Factor Endowment, Structural Coherence, and Economic Growth

Transcription

1 WP/12/165 Facor Endowmen, Srucural Coherence, and Economic Growh Naasha Xingyuan Che

2 2012 Inernaional Moneary Fund WP/12/165 IMF Working Paper STA Facor Endowmen, Srucural Coherence, and Economic Growh Prepared by Naasha Xingyuan Che Auhorized for disribuion by Mohammed El Qorchi June 2012 This Working Paper should no be repored as represening he views of he IMF. The views expressed in his Working Paper are hose of he auhor(s) and do no necessarily represen hose of he IMF or IMF policy. Working Papers describe research in progress by he auhor(s) and are published o elici commens and o furher debae. Absrac This paper sudies he linkage beween srucural coherence and economic growh. Srucural coherence is defined as he degree ha a counry's indusrial srucure opimally reflecs is facor endowmen fundamenals. The paper found ha a leas for he overall capial, he shares of capial inensive indusries were significanly bigger wih higher iniial capial endowmen and faser capial accumulaion. Moreover, here is a posiive relaionship beween a counry's aggregae oupu growh and he degree of srucural coherence. Quaniaively, he srucural coherence wih respec o he overall capial explains abou 30% of he growh differenial among sample counries. JEL Classificaion Numbers: O1, O4, E2 Keywords: Srucural Coherence, Economic Growh, Srucural Change, Facor Endowmen, Capial Accumulaion Auhor s Address: nche@imf.org

3 I. INTRODUCTION Alhough neoclassical growh models generally feaure balanced growh pah, in realiy he indusrial composiion of economies experience coninuous shifs, accompanied by massive reallocaion of labor and producion resources across secors. Invesigaions on he causes of srucural change have been mosly heoreical. A recen example is Acemoglu & Guerrieri (2008), who modeled srucural change as a resul of capial accumulaion. In heir wo-secor model, as capial becomes more abundan oupu increases in he capial-inensive secor, while he direcion of employmen composiion change depends on he elasiciy of subsiuion beween secors. 1 Ju, Lin & Wang (2008), focusing more on developing counries, arrived a similar conclusions: as capial accumulaes, a counry's indusrial srucure "upgrades" owards more capial-inensive indusries. Moreover, hey argue ha when he indusrial srucure is no coheren wih he capial endowmen level, i can lead o subopimal economic growh performance. 2 Ju, Lin & Wang's predicion abou he linkage beween srucural coherence and economic growh can also be derived from Acemoglu & Guerrieri (2008)'s framework, hough no explicily discussed in heir paper. The inuiion is sraighforward: in Acemoglu & Guerrieri's paper, oupu composiion change owards capial-inensive indusries is he naural resul of he agens' opimal decision as capial accumulaes. Hence, any arrangemen ha obsrucs he srucural change process owards alignmen wih facor endowmens is no an opimal choice and herefore has a negaive impac on long-run growh. Alhough i is beyond he scope of he curren sudy o idenify specific causes of srucural incoherence, he incoherence beween indusrial srucure and facor endowmen can be caused by such facors as over-resricive labor marke regulaion, lack of compeiion in cerain indusries, and echnology barriers, as idenified in relaed lieraure. 3 The major goal of his paper is o empirically examine he relaionship beween capial endowmen and indusrial srucure, and o esimae srucural coherence' impac on growh. Here is an overview of he main empirical resuls. For he overall capial, he daa shows ha he capial-inensive indusries' oupu and employmen sizes are larger when capial endowmen is higher, and growh in capial endowmen also leads indusrial srucure o shif owards capial-inensive indusries. Similar resuls apply, o various degrees, o deailed 1 There are oher explanaions of srucural change, o be sure. On he supply side, for example, Ngai & Pissarides (2007) models indusrial composiion change as a resul of uneven raes of TFP growh across secors. The demand side lieraure explains srucural change as a combined resul of nonhomoheic consumer preference and income growh (Echevarria (1997), Lainer (2000), Buera & Kaboski (2007)). Thus in he empirical regressions, I will conrol for hese oher facors ha poenially affec srucural change process. 2 In an earlier work, Hollis Chenery (1979) made a similar poin. He argues ha counries ha are shor on capial, in considering heir developmen policy, should choose indusries and producion echniques ha have low capial o oupu raio. 3 The linkage beween srucural change and aggregae economic performance have been discussed in some recen macroeconomic lieraure, such as, Nickell, Redding & Swaffield (2004), Rogerson (2007), van Ark, O'Mahony & Timmer (2008), and Baily (2001).

4 4 ypes of physical capial. 4 In erms of he relaionship beween srucural coherence and growh, he resuls show ha a counry's aggregae growh performance is significanly and posiively associaed wih he coherence level beween indusrial srucure and capial endowmen. In he counry-level regression, srucural coherence relaed o he overall capial explains abou 30% of he variaion in counry GDP growh. The indusry-level regression indicaes an effec of similar magniude. Moreover, he indusry-level resuls are mosly robus o changing he measuremen of capial inensiy and o conrols for oher indusry characerisics and srucural change deerminans. The paper is relaed o a large empirical inernaional rade lieraure ha aims o es Heckscher-Ohlin heorem and Rybczynski heorem. 5 Recen examples of his lieraure are Harrigan (1997), Reeve (2002), Romalis (2004) and Scho (2003). Some of hese papers found ha endowmen and change of endowmen in physical capial and/or human capial has a significan impac on rade paerns or indusrial srucure. 6 There are obvious differences in erms of he underlining heory beween he presen paper and mos of ha lieraure. Secoral srucural change induced by facor endowmen change is a process independen of wheher he counry is an open economy or no. Thus he presen paper covers all indusries in an economy, regardless of wheher he producs are considered radable or no. In erms of mehodology, mos of he endowmen-relaed rade sudies assume idenical capial inensiies of indusries across counries, or a leas he same capial inensiy ranking in differen counries. Thus he lieraure ofen uses indusry characerisics in one counry as proxies for all oher counries. Though a reasonable assumpion when counries are relaively similar, his assumpion is no necessarily rue as will be shown in Secion 3. 7 This paper allows he capial inensiy ranking of indusries o change across counries and over ime. The paper is also relaed o empirical invesigaions of allocaive efficiency across indusries and firms (e.g., Barelsman, Haliwanger & Scarpea (2008), Arnold, Nicolei & Scarpea (2008)). This srand of lieraure mainly focuses on efficiency in resource allocaion according o firm/indusry's produciviy level, insead of resource allocaion according o consisency wih facor endowmens. To my bes knowledge, he presen paper 4 My focus in his paper is mosly fixed physical capial. The mechanism examined here can apply o inangible capial, oo. Che (2009) argues ha he increasing imporance of inangible capial in he producion process is a cause of secoral srucural change in advanced economies. However, he es on inangible capial is difficul o execue in a cross-counry seing due o daa limiaions. 5 These heorems sae, respecively, ha differences in counries' expors are deermined by differences in heir facor endowmens, and ha a rise in he endowmen of a facor will lead o more han proporional oupu increase in secors ha use he facor inensively, given consan goods prices. 6 Fizgerald & Hallak (2002) gives an excellen review of recen empirical lieraure in rade ha is relaed o facor endowmens. 7 Lewis (2006) shows ha producion echniques wihin he same indusry vary even wihin US across differen regions according o he producion facor mix of he region. Sco (2003) finds ha capial abundan counries end o use more capial-inensive echniques in all indusries.

5 5 is he firs one o examine he impac of indusrial srucure-facor endowmen coherence on economic growh. The paper is organized as follows. Secion 2 provides a simple heoreical framework o explain he relaionship beween capial endowmen, srucural coherence and growh. Secion 3 discusses he daa and defines measures of variables. Secion 4 and 5 presen he empirical models, a counry and indusry level respecively, and discuss he esimaion resuls. More resricions o he indusry-level esimaion and robusness checks are added in Secion 6. Secion 7 concludes. II. AN ILLUSTRATIVE MODEL To examine he relaionship beween srucural coherence and growh, consider a simple wo-secor model adaped from Acemoglu & Guerrieri (2008). In he model economy, a single final good is produced by combining wo secoral goods, he elasiciy of subsiuion beween he wo secors equal o [0, ) : 1/ 1/ Y Y Y 1 1, 2 2, / 1 where There is one firm in each secor. Boh secors producion funcions are Cobb-Douglas wih capial and labor as producion inpus: Y ai 1 ai i, i, i, AK L (1) For simpliciy, le s assume ha he wo secors share he same produciviy level, A, while Secor 1 is more capial-inensive han Secor 2, i.e., a1 a2 0. Le he price of he final good P 1, hen he prices for he wo secoral goods can be expressed as 1/ 1/ Y Y P1, 1, and P2, 2 Y 1, Y 2, PY 1, 1, Thus he direcion of change in he raio of nominal oupu beween he wo secors, PY, 2, 2, corresponding o a change in he real oupu raio Y1, / Y 2, will depend on he value of. When >1, he nominal oupu raio moves in he same direcion as he real oupu raio, and he opposie is rue for <1. Assume ha labor is freely mobile beween he wo secors in any given period. Labor marke clearing implies

6 6 L L L (2) 1, 2, where L is he labor supply a ime, which is exogenously given. Capial is also mobile across secors. However, changes in he allocaion of capial G K / K s whenever resource are cosly. I manifess as a posiive adjusmen cos 1, 2, he raio beween he wo secors capial differs from a predeermined value s, which may be equal o, say, some hisorical raio beween K 1 and K 2. Capial marke clearing requires 1, 2, 1, 2, K K G K / K s K, (3) where K is he aggregae capial sock a ime. G 0 0, G' 0, and G" 0. Specifically, assume ha G() akes a quadraic form: K1, GK1,, K2, s K 2, 2 (4) where 0. The exisence of adjusmen cos inroduces fricion ino he cross-secor movemen of resources, hus can poenially aler he exen of secoral srucural change compared o he case of fricionless economy. Assume ha he markes are complee and compeiive. The equilibrium of he economy can be solved as a social planner s problem ha maximize he uiliy of he represenaive household, UC C K Y K. Given capial sock problem is o solve, subjec o he aggregae resource consrain for he economy: K in each period, he inra-emporal componen of he planner s max L1,, L2,, K1,, K2, 1/ 1/ Y Y Y 1 1, 2 2, / 1 (5) subjec o (1), (2), (3), and (4). Solving (5) requires he marginal producs of capial and labor in he wo secors being equal, which implies: 1/ 1/ Y Y 1, 1, 2, 2 2 K Y Y K 1, K 1, 1a1 / 1 s 2a2 / 1 s 2 Y 1, K1, K 2, K 2, Y 2, K2, K 2, K 2, (6)

7 7 1/ 1/ Y Y Y 1a 1a 1, 2, Y 1, L 1, Y 2, L2, Y (7) Denoe he share of capial and labor allocaed o he capial inensive secor (Secor 1) as K L, K K L L 1, 1, 1, 2, 1, 2, Then (6) and (7) imply ha 2 s K 11/ 1 2a Y 2 2, 2 1a 1 Y 1, s 1 (8) , 1 1 a 1 Y 1, a 1 a 1 a Y 11/ 1 1 K 2 s a a1 K s 2 s (9) Noice ha Y1, / Y 2, is equal o a1 a2 1a1 a21 a1a2 a2a1 (1 ) (1 ) K L, (10) Therefore, given he resource allocaion and unchanged, an increase in K will disproporionaely raise he real oupu of Secor 1 over Secor 2, and he opposie is rue when labor endowmen increases. Plugging (10) ino (8) and aking derivaive of boh sides of (8) wih respec o K, we arrive a he following proposiion describing he relaionship beween changes in cross-secor resource allocaion and changes in aggregae capial sock in any given period. PROPOSITION: In he saic equilibrium, 1a1a21 ln 0 1 ln K 1 a a (11)

8 8 where, when s K1, / K2,, can be expressed as a a a (12) The proposiion says ha when he elasiciy of subsiuion beween secors is greaer han one (which is he relevan scenario in our empirical invesigaion as Secion 4 and 5 will show), increasing aggregae capial sock will lead o capial being shifed o he capial inensive secor (Secor 1). From (9), we know ha is increasing in. Thus Secor 1 s labor share will also increase wih capial sock, when 1. However, he degree of his shif is subdued by he presence of srucural adjusmen cos, as is a posiive funcion of and i is easy o see from (11) and (12) ha ln / ln K 0 (13) Wha follows from (13), combined wih (10), is ha he secoral srucural change in erms of real oupu when capial sock increases is suppressed by he presence of srucural adjusmen cos: 2 ln( Y1, / Y2, ) 0, when 1. ln K The effec on he oupu of he final good is also sraighforward. When 0, he resource allocaion prescribed by he soluion o (5) achieves he maximized value of Y given he amoun of capial endowmen. In oher words, Y K 0 Y max K. Therefore, wih posiive srucural adjusmen coss, he increase in Y corresponding o an increase in capial sock is lower compared o he case of zero adjusmen cos: Y Y. K K 0 0 The main conclusion o draw from he heoreical discussion is wo folds. Firs, increasing capial endowmen is likely o be accompanied by srucural change owards he capial inensive secors and indusries in erms of real oupu. The change of indusrial composiion in erms of employmen and nominal oupu depends on he elasiciy of subsiuion beween

9 9 indusries. If he elasiciy is above uniy, hen nominal oupu shares and employmen shares of capial-inensive indusries will also rise as capial endowmen increases. Second, if for any srucural reasons he cross-secor reallocaion of resources is hindered, hen he indusrial srucure may become insensiive o he changes in facor endowmen. And his lack of responsiveness in indusrial srucure can lead o subopimal economic performance a he aggregae level. The subsequen par of he paper will empirically examine boh predicions. III. DATA AND VARIABLES The daa used in his paper is from he EU KLEMS daabase sponsored by he European Commission. The daabase provides indusry oupu, employmen, price, capial sock and invesmen daa from 1970 o 2005 for boh EU counries and several non-eu counries. 8 Table 1 liss he indusries covered, he cross-counry median growh raes of heir real oupu shares, employmen shares and nominal oupu shares over he 35-year period, and he crosscounry medians of indusry s overall capial inensiy. 9 Indusries are sored by heir median real oupu share growh. I is worh noing ha alhough he indusrial composiion change is differen for each counry, in general he real oupu composiion is shifing owards service indusries and a few more sophisicaed manufacuring indusries. This is consisen wih he sylized facs abou srucural ransformaion documened in he exising lieraure abou US and oher advanced economies. Employmen composiion has a similar rend o real oupu composiion, ye shows an even sronger shif owards service indusries. The median growh rae for nominal oupu shares has he same sign as employmen shares bu for seven indusries. Consisen wih common percepions, some indusries ha are radiionally perceived as labor inensive, such as exile and food indusries, have relaively low median capial inensiy. Somewha couner-inuiive, hough, cerain sereoypical capial-inensive manufacuring indusries, such as machinery and basic meals, do no have paricularly high median capial inensiy according o Table 1; in conras, service indusries such as social and personal services, healh, reail, finance and educaion show up as relaively capial inensive. The reason is ha alhough hese service indusries are no inensive in machinery capial, hey are generally more inensive in ICT capial and srucure capial, hus boosing heir overall capial inensiy scores. The opposie is rue for some basic manufacuring indusries ha rely heavily on machinery, bu are no paricularly inensive in he oher wo caegories of capial. On he whole, here is a posiive correlaion beween indusry s median real oupu share growh and median overall capial inensiy, wih a correlaion coefficien equal o 0.25 a 1% significance level. 8 The paper covers 15 counries: Ausralia, Ausria, Denmark, Finland, Germany, Ialy, Japan, Korea, Neherland, UK, USA, Czech, Porugal, Slovenia, and Sweden. Daa for he las 4 counries is only available saring he mid 1990s. 9 Capial inensiy is calculaed as indusry real capial sock over real oupu.

10 10 Table 1:Cross-counry median indusry size growh and capial inensiy Median share growh rae from 1970 o 2005 Median capial inensiy (Overall Indusry Real oupu share Employmen share Nominal oupu share capial sock/oupu) Texiles, Texile, Leaher And Foowear Mining And Quarrying Coke, Refined Peroleum And Nuclear Fuel Food, Beverages And Tobacco Consrucion Wood And Of Wood And Cork Hoels And Resaurans Oher Non-Meallic Mineral Manufacuring Nec; Recycling Pulp, Paper, Paper, Prining And Publishing Educaion Basic Meals And Fabricaed Meal Reail Trade Sale And Repair Of Moor Vehicles And Moorcycles Oher Communiy, Social And Personal Services Wholesale Trade And Commission Trade Real Esae Aciviies Transpor And Sorage Healh And Social Work Machinery, Nec Chemicals And Chemical Producs Elecriciy, Gas And Waer Supply Rubber And Plasics Transpor Equipmen Financial Inermediaion Elecrical And Opical Equipmen Rening Of M&Eq And Oher Business Aciviies Pos And Telecommunicaions * Real oupu, employmen and nominal oupu share growh is calculaed as log (share) in 2005 minus log (share) in Capial inensiy of indusry is calculaed as indusry s real overall capial sock divided by real oupu. The able repors he cross-counry medians of share growh and capial inensiy for each indusry. Figure1 and Table 2 presen he rend of aggregae labor income shares by counry. In 13 ou of he 15 counries covered, labor s share has declined over he sample period. This resul is consisen wih he fac ha he indusrial srucure of he sample counries is moving owards more capial inensive indusries The decline of labor income share in hese counries has been documened in previous lieraure. See, for example, Blanchard (1997), Benolila & Sain-Paul (2003), de Serres, Scarpea & de la Maisonneuve (2002), and Arpaia, Perez & Pichelmann (2009).

11 11 Figure 1: Evoluion of labor income share by counry AUS AUT DNK FIN Labor income share in aggregae value added GER ITA JPN KOR NLD SWE UK USA Table 2: Evoluion of labor income share over ime Aggregae labor income share counry % change: AUS AUT CZE n.a DNK FIN GER ITA JPN KOR NLD PRT SVN n.a SWE UK USA *Labor share measured as (1 CAP/VA) for code = TOT The overall capial endowmen of a counry is calculaed as he log of oal real fixed capial sock over oal labor. The overall capial sock consiss of differen ypes of capial, whose roles are arguably unique in he producion process and can be seen as differen producion facors. Examining he relaionship beween srucural change and hose deailed ypes of capial endowmen will allow us see if he heory s predicions can universally apply o differen producion facors. Therefore, in addiion o he overall capial, he paper examines hree deailed caegories of capial: ICT, machinery and non-residenial srucure. However, endowmen for hese deailed ypes of capial are more complicaed o measure. Alhough he absolue socks for all hree ypes of capial have been increasing over ime in all counries, heir relaive imporance in he oal capial sock has changed considerably. Figure 2 repors he share changes of each ype of capial in oal capial sock by counry. Noice ha ICT capial s imporance has risen in all counries while he share of srucure capial has almos universally declined. If we consider differen ypes of capial as differen

12 12 producion facors, he endowmen measure should ake ino accoun boh he absolue quaniy change in capial-x sock agains labor and is relaive change agains oher ypes of capial as well. Therefore, capial-x endowmen is calculaed as he log of capial-x sock x over oal labor muliplied by he share of capial-x ( K ) in he overall capial sock( K ) of counry j: K _ENDW ln K / L K / K x x x j, j j j j According o his definiion, he change in capial-x endowmen can be expressed as x x x K K K K_ENDW = x x K K K where K denoes he K / L raio. In oher words, he change in capial-x endowmen consiss wo pars: he percenage change in he value of and he difference beween he percenage changes of and of he overall capial-labor raio K. 0.3 Figure 2 Change of shares in oal capial by capial ypes UK FIN JPN PRT USA AUS AUT NLD GER DNK CZE SVN ITA KOR SWE Share change of machinery share change of srucure Indusry s capial sock o real oupu raio is used as he main measure of capial inensiy. 11 For robusness check, he paper also uses capial s income share in indusry value-added as an alernaive measure. Human capial inensiy is used as conrol variable in some of he regressions, which is measured by high-skill workers compensaion as a percenage of indusry s oal compensaion. Figure 3 plos indusry oupu share-weighed 11 Some sudies also used capial sock over value added raio as a measure of capial inensiy; see for example, Nunn (2007) and Ciccone & Papaioannou (2009). The wo measures are highly correlaed.

13 13 average capial inensiies a counry level for differen ypes of capial. For all ypes of capial he average inensiies differ across counries. Moreover, a leas in some counries, capial inensiies are no saionary. This is especially rue for ICT capial, he usage of which has experienced surges in all sample counries especially since he 1990s. Even wihin he same indusry, here are ofen big differences in capial inensiy across counries. This difference urns ou o be significanly relaed o he counries capial endowmens. Table 3 presens resuls of regressing capial inensiy on counry capial endowmen indusry by indusry for hree deailed ypes of capial. The regression coefficiens are posiive and highly significan for he majoriy of indusries. There can be differen facors causing he posiive correlaion. Since he indusry classificaion used here is fairly broad, wihin he same indusry differen counries may be specializing in very differen sub-indusries according o a counry s endowmen fundamenals. And even when differen counries are producing a similar produc or service, he echniques hey use can differ so as o ake advanage of he more abundan facor in he counry. The finding is consisen wih Blum (2010), who found ha a producion facor is more inensively used in all indusries of a counry when he facor becomes more abundan. Since cross-counry differences or ime rends in capial inensiy is no a focus of his paper, and because correlaion beween capial endowmen and indusry capial inensiy can poenially cause mulicolineariy in he regressions, he sandard score of capial inensiies insead of he raw capial-oupu raio is used in he acual esimaions. The sandard score is calculaed by normalizing an indusry s capial-x inensiy in counry j of ime wih he mean and sandard deviaion of capial-x inensiy of all indusries in counry j a ime. The capial inensiy score hus has he same disribuion wihin each counry and ime period, and measures he wihin-counry variaions of capial inensiy across indusries a a poin in ime. Figure 3: Capial inensiy by counry and ypes of capial ICT capial inensiy Machinery capial inensiy Srucure capial inensiy Overall capial inensiy

14 14 Table 3: Regression of capial inensiy on counry capial endowmen by indusry ICT capial Machinery capial Srucure capial Indusry code b 1 T value R square b 1 T value R square b 1 T value R square AB C E F H J L M N O * The esimaion equaion is capial inensiyi, j, b0, ib1, icapial endowmen j, e. The equaion is esimaed for every i, j, indusry i, andb 1 is he coefficien of capial endowmen. Table 4A: Summary saisics # of observaions Mean Sd. Dev. Min Max Counry variables Overall Capial endowmen ($mn) ICT capial endowmen Srucure capial endowmen Machinery capial endowmen Annual growh rae of GDP per worker Log GDP per worker ($mn) Indusry variables Real oupu share Employmen share Nominal oupu share * Overall capial endowmen of a counry is calculaed as he log of real overall capial sock over oal employmen raio. Endowmens of he deailed ypes of capial are measured as he log of capial-x sock over oal employmen raio imes he log of capial-x s share in he overall capial sock. Table 4B: Correlaion beween counry variables Capial GDP ICT Srucure Machinery Overall Capial endowmen 1.00 Log GDP per worker ICT endowmen

15 15 Srucure endowmen Machinery endowmen Table 4C: Correlaion beween indusry variables Overall capial ICT Srucure Machinery Human capial Value-added Overall capial inensiy 1.00 ICT inensiy Srucure inensiy Machinery inensiy Human capial inensiy Degree of value-added * All capial inensiies are in sandard score form. Table 4 liss summary saisics of main variables and heir correlaions. A number of correlaions are noeworhy. Firs, richer counries generally have higher capial endowmens. The correlaion beween per worker GDP and he four caogories of capial are 0.83, 0.42, 0.66 and 0.68 respecively, all significan a 1% level. I raises he quesion of wheher he capial endowmen variables are simply sand-in facors for counry s developmen sage. Second, indusries ha are inensive in overall capial, ICT and srucure capial also end o be human capial inensive. One explanaion for he posiive correlaions may be ha he sophisicaed indusries end o be inensive in muliple ypes of capial. hese quesions will be revisied laer in he robusness check secion. IV. COUNTRY LEVEL ANALYSIS A. Capial Endowmen and Indusrial Srucure Before empirically defining and analyzing srucural coherence, le s firs look a he general paerns in daa abou he relaionship beween capial endowmen and capial inensiy of he indusrial srucure. One conclusion from Secion 2 is ha here should be a posiive correlaion beween he wo when indusry size is calculaed as he real oupu share, since capial-inensive indusries grow bigger in erms of real oupu-- when capial endowmen increases. When indusry shares are calculaed in erms of employmen or nominal oupu, he relaionship beween capial endowmen level and capial inensiy of he indusrial srucure depends on, he elasiciy of subsiuion beween secors, as he elasiciy of subsiuion deermines he magniude of changes in he relaive price corresponding o real oupu changes. However, in realiy several facors can complicae he predicion. Firs, a real economy has more han wo indusries and he elasiciies of subsiuion across differen indusries can be differen. Second, as poined ou by Oulon (2001), many indusries produce inermediae goods ha do no arge end consumers, hus making he predicion by elasiciy-of-subsiuion-crieria hard o apply. Third, he counries in he sample are mosly open economies. Hence he domesic demand may have lile impac on goods prices, especially for radable indusries in small counries. Alhough hese facors complicae he predicion for he relaionship beween capial endowmen and employmen/nominal oupu share disribuion of he indusries, a leas i should be he case ha an indusry s

16 16 employmen share and nominal oupu share should move in he same direcion when endowmen changes. Table 5a and 5b repor correlaions among endowmens in differen ypes of capial and capial inensiy of indusrial srucure in erms of real oupu, employmen and nominal oupu. The capial x inensiy of indusrial srucure is measured in wo ways: (1) as COR( Yij,, K ij, ), he Spearman rank correlaion beween an indusry s capial-x inensiy score,, and indusry size ij Y ij, which is represened by he real oupu share, employmen share, and nominal oupu share of he indusry in n x he oal economy of counry j; (2) as i1 Kij, Yij,, he indusry-size-weighed average capial inensiy score across all n indusries of he economy. From now on, he paper will refer o he wo measures as correlaion measure and weighed average measure of he capial inensiy of indusrial srucure. 12 Keep in mind ha since ij is he sandard score of capial-x inensiy, i capures he ranking of capial inensiy of indusry i relaive o oher indusries wihin he same counry and ime period, independen of he average capial inensiy of he counry. The laer is iself a posiive funcion of he counry s capial endowmen, as shown in secion 3 and in Blum (2010). The resuls from Table 5a-b show ha for he overall capial, he capial inensiy of indusrial srucure, no maer wheher i is calculaed in erms of real oupu, employmen, or nominal oupu shares, is posiively correlaed wih capial endowmen level. All he correlaion coefficiens are significan a 1% level. In erms of magniude, he correlaion coefficien is highes for real oupu srucure, and lowes for he employmen srucure. These paerns in he daa are presen in using boh correlaion measure (Table 5a) and weighed average measure (Table 5b) for he capial inensiy of indusrial srucure. The capial inensiies using all hree indusry size measures are also posiively and significanly correlaed. Overall, hese resuls are consisen wih he assumpion ha he elasiciy of subsiuion beween indusries is generally greaer han 1. Table 5a: Correlaion beween capial inensiy of indusrial srucure and capial endowmen (correlaion measure) inensiy of real oupu srucure inensiy of real oupu srucure Overall Capial ICT Capial Machinery Capial Srucure Capial inensiy of employ men srucure inensiy of nominal oupu srucure Lagged endowm en inensiy of real oupu srucure inensiy of employ men srucure inensiy of nominal oupu srucure Lagged endowm en inensiy of real oupu srucure inensiy of employ men srucure inensiy of nominal oupu srucure Lagged endowm en inensiy of real oupu srucure inensiy of employ men srucure inensiy of nominal oupu srucure Lagged endowm en 12 The wo measures have heir respecive pros and cons. For example, he weighed average measure capures more variaions in capial inensiy of indusries han he correlaion measure, bu is sensiive o capial inensiy changes in individual indusries ha can be considered as ouliners. Therefore, empirical resuls using boh measures are repored in his paper.

17 17 inensiy of employmen srucure inensiy of nominal oupu srucure Lagged endowmen Table 5b: Correlaion beween capial inensiy of indusrial srucure and capial endowmen (weighed average measure) inensiy of real oupu srucure inensiy of employmen srucure inensiy of nominal oupu srucure inensiy of real oupu srucure Overall Capial ICT Capial Machinery Capial Srucure Capial inensiy of employ men srucure inensiy of nominal oupu srucure Lagged endowm en inensiy of real oupu srucur e inensiy of employ men srucure inensiy of nominal oupu srucure Lagged endowm en inensiy of real oupu srucure inensiy of employ men srucure inensiy of nominal oupu srucure Lagged endowm en inensi y of real oupu srucur e inensiy of employ men srucure inensiy of nominal oupu srucure Lagged endowm en Lagged endowmen The resuls for he deailed ypes of capial are somewha similar o hose for he overall capial. For boh ICT and srucure capial, capial inensiies of indusrial srucure are posiively correlaed wih capial endowmen levels. The correlaion coefficiens are significan a 1% level excep for he correlaion beween he non-residenial srucure capial inensiy calculaed using indusry employmen shares and he srucure capial endowmen, which is posiive bu no significan. In conras, he correlaions beween capial inensiy of indusrial srucure and capial endowmen are negaive for machinery capial, no maer which indusry size measure is used. Despie hese excepions, in general he resuls from Table 5a-b sugges ha he indusrial srucure ends o be more capial inensive when capial is more abundan. This is, however, a very general descripion of he daa. The counries ha have similar levels of capial abundance no necessarily share he same indusrial srucure in erms of capial inensiy. Wha happens if he capial inensiy level of a counry's indusrial srucure is no "coheren" wih he level of he counry's capial endowmen? Does he level of his coherence maer for a counry's growh performance? One way o answer hese quesions is o consruc a counry-level measure for he degree of coherence beween indusrial srucure and capial endowmen, and relae i o economic growh. The nex secion will implemen his approach.

18 18 B. Srucural Coherence and Growh The paper uses he erm srucural coherence o refer o he degree ha a counry s indusrial srucure aligns wih he counry s facor endowmen fundamenals. The endowmen-based srucural change heory predics ha he indusrial srucure will change owards more capial-inensive indusries when he endowmen of capial increases, given no disorions o he marke sysem and o individual incenives. However, as Secion 2 argues, when adjusmen cos associaed wih srucural change is high, he magniude of srucural change will be reduced and he aggregae growh performance negaively impaced. Empirically, previous sudies have shown ha he characerisics of srucural change have aggregae effecs on counries labor marke performance (Rogerson, 2007) and on aggregae produciviy (van Ark, O Mahony & Timmer, 2008; Duare & Resuccia, 2010). Bu lile empirical evidence exiss on wha kind of indusrial srucure faciliaes growh. This secion firs proposes a measure for srucural coherence a he counry level, and hen shows ha he measure can explain some of he cross-counry variaion in growh. Measuring Srucural Incoherence a he Counry Level The paper measures srucural coherence by is opposie srucural incoherence, ha is, he degree ha a counry s indusrial srucure deviaes from he opimal corresponding o he counry s capial endowmen level. The srucural incoherence (SI) index in erms of ype-x capial is measured as he absolue gap beween he sandardized capial-x inensiy score of a counry s overall indusrial srucure and he counry s capial endowmen level, also sandardized across counries. In oher words, he SI index can be expressed as SI k _inen k _endw (14) x x x j, j, j, 1 Here lower-cased leers are used o represen he sandard score of he acual variable. Thus he wo componens of he SI index respecively indicae where a counry is in erms of capial inensiy of indusrial srucure and capial endowmen, relaive o oher counries. This measure formulaes upon he idea ha he capial inensiy of he opimal indusrial srucure should be a sricly increasing funcion of a counry s capial endowmen level. Thus in he case of perfec srucural coherence, he SI index should be equal o zero; i.e., he level of he indusrial srucure s capial inensiy should be he same as he level of capial endowmen, in heir respecive disribuions. Again, o ake ino accoun he ime lags needed for he indusrial srucure o adjus o changes in capial endowmen, he capial inensiy and endowmen scores used are hose a he ending and beginning years of a 5-year window. Table 6 gives summary saisics of he SI index for he overall capial and hree deailed caegories of capial. In Version 1 of he SI index, he capial inensiy of a counry s indusrial srucure is measured as he rank correlaion beween indusries real oupu shares and indusries capial inensiies, while in Version 2, i is measured as he indusry-real-oupu-share-weighed average of indusry capial inensiies. Table 6 shows ha he wo versions of SI are of similar ranges. I is illuminaing o compare he srucural incoherence scores across counries and over ime. Figure 4 presens he ime rends of he SI score (Version 1) in erms of he overall

19 19 capial for 11 sample counries ha have relaively long ime-series daa. A few hings are worh noing. Among all counries, Japan has experienced he larges increase in srucural incoherence over ime, is SI scores close o zero in he 1970s and above 3 in The SI score has also increased since he 1980s in counries such as Ialy and Denmark, hough o a less degree. In conras, counries like US and Germany seem o have consisenly lowerhan-average SI scores. For US, he score has decreased from he beginning of he sample, and was especially low during he 1990s, a period of exraordinary economic growh for he counry. Germany s SI score periodically increased righ afer he re-unificaion bu decreased again in he lae 1990s. Table 6 Summary saisics of srucural incoherence (SI) scores Mean Sd. Dev. Min Max SI (version 1): Overall capial ICT Machinery Srucure SI (version 2): Overall capial ICT Machinery Srucure To see which of he wo componens of he SI score is driving he changes over ime, Figure 5 ploed he ime rends for he capial inensiy of indusrial srucure (correlaion measure) and capial endowmen ( k _endw j ) separaely for each counry. The cause for he dynamics in SI score is now clearer. For all sample counries, he capial endowmen has increased overime o various degrees. However, he rend of indusrial srucure is far less universal. For some counries such as US, Germany, and UK, he capial inensiy of indusrial srucure has risen along wih he movemen of capial endowmen, which resuls in seady or even decreasing srucural incoherence level overime. For he counries whose SI scores have been increasing, e.g. Japan, Ialy, and Denmark, he rise in srucural incoherence level is mainly caused by heir sicky indusrial srucure, i.e. he lack of upward movemen in he overall capial inensiy of he indusries, despie consisen capial accumulaion. Also noice ha compared o he US, all he coninenal European counries excep Germany appear o have less responsive indusrial srucure o he changes in capial endowmen. This is consisen wih previous sudies comparing he characerisics of srucural change beween US and EU counries. For example, van Ark, O Mahony & Timmer (2008) show ha he slower srucural ransformaion in European counries conribues o he lower labor produciviy growh in Europe compared o he Unied Saes.

20 20 Figure 4: Evoluion of srucural incoherence score by counry AUS AUT DNK FIN Srucural Incoherence Score (SI) GER ITA JPN KOR NLD UK US Figure 5: Decomposing he srucural incoherence score AUS AUT DNK FIN GER ITA JPN KOR NLD UK US Capial inensiy of indusrial srucure: cor(y, K) Capial endowmen: k_endw Srucural Coherence Effec on Growh The counry-level esimaion equaion for he relaionship beween srucural coherence and growh is

21 21 GROW 1 (15) k 1 x ' j, k, b1b2 SI j, b3 Z j, uj, k 0 where GROW j, k, is he real GDP growh rae of counry j from Year -k o. Equaion (15) relaes aggregae growh rae o he average srucural incoherence score over he same period, and a se of conrol variables Z j. Here Z j includes counries iniial GDP a -k, counries average physical capial invesmen inensiy, and counries average human capial inensiy as represened by he shares of high skilled and medium skilled workers in oal labor compensaion. The error erm includes counry fixed effec and an observaionspecific error. Table 7a and 7b repor he resuls of esimaing Equaion (15), using he wo versions of he SI index respecively. The sandard errors are adjused for heeroskedasiciy a he counry level. Column 1s of he wo ables display resuls for he overall capial wih he annual GDP growh as he dependen variable, i.e. k equals 1. The coefficien b 2 is negaive in boh versions of regressions, and has a -saisic of 4.75 and 2.63 respecively. According o he esimae in Version 1, decreasing srucural incoherence score from he 75 percenile (1.23) o he 25 percenile (0.46) of he disribuion is associaed wih 0.8 percenage poin increase in he annual GDP growh rae, which is abou 24% of he growh rae differenial beween he 25 percenile and 75 percenile counry-years. Column 3 and column 5 Table 7a-b repor regression resuls for he overall capial over 5- year (k=4) and 10-year (k=9) non-overlapping ime spans respecively. In boh cases, b 2 is negaive and significan. In Version 1, he -saisic of b 2 is equal o 2.15 for he 5-year esimaion and 3.42 for he 10-year esimaion. In Version 2, he -saisic is 2.25 and 2.61 for he 5-year and 10-year esimaions. To check ha he resuls are no driven by ouliers, Figure 6a-c display parial regression plos for he SI variable in Version 1. The hree graphs correspond o esimaes in Column 1, 3, and 5 of Table 7a respecively. I is clear from he plos ha he resuls are no driven by any paricular observaions. Column 2, 4, and 6 of Table 7a and 7b repor resuls for he hree deailed ypes of capial placed in he same regression. For machinery capial, he SI index is negaive and significan for all ime windows when he capial inensiy of indusrial srucure is calculaed as he oupu-share-weighed indusry capial inensiy (v2), bu is only significan in he annual regression in when he capial inensiy of indusrial srucure is calculaed as he rank correlaion beween indusry oupu share and capial inensiy (v1). Srucure capial s SI index is mosly negaive and significan in boh versions of regressions. However, he SI of ICT capial is never significan in any of he regressions. Regressing GDP growh on conemporaneous SI index raises he possibiliy of endogeneiy. For example, a negaive produciviy shock can bring down oupu growh rae, and a he same ime mess up he effeciveness of resource allocaion in he economy. To ake ino accoun such concerns, Equaion (15) is also esimaed using 2-sage Leas Square, wih he SI indices of lagged wo periods as insrumens for he curren period SI. The resuls

22 22 are shown in Table 8a-b for he wo versions of capial inensiy of indusrial srucure. The resuls indicae ha for he overall capial, he magniudes of he SI index are comparable o, if no larger han hose in he baseline regressions. For he deailed ypes of capial, he SI coefficiens for machinery capial are of he similar magniudes and significance levels o he baseline resuls; bu for srucure capial, he SI index now becomes mosly insignifican. In sum, he esimaes of Equaion (15) show ha a counry s GDP growh is negaively impaced by he degree of incoherence beween is indusrial srucure and is overall capial endowmen level. For deailed ypes of capial, he relaionship also exiss bu is no as clear. However, esimaions a he counry level do no exploi all he informaion conained in he daa. The nex secion will adop a differen approach, o examine he relaionship beween srucural coherence and growh based on an indusry-level regression seup. Table 7a Srucural coherence and growh: counry level regressions (v1) Dependen variable: real GDP growh rae k=1 k=4 k=9 (1) (2) (3) (4) (5) (6) Srucural incoherence index Overall capial *** ** ** (0.00) (0.02) (0.06) ICT (0.00) (0.02) (0.05) Machinery (MCH) ** (0.00) (0.02) (0.02) Srucure (STR) * ** (0.00) (0.02) (0.05) Conrol variables High skill 0.001** 0.001** *** ** 0.009* (0.00) (0.00) (0.00) (0.00) (0.01) (0.00) Medium skill 0.001*** 0.001** ** 0.012** 0.011** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) log(inv / GDP) (0.02) (0.02) (0.06) (0.07) (0.20) (0.29) log(gdp) ** *** ** *** (0.01) (0.01) (0.09) (0.06) (0.16) (0.11) N r * In consrucing SI scores, capial inensiy of indusrial srucure is calculaed as he Spearman rank correlaion beween indusry oupu share and indusry capial inensiy. Counry fixed effec esimaor is used in all regressions. Heeroskedasiciy-robus sandard errors are in he parenheses. Column 1-2 repor annual esimaes. Column 3-4 and Column 5-6 repor esimaes for non-overlapping 5- year and 10- year windows respecively. ***: p<0.01; **: p<0.05; *: p<0.1 Table 7b Srucural coherence and growh: counry level regressions (v2) Dependen variable: real GDP growh rae k=1 k=4 k=9 (1) (2) (3) (4) (5) (6) Srucural incoherence index Overall capial *** ** *** (0.00) (0.03) (0.05) ICT (0.00) (0.02) (0.07) Machinery (MCH) ** *** ** (0.00) (0.02) (0.05) Srucure (STR) * * (0.00) (0.01) (0.04) Conrol variables High skill 0.001* 0.002*** *** *** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Medium skill 0.001** 0.001** 0.002* 0.005** 0.011*** 0.012** (0.00) (0.00) (0.00) (0.00) (0.00) (0.01) log(inv / GDP)

23 23 (0.02) (0.02) (0.08) (0.09) (0.22) (0.26) log(gdp) ** *** *** *** (0.01) (0.01) (0.08) (0.05) (0.11) (0.08) N r * In consrucing SI scores, capial inensiy of indusrial srucure is calculaed as he indusry-oupu-share-weighed indusry capial inensiy. Counry fixed effec esimaor is used in all regressions. Heeroskedasiciy-robus sandard errors are in he parenheses. Column 1-2 repor annual esimaes. Column 3-4 and Column 5-6 repor esimaes for non-overlapping 5- year and 10-year windows respecively. ***: p<0.01; **: p<0.05; *: p<0.1 Table 8a Srucural coherence and growh: counry level regressions (v1), IV mehod Dependen variable: real GDP growh rae k=1 k=4 k=9 (1) (2) (3) (4) (5) (6) Srucural incoherence index Overall capial *** ** *** (0.00) (0.02) (0.04) ICT (0.00) (0.02) (0.12) Machinery (MCH) ** ** (0.00) (0.01) (0.03) Srucure (STR) * (0.00) (0.02) (0.09) Conrol variables High skill 0.001** 0.001*** *** (0.00) (0.00) (0.00) (0.00) (0.01) (0.01) Medium skill 0.001** 0.001** *** 0.015** (0.00) (0.00) (0.00) (0.00) (0.00) (0.01) log(inv / GDP) (0.02) (0.02) (0.08) (0.08) (0.23) (0.34) log(gdp) *** *** *** *** ** * (0.01) (0.01) (0.05) (0.05) (0.11) (0.13) N r Hansen J es (p value) * The SI scores of lagged wo periods are used as insrumens for he conemporaneous SI scores. ***: p<0.01; **: p<0.05; *: p<0.1 Table 8b Srucural coherence and growh: counry level regressions (v2), IV mehod Dependen variable: real GDP growh rae k=1 k=4 k=9 (1) (2) (3) (4) (5) (6) Srucural incoherence index Overall capial *** ** *** (0.00) (0.03) (0.05) ICT (0.00) (0.02) (0.07) Machinery (MCH) *** *** *** (0.00) (0.02) (0.04) Srucure (STR) (0.00) (0.02) (0.04) Conrol variables High skill 0.001** 0.002*** 0.007* 0.010*** *** (0.00) (0.00) (0.00) (0.00) (0.00) (0.00) Medium skill ** *** 0.013** (0.00) (0.00) (0.00) (0.00) (0.00) (0.01) log(inv / GDP) ** (0.02) (0.02) (0.10) (0.12) (0.27) (0.27) log(gdp) *** *** *** *** *** (0.01) (0.01) (0.07) (0.05) (0.07) (0.07) N r Hansen J es (p value) * The SI scores of lagged wo periods are used as insrumens for he conemporaneous SI scores. ***: p<0.01; **: p<0.05; *: p<0.1