Mind the gap! International comparisons of productivity in services and goods production

Transcription

1 Mind he gap! Inernaional comparisons of produciviy in services and goods producion Rober Inklaar, Marcel P. Timmer and Bar van Ark Universiy of Groningen and The Conference Board Paper o be presened a OECD Workshop on Produciviy Analysis and Measuremen Bern Ocober 2006 Session 4: New produciviy measures for OECD counries Iem 16b Absrac In his paper, we make a comparison of indusry oupu, inpus and produciviy growh and levels beween seven advanced economies (Ausralia, Canada, France, Germany, Neherlands, UK and U.S.). Our indusry-level growh accouns make use of inpu daa on labour quaniy (hours) and composiion (schooling levels), and disinguish beween six differen ypes of capial asses (including hree ICT asses). The comparisons of levels rely on indusry-specific purchasing power pariies (PPPs) for oupu and inpus, wihin a consisen inpu-oupu framework for he year Our resuls show ha differences in produciviy growh and levels can mainly be raced o marke services, no o goods-producing indusries. Par of he srong produciviy growh in marke services in Anglo-Saxon counries, such as Ausralia and Canada, may be relaed o relaively low produciviy levels compared o he U.S. In conras, services produciviy levels in coninenal European counries were on par wih he U.S. in 1997, bu growh in Europe was much weaker since hen. In erms of facor inpu use, he U.S. is very differen from all oher counries, mosly because of he more inensive use of ICT capial in he U.S. Acknowledgemens We kindly acknowledge Indusry Canada for funding his work under he projec Produciviy in Canada Relaive o US (conrac no. # ). The auhors are graeful o Jianmin Tang (Indusry Canada), Paul Schreyer (OECD) and paricipans a he Conference on Deerminans of Produciviy Growh a he Insiue of Advanced Sudies, Vienna (Sepember 15-16, 2006), NBER/CRIW Summer Insiue 2006 and he 5 h Annual Oawa Produciviy Workshop (Bank of Canada) for commens. Gerard Ypma (Universiy of Groningen) kindly provided assisance on he adequae use of he oupu PPPs in his paper. Only he auhors can be held accounable for he resuls presened in his paper.

2 1. Inroducion During he second half of he 1990s he comparaive growh performance of many OECD counries has undergone a marked change. For he firs ime since World War II labour produciviy growh in mos counries ha have been par of he European Union (EU) for a decade or more (he EU-15) has fallen behind he U.S. for a considerable lengh of ime. Whereas average annual labour produciviy growh in he U.S. acceleraed from 1.2 percen during he period o 2.3 percen during , EU-15 produciviy growh declined from 2.2 o 1.4 percen. The downward rend in he EU-15 was raher coninuous as growh slowed from 1.7 percen from o 1.1 percen from Jus like he U.S., various oher Anglo-Saxon counries, in paricular Ausralia and Canada, experienced a significan improvemen in produciviy growh during he lae 1990s (2.3 and 1.7 percen from respecively), bu since 2000 produciviy in hese counries slowed down again o growh raes comparable o he pre period (1.7 and 1.3 percen in respecively). In conras, U.S. produciviy acceleraed from 2.2 percen in o 2.5 percen in The sriking acceleraion in U.S. oupu and produciviy growh since he mid 1990s and in paricular he role of informaion and communicaion echnology (ICT) has been much discussed in he lieraure. 2 ICT had an impac on growh hrough a surge in ICT invesmen, srong produciviy conribuions from ICT-producing indusries and a more producive use of ICT in he res of he economy. Noably he marke services secor of he U.S. economy seems o have srongly benefied from he increase in ICT use. Unforunaely, here is much less agreemen on he reasons for he slower produciviy growh in Europe. Compared o he U.S., ICT invesmen, ICT producion and he producive use of ICT in Europe generaed less produciviy growh during he lae 1990s (van Ark e al., 2003; Inklaar e al., 2005, Eicher and Roehn, 2006). Bu he reasons for he limied impac of new echnology, innovaion and srucural reforms on economic growh in Europe are sill poorly undersood. The acceleraion and subsequen deceleraion in Ausralia and Canada has also puzzled various scholars (Parham, 2005; Rao e al., 2005). Cyclical slowdown migh be one reason, bu here may be oher reasons 1 See TCB/GGDC (2006) 2 See e.g. Oliner and Sichel (2000, 2002); Jorgenson and Siroh (2000); Jorgenson, Ho and Siroh (2005); Gordon (2003, 2004); Siroh (2002a); Triple and Bosworh (2004); Bosch and Siroh (2006); Basu and Fernald (2006) and Corrado, Lengermann, Barelsman and Beaulieu (2006). 2

3 relaed o shor erm shocks, he innovaion sysem and a possible slowdown in reforms in he labour and produc markes. This paper exends our previous sudies in his research field by providing a combined analysis of produciviy growh and levels by indusry. The major novely in his paper is he comparison of oupu, inpu and produciviy levels in a comprehensive framework. Looking a growh and levels ogeher enables one o beer undersand he differences in conribuions of inpus (labour composiion, ICT and non-ict capial) o oupu performance. In his paper we focus our analysis on a comparison of seven advanced economies (Ausralia, Canada, France, Germany, Neherlands, he Unied Kingdom and he Unied Saes). While our presen growh and level accouning daase provides deail on 26 indusries (of which 25 in he marke economy), we focus he discussion in his paper on he comparaive performance of hree major secors which consiue he marke economy: he ICT producion secor, goods-producing indusries and marke services. We find ha a disincion beween he Anglo-Saxon counries (Ausralia, Canada, UK and U.S.) and he coninenal European counries (France, Germany and he Neherlands) is useful. In he Anglo-Saxon counries marke services have been conribuing more o labour produciviy growh since 1995 han before. Only par of he increase in labour produciviy growh can be raced o higher direc conribuions from ICT invesmen. A large par of he acceleraion of growh in his secor is also due o higher oal facor produciviy growh. Srikingly, he coninenal European counries have no experienced his acceleraion in TFP growh. Evidence from oher sudies suggess ha his may be relaed o a more producive use of ICT in Anglo-Saxon economies, alhough his can be neiher confirmed nor rejeced on he basis of he indusry-level daa presened here. Our comparison of produciviy levels for 1997 suggess ha TFP levels in goods producion are relaively close beween counries. In conras, produciviy levels in marke services show relaively large gaps for Ausralia, Canada and he U.K. relaive o he U.S., whereas he coninenal European counries are a or even above he U.S. TFP level in marke services. This raises he possibiliy ha he hree Anglo-Saxon counries exhibi some kind of caching-up o he U.S. in paricular in marke services. Bu i also 3

4 raises quesions concerning he inerpreaion of he high produciviy levels in coninenal Europe. The paper proceeds as follows. In Secion 2 we ouline our basic growh and level accouning mehodology. In Secion 3, he daa and resuls of our growh accouns are presened a he level of hree major secors of he economy. Secion 4 briefly inroduces our level accouning mehodology, followed by a discussion of he level esimaes. In he concluding secion we summarize our main resuls, and indicae he areas for fuure research. 2. Growh and level accouning mehodology The produciviy analysis in his sudy is rooed in he radiion of naional accouning, inpu-oupu analysis and growh accouning, as pioneered by among ohers Simon Kuznes, Wassily Leonief, Rober Solow, Edward Denison, Zvi Griliches and Dale Jorgenson. Using hese echniques macroeconomic and indusry and aggregae measures of oupu can be decomposed ino he conribuions of inpus and produciviy. 3 In his secion we ouline he basic growh accouning mehodology and discuss how his framework is augmened o accoun for produciviy level differences across counries. To assess he conribuion of he various inpus o aggregae growh, we apply he growh accouning framework as developed by Jorgenson and associaes (see, for example, Jorgenson, Ho and Siroh, 2005). For each indusry gross value added (VA) is compued according o a producion funcion f using ICT capial services ( K ICT capial services ( K N ICT ), non- ) and labour services (L). Produciviy (A) is represened as a Hicks-neural augmenaion of aggregae inpus. The indusry producion funcion (indusry subscrips are omied) akes he following form: 4 N ICT ( L, K K ) VA = A f, (1) Under he assumpion of compeiive facor markes, full inpu uilizaion and consan reurns o scale, he growh of oupu can be expressed as he (compensaion share) weighed growh of inpus and oal facor produciviy, denoed by A, which is derived as a residual: 3 See Schreyer (2001) for a more deailed exposiion. 4 In equaions (1) o (5) we drop indusry subscrips i for sake of convenience 4

5 VA VA = v L L L + v N K K N N + v ICT K K ICT ICT A + A (2) where i v denoes he wo-period average share of inpu i in nominal value added. L N ICT Imposing consan reurns o scale implies v + v + v = 1. Capial services are defined in (3) as he aggregae of he individual capial socks weighed by he capial compensaion share: K K K j, = v j (3) K K j j, where K v j is he wo-period average share of asse ype j in oal nominal capial compensaion. For our growh accouns we use ICT capial services, which are calculaed by weighing each of he ICT capial socks by he share of he asse in oal ICT capial compensaion. Non-ICT capial services are calculaed analogously. We define he change in labour composiion as he difference beween he growh of labour inpu and he growh of oal hours worked: 5 q q L L = L, v h h L L h, h Lh, h h L h, L = L H H Here L is he labour inpu index, aggregaed over he h labour ypes using labour compensaion shares and H is oal hours worked, summed over he differen labour ypes. By rearranging equaion (2) he resuls can be presened in erms of average labour produciviy growh defined as he raio of oupu o hours worked, va = VA (4) H, he raio of capial services o hours worked, k = K H, labour composiion and TFP as follows: va va = v L q q L L + v N k k N N + v ICT k k ICT ICT A + A We name he decomposiion in equaion (5) an inpu decomposiion. As we also focus in his paper on he conribuion of indusries o aggregae labour produciviy growh, an indusry decomposiion measures indusry conribuions o GDP as follows (Siroh, 2002a): (5) 5 This is someimes referred o as labour qualiy growh such as in Jorgenson e al. (2005). However, he advanage of he erm labour composiion change is ha i does no imply ha workers wih lower wages have a lower qualiy. Insead, posiive labour composiion change only implies a shif owards workers wih higher wages and hence, higher marginal produciviy. 5

6 va va H va H v v v R va = VA i, VA i, VA i i + i i i va = + i i H i H, (6) 1, 1, 1 i vai, where VA vi is he wo-period average share of indusry i in aggregae value added. The erm in brackes in equaion (6) is he reallocaion of hours. I reflecs differences in he share of an indusry in aggregae value added and is share in aggregae hours worked. This erm will be posiive when indusries wih an above-average labour produciviy level show posiive employmen growh or when indusries wih below average labour produciviy have declining employmen shares. Obviously, one can go one sep furher by combining he inpu and indusry decomposiions of indusry labour produciviy. Labour produciviy growh a an aggregae level is hen decomposed ino inpu conribuions and a reallocaion erm for each indusry: L N ICT va q VA L i, k VA N i, k VA ICT i, A VA = i, vi vi vi vi vi vi vi + R (7) L N ICT va q k k A i i, i i, i i, i i, In similar fashion, he conribuion from indusry TFP growh o aggregae TFP growh can be calculaed: A A = v i VA i A A i, i, (8) Level accouning Comparing produciviy levels across counries is in many ways analogous o comparisons over ime. Under he assumpion of a common producion funcion across counries, equaions (2) o (8) can be used afer replacing ime subscrips and -1 by counry subscrips. Our level accouns differ from he growh accouns in wo respecs. Firs, while one ypically compares produciviy in one year wih produciviy in he nex or previous year, here is no such naural ordering of counries. Therefore he comparison should no depend on he counry ha is chosen as he base counry. There are various index number mehods ha can be used o make mulilaeral comparisons. We use he mehod suggesed by Caves, Chrisensen and Diewer (1982). This index mirrors he Törnqvis index approach used in our growh accouning, bu all counries are compared 6

7 o an arificial average counry (AC). 6 This average counry is defined as he simple average of all M counries in he se. For example, a mulilaeral index of capial services in counry c is given by: K K c j, c = v j (9) K j K AC 1 wih [ v + v ] j, AC v j = j, c j, AC v j, c 2 compensaion in counry c, asse j over all counries N and, he share of asse ype j in oal nominal capial v j AC = N v j, c, 1 he average compensaion share of capial K j AC = N K j, c, 1, he average capial sock. This mirrors equaion (3). A comparison beween wo counries, say Germany and he US, can be made indirecly: by firs comparing each counry wih he average counry and hen comparing he differences in German and US levels relaive o he average counry. Similarly, gaps in labour produciviy levels can be decomposed by mirroring equaion (5): va q k k + va L N ICT c L c N c ICT c c = v + v + v (10) L N ICT AC q AC k AC k AC AAC wih v s he inpu shares in value added averaged beween counry c and he average counry AC. Analogous o equaion (8), he conribuion of indusry TFP gaps o aggregae TFP gaps beween counry c and he average counry is given by A A A c VA i, c = vi (11) AC i Ai, AC Furhermore, while each counry s naional accouns presens informaion abou value added a consan prices over ime, such daa is no available for direcly comparing indusry value added across counries. As described in more deail below, our saring poin is o measure relaive prices (PPPs) for gross oupu a he indusry level. In combinaion wih Supply and Use ables, we use hese relaive gross oupu prices o implicily calculae relaive value added prices based on he following expression: A 6 In pracice his means applying an EKS procedure o he bilaeral Törnqvis indices. The resuls below are expressed by using he U.S. as he benchmark counry. I should be sressed ha he decomposiions of equaions (5)-(8) are noiceably less accurae approximaions o he underlying aggregaor funcions for level accouning han for growh accouning. 7

8 P P GO c GO AC II P Pc = w + ( 1 w) (12) II P P VA c VA AC AC where w is he share of value added in gross oupu averaged beween he counry and he average counry, and he superscrips GO, VA and II denoe gross oupu, value added and inermediae inpus, respecively. The relaive VA price levels are used o calculae implici relaive VA quaniy indices which are he oupu measures we use below. 3. Daa and resuls for growh accouning Daa For he growh accouns in his paper we developed a daabase on oupu and labour and capial inpus for seven counries, including Ausralia, Canada, France, Germany, he Neherlands, he Unied Kingdom and he Unied Saes, for 26 indusries covering he period 1987 o 2003 (see Table A.1). 7 The daa on value added and oal hours worked are based on he mos recen release of he GGDC (2006a) 60-Indusry Daabase, while he full indusry growh accouns are from he GGDC (2006b) Indusry Growh Accouning Daabase. 8 Oupu is defined as value added a consan prices and is aken from naional accouns complemened wih measures from indusrial and business surveys. Labour inpu is measured as hours worked defined as he oal number of persons employed (including self-employed) imes he average number of hours worked. For labour inpu in each counry we disinguish beween hree and seven ypes of educaional aainmen. For our capial inpu measures we use daa on invesmen in curren and consan prices for six asse ypes. Three asses refer o ICT goods (compuers, communicaion equipmen and sofware) and hree o non-ict goods (ranspor equipmen, oher (non-ict) machinery and equipmen and non-residenial srucures). 9 To esimae capial socks we 7 Togeher, he four European counries cover abou 70 percen of oupu in he EU-15. See O Mahony and van Ark (2003) and van Ark and Inklaar (2005) for figures a he aggregae level for all EU-15 counries. The Ausralian daa covers fiscal years running from July o June. Following OECD convenion, we allocae daa for July 2000 o June 2001 o he year All daa and more deailed source descripions are downloadable a 9 Residenial buildings are no aken ino accoun o allow for a sharper focus on he produciviy conribuion of business-relaed asses. Since mos of he oupus and inpus of he real esae indusry consiss of housing and impued rens from housing we would have o make an adjusmen for his. 8

9 use indusry-specific geomeric depreciaion raes for deailed asses in he U.S., provided in Fraumeni (1997), in combinaion wih indusry shares of hese asses from he BEA NIPA. The deflaors for ICT producing manufacuring indusries and ICT invesmen have been harmonised across counries as discussed by Inklaar e al. (2005), excep in he case of Canada and Ausralia which already use consan-qualiy price indices o a similar exen as he U.S. The series for he European counries were published earlier by Inklaar e al. (2005), bu were revised and updaed o 2003 using he same sources and mehods as before. For he Unied Saes, we have developed an enirely new daase, which is largely based on he laes releases by he Bureau of Economic Analysis (BEA) of GDPby-Indusry daa ha cover he period These daa are organized according o he NAICS 1997 classificaion sysem and are consisen wih he 2003 Comprehensive Revision of he Naional Income and Produc Accouns (NIPA). In addiion o he GDP by Indusry daa, we use numerous oher sources, mos noably indusry oupu and employmen daa from he BLS, o obain a complee se of growh accouns. For U.S. invesmen we used wo main sources from he BEA, namely he 1997 Capial Flow Table and he BEA Deailed Daa for Fixed Asses wih ime series for he period We made subsanial adjusmens o he U.S. daa o fi he ISIC rev. 3 indusrial classificaion, which is he basis of our inernaionally comparaive daabase. 10 Daa for Canada and Ausralia were developed specifically for he purpose of his paper. In he case of Canada he new NAICS-based use ables for he period have been used o consruc he oupu series. These daa were exrapolaed o 2003 using indusry oupu series. 11 Employmen daa were drawn mosly from he OECD STAN daabase (2004 release). Deailed ables on invesmen by indusry and asse ables from Saisics Canada were obained hrough Indusry Canada. For manufacuring indusries, we supplemened hose ables wih informaion from he final demand par of he inpuoupu ables. Ho, Rao and Tang (2004, Tables 10-12) provide daa on labour However, i is hard o separae impued rens from he oher oupu in he real esae secor. Hence we decided o leave he enire real esae indusry ou of our daase. 10 For he U.S., no educaional aainmen daa has been colleced for years afer 2000, so labour composiion change is assumed o be zero for he laes hree years. 11 Indusry oupu series for 2002 and 2003 are a consan prices. Curren price daa are esimaed using producer price indices for mos goods-producing indusries and he GDP deflaor for oher indusries. The 2002 inpu-oupu ables were released by Saisics Canada afer we compleed he daase for his paper. 9

10 composiion change for Canada, which were compiled in a broadly comparable fashion o our mehodology for he oher counries. Oupu series for Ausralia are mainly aken from he naional accouns, supplemened by indusry surveys. Daa on employmen are aken from he Ausralia s labour force saisics. Deailed invesmen by indusry and asse ables are par of he Ausralian naional accouns, and were supplemened wih daa from manufacuring surveys o disinguish invesmen by deailed manufacuring indusries. Invesmen in communicaion equipmen was spli off from he broader caegory (elecrical and elecronic equipmen) using inpu-oupu ables. For Ausralia we only had access o informaion from he naional accouns on labour composiion change for he aggregae marke secor from he naional accouns. This rae was applied o each of he individual indusries. Indusry decomposiion of labour produciviy growh On he basis of he daase described above, measures of labour produciviy growh and he conribuion of individual indusries and major indusry groups o aggregae produciviy growh can be calculaed. Table 1 shows he resuls of he indusry decomposiion from equaion (6) for he marke secor of he economy for he and periods. 12 The able shows a large acceleraion in produciviy growh in Ausralia, Canada and he U.S. for he marke secor as a whole since Labour produciviy growh in France and Germany slowed down subsanially, while growh in he Neherlands sagnaed. In he UK, growh also slowed down somewha, bu i remained high compared o he coninenal European counries. For he indusry decomposiion we presen he resuls for hree major indusry groups. 13 Firsly, he conribuion of ICT producion secor, which includes boh ICT 12 The exac definiion of he marke economy (or business secor) differs by counry and organizaion. We classify governmen (ISIC 75), educaion (ISIC 80) and healh (ISIC 85) as non-marke services, even hough some or even mos of educaion and healh may be operaed or owned by non-governmenal organizaions. The reason for excluding hese secors from he analysis is mainly due o severe measuremen problems in hese indusries. Moreover he inernaional comparabiliy of non-marke oupu measures is very weak, which makes affecs he inerpreaion of differences in growh performance a he oal economy level. Table 1 also shows he oal economy aggregaes, including non-marke services and hese resuls were more exensively analyzed in Timmer and van Ark (2005). 13 Reallocaion effecs are generally negaive, which suggess ha indusries wih an above-average labour produciviy level exhibi declining employmen shares. In earlier work, such as van Ark e al. (2003), we also disinguished beween indusries ha used ICT inensive and hose ha did no based on U.S. 10

11 manufacuring and services (communicaions, sofware, ec.), differs beween counries. The U.S. and he UK show raher high conribuions from ICT-producers o produciviy growh, whereas he coninenal European counries ake an inermediae posiion and Ausralia and Canada show very small conribuions. Secondly, he differences in average conribuions of oher goods-producing indusries (excluding ICT producing manufacuring) have been very small since Of course, here are differences beween conribuions of individual goods-producing indusries, bu mosly hese conribuions are posiive and small and herefore maer relaively lile for he aggregae performance. Finally, and mos imporanly, he differences in conribuions of marke services (excluding ICT services) are quie large beween he counries. All Anglo-Saxon counries show acceleraions in he conribuions from marke services o labour produciviy growh since 1995, ranging from 0.5 percenage poin acceleraion in he UK o 1.5 percenage poins in Ausralia. In he U.S., labour produciviy in marke services increased by 0.9 percenage poin from 0.9 per cen per year in o 1.8 per cen per year from In he Neherlands marke services showed a sligh increase in he conribuion of 0.2 percenage poin, bu France and Germany experienced a deceleraion of 0.1 and 0.6 percenage poins, respecively. Inpu decomposiion for marke services Using daa on capial services and labour composiion, labour produciviy growh raes for each indusry and indusry group are decomposed ino conribuions from ICT capial deepening, non-ict capial deepening, labour composiion change, TFP growh and reallocaion of hours worked (see equaion 7). Here we focus exclusively on an inpu decomposiion of labour produciviy growh in marke services which are he mos inensive users of ICT asses. The resuls are given in Table 2. All counries in he able show a moderae o subsanial acceleraion in he conribuion of ICT capial deepening in marke services. However, wha sands ou in Table 2 is he rapid acceleraion in TFP growh in marke services in he Anglo-Saxon counries (Ausralia, Canada, Unied esimaes of ICT invesmen relaive o oal invesmen. This approach necessiaes a somewha arbirary disincion ha has been criicized by, for example, Daveri (2004). More imporanly hough is he fac ha i has become less imporan o make his disincion because we now have acual measures of ICT use for counries ouside he U.S. 11

12 Kingdom and Unied Saes). Before 1995, TFP growh in marke services was negaive in hese counries, bu urned posiive afer In conras, all coninenal European counries (France, Germany and he Neherlands) showed deerioraing TFP growh in marke services afer Germany and he Neherlands experienced negaive TFP growh in marke services afer 1995 and France shows only a small posiive growh rae. Table 3, based on equaion (8), looks a he conribuion of individual service indusries o oal facor produciviy growh in marke services from , wih a breakdown for and The able shows a srong conribuion from wholesale rade in he Neherlands and he U.S., even hough i slowed down since 2000, in paricular in he Neherlands. Reail rade showed a paricularly srong performance in Ausralia and Canada from , bu he reail secor s conribuion o overall marke services TFP growh slowed down almos everywhere since Financial inermediaion was paricularly srong in he Unied Saes since 2000, bu no in oher counries. A disincive difference since 2000, however, is he srong improvemen of produciviy growh in business services in all four Anglo-Saxon economies. The caegory of oher marke services only showed a sizeable effec for Ausralia afer 2000, which is primarily due o a produciviy boom in consrucion. The precise reasons for he differences in produciviy dynamics of marke services are difficul o generalize. One possibiliy is ha in some counries invesmens in ICT do no lead o faser TFP growh whereas in oher counries hey do. There may be many reasons for his, among which cyclical effecs and he effecs of unmeasured inangible invesmens sand ou. Esimaes of ICT oupu elasiciies ha are larger han he marginal cos of ICT capial services would violae a basic growh accouning assumpion. However, Inklaar (2005) and van Ark and Inklaar (2005), in line wih, for example, Siroh (2002b), find ha indusry daa do no exhibi supra-normal reurns. There is even some evidence ha i akes a number of years of below-normal reurns before he producive reurns of ICT become high enough o ouweigh he cos An alernaive approach is o direcly esimae each of he oupu elasiciies separaely. For example, O Mahony and Vecchi (2005) find super-normal reurns on ICT, which hey aribue (a leas in par) o he reurns o unmeasured inangible invesmen, such as organizaional change or raining programs. An advanage of heir approach is ha no assumpions abou he oher oupu elasiciies have o be made, bu he disadvanage is ha i requires a raher resricive funcional form for he producion funcion (such as Cobb-Douglas) for he esimaion (see van Ark and Inklaar, 2005). 12

13 The slow upake of produciviy in marke services also emerges from sudies ha ake a more indusry-specific orienaion. For example, McGuckin, Spiegelman and van Ark (2005) analyze labour produciviy growh in European and U.S. wholesale and reail rade in deail. They find ha echnology adopion in Europe lags he U.S. by several years, which has been holding back European produciviy growh. They also find ha his lag can be (parly) aribued o sricer regulaions in European counries. 15 Bu he explanaion for he slowdown since 2000 has so far remained unexplained, and more work in his area (for reail and oher service indusries) is herefore required. To shed more ligh on he reasons for he large differences in he conribuion of marke services o produciviy growh one migh focus no exclusively on comparisons of growh raes bu also on comparaive levels of produciviy. For example, rapid TFP growh could poin owards caching-up and imiaion when saring from a relaively low produciviy level. Similarly, sagnaing TFP growh a a relaively high level of TFP migh be indicaive of a lack of innovaion. Therefore we urn o level comparisons for he remainder of his paper. 4. Daa and resuls for level accouning A level accouning approach o oupu and produciviy comparisons has no been widely applied, which is primarily due o he lack of adequae indusry-specific PPPs for oupu and inpus. PPPs are needed o adjus oupu and inpus for differences in relaive price levels beween counries. Since here is lile reason o assume ha hese price differences are negligible beween counries or he same across indusries, such PPPs have o be indusry-specific. Only few sudies have aemped o measure indusry-specific PPPs. Mos comparisons have ended o resric hemselves o he oal economy (e.g. Schreyer, 2006). Jorgenson, Kuroda and Nishimizu (1987) make use of specific PPPs for expendiure caegories from OECD o measure TFP for abou 30 indusries beween Japan and he U.S. These expendiure PPPs are adjused for differences in ranspor and disribuion margins across counries. Oher sudies have aimed o direcly measure producer-price based PPPs. Van Ark and Pila (1993) provided TFP level measures for 15 For a more deailed saisical analysis of growh differences in he rade secor, see Timmer and Inklaar (2005). 13

14 manufacuring indusries in Germany, Japan and he Unied Saes using a raio of uni values which are based on producer quaniies and values from producion censuses in differen counries. Similarly, Timmer (2000) measured TFP for manufacuring in Asian counries using a similar mehodology. Pila (1996), Mulder (1999), O Mahony (1999), and O Mahony and de Boer (2002) exended he indusry-specific PPP approach o measuring oupu and produciviy levels beyond manufacuring. Mos sudies had o rely on bilaeral comparisons of pairs of counries. I has also urned ou o be difficul o develop a consisen PPP mehodology across he various sudies because of differences in daa availabiliy. In paricular, value added is ofen deflaed by gross oupu price relaives, wihou aking ino accoun differences in prices of inermediae inpus (socalled single versus double deflaion). In his paper we make use of a new and comprehensive daase of indusry PPPs for 1997, in combinaion wih a benchmark se of Supply and Use ables from he relevan saisical offices. PPPs for value added are consruced by double deflaion of gross oupu and inermediae inpus wihin a consisen inpu-oupu framework (see equaion (12)). In addiion, relaive price raios for labour and capial inpu are developed. For a full discussion of he new indusry PPPs, he reader is referred o Timmer, Ypma and van Ark (2006). For he inegraion of gross oupu PPPs and he derivaion of inpu PPPs in a level accouning framework, deails are spelled ou in Inklaar and Timmer (2006). Below we only presen he mos imporan elemens of our mehodology. PPPs for gross oupu are defined from he producer s poin of view and are a basic prices, which measures he amoun received by he producer for a uni of a good or service produced. These PPPs have parly been consruced by way of uni value raios for agriculural, mining, manufacuring and ranspor and communicaion services. For oher indusries, PPPs are based on specific expendiure prices from Eurosa and he OECD, which are allocaed o individual indusries producing he specific iem. The value was adjused from expendiure o producer level wih relaive ranspor and disribuion margins and by adjusing for differences in relaive ax raes. Margins and ax raes were derived from benchmark supply and use ables for This se of gross oupu PPPs for 1997, covering 45 indusries a (roughly) 2-digi indusry level, has been made ransiive across counries by applying he mulilaeral EKS-procedure for a oal of 26 counries, which are all OECD counries bu also includes Taiwan). For his sudy he 14

15 gross oupu PPPs were hen are allocaed o he 26 indusries in Inpu-Oupu ables for Inermediae inpu PPPs are required in order o double deflae value added. These inpu PPPs should reflec he coss of acquiring inermediae deliveries, hence hey need o be based on purchasers prices. Assuming ha he basic price of a good is independen of is use, we can use he same gross oupu PPP for a paricular indusry, afer adjusmen for margins and ne axes, o deflae all inermediae deliveries from ha indusry o oher indusries. The aggregae inermediae inpu PPP for an indusry is hen derived by weighing is inermediae inpus a he gross oupu PPPs from he delivering indusries. Impors are separaely idenified for which exchange raes are used as PPP, hence assuming no price differences across counries for impored commodiies. To obain PPPs for capial and labour inpu, we follow he mehodology oulined by Jorgenson and Nishimizu (1978). The PPP for capial services is based on he expendiure PPP for invesmen from Eurosa and he OECD, adjused for differences in he user coss beween counries. The user cos of capial inpu depends on he rae of reurn o capial, he depreciaion rae and he invesmen price change. This daa is aken from he growh accouns discussed in he previous secion. The procedure o obain a PPP for labour is more sraighforward han for capial as i simply involves aggregaing relaive wages across differen labour ypes using labour compensaion for each ype as weighs. For his purpose we only disinguish beween wo labour caegories: workers wih a universiy degree or higher, and hose wihou. This limied number of skill ypes is due o difficulies in maching schooling sysems across he various counries. Indusry decomposiion of labour produciviy levels The firs sep in presening our level resuls, i o focus on he comparaive levels of labour produciviy for he same major indusry groups (ICT producion, oher goods producion and marke services) as in he firs par of he paper. Table 4 shows comparaive levels of labour produciviy, measured as value added per hour worked. The resuls for 1997 are for our benchmark year, and developed according o he mehodology oulined above. Table 4 shows ha he U.S. shows a labour produciviy level in he marke economy which is similar o ha of he hree coninenal European counries, ranging 15

16 from France a 96 per cen of he U.S. level o he Neherlands a 107 per cen. 16 The oher Anglo-Saxon counries show beween 15 and 20 percen lower levels han he U.S, ranging from Ausralia a 73 per cen of he U.S. level o Canada a 88 per cen. Table 4 also shows relaive labour produciviy levels in ICT producion, oher goods-producing indusries and oher marke services in The relaive produciviy levels in marke services correspond mos closely o hose a he level of he marke economy, wih produciviy levels in France, Germany and he Neherlands relaively close o ha of he Unied Saes, while Ausralia, Canada and he UK are a a large disance from he oher counries. Produciviy levels in goods-producing indusries are generally closer ogeher, wih he excepion of Canada which has high produciviy levels in some resource-inensive and heavy manufacuring indusries, like oil refining, meal producs and machinery. The high produciviy levels in ICT producion in France, Germany and he UK can mosly be raced o communicaions services, wih again smaller differences in ICT manufacuring. Table 4 also shows an exrapolaion of he benchmark resuls for labour produciviy in 1997 o Saring from he relaive labour produciviy levels for 1997, we have applied he relaive growh beween 1997 and 2003 a he aggregae and indusry group level (see Table 1). 17 The faser produciviy growh in he U.S. since 1997 has resuled in lower produciviy levels relaive o he U.S. for he oher counries, bu he cumulaive impac is sill relaively small over a period of only six years. Inpu decomposiion of level accouns for marke services The relaively low produciviy levels in marke services in he Anglo-Saxon economies (excep he U.S.) sheds addiional ligh on he possible reasons for rapid produciviy 16 The margin of error around level esimaes is non-negligible. Alhough (formal) research on his issue is limied, Schreyer (2006) arrives a a confidence inerval of (a leas) 10 percen using a simulaion exercise. Srikingly, he relaive oal economy levels are all higher relaive o he U.S. han he marke economy levels, signalling high relaive TFP levels in non-marke services in counries ouside he U.S. These differences are larges in European counries, like France and he Neherlands. As before, when dealing wih he growh, i is no clear how o inerpre he resuls for he non-marke secor given he subsanial measuremen problems ha remain. 17 The daa requiremens for a level comparison for one year are quie subsanial, and in many cases hese daa only become available wih a considerable lag. As a resul, a fully consisen level accouning exercise canno ye be done for a laer year han our benchmark year Exrapolaing he 1997 benchmark resuls forward is also no sraighforward since ideally his should be done a mos deailed level before reaggregaing using he CCD-mehod. As he daa requiremens for such an exrapolaion would be comparable o ha of a new benchmark, his is also no feasible a he momen. 16

17 growh in he marke services secor for hese counries, as was observed earlier. Rapid growh migh ypically be relaed o caching up-effecs as he indusries sared from low levels of produciviy, and benefied from echnological progress, innovaion and insiuional changes more han he coninenal European counries which sared from much higher levels. To explore his possibiliy furher, an inpu decomposiion of differences in he level of produciviy is useful. Table 5 shows an inpu decomposiion of he labour produciviy gap in marke services ino he conribuions of ICT capial inensiy, non-ict capial inensiy, labour composiion and TFP o he gap following equaion (10). For example, he (log) labour produciviy gap beween Ausralia and he U.S. is 39 percenage poins, 18 of which nine percenage poins are due o lower ICT inensiy, four percenage poins o lower non-ict inensiy and lower labour composiion, and 21 percenage poins due o lower TFP levels. A number of observaions sand ou from Table 5. Firsly, ICT capial inpu per hour worked is higher in U.S. marke services han anywhere else, and accouns for beween 3 and 10 percenage poins of he U.S. labour produciviy level advanage. In conras, non-ict capial levels are higher in he coninenal European counries, and accoun for he bulk of he produciviy advanage in marke services for France and Germany. In Germany, higher capial inensiy adds as much as 21 percenage poins o Germany s advanage relaive o he U.S., accouning for more han he oal labour produciviy gap. 19 This is parly compensaed by lower ICT capial inensiy and less skilled labour. In he Anglo-Saxon counries, non-ict capial inensiy in marke services is lower han in he U.S. Labour composiion levels in marke services are also below he U.S. levels, wih he excepion of Canada and he Neherlands where i is relaively close o he U.S. level. The major facor accouning for he produciviy gap beween he U.S. and he oher Anglo-Saxon counries is he lower TFP level in marke services in he laer counries. I accouns for beween 21 percenage poins of he Ausralian gap relaive o 18 The gap here is defined as he difference in he naural log of he levels, o say consisen wih he growh accouns and he decomposiion formulae which are also in erms of logs. 19 The qualiy of daa on invesmen by indusry and asse as well as he invesmen PPPs are areas for poenial improvemen of he resuls. However, any adjusmen would have o be implausibly large o accoun for he high non-ict capial inensiy level in Germany. 17

18 he U.S. and 26 percenage poins of he Canadian gap. In France and Germany, he TFP level is comparable o he U.S. Only in he Neherlands, he higher produciviy level in marke services relaive o he Unied Saes is largely due o a higher TFP level. Finally, Table 6 provides an indusry decomposiion of he TFP gaps in marke services relaive o he Unied Saes. In conras o he inpu decomposiion of marke services in able 5, produciviy gaps are now decomposed o he conribuion of produciviy gaps in he more deailed indusries, see equaion (11). So, for example, he 21 percenage poin TFP gap in marke services beween Ausralia and he U.S. is mainly due o he gap in business services, conribuing almos 15 percenage poins, followed by oher marke services a 7 percenage poins. In finance, he TFP level in Ausralia is higher han in he U.S., so is conribuion o he overall gap is posiive. There is clear evidence ha he produciviy level advanage of he coninenal European counries is primarily locaed in wholesale rade and o a lesser exen in reail rade. The Neherlands also shows a produciviy level advanage in ranspor and sorage and financial inermediaion. Financial inermediaion shows posiive conribuions from high produciviy levels in Ausralia, France and Germany. Srikingly, business services conribue negaively o he produciviy gap relaive o he U.S. for all counries. 6. Summary and concluding remarks In his paper we combined a growh and level accouning approach o compare produciviy a indusry level among seven advanced economies (Ausralia, Canada, France, Germany, Neherlands, he Unied Kingdom and he Unied Saes). Our analysis uses an indusry decomposiion for 26 indusries, of which 25 are par of he marke economy, and an inpu decomposiion of labour produciviy ino he conribuions from ICT capial inensiy, non-ict capial inensiy, labour composiion and oal facor produciviy. By looking a boh growh and levels ogeher, which is he major novely of his paper, we may ge a beer view of he exen o which fas (or slow) growh raes may be due o caching-up effecs. Our analysis shows ha marke services play a key role in explaining he sronger produciviy growh of he Anglo-Saxon economies (Ausralia, Canada, he UK and he U.S.). We also find ha he coninenal European counries (France, Germany and he 18

19 Neherlands) experienced a srong slowdown in marke services. Our inpu decomposiion shows ha he differenial growh performance is mosly srongly relaed o differences in TFP growh, whereas ICT capial deepening acceleraed everywhere since 1995 also in he counries wih slow produciviy growh. 20 An indusry decomposiion of marke services aribues a large role o slower produciviy growh in business services in Europe. The level analysis in he second par of he paper reveals ha oal facor produciviy gaps a he aggregae level mainly emerge for Anglo-Saxon economies (oher han he U.S.) which can also be largely raced o marke services. For he Anglo- Saxon economies lower TFP levels in marke services are he key explanaion for lower aggregae labour produciviy levels. An indusry decomposiion again poins in he direcion of he imporance of business services accouning for a large par of he produciviy shorfall relaive o he U.S. TFP levels in coninenal European counries are a or even subsanially above he U.S. produciviy level in marke services. The advanage for France and Germany is in par relaed o higher levels of non-ict capial inensiy, bu in he Neherlands higher TFP levels also play a role. The combinaion of low produciviy levels and high produciviy growh in marke services in he Anglo-Saxon economies (Ausralia, Canada and he Unied Kingdom), suggess a role for caching up on he U.S. produciviy level. This caching up may be relaed o a rapid adopion of ICT, which is an inpu of major imporance in marke services, and TFP growh. Even hough a direc relaionship beween ICT capial deepening and TFP growh canno be derived from he indusry daa as used in his sudy (Inklaar, 2005; van Ark and Inklaar, 2005), oher sudies using firm level daa or specific case sudies show ha he producive impac of ICT depends crucially on invesmens in inangible capial such as organizaional change and employee raining programs. Bu Brynjolfsson and Hi (2003), for example, argue ha i akes a subsanial amoun of ime for hese complemenary inangible invesmens o have heir full effec on produciviy. Reforms in produc and labour markes may also be needed o exploi he produciviy poenial of ICT (Nicolei and Scarpea, 2003). 20 A furher decomposiion of growh beween 1995 and 2003 ino he pre and pos-2000 period, shows ha he absolue conribuion of ICT capial deepening afer 2000 reurned o pre-1995 levels in mos counries. 19

20 The combinaion of high produciviy levels and slow produciviy growh in marke services in coninenal European counries may be due o an elemen of perversiy in he produciviy performance of many European service indusries. This perversiy may be relaed o underuilizaion of he consumer marke poenial. For example, if all he shopping needs o be done beween 9 am and 6 pm insead of having access o reail faciliies 24 hours per day, produciviy may urn ou higher in he former case. As a resul of rigid markes a higher uilizaion of labour and capial (shop floor) capaciy is realized. These observaions raise new quesions concerning an old debae on he need o adjus produciviy measures for users convenience, and he adjusmen of inpus for uilisaion raes, which goes beyond he scope of his paper. Finally, we sress ha a his sage, cauion is needed when drawing far-reaching conclusions from he inegraed produciviy growh and level comparisons as his ype inegraed measuremen is sill in is infancy. Some of he hypoheses abou indusry group and aggregae produciviy differences are no necessarily borne ou by he deailed indusry resuls. Measuremen and concepualizaion wihin a naional accouns framework sill offers much scope for improvemen. Some of he deailed indusry resuls sugges here is room o improve he comparabiliy and qualiy of he underlying daa. The daase for his paper provides a bridge beween our earlier daases on indusry and growh accouns and a more comprehensive growh and level accouns framework in he EU KLEMS projec which will be published in The resuls from he laer projec need o be awaied o find confirmaion a he level of individual indusries and for addiional counries of he broad rends skeched here. 20

21 References Ark, B. van and D. Pila (1993), Produciviy Levels in Germany, Japan and he Unied Saes Brookings Papers on Economic Aciviy, Microeconomics, 2, pp Ark, Bar van, Rober Inklaar and Rober H. McGuckin (2003), ICT and produciviy in Europe and he Unied Saes, Where do he differences come from? CESifo Economic Sudies, 49(3), pp , Auumn. Ark, Bar van, and Rober Inklaar (2005), Caching Up or Geing Suck? Europe s Troubles o Exploi ICT s Produciviy Poenial GGDC Research Memorandum, GD-79. Downloadable a Basu, Susano and John Fernald, (2006), Informaion Technology as a General Purpose Technology: Evidence from US Indusry Daa paper presened a he conference on he Deerminans of Produciviy Growh, Vienna Sepember 15-16, Bosch, Mahew and Kevin J. Siroh (2006), Informaion Technology and Produciviy Growh in he 2000s paper presened a he conference on he Deerminans of Produciviy Growh, Vienna Sepember 15-16, Brynjolfsson, Erik and Lorin Hi (2003), Compuing produciviy: firm-level evidence Review of Economics and Saisics, 85(4), pp Caves, Douglas W., Lauris R. Chrisensen and W. Erwin Diewer (1982), Mulilaeral Comparisons of Oupu, Inpu and Produciviy using Superlaive Index Numbers Economic Journal, 92, March, pp Corrado, Carol, Paul Lengermann, Eric Barelsman and J. Joseph Beaulieu (2006), Modeling Aggregae Produciviy a a Disaggregae Level: New Resuls for U.S. Secors and Indusries paper presened a he conference on he Deerminans of Produciviy Growh, Vienna Sepember 15-16, Daveri, Francesco (2004), Delayed IT usage: Is i really he drag on Europe s produciviy? CESifo Economic Sudies, 50(3), pp Eicher, Theo S. and Oliver Roehn (2006), Sources of he German Produciviy Demise; Tracing he Effecs of Indusry Level ICT Invesmen paper presened a he conference on he Deerminans of Produciviy Growh, Vienna Sepember 15-16, Fraumeni, Barbara M. (1997), The Measuremen of Depreciaion in he U.S. Naional Income and Produc Accouns Survey of Curren Business, July GGDC (Groningen Growh and Developmen Cenre) (2006a), 60-indusry Daabase, Sepember 2006, downloadable a GGDC (2006b), Indusry Growh Accouning Daabase, Sepember 2006, downloadable a 21

22 Gordon, Rober J. (2003), Exploding Produciviy Growh: Conex, Causes, and Implicaions Brookings Papers on Economic Aciviy, 2, pp Gordon, Rober J. (2004), Why was Europe Lef a he Saion When America s Produciviy Locomoive Depared? NBER Working Paper, no , Augus. Ho, Mun, Someshwar Rao and Jianmin Tang (2004), Sources of Oupu Growh in Canadian and U.S. Indusries in he Informaion Age in Dale W. Jorgenson (ed.), Economic Growh in Canada and he Unied Saes in he Informaion Age, pp , Indusry Canada: Oawa. Inklaar, Rober, (2005), Perspecives on Produciviy and Business Cycles in Europe, PhD hesis, Universiy of Groningen. Inklaar, Rober, Mary O Mahony and Marcel P. Timmer (2005), ICT and Europe's Produciviy Performance; Indusry-level Growh Accoun Comparisons wih he Unied Saes Review of Income and Wealh, 51(4), pp Inklaar, Rober, and Marcel P. Timmer (2006), Inernaional Comparisons of Indusry Oupu, Inpus and Produciviy Levels: Mehodology and New Resuls paper prepared for he Inermediae Inpu-Oupu Meeings 2006 on Susainabiliy, Trade & Produciviy in Sendai, Japan. Jorgenson, Dale W. and Mieko Nishmizu (1978), U.S. and Japanese Economic Growh, : An Inernaional Comparison Economic Journal, 88(352), pp Jorgenson, Dale W., Masahiro Kuroda and Mieko Nishimizu (1987) Japan-U.S. Indusry-Level Produciviy Comparisons, Journal of he Japanese and Inernaional Economies, 1(1), pp Jorgenson, Dale W. and Kevin J. Siroh (2000), Raising he Speed Limi: U.S. Economic Growh in he Informaion Age Brookings Papers on Economic Aciviy, 1, pp Jorgenson, Dale W., Mun Ho and Kevin J. Siroh. (2005) Informaion Technology and he American Growh Resurgence, MIT: Cambridge. McGuckin, Rober H., Mahew Spiegelman and Bar van Ark (2005), The Reail Revoluion: Can Europe Mach he U.S. Produciviy Performance? Perspecives on a Global Economy, The Conference Board: New York. Mulder, Nanno (1999), The Economic Performance of he Service Secor in Brazil, Mexico and he USA, A Comparaive Hisorical Perspecive, GGDC monograph series no. 4, Groningen: Groningen Growh and Developmen Cenre. Nicolei, Giuseppe and Sefano Scarpea (2003), Regulaion, Produciviy and Growh: OECD Evidence, Economic Policy, no. 36, pp. 9-72, April. Oliner, Seven D. and Daniel E. Sichel (2000), The Resurgence of Growh in he Lae 1990s: Is Informaion Technology he Sory Journal of Economic Perspecives, 14(4), p