International Produtivity Differenes, Infrastruture, and Comparative Advantage For Submission to the Review of International Eonomis Manusript 4349 Revised, February 2006 Abstrat This paper provides an empirial analysis of the effet of infrastruture on industry-level produtivity and international speialization, as suggested by Clarida and Findlay s (1992) model. Attempting to ontrol for simultaneity with a three-stage least squares estimation strategy, we find that publi infrastruture helps to explain patterns of omparative advantage and international speialization.
Before we an plae muh faith in the hypothesis of systemati tehnologial differenes [between ountries], we will need a lot more work to determine the soure of these differenes. Is it infrastruture? Is it organization forms? Or what? (Leamer and Levinsohn, 1995, p. 1360) Introdution There is a substantial body of evidene that produtivity differs between ountries at the industry level. A number of studies have doumented large and persistent setoral differenes in labor produtivity between ountries (e.g. Van Ark and Pilat, 1993). More reently, several studies have shown that there are also large international disparities in total fator produtivity (TFP) (e.g. Bernard and Jones 1996, Hall and Jones 1999, Harrigan 1999). Other studies have shown that setoral produtivity differenes aross ountries in turn affet international speialization in standard Riardian and neolassial models of international trade (Harrigan 1997, Golub and Hsieh 2000). In this paper, we investigate two related questions: 1) the extent to whih differenes in TFP aross ountries at the setoral level an be attributed to differenes in the availability of physial infrastruture, and 2) the effets of infrastruture on international speialization of prodution. While an affirmative answer to the first question implies an affirmative answer to the seond one, the latter is also worth investigating beause there may be hannels of ausation from infrastruture to international speialization beyond those operating from relative produtivity, as well as problems of measuring relative produtivity that ould obsure the relationship. Casual observation in developing ountries suggests that poor infrastruture ontributes to low produtivity. Power outages, weak teleommuniations systems, lak of adequate roads are impediments to investment, growth, and poverty alleviation in these ountries. (World Bank 2002, Chapter 8) A number of studies have examined the effet of infrastruture on aggregate 1
output, beginning with Ashauer (1989). Reent studies inlude Holtz-Eakin (1994), Fernald (1999), Roller and Waverman (2001) and Canning (2001). Gramlih (1994) provides a survey. In addition to the effets on aggregate produtivity absolute advantage, some reent theoretial literature has suggested that infrastruture is likely to influene setoral produtivity omparative advantage. Jones (2000) emphasizes the importane of servie links in lowering transations osts and thus failitating international speialization and the loation hoies of footloose industries. That is, the quality of publi servies matters for investment deisions in suh industries. In a series of papers, Ronald Findlay and his o-authors have modeled the effet of infrastruture on produtivity and omparative advantage (Findlay and Wilson 1987, Clarida and Findlay 1992). The relationships between infrastruture, produtivity, and international speialization are of onsiderable poliy relevane. Poliy plays a deisive role in the provision of infrastruture due to the publi good and natural monopoly dimensions of infrastruture, either diretly via publi investment or indiretly via the regulatory environment. Government therefore may influene both absolute advantage and omparative advantage through its poliies towards infrastruture. Indeed, many ountries pledge speifi infrastruture improvements as part of a pakage of targeted investment inentives. 1 Despite the importane of this issue, to our knowledge there have been no previous empirial studies of the effet of infrastruture on omparative advantage. The endogeneity of infrastruture provision, however, raises diffiult problems of identifying the effets of infrastruture on omparative and absolute advantage insofar as governments hoose infrastruture onditional on the harateristis of their ountry. Highly 1 For example, Costa Ria sueeded in attrating a maor investment from Intel in semiondutors in part by ommitting to funding a new eletrial substation (Moran 2002, p. 40). 2
produtive ountries an afford higher levels of infrastruture, so the diretion of ausation runs in both diretions. Also, ountries with natural omparative advantages in infrastrutureintensive industries may tend to invest more in infrastruture, as Clarida and Findlay s (1992) model suggests. In our empirial analysis, we ontrol for the endogeneity of infrastruture provision using the insights provided by a model inspired by Clarida and Findlay (1992). We find that after ontrolling for unobserved heterogeneity aross ountries and the potential endogeneity of infrastruture, there remains a statistially signifiant effet of infrastruture on industry-level produtivity and omparative advantage. The remainder of this paper is organized into four setions. Setion 1 presents the model and the empirial strategy. Setion 2 desribes the onstrution of our TFP and infrastruture data. The results are presented in setion 3. Setion 4 onludes. 1. Coneptual Framework Consider a small open eonomy that produes J goods, indexed by, using onstant returns to sale tehnologies from M primary fators, indexed by m. Sine the ountry is small, it takes the J-vetor of world pries P as given. All produt and fator markets are perfetly ompetitive. As is well-known, the ompetitive equilibrium of this eonomy maximizes the value of final output. A ommon formulation of this problem (see Dixit and Norman 1980, p. 31) is r( P, V ) = max x { P x ( x, V ) feasible} = P x( P, V ) (1) 3
where r(p,v) is known as the revenue funtion and x(p,v) is the vetor of net output supplies that maximize the value of national inome. If fators are ontinuously substitutable and M J, then the vetor of net output supplies is given by r( P, V ) x ( P, V ) =, = 1,..., J. (2) P In addition to the assumptions impliit in the revenue funtion, suppose that the prodution tehnology for good an be written as x = θ f ( v ) = θ ~ x, = 1,..., J, (3) where θ is a produtivity shift parameter defined relative to some base time period or ountry s output x~ and v is an M-vetor of fator inputs in good. An inrease inθ represents a Hiks neutral produtivity improvement. Dixit and Norman (1980, p. 138) show that under these irumstane, the revenue funtion (1) has the partiular form r( Θ P, V ), where Θ = diag θ,..., θ ), suh that hanges in ( 1 J an industry s produtivityθ affet prodution in the same way as hanges in P. This formulation offers a oneptually straightforward mehanism for introduing produtivity differenes aross ountries as a soure of omparative advantage, and it has proven useful in Harrigan s (1997) empirial work. We extend this framework by deomposing international produtivity differentials into (i) inherent tehnologial differenes and (ii) stoks of infrastruture. 4
We assume that the Hiks neutral produtivity shifter in industry an be written θ = ω g (I ), (4) whereω is a tehnologial parameter inherent to industry, I is the stok of infrastruture available in the eonomy, and ( ) g infrastruture into produtivity. The is an inreasing funtion that maps the availability of g ( ) funtion is the same aross ountries but speifi to industry. 2 Infrastruture is a soure of omparative advantage as its industry-speifi produtivity effets interat with international differenes in fator endowments. The government is assumed to provide infrastruture using a Leontief tehnology with a vetor of fixed unit input requirements Γ with typial element λm indiating the amount of input m required to produe one unit of infrastruture. Sine government provision of infrastruture onsumes resoures, not all of the ountry s endowments will be available to private firms for the prodution of final goods. The net vetor of fator endowments available for this purpose is given by NV = V Γ I. (5) 2 The assumption that the mapping of infrastruture to an observed produtivity level is the same aross ountries is neessary to identify the effets of infrastruture on produtivity. Infrastruture remains a soure of omparative advantage given that observed θ and hene omparative advantage is affeted by differenes in infrastruture provision. This assumption is analogous to the assumption frequently made in the empirial literature on the Heksher-Ohlin model that tehnologies are idential aross ountries or differ by an equal proportion aross industries (e.g.,, Trefler 1995). 5
If G( I) = diag( g1( I),..., g ( I)) and Ω = diag ω,..., ω ), then the revenue funtion an be written J ( 1 J r ( Ω G I) P, V Γ I ) (. (6) Given I, the revenue funtion in (6) has all the standard features. Notably, the gradient of r( P Ω G( I), V Γ I ) with respet to P yields the net supply of eah final good produed by the eonomy: ( Ω G( I) P, V Γ I ) r x ( Ω G( I) P, V Γ I) = (7) P A welfare maximizing government will hoose the level of I that maximizes (6). An inrease in I is expeted to inrease produtivity in all setors, thereby raising absolute advantage, but to an extent that ould vary aross setors, therefore influening omparative advantage. The size of the produtivity effet assoiated with an inremental inrease in I depends on other harateristis of the ountry: its fator endowment V and its tehnology as summarized by Ω. The ost of an inremental inrease in I is the real resoures onsumed Γ. The optimal level of infrastruture depends on a ountry s absolute and omparative advantage, and in turn further influenes the ountry s pattern of international speialization. For example, the government of a ountry with fator endowments and tehnologies that onfer a omparative advantage in industries in whih roads are relatively important may provide a larger stok of roads, thereby further reinforing the omparative advantage in road-intensive industries. 6
2. Empirial Strategy Our obetive is to estimate the relationships: (1) between unobserved TFP Θ and infrastruture, and (2) between infrastruture and industrial speialization. We will do so using data on observed TFP and industrial speialization aross ountries, industries, and years. In this setion, we speify an eonometri model apable of ahieving these goals that follows from the oneptual framework presented in the previous setion. First, onsider the relationship between infrastruture and observed produtivity implied by equation (4). To make equation (4) operational, we assume that the elastiity of produtivity with respet to infrastruture provision is onstant, so that (4) an be represented in logarithms as lnθ t = β ln I + lnω + e t t t. (8) Aording to (8), observed TFP ( ) θ in industry in year t and ountry is a funtion of t infrastruture in that ountry at time t ( I ), unobserved tehnologial ability ( ω ) e. stohasti measurement error ( ) t t t, and a To relate infrastruture to observed industrial speialization aross ountries, we follow Harrigan (1997) in assuming that the revenue funtion given by (6) an be approximated by a seond-order translog: ln r( ΘP, NV ) = a + t b 00 ln nv + + 1 2 a 0m m m m i ln nv 1 2 0 k lnθ P + b mi m i a ln nv + k lnθ P m m lnθ P k lnθ P k ln nv m, (9) 7
where the summations over and k run from 1 to J goods, and the summations over m and i run from 1 to M fators. nv m is the ountry s endowment net of the quantity of resoures onsumed by infrastruture prodution of fator m. Symmetry of ross effets requires a k = a k and b mi = b im for all, k, i, and m. For expositional onveniene, ountry supersripts and time subsripts have been suppressed. It should be remembered that θ is a funtion of inherent produtivity ω and the influene of infrastruture as in (8). Linear homogeneity of the revenue funtion requires m a b 0 mi = = 0, 1, b0m m m m = 1, = 0. a k = 0, Differentiating equation (9) with respet to an industry prie, P, we obtain s m lnnvm + a k ln Pk + = a0 + a lnθ, (10) m k k k where s is the share of the value of output in industry in total GDP. 3 An important feature of (10) is that prie hanges and hanges in produtivity in one industry have general-equilibrium effets on other industries (aptured by a k ). 3 This equation is idential to Harrigan s (1997) equation (8) exept that fator supplies are net of usage in infrastruture. 8
Substituting (8) into (10) and adding bak in ountry supersripts and time subsripts yields, s t = a + 0 k + a m k e kt m. lnnv mt + k a k ln P kt + k a k β k ln I t + k a k lnω kt (11) Infrastruture influenes industry-level produtivity as aptured by β and these hanges in produtivity in turn have general equilibrium effets through the produtivity oeffiients, a k. Equation (11) implies the following empirial speifiation s t = a ln 0 + m lnnvmt + α I t + υ t (12) m whereα = a k β k is the effet of the level of infrastruture on the share of industry in total k GDP and = t a k Pkt + a k lnωkt + k k k υ ln a e + ς (13) k kt t is unobserved, onsisting of the prie vetor P, the vetor of tehnologial apabilities Ω, and measurement error from both the produtivity equation and the speialization equation ς t, respetively. e kt and 9
In estimating equations (8) and (12), we fae an identifiation problem aused by the fat ω t that tehnologial abilities appears in the unobservables. As our model shows, a ountry s hoie of infrastruture is a funtion of both its fator endowments and these unobserved tehnologial abilities. Our identifiation strategy begins by time differening (5) and (12), whih yields υ t Δ lnθ + t = β Δ ln I t η t (8 ) where η = Δ lnω + Δe and t t t Δs = Δ lnnv + α Δ ln I t m m mt t + Δυ. (12 ) t Time differening eliminates any time-invariant, ountry-industry speifi effets that influene both infrastruture and our two sets of dependent variables, observed TFP and output shares. Note also that only the innovation in unobserved tehnologial ability, Δ lnω t, appears in Δυ t. Innovations in unobserved tehnologial ability are far less preditable than levels, and thus less likely to affet governments infrastruture plans. As a result, the degree of endogeneity is lower in a speifiation estimated in differenes than in a speifiation estimated in levels. To formalize the idea that infrastruture planning lags advanes in tehnologial apabilities, we assume that infrastruture provision takes the form of the partial adustment model given by 10
ln ~ ( ln I I ) I t ln I t 1 + Ψ t 1 ln t 1 =, (14) ~ where I 1 is the optimal level of infrastruture for ountry. The oneptual framework t presented in the previous setion suggests that a ountry s optimal level of infrastruture is a funtion of the vetor of its fator endowments V and its tehnologial apabilities Ω. Therefore, we assume ~ ln I = φ lnv + δ lnω, (15) t m m mt t whereφ m andδ are unknown oeffiients. Substituting (15) into (14), time differening the resulting expression, yields the following first stage equation: ( 1 Ψ) Δ ln I + Ψ Δ V u Δ t = t 1 m ln mt 1 m ln I φ +. (16) t Note that u = Ψ δ lnω + ε, where is an error term assoiated with measurement of the t t t ε t dependent variable. If inherent tehnologial ability follows a random walk, then lagged growth rates in endowments and infrastruture are appropriate instruments for urrent hanges in infrastruture stoks. While we rely hiefly on instrumental variables to identify the effet of infrastruture on observed TFP and industrial speialization, the panel data nature of our data allows us to ontrol 11
for ommon unobservables. First, to ontrol for tehnology shoks by industry that are ommon aross ountries, we add time-industry fixed effets to (8 ). Seond, to ontrol for unobserved hanges in pries, exogeneous TFP levels ω t, and government poliies that might also be orrelated with both infrastruture provision and industry output shares aross ountries, we add ountry-industry-speifi interepts and industry-time time effets to equation (12 ). Finally, while we do not observe pries, we an proxy for hanges in the relative prie of traded goods by inluding hanges in the logarithm of the real exhange rate (RER). Equations (16), (8 ), and (12 ) respetively form an eonometri model that speifies a link between (i) fator endowments and the level of infrastruture, (ii) infrastruture and observed produtivity, and (iii) infrastruture and the struture of international speialization. Note that the unobserved produtivity levels appear in the unobservable terms in all three equations: ω t Produtivity Equation (8 ): η = Δ lnω + Δe, t Speialization Equation (12 ): υ = t a k ln Pkt + a k lnωkt + a kekt + ς t, Infrastruture Equation (16): u = Ψ δ ln ω + ε. t k t Given this feature of our system of equations, we estimate our system via three stage least squares. Further, following Harrigan (1997) we impose the linear homogeneity requirements m = 0 m for the international speialization equation (12 ). t t k t k 3. Data and Measurement 12
We now desribe our data. These data fall into two ategories, measures of output and total fator produtivity, and measures of fator endowments and infrastruture. Further details are provided in Appendix 1. 3.1 Total Fator Produtivity We alulate international differenes of setoral TFP for ten industries and eighteen developed and developing ountries over the years 1979-97. The data used to onstrut our measures of TFP are from the United Nations Industrial Development Organization (UNIDO) INSTAT database. From this database we obtain measures of value-added, employment, and apital formation by manufaturing ISIC ategory. The ountries inluded in our data were seleted aording to data availability and geographi and eonomi diversity. 4 Following muh of the literature on produtivity omparisons, we use the TFP index proposed by Caves et al (1982), whih is defined as lnω t ( ln X ln X ) σ ( ln L ln L ) (1 )( ln K ln K ) = σ t t t t t t t t, (17) where X is value-added, L is total hours of labor, and K is apital stok. In eah of the parentheses, there are individual ountry variables relative to the mean value of this variable aross ountries in the sample. Finally, σ t = (q t + q t )/2 is an average of the labor ost share in ountry and the ross-ountry average. This index is widely used beause it is superlative in 4 The eighteen ountries are Austria, Canada, Colombia, Finland, Frane, Hong Kong, India, Indonesia, Italy, Japan, Korea, Netherlands, Singapore, South Afria, Spain, Turkey, the United Kingdom, and the United States of Ameria. The ten industries are Food Produts (ISIC 31, less 312), Textiles (ISIC 32), Wood Produts (ISIC 33) Paper Produts ISIC (34), Chemials (ISIC 35), Metals (ISIC 381), Mahinery (ISIC 382), Eletronis (ISIC 383), Transport Equipment (ISIC 384), and Instruments (ISIC 385). 13
the sense that it is exat for a translog funtional form, and it is transitive so that the hoie of the base ountry (here a geometri average of sample ountries) is unimportant. In implementing (17), we adust eah ountry s number of workers by industry and year as reported in UNIDO by the average number of hours per worker by ountry and year for total manufaturing as reported by the International Labor Organization. Our industry-ountry-year apital stok data are onstruted via the perpetual inventory method using real dollar gross fixed apital formation as reported by UNIDO. The resulting data are then adusted for year-to-year variations in apaity utilization. Finally, labor ost share data were omputed as the ratio of payments to labor in total value-added. Following Harrigan (1997, 1999), we smooth the relatively volatile labor ost oeffiients using the fitted values from a regression of labor shares on apital-labor ratios. 5 3.2 Output Shares, Infrastruture, and Endowments The study also requires data for fator and infrastruture endowments and measures of industrial speialization over time. Output shares were onstruted from the UNIDO dataset, whih provide data to alulate industry shares of manufaturing, and the World Bank s World Development Indiators, whih provide output shares of GDP in manufaturing, agriulture and servies. Our five fator endowments inlude physial apital, arable land, and low, medium, and high skilled labor. Aggregate physial apital stoks were onstruted using the perpetual inventory method using real gross fixed apital formation data from the World Development Indiators. Arable land and total labor fore partiipation are also from this soure. Finally, shares of the working population that fall into low, medium, and high eduation workers ome from the Barro-Lee dataset. 5 Appendix Table 1 presents summary results for levels and rates of hange of TFP for the ountries in the sample. 14
Our three measures of infrastruture are total length of the road network in miles, number of telephone lines in use, and eletrial power generating apaity. All three measures are taken from Canning (2001) and extended using data from the World Development Indiators. Our hoie of infrastruture variables is largely governed by availability over time and ountries and are similar to those used in other studies. It an be argued, however, that while roads are pure publi goods, teleommuniations and eletri power are mostly private, albeit subet to inreasing returns in some areas. Nonetheless, the natural monopoly harater of some teleommuniations and eletriity servies, ombined with their importane as shared inputs in all setors, has entailed a ritial role for government poliy in determining their supply, both diretly and through regulation of private utilities. We have therefore deided to inlude all three measures of infrastruture. Ideally our measures of all three infrastruture variables would apture some element of quality (failed telephone alls, power outages, et.). Suh data does not exist in the neessary panel format, however, and we are thereby fored to use quantity measures unadusted for quality. Road network length and number of telephone lines are eah normalized by the size of the labor fore, while eletrial power generating apaity is normalized by the real dollar value of the national apital stok. 6 Our theory suggests that the relevant measure of endowments is net of the primary fator requirements in the onstrution, maintenane, and operation of infrastruture. We ompute unit fator input requirements for roads, teleommuniations, and eletrial generating apaity for the United States and assume that these apply to all ountries. 7 We further assume that the fator 6 In priniple, one would want to normalize roads also by the geographi size of the ountry. However, sine our eonometri speifiations relate hanges in log infrastruture to hanges in log TFP, the effet of any time invariant variable is neessarily unidentified. 7 The assumption that fator requirements are the same aross ountries is unlikely to matter muh in pratie for net fator supplies, given that fator use in infrastruture aounts for a small part of the apital stok. For the United States, for example, infrastruture aounts for about 10-15 perent of the total apital stok and less than 5 perent of employment. 15
proportions of the onstrution, teleommuniations, and utility setors of the U.S. eonomy ould be used as proxies for those of roads, teleommuniations, and eletrial generating apaity, respetively. Unit fator input requirements in U.S. infrastruture were alulated based on data from the Department of Commere for labor and apital input, and from the Current Population Survey of the Census Department for skill omposition orresponding to the Barro-Lee ategories. 4. Results We report estimation results for equations (13)-(15): the growth of infrastruture, and the effets of infrastruture on produtivity and omparative advantage as measured by output shares. The results reported in this setion all orrespond to 3-year time differenes, whih yield approximately 5 observations for eah ountry-industry pair and a total of 88 ountry-year observations for eah industry. All instruments are lagged four years to avoid any overlap with the dependent variable. The use of differenes rather than levels, while aiding in identifiation, eliminates the information ontained in levels and makes it more diffiult to obtain easily interpreted and statistially-signifiant results. Appendix 2 reports results for developed and developing ountries separately. We first estimated (14) for eah industry using the instruments suggested by (13) via GMM and onduted Hansen onditional moment tests and Sargan tests to assess our assumption that lagged endowments and infrastruture growth rates are appropriate instruments. In all ten industries onsidered in our study, we were able to reet the hypothesis that our instruments are invalid at very high levels of onfidene. 16
4.1 Infrastruture Growth The oeffiient estimates of the first-stage equation (13) are shown in Table 1. Eah olumn orresponds to an infrastruture type and eah row to an independent variable. The Table reveals that the growth rates of both road length per worker (Roads) and telephone lines per worker (Teleom) are losely related to lagged growth rates of endowments and infrastruture measures, partiularly their own lagged growth rate, while the growth rate of eletrial generating apaity (Power) is not as losely linked to the orresponding variables. Interestingly, there appear to be signifiant differenes in the types of lagged endowments that matter, with Capital partiularly important for Roads, Low and Moderately skilled labor important for Teleoms, and High skilled labor relatively important for Power. The onnetion between roads and apital is plausible insofar as roads involve muh higher apital-labor ratios than teleommuniations and eletriity. There is insuffiient information on the skillomposition of the labor fore used in infrastruture provision to assess the relative skillintensities of teleommuniations and eletriity provision. Separating the sample into developed and developing ountries has very little effet on the relative magnitude of the effets of the various fator-endowment and lagged-infrastruture variables. 8 4.2 Growth Rates of TFP The estimates of equation (14), whih relates growth rates in industry TFP to growth rates of infrastruture, are shown in Table 2. The table is organized with eah row orresponding to an industry (dependent variable) and eah olumn orresponding to an infrastruture measure 8 Results for the stage 1 equation for developed and developing ountries separately are available from the authors upon request. Other results for these groups of ountries are reported in Appendix 2. 17
(independent variable). Standard errors for the oeffiient estimates are shown in parentheses. Finally, the last two rows show the hi-squared values and their orresponding p-values for two types of tests of ross-equation oeffiient restritions on that olumn s infrastruture variable. The first is the test of the null hypothesis that the oeffiients on hanges in infrastruture are ointly zero, while the seond is the test of the hypothesis that the oeffiients are the same aross equations. We begin with several observations on the general features of the results. The first is that every industry displays a statistially signifiant, positive relationship between TFP growth and growth in at least one type of infrastruture and in no industry is there a statistially signifiant, negative relationship between the growth rates of infrastruture and TFP. Seond, as indiated by the Chi-squared and orresponding p-values reported in the seond row from the bottom in Table 2, the test that all oeffiients are equal to zero aross industries an be reeted at standard levels for all three types of infrastruture. Third, of the three infrastruture types, Roads appears to have the most pervasive effets aross industries. An inrease in the length of the road network per worker is assoiated with a statistially signifiant inrease in TFP in nine of the ten industries, while telephone lines per worker is assoiated with a statistially signifiant inrease in only Transport Equipment and Sientifi Instruments, and eletrial generating apaity per dollar of apital is assoiated with a statistially signifiant inrease in TFP in only Food Produts and Chemials. Taken as a whole, these results are highly suggestive of a positive link between infrastruture provision, partiularly with respet to transportation, and industrial produtivity. We now turn to the impliations of our results for the hypothesis that infrastruture affets omparative advantage through its differential impat on setoral TFP. While the 18
oeffiients estimated aross industries vary from industry to industry in terms of both magnitude and statistial signifiane, the standard errors of the oeffiients are large relative to their differenes aross industries. The last row in the table reports the Chi-squared statisti and its orresponding p-value for the test of the hypothesis that the oeffiients are the same aross industries. The results of these tests indiate that the hypothesis that the effet of Roads is uniform aross industries annot be reeted at standard levels of signifiane, while the hypothesis that the oeffiients on Teleoms and Power are the same an be reeted at reasonable levels. The results suggest that Road provision, while a soure of overall produtivity differenes aross ountries, i.e., absolute advantage, does not appear to be a lear soure of omparative advantage, at least aross manufaturing industries. Teleoms and Power affet omparative advantage within manufaturing for some setors, but the effets are statistially insignifiant for about half the industries. It is important to remember, however, that the relatively weak results for Teleoms and Power probably reflet in large part the relatively impreise estimates of our oeffiients, whih stems in part from our small sample size and the differening of the variables. In addition, the effets of unmeasured quality variations in eletriity distribution and teleommuniations may be more important than the quantity differenes used here. More generally, the data for developing ountries are likely to be inferior to that of developed ountries for both infrastruture and produtivity measures. Finally, as noted earlier, roads are the losest to the onept of infrastruture as a pure publi good, whereas eletriity and teleommuniations servies are inreasingly provided by private suppliers, and therefore fit the model less losely. Tables 4A and 4B in Appendix 2 report results on the effets of infrastruture on TFP for developed and developing ountries separately. The results for developed ountries are 19
onsiderably stronger than those for developing ountries, and for the ombined sample, with both Roads and Power having a generally muh learer positive effet on TFP for developed ountries than for developing ountries, probably refleting poorer quality of the data for the latter. 4.3 Industrial Speialization We now turn to the relationship between infrastruture and industrial speialization. 9 Table 3 is organized in the same fashion as Table 2 with eah row orresponding to an industry equation and eah olumn orresponding to an independent variable. The first three olumns are our three infrastruture measures. The next five olumns orrespond to fator endowments net of infrastruture provision. The ninth olumn is the measure of the real exhange rate, where an inrease orresponds to a real depreiation. To better understand the effet of infrastruture (and primary fator endowments) on industrial speialization, we inluded both the agriultural setor (first row) and the servie setor (the twelfth row). Finally, the last two rows report Chi-squared and p-values for tests of oint hypotheses aross equations. The seond row from the bottom orresponds to the Chi-squared and orresponding p-value for the test that all the oeffiients for the appropriate input are zero, while the last row orresponds to the Chi-squared and orresponding p-value for test that all the oeffiients for the tradeable goods industries (all exept servies) are the same. Several general observations are in order regarding the oeffiients on primary fator endowments. First, an inrease in physial apital is generally assoiated with inreased output in the servie setor and lower output in manufaturing. More generally, the endowment 9 Reall that all twelve industry output share equations were estimated simultaneously with eah other and with the ten industry TFP equations. 20
variables seem to have opposite effets on manufaturing and servies. Seond, an inrease in least skilled workers is assoiated with greater agriultural output and an inrease in the output of several manufaturing industries while also reduing the output of servie industries. Finally, all fator endowment variables are statistially signifiant in the sense that a oint test that they are olletively zero is resoundingly reeted for eah endowment type. 10 The hypothesis that the oeffiients on fator endowments for the traded goods setors are the same is also sharply reeted for eah endowment variable. Next, we briefly disuss the oeffiients on hanges in the real exhange rate, RER. As expeted, an inrease in the relative prie of traded goods in terms of nontraded goods is assoiated with an inrease in the output of a number of manufaturing industries, suh as Textiles, Wood Produts, Chemials, Instruments, and a derease in the output of nontraded servies. We now turn to the estimated effet of our three infrastruture variables on the struture of industrial prodution. The output shares of eight industries appear to be influened by the availability of at least one type of infrastruture. Further, all three infrastruture types are found to be statistially different from zero and to be statistially different from eah other as indiated by the Chi-squared values shown in the seond-to-last row of Table 3. Fousing first on Roads, we find that an inrease in infrastruture is assoiated with a shift in output out of manufaturing industries and into servies, while the evidene of shifts of output within manufaturing is more subtle. Within manufaturing it appears that an inrease in Roads is assoiated with a derease in Textiles, Metal Produts, and Mahinery, relative to Eletronis, Chemials and perhaps Wood Produts. As indiated by the Chi-squared and their 10 Note that the same test (not shown) indiates that the oeffiients on only the manufaturing industries are statistially different from zero at standard levels. Further, a test that the oeffiients on endowment variables for the manufaturing industries are the same is also resoundingly reeted. 21
orresponding p-values shown in the last row of Table 3, the hypothesis that the oeffiients on Roads aross the eleven tradable-goods industries are the same is sharply reeted: growth in the road network appears to be assoiated with omparative advantage in this speifiation. The results for Teleoms per worker are very different from those of Roads. While the oeffiient on this variable is statistially signifiant in three of twelve industries, it is found not to be statistially different from zero in the test of oint signifiane (see seond to last row). Similarly, the oeffiients are also found not to be different aross industries within the traded goods setor. These results suggest that teleommuniations do not appear to be a signifiant soure of omparative advantage. These weak results for teleommuniations ould reflet the rude nature of the measure of teleommuniations used here, notably the omission of the quality of the ost and quality of teleommuniations servies. Finally, we turn to Power, shown in the third olumn of Table 3. The oeffiients on Power are statistially different from zero aross industries as indiated by the test statisti in the seond to last row. Similarly to physial apital, an inrease in eletrial generating apaity appears to be assoiated with a shift of prodution out of manufaturing into servies, with the three manufaturing industries with statistially-signifiant oeffiients being negative and the statistially signifiant oeffiient in servies positive. Nevertheless, there are very substantial differenes in the oeffiients within manufaturing, and the Chi-squared value reported in the last row strongly suggests that the oeffiients are not idential aross traded goods industries. Hene, the availability of eletrial generating apaity also appears to be assoiated with omparative advantage. Separating the sample into developed and developing ountries yields rather divergent results, perhaps refleting differenes in the nature of the produts produed as well as greater 22
measurement problems in developing ountries, as noted above. Overall, it appears that infrastruture provision does affet the pattern of speialization but the results are less lear and robust than the effets of infrastruture on TFP. 5. Conlusion Infrastruture is one of the main andidates for explaining overall international disparities of produtivity between ountries absolute advantage. Clarida and Findlay (1992) also proposed that infrastruture ould be a soure of omparative advantage, i.e., industry-level produtivity variations aross ountries. We alulate total fator produtivity for eighteen developed and developing ountries and ten manufaturing industries, and study the effets of supplies of roads, teleommuniations and eletri power on international variations in setoral TFP. The analysis is ompliated by the endogeneity of infrastruture provision: ountries with a omparative advantage in a partiular setor may invest relatively more in infrastruture that raises produtivity in that setor. This paper therefore goes to onsiderable lengths to address identifiation and spurious orrelation, by fousing on growth rates rather than levels of produtivity and infrastruture, and through the use of a three-stage least squares estimation proedure. The use of differenes rather than levels, while aiding in identifiation, eliminates the information ontained in levels and makes it more diffiult to obtain statistially-signifiant results. It is therefore enouraging that we still find onsiderable evidene that inreased provision of infrastruture affets total fator produtivity. As expeted, inreased provision of infrastruture affets absolute advantage insofar as it tends to raise TFP in most setors, with road networks having a partiularly strong effet on TFP. Our results are therefore onsistent with the oneture by Leamer and Levinsohn (1995) that 23
some of the apparent differenes in tehnology aross ountries are attributable to infrastruture. We also found that the effet of infrastruture on total fator produtivity varied onsiderably aross setors as suggested by the Clarida-Findlay (1992) model, onfirming infrastruture as a soure of omparative advantage. This onlusion is reinfored by our finding that hanges in the availability of infrastruture alter industrial speialization, although our results here are less robust. Further researh should seek to refine the measures of infrastruture to inlude quality as well as quantity indiators, suh as power outages and failed telephone onnetions. Improved indiators might help to improve the results for teleommuniations and power, whih were less suessful than for roads. The soures of the different results for developed and developing ountries should also be explored further. 24
Referenes: Ashauer, David. 1989. Is Publi Expenditure Produtive? Journal of Monetary Eonomis 23(2). pp. 177-200. Barro, Robert and Jong-Wha Lee. 2001. International Data on Eduational Attainment: Updates and Impliations. Oxford Eonomi Papers 53(3). pp. 541-563. Bernard, Andrew B., and Charles Jones. 1996. Comparing Apples to Oranges: Produtibity Convergene and Measurement aross Industries and Countries. Amerian Eonomi Review 86(5). pp. 1216-1238.. Canning, David. 2001. The Contribution of Infrastruture to Aggregate Output. Mimeo, World Bank. Clarida, Rihard, and Ronald Findlay. 1992. Government, Trade and Comparative Advantage. Amerian Eonomi Review 82(2). pp. 122-127. Fernald, John. 1999. Roads to Prosperity? Assessing the Link between Publi Capital and Produtivity. Amerian Eonomi Review 89(3). pp. 619-638. Findlay, Ronald and John Wilson. 1987. The Politial Eonomy of the Leviathan. In Eonomi Poliy in Theory and Pratie. ed. Assaf Razin and Sadka Efraim. St. Mratin s Press. pp. 289-304. Golub, Stephen S.. and Hsieh, Chang-Tai. 2000. Classial Riardian theory of omparative advantage revisited. Review of International Eonomis, 8(2), 221-234. Gramlih, Edward, 1994. Infrastruture Investment: A Review Essay, Journal of Eonomi Literature 32(3). pp. 1176-1196. Hall, Robert, and Charles Jones. 1999. Why do Some Countries Produe so muh More Output than Others. Quarterly Journal of Eonomis 114(1). pp. 83-116. 25
Harrigan, James. 1997. Tehnology, Fator Supplies, and International Speialization: Estimating the Neolassial Model. Amerian Eonomi Review 87(4). Pp. 475-494. Harrigan, James. 1999. Estimation of Cross-Country Differenes in Industry Prodution Funtions. Journal of International Eonomis 47(2).pp. 267-293. Holtz-Eakin, Douglas. 1994. Publi-Setor Capital and the Produtivity Puzzle. Review of Eonomis and Statistis 76(1). pp. 12-21. Jones, Ronald. 2000. Globalization and the Theory of Input Trade. MIT Press. Leamer, Edward, and James Levinsohn. 1995. Testing Trade Theories and Prediting Trade Flows. In Handbook of International Eonomis vol. 3. ed. By Gene Grossman, and Kenneth Rogoff. North Holland. Moran, Theodore H. 2002. Beyond Sweeatshops; Foreign Diret Investment and Globablization in Developing Countries. Brookings. Roller, Lars-Hendrik and Leonard Waverman. 2001. Teleommuniations Infrastruture and Eonomi Development: A Simultaneous Approah. Amerian Eonomi Review, 95(4). pp. 1029-46. Trefler, Daniel. 1995. The Case of the Missing Trade and Other HOV Mysteries. Amerian Eonomi Review 85(5). pp. 1029-1046. Van Ark, Bart, and Dirk Pilat. 1993. Produtivity Levels in Germany, Japan, and the United States: Differenes and Causes. Brookings Papers on Eonomi Ativity: Miroeonomis, 2. pp. 1-48. World Bank. 2002. World Development Report: Building Institutions for Markets. 26
Table 1: Growth in Infrastruture as a funtion of Growth in Lagged Infrastruture and Lagged Endowments Roads Teleom Power Lagged Infrastruture Roads 0.49*** (0.07) 0.23** (0.11) -0.26** (0.13) Teleom 0.07** (0.04) 0.52*** (0.06) 0.15** (0.08) Power 0.05 (0.04) 0.21*** (0.06) 0.06 (0.07) Lagged Endowments Capital 0.17*** (0.06) 0.06 (0.10) -0.09 (0.12) Arable Land 0.03 (0.05) -0.14* (0.09) 0.16* (0.11) Low Skill Lab. -0.09** (0.04) 0.16** (0.07) 0.11 (0.08) Mod. Skill Lab. -0.03 (0.05) 0.30*** (0.08) 0.14* (0.09) High Skill Lab. -0.00 (0.03) 0.03 (0.05) 0.13** (0.05) R-sq. 0.57 0.68 0.29 Chi-2 (p-value) 131.09 (0.000) 202.09 (0.000) 37.73 (0.003) Notes: N=88. Std Errors in Parens. Coeffiients on time dummies suppressed. Supersripts *, **, and *** indiate that oeffiient is individually signifiant at 10, 5, 1 perent level respetively. 27
Table 2: TFP Growth as a funtion of Infrastruture Growth Industry Roads Teleoms Power Food 0.66 *** (0.21) -0.08 (0.12) 0.35 ** (0.14) Textiles 0.70 *** (0.19) 0.11 (0.11) 0.08 (0.13) Wood 0.78 *** (0.20) -0.12 (0.11) 0.08 (0.13) Paper 0.48 ** (0.19) 0.16 (0.11) 0.12 (0.13) Chemials 0.61 *** (0.21) -0.05 (0.11) 0.44 *** (0.13) Metal Prod. 0.62 *** (0.21) -0.01 (0.12) 0.07 (0.14) Mahinery 0.89 *** (0.20) 0.15 (0.11) -0.06 (0.13) Eletronis 0.62 *** (0.22) 0.10 (0.12) -0.10 (0.15) Transport 0.72 *** (0.28) 0.32 ** (0.15) 0.17 (0.19) Instruments 0.19 (0.27) 0.38 ** (0.15) -0.29 (0.19) Test-Coeff. Jointly Zero (p-value) 35.98 (0.000) 21.24 (0.020) 35.35 (0.000) Test-Coeff. Equal (p-value) 12.38 (0.193) 21.23 (0.012) 31.27 (0.000) Notes: N=88 for all industries. Standard errors in parentheses. Supersripts *, **, and *** indiate that oeffiient is individually signifiant at 10, 5, 1 perent level respetively. Coeffiients on time dummies are suppressed. 28
Table 3: Change in Industrial Speialization as a funtion of Change in Infrastruture and Fator Endowments Road Tele. Power Capital Land Low Skill Labor Agr. -1.9 2.6-1.9-2.0 0.1 1.7 * (3.5) (1.8) (3.0) (2.3) (0.9) (1.1) Food -0.9-1.3 ** 0.4-1.2 0.6 * 0.1 (1.3) (0.7) (1.1) (0.8) (0.3) (0.4) Textile -7.7 *** 1.4 * 1.2-1.5 0.4 1.5 *** (1.7) (0.9) (1.5) (1.1) (0.4) (0.5) Wood -0.2 0.9-4.1 *** -2.4 *** 0.1 0.7 * (1.2) (0.6) (1.0) (0.8) (0.3) (0.4) Paper -1.1-0.1 1.4-0.5 0.2 0.8 ** (1.2) (0.6) (1.1) (0.8) (0.3) (0.4) Chem. 1.6-0.9-0.8-2.9 *** 1.2 *** 1.7 *** (1.6) (0.8) (1.4) (1.0) (0.4) (0.5) Metal -1.7 ** 0.5-1.3 * -0.9 0.3 0.7 *** (0.9) (0.4) (0.7) (0.6) (0.2) (0.3) Mah. -4.1 * 1.8 ** -3.6 *** -1.3-2.2 ** 1.4 ** (2.4) (0.8) (1.3) (1.0) (0.9) (0.6) Elet. 7.8 * -1.6 0.2-3.9 *** 4.8 *** 0.9 (4.8) (1.4) (2.4) (1.8) (1.9) (1.1) Trans. -1.2 0.4-0.5-0.9-0.1 0.9 ** (1.2) (0.6) (1.0) (0.7) (0.3) (0.4) Si. -0.9 ** -0.1 0.2 0.1 0.1 0.0 Inst. (0.4) (0.2) (0.4) (0.3) (0.1) (0.1) Servie 7.2-1.7 10.8 * 12.2 ** -3.2 * -6.5 *** (3.2) (3.7) (6.2) (4.8) (2.0) (2.3) All 28.97 14.66 26.31 32.17 28.74 37.28 Zero? (0.00) (0.26) (0.01) (0.00) (0.00) (0.00) Traded 24.44 14.22 23.56 30.04 23.88 34.42 Mod. Skill Labor -0.2 (1.5) -0.1 (0.6) 1.1 (0.7) 1.0 ** (0.5) -0.4 (0.5) 1.3 * (0.7) 0.4 (0.4) 1.2 (0.9) 1.9 (1.7) 0.2 (0.5) 0.0 (0.2) -3.9 (3.1) 22.94 (0.03) 28.20 High Skill Labor 0.4 (1.0) 0.7 * (0.4) -1.4 *** (0.5) 0.5 * (0.3) 0.0 (0.3) -1.3 *** (0.5) -0.6 *** (0.2) 1.0 ** (0.5) -3.6 *** (0.8) -0.1 (0.3) -0.2 (0.1) 1.4 (2.1) 73.69 (0.00) 70.00 (0.00) RER 0.6 (0.6) -0.1 (0.2) 0.7 ** (0.2) 0.6 *** (0.1) -0.2 (0.2) 0.8 *** (0.3) 0.2 (0.2) 0.4 (0.3) 0.2 (0.5) 0.0 (0.2) 0.2 *** (0.1) -4.9 *** (1.3) Equal? (0.00) (0.16) (0.01) (0.00) (0.01) (0.00) (0.00) Notes: N=88 for all industries. Standard errors in parentheses. *, **, *** indiate oeffiient is individually signifiant at 10, 5, 1 perent level respetively. Coeffiients on ountry interepts and ountry time trends are suppressed. 29
Appendix 1: Data Soures and Methods 11 Produtivity. Our primary soure of data is the United Nations Industrial Development Organization (UNIDO) Industrial Statistis Database (INSTAT) at the 3-digit SIC level. This database ontains information on prodution, value added, apital formation, labor ompensation and employment for a large group of ountries. To fill the missing data and to provide a onsisteny hek we also made extensive use of the OECD s Strutural Analysis Industrial Database (STAN). The OECD strives for greater onsisteny and ompleteness than UNIDO but only overs OECD ountries, whih now inludes Korea and Mexio. Output is measured as real manufaturing value added per employee, onverted to $U.S. at an equilibrium or purhasing power parity (PPP) exhange rate. We used the aggregate manufaturing value added deflator, obtained from the World Development Indiators (WDI) to alulate real value added. Unfortunately, there are no available estimates of absolute PPP exhange rates for manufaturing for a number of ountries. As an alternative, we used the average real exhange rate over the sample period, using manufaturing value added deflators, as a rough indiator of the equilibrium real exhange rate. In omputing apital stoks, we followed Harrigan s (1999) in assuming a useful life of apital goods of ten years (T = 10) and a depreiation rate of 15 per ent (δ = 0.15). Finally, we adusted for varying apital utilization over time by estimating a trend for output by ountry and industry and adusting the apital stok figures for deviations from this trend. Table 1A onsiders the level of TFP by ountry and industry for 1988, whih is lose to the mid-point of our 1979-1997 sample. The overall levels of alulated TFP mostly aord well with intuition. India s TFP is of the order of 10 perent of the U.S. level, and the other 11 Further details on soures and methods are available from the authors upon request. 30
developing ountries in Table 1A suh are generally at 10-40% of the U.S. level. Japan s TFP is the losest to the United States and even exeeds the latter in two setors, while Canada s TFP approahes the level of the United States in a number of setors. Note that the levels of TFP vary onsiderably within ountries as well. For instane, Singapore s TFP is low in Food Proessing and Textiles but relatively high in Eletronis and Mahinery while TFP in the UK is relatively low in Chemials and high in Mahinery and Transport Equipment. Appendix Table 1B shows the average annual growth rates of TFP relative to the United States over our sample period, illustrating large disparities aross ountries, with a number of ountries TFP growing more rapidly than the United States, partiularly in East Asia, whereas others, notably in Western Europe, grow more slowly than the United States. Infrastruture. Appendix Table 2A and 2B show levels and rates of hange of fator endowments and infrastruture. The infrastruture struture levels onfirm intuition that the wealthier ountries have the largest relative stok of infrastruture, with the less developed ountries, suh as India, Indonesia, South Afria, and Colombia, having muh lower levels of infrastruture. Given that our estimation strategy will fous on differenes aross time, we show the average annual growth rates of infrastruture and endowments in Table 2B. These data reveal that it is indeed the least developed ountries that inreased the size of their infrastruture stok most substantially over the sample period. In partiular, Indonesia, one of the ountries with the most rapid produtivity growth, led all other ountries in expanding teleommuniations and eletrial generating apaity. Korea and Turkey, two other ountries enoying rapid produtivity growth, also aumulated infrastruture at a partiularly rapid pae. Infrastruture growth apparently did not guarantee produtivity growth, however, with India being a 31