Efficiency change and growth in productivity: the Asian growth experience

Efficiency change and growth in productivity:the Asian growth experience

Ching-Cheng Changa,*, Yir-Hueih Luhb

aAssociate Research Fellow, Institute of Economics, Academia Sinica, No. 128, Yen-chiou Yuan Road, Sec. 2,Taipei 115, Taiwan

aAssociate Professor, Department of Agricultural Economics, National Taiwan University, Taipei, Taiwan.bProfessor, Department of Economics, National Tsing Hua University, Hsin-Chu, Taiwan

Received 1 July 1998; received in revised form 1 September 1999; accepted 1 December 1999

Abstract

This paper focuses on identifying the sources of productivity growth in ten Asian economiesincluding China, Japan, the NIEs and the ASEAN-4. We calculate productivity growth and itscomponents using distance-function-based Malmquist productivity indexes following Fa¨re,Grosskopf, Norris, and Zhang (1994a). Hong Kong and Singapore are found to have thecapabilities to shift the grand frontier of the APEC economies. But the productivity divergencemight have occurred since the 70’s. The FDI contributes to the Asian growth either throughcatching-up or through technological innovations when a sufficient learning capacity is available inthe host economy. © 2000 Elsevier Science Inc. All rights reserved.

JEL classification:C43; D24

Keywords: Productivity growth; Efficiency; Technical change; Human capital

1. Introduction

This paper intends to use the Malmquist index to identify the sources of productivitygrowth in ten Asian economies including China, Japan, the East Asian Newly Industrializedeconomies (NIEs) and the Association of South East Asian Nations (ASEAN). The strongpatterns of convergence in total factor productivity (TFP) for the OECD countries are well

* Corresponding author. Tel.:1886-2-2782-2791; fax:1886-2-2785-3946.E-mail address:[email protected]. sinica.edu.tw (C-C Chang).

Journal of Asian Economics 10 (2000) 551–570

1049-0078/00/$ – see front matter © 2000 Elsevier Science Inc. All rights reserved.PII: S1049-0078(00)00032-4

documented and have been widely discussed in the literature (e.g., Baumol, 1986; Dowrick& Nyugen, 1989). However, the causes of rapid Asian economic growth and its sustainabilityhave generated considerable debate among the scholars since the early 90’s (e.g., WorldBank, 1993; Krugman, 1994; Kim & Lau, 1994, 1995; Young, 1992, 1994, 1995; Chen,1997). The key questions such as what is the engine of growth of the NIEs or the ASEAN-4and what are the major differences and similarities in the growth patterns between the regionswarrant further discussion.

Based on the economic theory of production, productivity is generally defined in terms ofthe efficiency with which inputs are transformed into outputs in the production process.Indexes of productivity, therefore, are simply the ratios of an aggregate output index to anindex of either single or total factor use. The TFP approach seeks to evaluate the independentinfluences of technical change and factor substitution. Even though it represents a clearimprovement over the single factor measures by accounting for changes in the quantity andquality of all production factors, the conventional Divisia indexing procedure has the samedifficulty in distinguishing movements towards the production frontier from shifts in theproduction frontier itself.

Theory underlying the Divisia formula consists of a production function describing thetechnical substitutability between inputs in the production process. Since the productionfunctions are defined on continuous time and economic data generally comes in a discretetime form, one of the major problems of measuring productivity in practice concerns findingthe appropriate method to make a discrete approximation to the Divisia formula.

The most popular form of approximation for the Divisia formula in the past is theTornqvist index. The To¨rnqvist index calculates TFP growth based on information concern-ing prices, and uses cost/revenue shares as weights to aggregate inputs/outputs. However,when calculating the To¨rnqvist index, observed output is assumed to be equivalent to frontieroutput. Consequently, decomposition of the growth in productivity into the movementstowards and shifts in the production frontier is not possible.

The other TFP index that is defined consistently directly for discrete data is the Malmquistindex. The Malmquist index has gained considerable popularity in recent years since Fa¨re,Grosskopf, Norris, and Zhang (1994a) proposed applying the linear-programming approachto calculate the distance functions that make up the Malmquist index. There may be threereasons for this increasing popularity. First of all, since the data envelopment type of analysiscan be directly applied to calculate the index, the Malmquist index has the advantage ofcomputational ease. Second, calculation of the Malmquist index does not require informationon cost or revenue shares to aggregate inputs or outputs. Consequently, the Malmquist indexis less data-demanding than the To¨rnqvist index. Finally, the Malmquist productivity-changeindex is more general in that it allows for further decomposition of productivity into changesin efficiency and changes in technology. This further decomposition is important for facil-itating a multilateral comparison that may help explain and characterize the differences andsimilarities in growth patterns for different regions.

The remainder of the paper is organized as follows. In the next section, we give a briefreview of the distance-function-based Malmquist productivity indexes and the linear pro-gramming method outlined in Fa¨re, Grosskopf, and Lovell (1994b). The third sectiondescribes the empirical data. The data of our multilateral comparison comes from the Penn

552 Chang and Luh / Journal of Asian Economics 10 (2000) 551–570

World Tables over the two decades 1970–80 and 1980–90. Section 4 presents the empiricalresults. Since the method constructs a best-practice frontier from the sample economies, theresults not only allow us to compare the pattern of productivity growth and its componentsbut also to identify economies shifting the frontier over time (i.e., the “innovators”). Section5 discusses the role of human capital and knowledge in TFP growth among these economies.The concluding comments are offered in the last section.

2. Decomposition Of the Malmquist Index

For each time periodt 5 1,. . .,T, the Malmquist index is based on the output distancefunction defined as1

DT~ xt, yt! ; inf $u : ~ xt, yt/u ! [ ST% (1)

where superscriptT denotes the technology reference period,ST is the technology set,xt isa vector of inputs andyt is a vector of outputs. Following Fa¨re et al. (1994a), the Malmquistproductivity-change index is defined as the geometric mean of two distance-function-basedMalmquist productivity indexes and is of the following form,

M~ xt11, yt11,xt, yt! 5 @Mt z Mt11#1/2

5 FSDt~ xt11, yt11!

Dt~ xt, yt! DSDt11~ xt11, yt11!

Dt11~ xt, yt! DG1/2

(2)

In Eq. (2), the Malmquist productivity index with technology in periodt as the referencetechnology is defined as2

Mt 5Dt~ xt11, yt11!

Dt~ xt, yt!(3)

where the distance function in the numerator,Dt( xt11, yt11), measures the maximal pro-portional change in output required to make (xt11, yt11) feasible in relation to the tech-nology att 5 1, whereas the distance function in the denominator,Dt( xt, yt), measures thereciprocal of the maximum proportional expansion of the output vectoryt given input vectorxt. The output distance functions are assumed to be closed, bounded, convex and to satisfystrong disposability properties.

Similarly, the Malmquist productivity index with technology in periodt 1 1 as thereference technology is defined as

Mt11 5Dt11~ xt11, yt11!

Dt11~ xt, yt!(4)

In Eq. (4), the distance function in the denominator,Dtt( xt, yt), measures the maximalproportional change in output required to make (xt, yt) feasible in relation to the technologyat t 1 1, whereas the distance function in the numerator,Dt11( xt11, yt11), measures the

553Chang and Luh / Journal of Asian Economics 10 (2000) 551–570

reciprocal of the maximum proportional expansion of the output vectoryt 1 1 given inputvectorxt 1 1.

The Malmquist index can be calculated through the linear-programming approach out-lined in Fare, Gorsookopf, and Lovell (1994b). Since for each sample economy there is onlyone aggregate output, the output distance function is equivalent to a frontier productionfunction in the sense that the frontier gives the maximum output. Therefore, the nonpara-metric programming technique involves constructing a world or best-practice frontier fromthe data in the sample, and then compares individual economy to the frontier. The frontiertechnology in periodt, St, can be constructed from the data as

St 5 $~ xt, yt! : ymk9,t # O

k51

K

zk,tymk9,t, m 5 1,. . .,M;

Ok51

K

zk,txnk,t # xn

k9,t , n 5 1,. . .,N;

zk,t $ 0, , k 5 1,. . .K} (5)

wherezk,t is an intensity variable indicating at what intensity economyk may be employedin production.

To calculate the Malmquist index of productivity change for economyk9, the linear-programming approach solves for four different distance functions that make up the index,that is, Dt( xk9,t, yk9,t), Dt11( xk9,t11, yk9,t11), Dt( xk9,t11, yk9t11), and Dt11( xk9,t, yk9t).The output distance functions are reciprocal to the output-based Farrell measure of technicalefficiency. Calculating the Malmquist index relative to the constant-returns-to-scale (CRS)technology, the four different linear programming problems can be stated as

~Dct 1 j~ xk9,t1j, yk9,t1j!! 5 maxuk9

subject to

uk9 ymk9,t 1 j # O

k51

K

zk,t1i ymk,t 1 i, m 5 1,. . .,M; (6)

Ok51

K

zk,t1ixnk,t 1 i # xn

k9,t 1 j, n 5 1,. . .,N;

zk,t1i $ 0, k 5 1,. . .,K;

where (i , j ) 5 (0,0) for solving for (Dct ( xk9,t, yk9,t))21;

(i , j ) 5 (1,1) for solving for (Dct11( xk9,t11, yk9,t11))21;

(i , j ) 5 (0,1) for solving for (Dct ( xk9,t11, yk9,t11))21; and

(i , j ) 5 (1,0) for solving for (Dct11( xk9,t, yk9,t))21.

The subscriptc denotes the CRS benchmark technologies. The assumption of CRS technol-ogy can be relaxed to allow for variable returns to scale (VRS) by adding the followingconstraint:


Ok51

K

zk,t # 1 (7)

The Malmquist productivity-change index in Eq. (2) can be decomposed into the change inrelative efficiency and shift in technology over time (Fa¨re et al., 1992). In empiricalapplications, there are alternative ways to decompose the same Malmquist index (Ray &Desli, 1997; Fa¨re, Grosskopf, & Norris, 1997a). Here, we employ the decompositionproposed by Ray and Desli (1997) in which VRS is assumed to be the benchmark technologywhen calculating technical change.3

Let subscriptn denote the VRS technologies. Following Banker (1984), the ratio ofDn

t ( xt, yt) andDct ( xt, yt) defines the scale efficiencySEt (xt, yt) for economyk in period t.

The Ray and Desli decomposition is written as follows:

M~ xt11, yt11,xt, yt! 5

@Technical change]3[pure efficiency change]3[scale efficiency change]

FS Dnt ~ xt, yt!

Dnt 1 1~ xt, yt!DS Dn

t ~ xt11, yt11!

Dnt 1 1~ xt11, yt11!DG

1/2FDnt 1 1~ xt11, yt11!

Dnt ~ xt, yt! G (8)

FSSEt~ xt11, yt11!

SEt~ xt, yt! DSSEt11~ xt11, yt11!

SEt11~ xt, yt! DG1/2

The geometric mean of the two ratios inside the first brackets in Eq. (5) can be interpretedas the technical change component, which measures the shift in the frontier over time. Howmuch the world frontier shifts at each economy’s observed input mix is measured by thiscomponent. The improvements in this technical-change component can be interpreted asproviding evidence of innovation (Fa¨re et al., 1994) for the economy considered. A furtherexamination of this component thus allows for identifying the innovators.

The expression in the second bracket illustrates the change in “pure” technical efficiencybased on the VRS technology, and thus measures the extent to which observed productionis getting closer (or farther) from the frontier. For a multilateral analysis, the frontier is a“grand” or “world” frontier, which is constructed by the best practice economies in thesample. The efficiency change component, therefore, captures the performance relative to thebest practice in the sample and can be interpreted as the catching-up effect.

Scale efficiency in a given period captures the deviations between the variable returnstechnology and the constant returns technology at observed input levels. The third bracketmeasures changes in scale efficiency between periodt and t11 due to a movement towardor away from the point of optimal scale. It is a geometric mean of the ratios of scaleefficiencies of the two bundles (xt, yt) and x(xt11, yt11) using in turn the VRS technologiesfrom the two periods as the benchmark (Ray and Desli).

Following Fare and Grosskopf (1996) and Fa¨re et al. (1997b), the technical change indexcan also be decomposed into the product of a magnitude index and a bias index as follows:


Technical Change

5 @Magnitude Index] 3 [Bias Index] (9)

5 F Dnt ~ xt, yt!

Dnt 1 1~ xt, yt!G 3 F Dn

t ~ xt11, yt11!

Dnt 1 1~ xt11, yt11!

3Dn

t 1 1~ xt, yt!

Dnt ~ xt, yt! G .

The magnitude index measures the magnitude of technical change along a ray using data forperiodt. The bias index measures the bias of technical change as the ratio of the magnitudeof technical change along a ray through periodt11 data to that along a ray through periodt data. Therefore, if the magnitude indexes are the same during the two periods, there wouldnot be any contribution of the bias index to productivity change.

The bias index can be further expressed as the product of an output bias index and an inputbias index as follows (Fa¨re et al., 1997b):

Bias Index

5 @Output Bias Index] 3 [Input Bias Index]

5 F Dnt ~ xt11, yt11!

Dnt 1 1~ xt11, yt11!

3Dn

t 1 1~ xt11, yt!

Dnt 1 1~ xt11, yt!G

1/2

(10)

3 FDnt 1 1~ xt, yt!

Dnt ~ xt, yt!

3Dn

t ~ xt11, yt!

Dnt 1 1~ xt11, yt!G

1/2

.

For the one-output case as in our study, output bias index will be equal to one. The input biasindex is a geometric mean of the shift in technology between periodt andt11 evaluated atperiodt input-output level and the shift in technology observed in periodt11 input level withoutput remaining at periodt level. Since the output level does not change, the only thing thatchanges is the input level. Furthermore, if the shift in technology is not neutral, the squareroot of either output bias index or input bias index will not equal unity.

It is important to note that when solving forDct ( xk9,t11, yk9,t11), Dc

t11( xk9,t, yk9,t),. . .,etc., the linear-programming problems involve observations from both periodt and periodt11 because the reference technology relative to which the given input-output mix isevaluated is constructed from observations at the other period. In our empirical work, foreach economy, twelve linear programming problems are solved for every two consecutiveyears. Therefore, a total of 4,560 linear programming problems are solved for our results.

3. Data

Recognizing the fact that the ten selected Asian economies are all APEC members, oursample includes nineteen Pacific Rim economies including Australia, Canada, Chile, China,Columbia, Hong Kong, Indonesia, Japan, Korea, Malaysia, Mexico, New Zealand, PapuaNew Guinea, Peru, the Philippines, Singapore, Taiwan, Thailand, and the United States forwhich consistent quantity data are collected. The Vietnam and Russia are excluded due to


lack of data. The data of our multilateral comparison comes from the Penn World Tables,which are available through the internet website: www.bized.ac.uk/dataserv/datahome.htm,over the 1965-90 period. Details on the procedures used to create the data set can be foundin Summers and Heston (1991). The method used involves constructing a best-practicefrontier from the sample.

Our model is a simple single-output, two-input model. Real gross domestic product (GDP)is used as the output measure whereas labor and non-residential capital are two aggregateinput proxies. The real GDP measure is real in the sense that it is measured in 1985international prices. The two input proxies are not directly available from the data set. Thelabor variable is calculated by dividing real GDP (RGDP) with real GDP per worker(RGDPW).4 The non-residential capital stock variable (K) is retrieved by multiplyingnonresidential capital stock per worker by labor. However, for China, Indonesia, Malaysia,Papua New Guinea, and Singapore, the data on nonresidential capital stock per worker is notavailable. They are calculated using the perpetual inventory method, that is,

Kt 5 I t 1 ~1 2 d! Kt21, (11)

whereIt denotes non-residential gross investment5 at timet andd is the depreciation rate. Wehave assumed a 10% depreciation rate for these five economies. Since their initial capitalstock, K1965, is also not available, it is estimated by 0.5 (which is the sample mean of theratio of non-residential capital to real GDP in 1965) times their real GDP in 1965.

The average annual growth rates of real GDP, capital and labor for each sample economyare reported in Table 1. To take into account possible structural changes, our discussion isbased on two separate periods, that is, 1970-80 and 1980-90. The summary statistics indicatethat growth in real GDP on average is 6 percent per year over the 1970-80 period. Growthin the second period, however, is less satisfactory. As seen in Table 1, for all sampleeconomies, growth of real GDP over time is lower, on average, in the second period than thatin the first period. Consequently, growth in real GDP averaged 4 percent over the 1980-90period. It is worth noting that, for both periods, the economy having the highest averageannual growth in real GDP among the 19 APEC members is in Asia. Singapore and Korea,respectively, had the highest average annual growth in real GDP over the first and secondperiods. Capital exhibits a similar growth pattern with Singapore, Indonesia, Malaysia, Koreaand Taiwan having the highest average growth rate over the first and second period. On theother hand, the average labor growth is rather modest and even across all APEC economies.

Table 2 presents the capital-labor ratios for the years 1970, 1980 and 1990. The ratios ofmost of the APEC economies increase over time. In 1970, only Australia, Canada, NewZealand, and USA had high (above 2.0) capital-labor ratio. In 1980, Japan and Singaporeemerged quickly into the high-ratio group. By 1990, Singapore ranked the second and Japanranked the fourth among the 19 economies. Taiwan also emerged into the high-ratio group.

4. Decomposition results

Geometric means of the Malmquist productivity-change indexes and the two componentsof growth for each sample economy are listed in Table 3. Annual results are available from


the authors upon request. Values exceeding unity imply improvements in the relevantperformance, while values less than unity denote regress or deterioration in performance. Itis clear from the table that the average performance of each economy over the second periodis better than in the first period, except for the Philippines, Indonesia, and four of the SouthAmerican economies. As for an economy-to-economy comparison, Hong Kong has thehighest TFP change in the sample during both periods. The rates of productivity change forKorea and Singapore are also quite high for the 1980-90 period. Basically, our empiricalfinding is in accordance with the Young (1994) study that the aggregate productivity growthin the NIEs is not extraordinarily high with the possible exception of Hong Kong and Korea.Our results, however, are contradictory to those in Kim and Lau (1994, 1995) where theyfound no productivity growth in the NIEs. A possible reason for the apparent difference isthat we do not have human capital as the input variable. Therefore, part of the growth inproductivity may be due to the effect of human capital. We will explore this issue in the nextsection.

Further comparison can be made based on either the NIEs, including Hong Kong, Korea,Singapore and Taiwan, or the ASEAN-4, including Indonesia, Malaysia, Philippines andThailand. Results in Table 3 suggest that the NIEs does not always outperform the ASEAN-4

Table 1Average annual growth rate of real GDP, capital, and labor: 1970–1980 and 1980–1990

Real Gross Domestic Product Labor Capital

1970–1980 1980–1990 1970–1980 1980–1990 1970–1980 1980–1990

East Asian economiesChina 0.05 0.05 0.02 0.02 0.10 0.07Japan 0.04 0.04 0.01 0.01 0.10 0.06

NIEsHong Kong 0.09 0.07 0.04 0.02 0.08 0.02Korea 0.08 0.09 0.03 0.02 0.12 0.08Singapore 0.10 0.07 0.04 0.02 0.16 0.08Taiwan 0.09 0.08 0.03 0.02 0.14 0.09

ASEAN-4Indonesi 0.08 0.06 0.02 0.02 0.14 0.12Malaysia 0.08 0.06 0.04 0.03 0.14 0.08Philippines 0.06 0.02 0.02 0.03 0.04 0.03Thailand 0.06 0.07 0.03 0.02 0.10 0.06

Other APEC economiesIndustrialized

Australia 0.03 0.03 0.02 0.02 0.04 0.04Canada 0.05 0.03 0.03 0.01 0.05 0.05New Zealand 0.02 0.02 0.02 0.02 0.04 0.03USA 0.03 0.03 0.02 0.01 0.04 0.03

DevelopingChile 0.02 0.03 0.02 0.02 0.03 0.06Columbia 0.06 0.03 0.03 0.03 0.06 0.03Mexico 0.07 0.02 0.04 0.03 0.08 0.02Papua N. Guinea 0.02 0.00 0.02 0.02 0.03 0.02Peru 0.03 20.01 0.03 0.03 0.04 0.02

Average 0.06 0.04 0.03 0.02 0.08 0.05

Sources: Calculated from Penn World Tables (Summers & Heston 1991).


during the 1970-80 period except Hong Kong. The average performance of the Philippinesis even better than three of the NIEs, that is, Korea, Singapore and Taiwan. However, for the1980-90 period, the NIEs show significant improvements in productivity over time withMalmquist indexes all greater than unity. In addition, it is over this period that the NIEsoutperform the ASEAN-4 since the Malmquist indexes for three ASEAN-4 economies arelower than one. Thailand is the only ASEAN-4 to have a positive productivity growth duringthe same period.

An intra-group comparison suggests that among the NIEs, for the 1970-80 period, HongKong has the highest TFP change. For the 1980-90 period, Hong Kong still outperforms theother NIEs. Korea and Singapore show an outstanding progress and follow closely in order.Among the ASEAN-4, it is clear from Table 3 that the order based on the averageperformance remains largely unchanged during the two periods, except that Thailand re-places Philippines and becomes the top performer.

Turning to the sources of productivity growth, the results in Table 3 suggest that duringthe 1970’s, most of the Asian economies experience either technical regress or efficiency loss,and thus deterioration in productivity. The only exceptions are Singapore, Hong Kong and thePhilippines who perform rather well in catching up. Productivity in Singapore deterioratesbecause technical regress dominates efficiency gain. As for Hong Kong and the Philippines, thegain in efficiency dominates technical regress leading to positive growth in productivity.

Table 2Capital-Labor ratio: 1970, 1980, 1990

1970 1980 1990

East Asian economiesChina 0.09 0.19 0.29Japan 0.94 2.21 3.65

NIEsHong Kong 0.86 1.22 1.28Korea 0.39 0.98 1.80Singapore 0.79 2.35 4.14Taiwan 0.50 1.37 2.57

ASEAN-4Indonesia 0.09 0.27 0.65Malaysia 0.43 0.91 1.33Philippines 0.32 0.37 0.37Thailand 0.18 0.34 0.49


Australia 2.56 3.11 3.79Canada 2.31 2.89 4.27USA 2.33 2.76 3.47

DevelopingChile 0.63 0.70 0.95Columbia 0.84 1.18 1.27Mexico 0.92 1.40 1.29Papua N. Guinea 0.32 0.35 0.22Peru 0.81 0.90 0.88

Sources: Calculated from Penn World Tables (Summers & Heston 1991).


During the 80’s, the results in Table 3 indicate Japan and all four economies of the NIEsexperience efficiency progress. Once again, the catching-up effect outperforms the innova-tion effect, except in Hong Kong. In comparison, some of the ASEAN-4 experience technicalprogress but no efficiency gain in this period. China and Indonesia remain lack of progressin both technical and efficiency aspects. Thailand, being an exception in ASEAN-4, expe-riences productivity growth largely from technical progress.

In order to characterize the differences and similarities in growth patterns for the Asianeconomies, further decomposition results of efficiency change and technical change arereported in Tables 4 and 5. For the NIEs, Hong Kong registers pure efficiency gains in the70’s, followed by technical progress in magnitude. In comparison, Singapore outperformsthe other economies in enhancing scale efficiency during both decades. The other two NIEs,Korea and Taiwan, show significant progress in the 80’s, but the sources of their growth arequite different. The sources for Korea are mainly from the catching-up effect, one half pureefficiency and the other half scale improvement. As for Taiwan, technical progress isidentified as one of the major sources of growth. Therefore, technical progress does not playan equal role among the NIEs.

Table 3Decomposition of Malmquist Productivity Index

Malmquist Technical change Efficiency change

1970–1980 1980–1990 1970–1980 1980–1990 1970–1980 1980–1990

East Asian economiesChina 0.959 0.980 0.985 0.993 0.974 0.987Japan 0.968 1.003 0.988 0.997 0.979 1.005

NIEsHong Kong 1.024 1.047 0.980 1.035 1.045 1.012Korea 0.970 1.034 0.988 0.997 0.981 1.037Singapore 0.974 1.031 0.917 0.994 1.063 1.037Taiwan 0.975 1.011 0.987 1.003 0.988 1.008

ASEAN-4Indonesia 0.954 0.951 0.970 0.998 0.983 0.953Malaysia 0.962 0.992 0.983 1.011 0.978 0.981Philippines 1.017 0.992 0.986 1.010 1.031 0.983Thailand 0.970 1.007 0.982 1.007 0.988 1.000


Australia 1.008 1.011 1.003 1.013 1.005 0.998Canada 1.008 1.018 0.997 1.014 1.011 1.004New Zealand 0.987 1.000 0.984 0.976 1.003 1.025USA 0.997 1.007 1.007 1.009 0.989 0.998

DevelopingChile 0.992 0.974 0.990 1.006 1.002 0.968Columbia 1.002 1.001 0.993 1.006 1.008 0.995Mexico 0.996 0.994 0.993 1.000 1.003 0.994Papua N. Guinea 0.989 0.984 0.993 0.985 0.996 0.999Peru 0.990 0.973 0.993 1.014 0.997 0.960

Note: These numbers are the goemetric means of annual productivity indexes in each economy.


For the ASEAN-4, the Philippines is the only economy with positive productivity growthin the 70’s due to pure efficiency improvements. In the 80’s, Thailand becomes the soleeconomy with positive growth in both technical change and pure efficiency. The Philippinesand Malaysia exhibit productivity regress due to deterioration in pure efficiency, despite theirprogress in technical change. Indonesia experiences technical regress in the 70’s, followedby pure efficiency loss in the 80’s. Likewise, pure efficiency loss is the main reason whyChina exhibits productivity regress in both decades. Therefore, in order to promote economicgrowth in the ASEAN-4 and China, the priority should be in their catching-up capability.

Japan exhibits a different growth pattern. In the 70’s, it has shown productivity regress dueto deterioration in both catching-up and technical regress. Situation improved in the 80’smainly due to the scale efficiency factor. Therefore, scale efficiency seems to be the only andmost important driving force for Japan’s economic growth in those two decades.

According to proposition by Fa¨re et al. (1997b), input bias makes no contribution toproductivity change under a very stringent set of conditions including constant-returns-to-scale technology and implicit Hicksian input-neutral technical change. Our results imply thatChina and the Philippines exhibit constant-returns-to-scale technologies as well as input-neutral technical changes. But increases in input bias contribute positively to the perfor-mance of Hong Kong, Japan, Malaysia, Singapore and Thailand in technical progress.

Table 4Decomposition of efficiency change

Scale efficiency Pure efficiency

1970–1980 1980–1990 1970–1980 1980–1990

East Asian economiesChina 0.998 0.998 0.976 0.989Japan 1.015 1.008 0.965 0.998

NIEsHong Kong 1.008 1.002 1.037 1.009Korea 0.999 1.017 0.982 1.019Singapore 1.063 1.037 1.000 1.000Taiwan 0.999 1.008 0.989 1.000

ASEAN-4Indonesia 0.990 0.994 0.994 0.959Malaysia 1.003 1.001 0.975 0.980Philippines 0.997 0.997 1.034 0.985Thailand 0.992 0.993 0.996 1.007


Australia 1.001 1.001 1.004 0.997Canada 0.999 1.000 1.012 1.004New Zealand 1.003 1.025 1.000 1.000USA 0.989 0.998 1.000 1.000

DevelopingChile 1.002 1.004 1.000 0.964Columbia 1.001 0.998 1.008 0.996Mexico 1.003 0.994 1.000 1.000Papua N. Guinea 0.996 0.999 1.000 1.000Peru 0.998 0.998 0.999 0.961


In addition to investigating whether it is change in efficiency or technology that contrib-utes to the growth in productivity for the Asian economies, decomposition of the productivitychange allows identifying the innovators who actually cause the best-practice frontier toshift. Like Fare et al. (1994), we use the following conditions to identify the innovators undertwo alternative benchmark assumptions,

TCk . 1,

Dt~ xk,t11, yk,t11! . 1,

Dt11~ xk,t11, yk,t11! 5 1.

Economies satisfying the three conditions outlined above can be regarded as having con-tributed to a shift in the frontier between periodt andt 1 1. Among the 17 OECD countries,Fare et al. (1994) found that the United States was the sole innovator over the 1979-88period. We would like to note that as to who the innovators are might be sensitive to thedifferent content and time span of the OECD sample. In the present study, we find that theUnited States is not the sole innovator over time among the 19 APEC member economies.As seen in Table 6, under the CRS benchmark, Hong Kong also shows its capability to shiftthe grand frontier of the APEC economies during the 1980-90 period. If the benchmark

Table 5Decomposition of technical change

Magnitude Input bias

1970–1980 1980–1990 1970–1980 1980–1990

East Asian economiesChina 0.985 0.993 1.000 1.000Japan 0.989 0.995 1.000 1.003

NIEsHong Kong 0.980 1.028 1.000 1.007Korea 0.988 0.998 1.000 0.999Singapore 0.898 0.992 1.020 1.002Taiwan 0.987 1.003 0.999 1.000

ASEAN-4Indonesia 0.995 0.998 1.016 0.999Malaysia 0.983 1.010 1.001 1.001Philippines 0.986 1.009 1.000 1.000Thailand 0.981 1.007 1.001 1.000


Australia 1.003 1.013 1.000 1.000Canada 0.993 1.014 1.004 1.000New Zealand 0.970 0.954 1.014 1.023USA 0.988 0.992 1.020 1.017

DevelopingChile 0.984 1.003 1.006 1.003Columbia 0.993 1.006 1.001 1.000Mexico 0.989 0.998 1.004 1.002Papua N. Guinea 1.018 1.002 0.976 0.983Peru 0.992 1.015 1.001 1.000


technology is relaxed to variable returns to scale, then both Hong Kong and Singapore jointhe U.S. as the APEC innovators in the 80’s.6 This result is quite inspiring because it impliesthat the NIEs not only are good at moving towards the frontier, they also have potential tobecome innovative at the same time. One more important observation pointed out by Fa¨re etal. (1994a) is that United States is the economy with the highest ratio of output to capital inthe sample. To see if this observation also applies to the Asian innovators like Hong Kong,we plot the output per labor against the capital-labor ratio for the 19 APEC economies. Fig.1is consistent with the Fa¨re et al. (1994a) observation because Hong Kong not only has highoutput to capital ratio but also is the only Asian economy that locates above the diagonal line.

Finally, we compare the engine of growth for the three major innovators in APEC region.Our results show that the input-biased technical change is the most important source of TFPgrowth for the United States. Singapore’s TFP growth is mainly driven by scale efficiency.Hong Kong’s TFP growth has transformed from catching-up in the 70’s into technologicalinnovations in the 80’s. Therefore, there seems to be no standardized model of growthexisted.

5. Role of human capital in long-run TFP growth

The discussions in the previous section ignore the role of human capital since onlyphysical capital is considered in our empirical analysis. In this section, we explore theimplications of our results for the endogenous growth models which emphasize the impor-tance of human capital and knowledge acquisitions in long-run growth (Romer, 1990). Our

Table 6Countries shifting the frontier, 1970–1990

Year Constant returns benchmark Variable returns benchmark

1970–1971 Chile, USA Chile, USA1971–1972 USA Mexico, New Zealand, USA1972–1973 USA —1973–1974 — —1974–1975 — —1975–1976 — —1976–1977 USA Chile, USA1977–1978 USA Mexico, USA1978–1979 — Mexico1979–1980 Mexico Mexico1980–1981 Chile Chile1981–1982 — —1982–1983 Hong Kong, USA —1983–1984 Hong Kong Hong Kong1984–1985 — Singapore, USA1985–1986 — Hong Kong1986–1987 Hong Kong Hong Kong1987–1988 Hong Kong, USA Hong Kong, Mexico, Singapore, USA1988–1989 USA USA1989–1990 Hong Kong Singapore


empirical framework is similar to those of Barro (1991), Barro and Lee (1994), andBorensztein, De Gregorio, and Lee (1998) which relate the TFP growth to two sets ofvariables. The first set is the state variables, which include the initial GDP per capita, the ratioof foreign direct investment (FDI) to GDP, stock of human capital in the form of educationalattainment (schooling), and the interaction between FDI and schooling. In doing so, weassume that education and FDI are the main vehicles to accumulate human capital and toacquire knowledge. The second set is the control (or policy) variables, which consist of theratio of real government consumption (net of defense and education) to real GDP, the blackmarket premium on foreign exchange (as a proxy for market distortion), measure of civilliberty7, terms of trade shock, . . ., etc.

Data on FDI and GDP come mostly from theBalance of Payments Statistics YearbookandInternational Statistics Yearbookpublished by the International Monetary Fund.8 We use thegross concept (i.e., sum of inflows and outflows) on FDI because both of them have thepotential to close the technological gap. For the education variable, we use the average yearsof schooling in total population over age 25 constructed by Barro and Lee (1993, 1994). Datafor the policy variables also come from Barro and Lee (1994).

Our regression is based on panel data for the two decades 1970-80 and 1980-90. Oursample includes 19 APEC member economies. The seemingly unrelated regression method(SUR) is adopted on the panel estimation. The Malmquist TFP growths over each decade (aslisted in Table 3) are used as dependent variables. Each equation has a different constantterm. Other coefficients are constrained to be the same for the two decades. The estimation

Fig. 1. Output-labor ratio and capital-labor ratio for APEC economics


allows for different error variances in each equation and for correlation of these errors acrossequations.

The regression results are shown in Table 7. Our first finding is that both FDI andschooling have positive direct effects on TFP growth. The interaction between FDI andschooling is also positive. Although the coefficients of schooling and the interactive term arenot statistically significant, these results are quite consistent with the theory and empiricalstudies on technology diffusion from FDI postulated by Findlay (1978), Wang (1990) and theothers. It also provides an empirical evidence that the flow of advanced technology broughtalong by FDI increases the TFP growth in the APEC region. Education is another comple-mentary factor to the TFP growth.

We conduct further regression analyses where the dependent variables are the twocomponents of TFP, i.e., technological progress and efficiency change, respectively. FromTable 7, we see that FDI has direct and positive contribution to efficiency. This, along withthe Young (1992) remark that Singapore has benefited from the sizable inflow of FDI sincethe 60s, explains why Singapore’s TFP growth is mainly driven by the efficiency factor. Onthe other hand, our results indicate that FDI can only promote technical progress (orinnovations) through advances in education levels. This implies that the innovative knowl-edge acquired by international transfer is embodied in labor force. Therefore, the growthexperience in APEC region tells us that FDI contributes to TFP growth either through

Table 7Regression results for productivity growth, FDI and education, panel of two decades 1970–80 and 1980–90,19 APEC economies

Independent variables Explanatory variable

TFP growth Efficiency change Technical change

Log (initial per capita GDP) 0.0163** 0.0151 0.0065(2.216) (1.548) (0.804)

Schooling 0.0011 0.0036 20.0039(0.502) (1.212) (21.555)

FDI 0.5041* 0.7445* 20.2800(1.946) (2.145) (21.030)

Schooling3 FDI 0.0119 20.0714 0.0960*(0.241) (21.062) (1.803)

Terms of trade shock 20.0666 0.0741 20.1401*(20.736) (0.599) (21.904)

Government consumption 20.111** 20.0928 20.0004(22.694) (21.662) (20.010)

Black market premium 0.0040 0.0046 20.0029(0.677) (0.587) (20.672)

Civil liberty 0.0071** 0.0088** 20.0020(2.797) (2.677) (20.738)

Degree of freedom 26 26 26Adjusted R2 0.4963 0.4140 0.3019

Note: The t-values are in parenthesis. The constant terms are not reported.* Significant at 10% level.** Significant at 5% level.


catching-up or through innovations when a sufficient learning capacity of the advancedtechnologies is available in the host economy.

Our second empirical finding is that the initial GDP level has a positive and significantimpact on TFP growth.9 This seems to indicate that TFP tends to accelerate only in thoseeconomies which already have higher income levels. Despite the fact that the income levelsamong the APEC economies have been converging during the post-war period (Drysdale &Huang, 1997), our results show that perhaps the TFP divergence has occurred since 1970’s.Similar divergent pattern exists in the cases of technological catch-up and innovations. Thesedivergent patterns signal that less developed economies may not take advantages of their“backwardness” in accelerating TFP growth.

As for the policy distortion variables, the real government consumption has negative andsignificant impact on TFP growth and its components as expected. However, the othermarket distortion factors like terms of trade shock and black market premium have nosignificant impact on TFP growth. On the other hand, the civil liberty coefficients aresignificantly positive in both the TFP and efficiency regressions, meaning that more libertiesare bad for TFP growth and technological catch-up in the APEC region. Although moreliberties have potentials to reduce transaction costs (or market inefficiencies), it might alsobring in certain degree of political instability and become obstacles to TFP growth. Thereason behind this is certainly an interesting topic for further study.

We also include a regional dummy for the ten Asian economies as one of the regressors.The results (not reported here) are largely unaffected by adding this regional dummy. Thesedummies enter with statistically significant positive coefficients only when TFP is theexplanatory variable. This is consistent with our expectation that these ten economiesoutperform the rest of the APEC economies in TFP growth after controlling the state andpolicy factors. Alternatively, a dummy for the NIEs is also included as a regressor. Thecoefficient of this dummy is not only significantly positive in the TFP regression, but also inthe efficiency case. This provides some insights into how NIEs outperform the others in TFPgrowth.

6. Concluding comments

The eruption of the Asian financial crisis in July 1997 has stimulated another wide debateover the sources of Asia’s rapid growth and its sustainability. Some pointed to lack oftechnological progress and innovation, while others blamed failure to achieve efficiency inthe allocation of key resources. This paper attempts to address these issues by identifying thesources of productivity growth in ten Asian economies including China, Japan, NIEs and theASEAN-4. We calculate productivity growth and its components using distance-function-based Malmquist productivity indexes and linear programming methods following Fa¨re et al.(1994a). We include 19 APEC member economies to construct our benchmark frontier forwhich consistent quantity data are collected from the Penn World Tables (Mark 5) over the1970-80 and 1980-90 periods. The productivity growth can then be decomposed into changesin efficiency (the catching-up effect) and technical progress (the innovation effect). Regres-


sion analyses are also conducted to investigate the role of FDI and education in pushingtoward the frontier as well as moving the entire frontier.

Our results indicate that the United States is not the sole innovator among the 19 APECmember economies. Rather, over the 80’s, Hong Kong and Singapore have shown theircapability to shift the grand frontier of the APEC economies. This result is quite inspiringbecause it implies that the NIEs not only are good at moving toward the frontier, they havethe potential to be the innovators.

Based on the findings of Young (1992, 1994, 1995) and Kim and Lau (1994), Krugmanargues that TFP is of little importance in explaining the East Asian growth miracle and,therefore, East Asian’s high growth rate is unlikely to be sustained in the future. Kim and Lau(1996), in a more recent study, confirm once again that TFP has been relatively unimportantin the Asian Pacific economies. Japan, Hong Kong and Singapore are the only threeexceptions in the region. However, TFP is found to be important contributor to East Asianeconomic growth in other empirical studies. For example, Drysdale and Huang (1997) usethe Penn World data to identify the sources of output growth in APEC economies. From theirregression results, TFP growth and factor accumulation are found to be equally important tooutput growth for Hong Kong, Japan, Taiwan, Korea, Indonesia and Thailand, but lessimportant in Singapore and Malaysia. In fact, using the same data set, the Young (1994)simple regression results also find low TFP growth for Singapore and Malaysia, while HongKong, Thailand, Taiwan and Korea are among the top performers.

In comparison to the existing literature on the East Asian miracle growth, our results seemto reconcile some of the arguments over Krugman’s 1994 article. On the one hand, ourresults are in accordance with the Krugman’s argument in that NIEs as a whole outperformthe others only in catching-up with the frontiers. On the other hand, despite the differencesin methodology, sample size, data, assumptions on capital stock depreciation rates, . . .etc.,we find that the overall TFP growth in ten East Asian economies is indeed better than theothers. We also found that the productivity divergence might have occurred since the 70’s.

Moreover, our results exceed the existing literature by exploring further into the sourcesof TFP growth. Our TFP decompositions indicate that the input-biased technical change isthe growth engine for the United States. Hong Kong exhibits a quite unique pattern for beingconsistently efficient with technical progress. Scale efficiency is found to be the major sourceof growth in Singapore’s economy. This may also explain why some literature have founddifferent performances between Hong Kong and Singapore in TFP growth.

Besides identifying the sources of TFP growth, the strengths and weaknesses of eacheconomy, our analysis carries further into the more fundamental issue on knowledge andhuman capital accumulation. Our regression results indicate that FDI contributes to TFPgrowth either through catching-up or through technological innovations. But the latter comeswith a condition that a sufficient learning capacity of the advanced technologies is availablein the host economy. Therefore, the innovative knowledge acquired by international transferis embodied into labor force to promote TFP growth in the Asian economies.

Finally, the policy implications from this study are twofold. First, the policies towardoutward-looking and less government interventions are critical to the TFP growth and shouldbe continued in the future. On the other hand, the economic growth in East Asia will be able


to sustained only if more investment in human capital (e.g., education) are devoted topromote technological innovations.

Notes

1. According to Fa¨re and Lovell (1978), the distance functions are reciprocals of Debreu(1951)-Farrell (1957) technical efficiency measures. Either input or output distancefunctions can be used to define the Malmquist TFP index. For this paper, we use theoutput distance function. The output distance function characterizes the productiontechnology by considering a maximal proportional expansion of outputs with giveninputs.

2. This is the Caves, Christensen, and Diewert (1982) version of the Malmquist produc-tivity change.

3. Either constant returns to scale (CRS) or VRS technology can be accepted as abenchmark in identifying the most productive scale (size) and deviations from thatoptimal scale (size). In the Kim and Lau (1996) study on ten Asian-Pacific countriesand four western industrialized countries, the test for CRS assumption is rejected underthe aggregate meta-production function specification. Therefore, we adopt the VRSassumption following one referee’s suggestion.

4. According to the appendix of Penn World Table 5.6 (PWT), the worker for thisvariable is a census definition based on economically active population. This definitionimplies that our labor variable is based on total workforce rather than actual number oflabor employed. Thus, our study may overstate the actual amount of labor.

5. The non-residential investment is estimated by first multiplying the real investmentshare by real GDP. Then, we have to separate the residential investment from thenon-residential one. Because PWT 5.6 only has data on residential and non-residentialcapital stock, we assume that the two investment figures are proportional to those incapital stock. Therefore, we calculate the ratios of non-residential capital stock to totalcapital stock for the rest fourteen economies. The sample mean is 0.72. Thailand’sratios, which range from 0.72 to 0.81, are used for these five economies to estimatetheir non-residential investment.

6. Under variable returns to scale benchmark, Mexico and Chile are also identified asinnovators during the 70’s, but they no long are during the 80’s.

7. The measures for civil liberties is a subjective index from 1 (most freedom) to 7 (leastfreedom). The Barro and Lee (1994) data set does not report this index for Hong Kongexcept the average of 1985-89. Here, we assume that there is no change during theperiod from 1970 to 1985 in Hong Kong.

8. Data on FDI comes from the table titled “Direct Investment Abroad and DirectInvestment in the Reporting Economy” in theBalance of Payment Statistics Yearbook(International Monetary Fund [IMF]). But data on Hong Kong and Taiwan is notavailable in the IMF publications. Therefore, the FDI data reported in theWorldInvestment Directory,1992(United Nations) is used for these two economies. The GDPdata for Hong Kong come from theKey Indicators of Developing Asia and Pacific,


(Asian Development Bank). Taiwan’s GDP data come from theTaiwan StatisticalData Book, 1997, (Council for Economic Planning and Development). China’s GDPdata come fromWorld Tables, 1995(World Bank) starting from the year 1973.

9. We have tried many different specifications in our regression. The coefficient of thisvariable always turns out to be positive.

References

Asian Development Bank. Key indicators of developing Asia and Pacific (Asian Development Bank, Manila)various issues.

Banker, R.D. (1984). Estimating most productive scale size using data envelopment analysis.European Journalof Operational Research 17(1), 35–44.

Barro, R.J. (1991). Economic growth in a cross section of countries.The Quarterly Journal of Economics 106(2),407–443.

Barro, R., & Lee, J-W. (1993). International comparison of educational attainment.Journal of MonetaryEconomics 32(3), 361–394.

Barr, R., & Lee, J-W. (1994). Sources of economic growth.Carnegie Rochester Conference Series on PublicPolicy 40, 1–46.

Baumol, W.J. (1986). Productivity growth, convergence, and welfare: what the long-run data show.AmericanEconomic Review 76(5), 1072–1085.

Borenszstein, E., De Gregorio, J, & Lee, J-W. (1998). How does foreign direct investment affect economicgrowth?Journal of International Economic 45(1), 115–135.

Caves, D.W., Christensen, L.R., & Diewert, W.E. (1982). The economic theory of index numbers and themeasurement of input, output, and productivity.Econometrica 50(6), 1393–1414.

Chen, E.K.Y. (1997). The total factor productivity debate: determinants of economic growth in East Asia.AsianPacific Economic Literature 11(1), 18–38.

Council for Economic Planning and Development. (1997).Taiwan Statistical Data Book, 1997.Taipei: Councilfor Economic Planning and Development.

Debreu, G. (1951). The coefficient of resource utilization.Econometrica 19(3), 273–292.Dowrick, S., & Nyugen, D. (1989). OECD comparative economic growth 1950-85: catch-up and convergence.

American Economic Review 79(5), 1010–1030.Drysdale, P., & Huang, Y. (1997). Technological catch-up and economic growth in East Asia and the Pacific.The

Economic Record 73(2), 201–211.Fare, R., & Lovell, C.A.K. (1978). Measuring the technical efficiency of production.Journal of Economic Theory

19(1), 150–162.Fare, R., Grosskopf, S., & Lovell, C.A.K. (1992). Productivity change in Swedish pharmacies 1980-1989: a

nonparametric Malmquist approach.Journal of Productivity Analysis 3(3), 85–101.Fare, R., Grosskopf, S., Norris, M., & Zhang, Z. (1994a). Productivity growth, technical progress, and efficiency

change in industrialized countries.American Economic Review 84(1), 66–83.Fare, R., Grosskopf, S., & Lovell, C.A.K. (1994b).Production Frontiers.Cambridge: Cambridge University

Press.Fare, R., & Grosskopf, S. (1996).Intertemporal Production Frontiers: With Dynamic DEA.Boston/London/

Dordrecht: Kluwer.Fare, R., Grosskopf, S., & Norris, M. (1997a). Productivity growth, technical progress, and efficiency change in

industrialized countries: reply.American Economic Review 87(5), 1040–1043.Fare, R., Grifell-Tatje, E., Grosskopf, S., & Lovell, C.A.K. (1997b). Biased technical change and the Malmquist

Productivity Index.Scandinavian Journal of Economics 99(1), 119–127.Farrell, M.J. (1957). The measurement of productive efficiency.Journal of the Royal Statistical Society.Series

A: General. 120, 253–281.


Findlay, R. (1978). Relative backwardness, direct foreign investment, and the transfer of technology: a simpledynamic model.Quarterly Journal of Economic, 92(1), 1–16.

International Monetary Fund.Balance of Payment Statistics Yearbook.Various issues. Washington DC: Inter-national Monetary Fund.

International Monetary Fund.International Statistics YearbookVarious issues.Washington DC. InternationalMonetary Fund.

Kim, J.-I., & Lau, L.J. (1994). The source of growth of the East Asian Newly Industrialized Economies.Journalof Japanese and International Economies 8(3), 235–271.

Kim, J.-I., & Lau, L.J. (1995). The role of human capital in the economic growth of the East Asian NewlyIndustrialized Countries.Asia-Pacific Economic Review 1(1), 3–22.

Kim, J.-I., & Lau, L.J. (1996). The sources of Asian Pacific economic growth.Canadian Journal of Economics26, 448–454.

Krugman, P. (1994). The myth of Asia’s miracle.Foreign Affairs 73(6), 62–78.Ray, S.C., & Desli, E. (1997). Productivity growth, technical progress, and efficiency change in industrialized

countries: comment.American Economic Review 87(5), 1033–1039.Romer, P.M. (1990). Endogenous technological change.Journal of Political Economy 98(5), Part 2, S71–S102.Summers, R., & Heston, A. (1991). The Penn World Table (Mark 5): an expanded set of international

comparisons, 1950–87.Quarterly Journal of Economics 106(1), 1–41.United Nations. (1992).World Investment Directory 1992, Volume 1, Asia and the Pacific.New York: United

Nations.Wang, J-Y. (1990). Growth, technology transfer, and the long-run theory of international capital movements.

Journal of International Economics 29(3–4), 255–271.World Bank. (1993). The East Asian Miracle: Economic Growth and Public Policy. Washington DC: World

Bank.World Bank. (1995).World Tables 1995. Washington DC: World Bank.Young, A. (1992). A tale of two cities: factor accumulation and technical change in Hong Kong and Singapore.

In O.J. Blanchard & S. Fisher (Eds.),NBER Macroeconomic Annual 1992(pp. 13–54). Cambridge, MA: MITPress.

Young, A. (1994). Lessons from the East Asian NICs: a contrarian view.European Economic Review 38(3–4),964–973.

Young, A. (1995). The tyranny of numbers: confronting the statistical realities of the East Asian growthexperience.Quarterly Journal of Economics 110(3), 641–680.


Documents

Efficiency change and growth in productivity: the Asian growth experience