.g., a 100% ad valorem tariff isrepresented by −1). “Export Share” is the share of industry exports in output in 1989, based on the World Bank’sTrade and Production database. All specifications include year and industry dummy variables. Standard errors are inparentheses, and are clustered at the state-industry level. *, **, and *** represent significance at the 10%, 5% and 1%levels, respectively.

53

Table 12: Tariffs and Labor-Intensive Industries

Panel A: Firm Size(1) (2) (3) (4) (5) (6)All All Formal Formal Informal Informal

Fall in Final Goods Tariffs 0.46** 0.47** -0.30*** -0.32*** 0.48** 0.53**(0.19) (0.21) (0.061) (0.062) (0.19) (0.21)

Fall in Final Goods Tariffs x Comp. Adv. -0.0026 0.47** -0.090(0.15) (0.19) (0.15)

Fall in Input Tariffs 1.03 1.03 0.98*** 1.04*** 1.18 1.12(1.00) (1.01) (0.21) (0.21) (1.05) (1.05)

Observations 273185 273185 93768 93768 179417 179417R2 0.408 0.408 0.194 0.194 0.400 0.400

Panel B: Total Factor Productivity

(1) (2) (3) (4) (5) (6)All All Formal Formal Informal Informal

Fall in Final Goods Tariffs 0.30*** 0.22* -0.11*** -0.11*** 0.31*** 0.24*(0.093) (0.12) (0.024) (0.024) (0.097) (0.12)

Fall in Final Goods Tariffs x Comp. Adv. 0.15* 0.025 0.13(0.086) (0.063) (0.087)

Fall in Input Tariffs 0.57 0.68 0.59*** 0.60*** 0.60 0.69(0.58) (0.62) (0.095) (0.096) (0.61) (0.65)

Observations 271728 271728 93586 93586 178142 178142R2 0.086 0.087 0.293 0.293 0.088 0.089

Panel C: Employment

(1) (2) (3) (4) (5) (6)All All Formal Formal Informal Informal

Fall in Final Goods Tariffs 0.069 0.079 -0.11*** -0.13*** 0.071 0.11(0.083) (0.068) (0.029) (0.030) (0.083) (0.070)

Fall in Final Goods Tariffs x Comp. Adv. -0.018 0.46*** -0.065(0.075) (0.095) (0.076)

Fall in Input Tariffs 0.12 0.11 0.32*** 0.38*** 0.20 0.16(0.27) (0.26) (0.11) (0.11) (0.27) (0.26)

Observations 273185 273185 93768 93768 179417 179417R2 0.188 0.188 0.149 0.149 0.150 0.150

Final goods tariffs and input tariffs are measured as a fraction of value, and all values are negative, so that the co-efficients may be interpreted as the effects of a fall in final goods or input tariffs (e.g., a 100% ad valorem tariff isrepresented by −1). “Comp. Adv.” is a dummy variable that takes on a value of 1 if an industry’s capital-employeeratio in 1989 was lower than the median capital-employee ratio, 0 otherwise. All specifications include year and indus-try dummy variables. Standard errors are in parentheses, and are clustered at the state-industry level. *, **, and ***represent significance at the 10%, 5% and 1% levels, respectively.

54

A Appendix: Measuring Productivity

I construct a measure of total factor productivity (TFP) for each firm using an index number methodsuggested by Aw et al. (2001), who show that the log of TFP (hereafter simply referred to as TFP)for firm i in industry j in year t can be calculated as follows:

TFPijt = (qijt − qjt)︸ ︷︷ ︸deviation from avg. q

+t∑

r=2

(qjr − qjr−1)︸ ︷︷ ︸yearly change in q

−[ K∑

k=1

12

(Skijt + Sk

jt)(kijt − kjt)︸ ︷︷ ︸deviation from avg. k

+t∑

r=2

K∑k=1

12

(Skjr + Sk

jr−1)(kr − kr−1)︸ ︷︷ ︸yearly change in k

]

whereqijt=log of outputSkijt=cost share of input kkijt=log of input kA firm’s TFP is the deviation in its output from average output in that year, along with how averageoutput in that year differs from the base year, minus the deviation of the firm’s inputs from averageinputs in that year, along with how average inputs in that year differ from the base year. Bars overvariables indicate average values within a particular industry and/or year.

Output is calculated as discussed in Section 4. To deflate raw material inputs, I use India’sIOTT to calculate the average deflator for each industry, using the technique described for inputtariffs (Section 3). Fuel and electricity are deflated by the energy price index and are added to rawmaterial inputs to calculate total material inputs. To calculate cost shares, I estimate total inputcosts as follows. Labor costs are the wage bill, and are deflated using the consumer price index.25

Material costs are total costs of fuel, electricity, and raw materials, deflated as discussed above.Capital costs are constructed by multiplying real capital stock by the rental rate of capital. Forcapital rental rates, I use information on rented capital value and payments from the informal firmsurveys; I calculate the average rental rates among firms that rent capital in each industry and year,and apply the average rental rate to all firms in that industry/year. Given the inherent difficulties incalculating capital rental rates, particularly when using only informal sector data, I also calculateTFP using revenue shares instead of cost shares. These two measures of TFP are highly correlated(ρ=0.936). Section 5.4 shows that the results are robust to the use of cost shares or revenue shares,as well as to several other measures of productivity.

25For own-account enterprises in the informal sector (those that do not hire non-family labor), I impute per-employeecost as the average cost per employee among informal firms that do hire labor in the same state and year.

55