DISCUSSION PAPER SERIES IZA DP No. 2808 Produciviy and Trade Orienaion in UK Manufacuring Marian Rizov Paric Paul Walsh May 2007 Forschungsinsiu zur Zuunf der Arbei Insiue for he Sudy of Labor
Produciviy and Trade Orienaion in UK Manufacuring Marian Rizov Middlesex Universiy Business School and Triniy College Dublin Paric Paul Walsh Triniy College Dublin and IZA Discussion Paper No. 2808 May 2007 IZA P.O. Box 7240 53072 Bonn Germany Phone: +49-228-3894-0 Fax: +49-228-3894-180 E-mail: iza@iza.org Any opinions expressed here are hose of he auhor(s) and no hose of he insiue. Research disseminaed by IZA may include views on policy, bu he insiue iself aes no insiuional policy posiions. The Insiue for he Sudy of Labor (IZA) in Bonn is a local and virual inernaional research cener and a place of communicaion beween science, poliics and business. IZA is an independen nonprofi company suppored by Deusche Pos World Ne. The cener is associaed wih he Universiy of Bonn and offers a simulaing research environmen hrough is research newors, research suppor, and visiors and docoral programs. IZA engages in (i) original and inernaionally compeiive research in all fields of labor economics, (ii) developmen of policy conceps, and (iii) disseminaion of research resuls and conceps o he ineresed public. IZA Discussion Papers ofen represen preliminary wor and are circulaed o encourage discussion. Ciaion of such a paper should accoun for is provisional characer. A revised version may be available direcly from he auhor.
IZA Discussion Paper No. 2808 May 2007 ABSTRACT Produciviy and Trade Orienaion in UK Manufacuring * Wihin a srucural model we explicily allow for he rade orienaion of companies o esimae produciviy dynamics wihin 4-digi UK manufacuring indusries. We use he FAME daa on UK companies over he period 1994-2003. Following Acerberg e al. (2005) we adjus he algorihm in Olley and Paes (1996) by augmening invesmen and exi decisions o allow for exogenous demand shocs by rade orienaion, assuming ha labour and capial are sae variables, and produciviy follows a firs-order Marov process. We exend he framewor furher by allowing exporing o be an addiional conrol variable ha is driven by lagged produciviy as in Meliz (2003), leading produciviy o follow a second-order Marov process. We find ha over he period of inroducion of he Euro improvemens in aggregae produciviy were driven by exporers mainly by mare share reallocaions away from inefficien and owards efficien expor companies. Aggregae produciviy also benefied from improvemens in produciviy of non-exporers bu was driven by improvemens wihin companies raher han by mare share reallocaions. In a period of susained real exchange rae appreciaion boh expor cleansing and compeiive pressure on non-exporers seem o have conribued o improvemens of produciviy in he UK manufacuring. JEL Classificaion: F14, D24 Keywords: produciviy dynamics, srucural model, rade orienaion, manufacuring companies, UK Corresponding auhor: Paric Paul Walsh Deparmen of Economics Triniy College, Dublin Dublin 2 Ireland E-mail: ppwalsh@cd.ie * This research was underaen in he Insiue for Inernaional Inegraion Sudies (IIIS), Triniy College, Dublin, as par of he research projec Evaluaing he impac of globalisaion using micro daa. The paper has benefied from discussions wih Gauam Gowrisanaran and Ariel Paes on he Olley and Paes (1996) esimaion algorihm. The paper was presened a EARIE 2005 and o he Inernaional Comparaive Analysis of Enerprise (micro) Daa (CAED) conference organised by he U.S. Bureau of Census, he Federal Reserve Ban of Chicago, and OECD in 2006. We han paricipans for heir commens.
1 INTRODUCTION The co-exisence of exporing wih non-exporing companies wihin 4-digi indusries is a srong feaure of our UK daa. Bernard e al. (2003) ouline he same fac for he US. The primary focus of he paper is o embed he role of company rade orienaion ino our srucural esimaion algorihm of produciviy dynamics wihin 4-digi UK manufacuring indusries. Afer esimaing produciviy dynamics for each company we documen and analyse he naure of he aggregaion over companies by rade orienaion in order o undersand aggregae produciviy movemens. Inegraing rade informaion in produciviy esimaion is achieved by adaping he algorihm developed in Olley and Paes (1996). Following Acerberg e al. (2005) we adjus he Olley-Paes algorihm o allow exogenous demand shocs by rade orienaion o agumen invesmen and exi decisions, assuming ha labour and capial are sae variables and produciviy follows a firs-order Marov process. We exend he framewor furher by allowing exporing o be an addiional conrol variable ha is driven by lagged produciviy as in Meliz (2003), leading produciviy o follow a second-order Marov process. We apply he modified algorihm o an unbalanced panel of UK exporing and non-exporing manufacuring companies, wih annual observaions for he period 1997-2001, and esimae ime varying produciviy for companies wihin each 4-digi indusry, over he 1997-2001 period. Our approach brings ogeher wo srands of lieraure on produciviy and exporing. In he firs srand, sudies esimae company oal facor produciviy, in a firs sep, and in a second sep hey proceed o lin produciviy o exporing and conribuions o aggregae produciviy. 1 I is our view ha esing for a relaionship beween exporing and (unobservable) produciviy, ex-pos, is admiing ha here is informaion ha should have been used in he srucural model of he unobservable while esimaing he producion funcion. Indeed heory and empirical evidence on selecion mechanisms guide us. Meliz (2003) employs sun coss associaed wih exporing ha lead o high produciviy companies selecing o exporing. Indeed, a second srand of empirical lieraure confirms his. Robers and Tybou (1997) for Colombia, Bernard and Jensen (2001) for he US, and Bernard and Wagner (2001) for Germany, esimae selecion o exporing regressions and documen ha sun expor-mare enry coss seem imporan enough o generae immense persisence in company expor mare paricipaion. Our daa also confirm his paern of persisence in expor mare paricipaion. 1 Bernard and Jensen (1999), Pavcni (2002), Lopez-Cordova (2002), and Fernandes (2001), for example, apply Olley and Paes (1996) algorihm o approximae produciviy in he firs sep and correlae i wih rade in a second sep. 2
Given his evidence, as argued in Van Biesebroec (2003), one should allow for rade orienaion when esimaing he parameers of he producion funcion. 2 We wish o go a sep furher, following Acerberg e al. (2005), by allowing exporing o be an addiional conrol variable ha is driven by lagged produciviy (Meliz, 2003), leading o a srucural model ha esimaes company produciviy dynamics as a second-order Marov process. The model generaes consisen esimaes of he coefficiens on labour and capial, amongs oher observables. An unbiased produciviy index for exporers and non-exporers is baced ou as a residual. Thus, we mae a conribuion o he efficiency and rade debae, adding new evidence from he UK. 3 We use OLS, GLS and Olley-Paes esimaors of produciviy ha do no allow for rade orienaion of companies as our counerfacuals. Our esimaes of produciviy ha allow for endogenous rade orienaion show clear and persisen differenials in produciviy by rade orienaion and over-ime. Using he Olley and Paes (1996) decomposiions, we are also able o demonsrae ha he improvemens in aggregae produciviy of he UK manufacuring, during he period of inroducion of he Euro, was mainly driven by mare share reallocaions away from inefficien and owards efficien expor companies, alongside gains in produciviy wihin nonexporer companies. In a period of susained real exchange rae appreciaion boh expor cleansing and compeiive pressure on non-exporers seem o have increased produciviy of he UK manufacuring, see Harris (2001) for a lieraure review of hese mechanisms. The remainder of he paper is srucured as follows. Secion 2 provides a brief overview of daa. Secion 3 oulines our behavioural model and he esimaion mehodology used in he paper. Our regression resuls are repored in secions 4. In secion 4 we also underae our analysis of aggregae produciviy while in Secion 5 offer conclusions. 2 Van Biesebroec (2003) and De Loecer (2004) also consider adaping he algorihm developed in Olley and Paes (1996) o allow for exporing, however, heir approaches differ from ours in he way rade orienaion informaion is incorporaed ino heir model of he unobservable. 3 In summary, he lieraure on efficiency and exporing comprises several papers covering various counries: Aw and Hwang (1995) and Aw, Chen, and Robers (2001) on Taiwan; Bernard and Jensen (1995; 1999) on he US; Clerides, Lach and Tybou (1998) on Colombia, Mexico and Morocco; Bernard and Wagner (1997) on Germany; Kraay (1999) on China; Casellini (2001) on Ialy; Delgado, Farinas and Ruano (2002) on Spain; Pavcni (2002) on Chile. On he UK he only exising sudy ha we are aware of is by Girma, Greenaway and Kneller (2004) covering he period 1988-1999. The sudies cover a range of ime periods and use a variey of mehodologies. Imporanly, every single sudy finds ha exporers have higher produciviy han non-exporers - a relaionship ha goes beyond size. They also ypically find ha exporing companies are bigger, more capial inensive and pay higher wages. The lieraure does disagree on he self-selecion versus learning hypohesis. Casellini (2001) repors some evidence suggesing ha he produciviy of exporing companies may increase wih increases in expor inensiy. For Chinese companies, Kraay (1999) repors evidence of learning by exporing as well as Van Biesebroec (2003), for exporers in Africa. Ineresingly, Girma, Greenaway and Kneller (2004) is he only sudy ha suppors he learning hypohesis for a developed mare economy he UK. The evidence in Delgado, Farinas and Ruano (2001) is inconclusive and Bernard and Jensen (1995, 1999), Bernard and Wagner (1997), Clerides, Lach and Tybou (1998) and Aw and Hwang (1995) explicily es for bu fail o find any evidence o suppor he learning by exporing hypohesis. 3
2 THE FAME DATA According o Bureau van Dij, FAME is he mos comprehensive daabase of UK companies available. Daa cover all companies filing a he Companies House in he UK and informaion comprises deailed financial saemens, ownership srucure, aciviy descripion, direc expors, various financial raios and credi scores. 4 The daase used in our analysis conains annual records on more han 80,000 manufacuring companies over he period 1997-2001. The coverage of he daa compared o he aggregae saisics repored by he UK Office for Naional Saisics is as follows: sales - 86%, employmen - 92%, and expors - 100%. The manufacuring secors are idenified on he bases of he curren 2003 UK SIC a he 4-digi level and range beween 1513 and 3663. All nominal moneary variables are convered ino real values by deflaing wih he appropriae 4-digi UK SIC indusry deflaors aen from he Office for Naional Saisics. We use PPI o deflae sales and cos of maerials, and asse price deflaors for capial and fixed invesmen variables. 5 The descripive saisics repored in Table 1 are calculaed from he FAME sample of manufacuring companies on he basis of company averages. We firs loo a he prevalence of exporing among UK manufacuring companies. A one exreme, companies could expor he same share of heir oal oupu. A he oher, a few gian companies would accoun for all expors. In fac, ou of roughly 80,000 companies in he original sample only 15.6 percen repor expor sales over he period of analysis. Previous wor has sough o lin rade orienaion wih indusry. I urns ou ha exporing companies are quie spread ou across indusries. Figure 1 plos he disribuion of indusry expor inensiy: each of he 215, 4-digi manufacuring indusries represened in he sample is placed in one of he 10 bins according o he percenage of companies in he indusry ha expor. In almos all he indusries, he fracion of companies ha expor lies beween 10 and 50 percen. Hence, nowing wha indusry a company belongs o would no answer wih sufficien cerainy wheher i expors. This fac, similar o he findings of Bernard e al. (2003) for he US manufacuring, suggess ha indusry has less o do wih exporing han sandard rade models migh sugges. 4 FAME is a combinaion of high qualiy informaion from Jordans wih easy o use sofware which has been developed by Bureau van Dij Elecronic Publishing (BvD). The financial breadown of he companies in he differen FAME modules is as follows: FAME A - Turnover > 1.5 million or Profis > 150,000 or Shareholder Funds > 1.5 million; FAME B - Turnover > 500,000 and < 1.5 million or Shareholder Funds > 500,000 and < 1,500,000 or Fixed Asses or Curren Asses or Curren Liabiliies or Long Term Liabiliies > 500,000; FAME C - Fixed Asses or Curren Asses or Curren Liabiliies or Long Term Liabiliies > 150,000 and < 500,000; recenly formed companies and oher companies where full financial informaion is no available are also included in his module. 5 Kaayama, Lu, and Tybou (2003), and relaed sudies, argue ha as producion funcions should be a mapping of daa on inpus and oupus, sudies using revenues and expendiure daa as proxies would produce biased produciviy measures. As in his sudy, mos use indusry level deflaors for oupu, raw maerial and capial asses o ge bac he quaniy daa needed. I is clear ha inpus and oupus can be priced differenly for exporers and non-exporers wihin narrowly defined indusries. We noe, however, ha allowing for endogenous rade orienaion in he unobservable will conrol, o a cerain degree, for persisen exchange rae adjused pricing gap beween exporers and non-exporers in heir use of inpus and heir oupus wihin 4-digi indusries. Time dummies can conrol for movemens in he real effecive exchange rae over-ime wihin exporing and non-exporing samples. 4
No only are companies heerogeneous in wheher hey expor, hey also differ subsanially in various crude measures of produciviy. Table 1 repors mean differences, segregaing exporers from non-exporers, and sandard deviaions characerising he disribuions across companies of value added per worer relaive o he overall mean. Similarly, he disribuions across exporing and non-exporing companies of value added per worer relaive o he 4-digi indusry mean are characerised. While differences across indusries cerainly appear in he daa, wha is surprising is how lile indusry explains abou exporing and produciviy. Hence, a saisfacory explanaion of company level behaviour mus go beyond he indusry dimension. Therefore, we consequenly pursue an explanaion of hese facs ha bypasses indusries and goes direcly o samples defined by rade orienaion a he company level. Table 1 also shows he imporance of expor mares for he companies ha do expor. Ineresingly, he vas majoriy of exporers expor less han 30 percen of wha hey produce. Less han 10 percen of he exporing companies expor more han 70 percen of heir producion. Even for he minoriy of companies ha do expor, domesic sales dominae. An answer o hese facs is documened in Table 1 - exporers are much larger. They are almos 4 imes he size of nonexporing companies on average, even when expor revenues are excluded from he calculaion. While only 15.6 percen of manufacuring companies repor ha hey consisenly expor, hese companies accoun for almos 75 percen of he oupu of UK manufacuring. In his paper our goal is o esimae oal facor produciviy (TFP) in a consisen manner, o documen he TFP gaps and o cas ligh on he naure of hese gaps beween exporers and nonexporers, wihin 4-digi indusries. In addiion we ry o explain movemens in aggregae produciviy. The sraegy of our empirical analysis implies ha we will run regressions wihin 4- digi indusries, by sub-samples defined according o company expor saus. This leaves us wih he 41 larges 4-digi indusries, wih sufficien number of observaions o run regressions for exporing and non-exporing sub-samples. These larges 4-digi indusries accoun for almos 80 percen of he UK manufacuring sales in our daa. In erms of he smalles esimaed sample, afer lags are applied and observaions wih missing values deleed, here are 24,338 remaining observaions for 6,722 companies. The coverage of he daa from his sample compared o he aggregae saisics is 58% for expors, and 56% for employmen. The correlaions beween he aggregae saisics series and he esimaed sample series are as follows: value added (used in he regressions as dependen variable) - 0.94, employmen - 0.97, expors - 0.95. 5
In Table 2 we documen summary saisics of regression variables. Exporing companies are older, bigger in erms of value added, employmen and capial, and inves more. 6 The deailed definiions of regression variables are as follow: Value added is oal annual sales adjused for changes in invenories, minus maerial coss in housands of pounds serling. We assume ha maerials used are in a consan proporion of oupu. Expors is he repored value of direc expors, in housands of pounds serling, recorded annually. The problem of poenial undercouning, due o he fac ha indirec expors are no included in his measure is discussed in Bernard and Jensen (1995). Labour is full-ime equivalen number of employees, recorded annually. Age is consruced by using year of incorporaion as a saring poin. Capial is measured as oal fixed asses by boo value, in housands of pounds serling, recorded annually. Invesmen is consruced from he annually observed (for each period, ) capial soc, K and depreciaion, δ using he perpeual invenory mehod: I =K +1 -(1-δ)K. 3 BEHAVIOURAL FRAMEWORK AND ESTIMATION METHODOLOGY To esimae produciviy we relay on a behavioural framewor which builds on models of indusry dynamics by Ericson and Paes (1995) and Hopenhayn (1992) wih applicaions o firm expor decisions as in Meliz (2003). Alongside he economeric modelling ideas in Acerberg e al. (2005), he framewor underlines our esimaion sraegy and helps us specify iming and relaional assumpions for he company decisions in a manner similar o Olley and Paes (1996). 7 The innovaion in our approach is ha we exend he Olley and Paes (1996) framewor by explicily allowing mare srucure (facor mares, demand condiions and prices) o differ across exporing and non-exporing companies. Furhermore, we relax he ofen criicised assumpion abou labour being variable and non-dynamic inpu, i.e., ha he choice of labour in period has no implicaions for he fuure of he company choices. Considering he feaure of he labour mares in developed European counries wih srong employmen proecion laws, i is unliely ha hiring decisions do no have long-erm implicaions for he company. 3.1 Esimaion mehodology assuming exogenous exporing As in Olley and Paes (1996) he log-linear producion funcion o be esimaed is y = β 0 + β + β a + β l + ω + η, (1) a l 6 I is worh noing ha expor saus is persisen over ime as only 9 percen of exporing companies swich beween exporing and non-exporing saes, in our sample during he period of analysis. We mar a company as an exporer if we observe in he daa exporing by he company in any year wihin a 3-year moving window. 7 Levinsohn and Perin (2003) modify he Olley and Paes (1996) approach by using inermediae inpus, such as elecriciy or fuel usage insead of invesmen which have he advanage of more efficien use of he daa. See Acerberg e al. (2005) for a criique of his approach. 6
where he log of company j's value added a ime, y, is modelled as a funcion of he logs of ha company s sae variables a, namely age, a, capial,, and labour, l. Invesmen demand, i deermines he capial soc a he beginning of each (nex) period. The law of capial accumulaion is given by + 1 = (1 δ ) + i. The error srucure is comprised of a sochasic componen, η, wih zero expeced mean, and a componen ha represens unobserved produciviy, ω. Boh ω and η are unobserved, bu ω is a sae variable, and hus affecs company s choice variables. On he oher hand η has zero expeced mean given curren informaion, and hence does no affec decisions. Company s single period profi funcion is π, a, l, ω, e ) c( i, e ), where boh π(.) ( and c(.) depend on e, which represens he economic environmen ha companies face a a paricular poin in ime; e could capure inpu prices, characerisics of he oupu mare, or indusry characerisics. As in Olley and Paes (1996) all hese facors are allowed o change over ime; imporanly, in our exension we allow he facors o also vary across companies according o heir exporing saus. Including mare srucure variaion in he sae space reflecs some of he compeiive richness of he Marov-perfec dynamic oligopoly models such as Ericson and Paes (1995). The company maximizes is expeced value of boh curren and fuure profis according o V(, a, l Φ(, a, l, ω, e ),, ω, e ) = max maxi 0{ π(, a, l, ω, e ) c( i, e ) + (2) βe[ V( + 1, a + 1, l + 1, ω + 1, e + 1), a, l, ω, e, i ]}. The Bellman equaion explicily considers wo company decisions. Firs is he exi decision; Φ(, a, l, ω, e ) represens he sell-off value of he company. Second is he invesmen decision, i, which solves he inerior maximizaion problem. Under he assumpion ha equilibrium exiss and ha he difference in profis beween coninuing and exiing is increasing in ω we can wrie he opimal exi decision rule as 1 if ω ω (, a, l ) Χ = (3) 0 oherwise and he invesmen demand funcion as, i = i, a, l, ω, e ). (4) ( Produciviy, ω is assumed o be deermined by a family of disribuions condiional on he informaion se a ime -1, J -1, which includes pas produciviy shocs. Given his se of disribuions, boh he exi and invesmen decisions will crucially hinge upon he companies percepions of he disribuion of fuure mare srucure condiional on curren informaion (pas produciviy). The decisions ha he companies ae will in urn generae a disribuion for he 7
fuure mare srucure (Masin and Tirole, 1988). In our behavioural framewor we explicily inroduce rade orienaion by companies in e. Decisions on wheher o inves and o exi he mare will depend explicily on wheher a companies expors or no. From he summary saisics of our daa (Table 2) and he findings of numerous empirical sudies we now ha exporers inves more per worer and are also less liely o exi compared o non-exporing companies. 8 A fundamenal problem afflicing invesmen and hus produciviy measuremen is ha companies are made up of differen produc mixes. In he absence of produc-specific daa, which is a ypical problem for micro daases available, consisen esimaes of company produciviy can be obained by allowing he parameers of he producion echnology o vary across companies maing differen (ypes of) producs. The idenifying informaion ha we use here o sor companies by produc ypes is he companies exporing saus. As argued above, exporers differ from nonexporers by boh he producion echniques hey employ and he demand characerisics hey face. Since we deflae value added wih an indusry-wide PPI, we do no conrol for he fac ha oupu and facor prices migh be differen and/or evolve differenly over ime for exporing and non-exporing companies. Therefore we drop his assumpion and incorporae he exporing informaion in he invesmen and survival equilibrium equaions. More formally, we explicily allow ha exporing companies face differenial mare srucures and facor prices when decisions are made abou invesmen and exi from he mare. Paes (1994) discusses condiions under which he invesmen demand funcion (equaion (4)) is sricly monoonic in ω. Under such condiions invesmen can be invered o generae ω = h i, e,, a, l ). (5) ( Then subsiuing equaion (5) ino he producion funcion (1) gives us, y = β ) + 0 + β + β aa + β ll + h ( i, e,, a, l η. (6) Equaion (6) can be esimaed as in Olley and Paes (1996) wih semi-parameric mehods ha rea he inverse invesmen funcion h (.) non-paramerically, using eiher polynomial or ernel. The non-parameric reamen, however, resuls in collineariy and requires he consan,, a and l erms o be combined ino funcion φ i, e,, a, l ) such ha equaion (6) becomes ( y = φ ( i, e,, a, l ) + η. (7) I is imporan o noe ha he abiliy o inver invesmen depends no only on he sric monooniciy in ω bu also on he fac ha ω is he only unobservable in he invesmen equaion. This scalar unobservable assumpion implies ha here can be no measuremen error in he 8 Noe ha he invesmen policy funcion in Olley and Paes (1996) is a soluion o a complicaed dynamic programming problem and depends on all he primiives of he model lie demand funcions, he specificaion of sun coss, form of conduc in he indusry and oher facors as recenly clarified by Acerberg e al. (2005). All hese facors are allowed here o be differen and evolve differenly over ime, for exporing and non-exporing companies. 8
invesmen equaion, no unobserved differences in invesmen coss across companies and no unobserved separae facors ha affec invesmen across companies bu no producion. By inroducing exporing informaion in he invesmen funcion we are able o conrol for such differences and o relax he above assumpions. In secion 3.2 we relax he scalar unobservable assumpion following Acerberg e al. (2005) and furher discuss implicaions for our esimaion algorihm. The capial, age and labour coefficiens are idenified in he second sage of he Olley-Paes procedure. Firs, noe ha even hough we do no idenify any inpu coefficiens in he firs sage we are able o esimae φˆ for use in he second sage. We express ω as ˆ ω ˆ = φ β 0 β β a a β l l. Second, he firs sage of he Olley-Paes procedure is no affeced by endogenous selecion because φ fully conrols for he unobservable; by consrucion, η represens unobservables ha are no nown by he company before inpu and exi decisions are made. In conras, he second sage of he esimaion procedure is affeced by endogenous exi. From equaion (3) i is eviden ha he exi decision in period depends direcly on ω. Using he assumpion ha ω follows an exogenous firs-order Marov process, we can decompose ω ino is condiional expecaion given he informaion nown by he company a -1 and a residual: ω = E[ ω J 1 ] + ξ = E[ ω ω 1] + ξ = g( ω ) + ξ. 1 (8) By consrucion ξ is uncorrelaed wih J -1 and hus wih, a and l which are funcions of only he informaion se a ime -1. Nex from equaion (1) and subsiuing equaion (8) above we can wrie y = β ) + 0 + β + β aa + β ll + g( ω 1 + ξ η. (9) Now o correc for endogenous selecion (exi) les ae expecaions of equaion (9) condiional on boh he informaion a -1 and on X =1 (i.e., surviving ill he nex period). We can wrie E[ y J 1, Χ = 1] = β 0 + β + β aa + β ll + E[ ω J 1, Χ = β + β + β a + β l + g( ω, ω (, a 0 a l 1 = 1], l )). We do no direcly observe ω, a, l ) and o conrol for i we use daa on observed exi. This ( means ha he probabiliy of being in he daa a period condiional on informaion nown a -1 is: Pr( Χ = 1 J 1 ) = Pr( ω ω ( = ϕ ( ω 1, ω ( = ϕ ( i, e, 1 1, a, l, a, l, a 1 1 ) J 1 ) )) = ϕ ( ω, l ) = P 1 1,., a, l ) (10) (11) 9
Equaion (11) represens he second sep of our esimaion algorihm and can be esimaed nonparamerically using probi model wih a (4 h -order) polynomial or ernel as in Olley and Paes (1996). Noe ha we exend he sae variable se wih exporing saus informaion which is an imporan deerminan of companies decision o exi. For a company characerised by ( 1 1 1 1 1 i, e,, a, l ) we are able o generae a consisen esimae of he probabiliy of he company surviving o period, Pˆ. As long as he densiy of ω given 1 ω is posiive in he area around ω (, a, l ), we can inver o wrie ω (, a, l ) as a funcion of ω 1 and P : ω, a, l ) = f ( ω, P ). ( 1 and E[ y Subsiuing his equaion ino equaion (10) and using expressions for esimaed valus, ω 1 Pˆ gives us J 1, Χ = 1] = β 0 + β + β aa + βll + g( ω 1, f ( ω 1, P )) = β 0 + β + β aa + βll + g ( ω 1, P ) = β + β + β a + β l + g ( ˆ φ β β β a 0 which afer removing he expecaions operaor becomes y a l 1 0 1 a 1 β l l 1 β ˆ, ˆ + β aa + β ll + g ( φ 1 β 1 β aa 1 β ll 1 P ) + ξ + η, (12) = where he wo β 0 erms have been encompassed ino he non-parameric funcion g. Equaion (12) represens he hird (las) sep of our esimaion algorihm and can be esimaed wih NLLS, approximaing g wih eiher polynomial or ernel. I is also imporan o noe ha he idenificaion of he labour coefficien in he las raher han in he firs sep of he algorihm requires maing assumpions how curren labour responds o he curren realisaions of ξ. One possible reamen is ha labour is fixed before he realisaion of ξ, which is he same assumpion as for capial. This implies ha curren labour is no correlaed wih curren innovaion in produciviy and β l can be idenified in he hird sep. A second, and more realisic, assumpion is ha curren labour can respond o curren innovaions in produciviy. We sill can obain esimaes of β l using equaion (12) and he fac ha lagged values of labour (l -1 ) should be uncorrelaed wih ξ which follows from he informaion srucure of he model., Pˆ ), ˆ 3.2 Esimaion mehodology assuming endogenous exporing The assumpion of a scalar unobservable sae variable can be relaxed. In he preceding analysis we discuss how inroducing exporing informaion resuls in relaxing he original Olley-Paes assumpion in wo ways. Firs, we allow invesmen o depend on an unobservable demand shoc ha varies across companies according o heir exporing saus, in addiion o he ω process. If we assume ha he demand shoc, α, also follows a firs-order Marov process, independen of he 10
process ω hen he invesmen funcion will be a funcion of boh unobservables: i = i, a, l, ω, α ). 9 Using he exporing informaion as a conrol ogeher wih observed ( invesmens allows us o subsiue for ω in he firs sep of he esimaion algorihm. Specifically, i is reasonable o assume ha he exporing informaion conains companies pricing decisions. If we label he bivariae policy funcion (i, e ) as Ψ and assume as in Acerberg e al. (2005) ha i is bijecure in (ω, α ) condiional on (, a, l ), he policy funcion can be invered o form ω = Ψ 1 (, a, l, i, e ). Then we can proceed wih he firs sep of esimaion as in secion 3.1. For he hird sep, since α progresses independenly of ω he company s condiional expecaion of ω given J -1 only depends on ω -1. Thus he hird sep can again be specified as equaion (12). The advanage of using exporing informaion (inerpreed here as a conrol for demand shocs) besides being an imporan deerminan of invesmen also generaes independen variance in φˆ and hus improves he precision of our esimaes. However, assuming ha exporing saus of he company is compleely independen process from he evoluion of ω is no saisfacory. Besides demand side, informaion for exporing saus is relaed o imporan supply side characerisics. A large number of empirical sudies show ha here is a srong selecion mechanism where more producive companies ener foreign mares. Meliz (2003) builds his sylised fac ino a heoreical model of firm produciviy and rade orienaion. Thus, a more realisic assumpion is ha a company s exporing saus a ime depends on he company s produciviy in previous periods. These consideraions also sugges ha produciviy follows more complicaed process han previously assumed. We implemen his fac hrough a second-order Marov process describing produciviy. The invesmen demand equaion hen becomes: i = i ω, ω,, a, l ). The presence of wo unobservables complicaes ( 1 inverabiliy of he invesmen equaion and requires addiional assumpions. The soluion relies on again using a second observable conrol of he company decision problem. The conrol ha we use is he company s exporing saus, which can be inerpreed as an indicaor of company s invesmen in developing higher qualiy producs, adverising, and building disribuion newors in foreign mares. Then we can formulae as in Acerberg e al. (2005) he bivariae policy funcion: 9 I is also reasonable o assume ha here are wo Marov processes ha are inerrelaed. The company s condiional expecaion of ω given J -1 hen depends on boh ω 1 and α 1. The informaional demand in his case is much higher hough. We will need hen o esimae α 1 using deailed demand side daa as in Berry e al. (1995) which we do no have. 11
i e = Ψ ω, ω,, a, l ). Under cerain condiions he Ψ can be invered in ( 1 1 ω = Ψ ( i, e,, a, l ). ji ω o obain However, we can also inerpre exporing saus as a sae variable represening he company s echnology, i.e., he soc of higher qualiy producs, now-how, and disribuion newors in foreign mares. The company s echnological and logisic asses evolve as a resul of decisions, affeced by he observed produciviy in previous periods. Furhermore, he exporing saus will also be affeced by oher company-specific facors such as ype of ownership and corporae governance which generally should no be correlaed wih he invesmen decisions bu may affec demand. Following he argumen we can specify he curren exporing saus as a nonparameric funcion of i 1 1, a 1, l 1 and a vecor of oher company-specific characerisics,, x' such as ype of ownership and corporae governance. Esimaing an exporing equaion and using he propensiy o expor, ê ji as a conrol insead of e ji allows as o rea he expor decision as an endogenous one. This reamen also implies ha we implicily consider wo sources of produciviy growh, one evolving as a conrolled Marov process, and one as an exogenous Marov process. This is he closes empirical approximaion of our behavioural framewor. The firs sep of our esimaion algorihm remains he same as before - he non-parameric funcion conrolling for curren produciviy is specified as a polynomial, including he exporing informaion. The hird sep is modified and now becomes ˆ~ y b + b a + g ˆ φ b b a b l, ˆ φ b b a b l, P ) + ξ + η, = a ( 1 1 a 1 l 1 2 2 a 2 l 2 where φˆ variables are obained from he firs sage esimaes a -1 and -2 periods. Because he condiional expecaion of ω given J -1 now depends on ω 1 and ω 2, we need o use esimaes of φˆ from wo prior periods. Conrolling for endogenous selecion has o be modified accordingly as well; noe he (13) change of noaion ( Pˆ~ insead of Pˆ ). Thus, he second-sep equaion (11) will become ~ P = ~ ϕ ( ω 1, ω 2, ω ( = ~ ϕ ( i, i, e, e 1 2 1 2, a, l, l )) ~ = ϕ ( ω, l,, a 1 2 2 1, ω ), 2 2,, a, l ) (14) where he second equaliy holds because of equaion (5) and he fac ha and a are deerminisic funcions of i -1, -1, and a -1, -1 and a -1 of i -2, -2, and a -2, ec. In erms of verifying wheher enering foreign mares maes companies more producive, we have filered ou mare srucure specific shocs ha are differen for exporers (lie demand condiions, facor mares, exi barriers, ec.) and do no aribue hem o produciviy gains by 12
exporers. 10 However, hese facors remain consan across exporers wihin a given indusry and a ime period. On a more concepual level, mare condiions migh jus be one of he driving forces behind he learning process. So if we sill find evidence for produciviy gains by exporers, i would mean ha addiional company-specific facors play a significan role in maing exporers more producive once hey have sared exporing, e.g., conacs wih foreign buyers, locaion, desinaion of expors. Noe ha his effec is purified from differences in price rends, facor prices, and mare condiions common o all exporers wihin an indusry. 4 ESTIMATION RESULTS AND ANALYSIS OF AGGREGATE PRODUCTIVITY We run separae regressions for each of he op 41 4-digi indusries. In Table 3 we repor weighed averages, using value added as weigh, of he esimaed coefficiens from hese regressions. In addiion, we also repor weighed averages, using value added as weigh, of log company level produciviy, ω, wih and wihou he firs sep regression error for resuls form all esimaors used. We compue produciviy measures aggregaing over exporing and non-exporing sub-samples and over 4-digi indusries, firs, where produciviy a he company level conains he regression error by company. In he second produciviy measure he firs sep regression error is deduced such ha we are lef wih he pure deerminisic par of TFP, i.e., ω. The difference beween he wo measures is very small and we furher focus our discussion on he produciviy measure conaining he regression error, which can be inerpreed as sochasic learning process. Firs, in Table 3 we repor our regressions where expor saus of a company is no considered. In his conex we repor OLS, GLS (wihin group), and he sandard Olley-Paes (wihou exporing informaion) esimaors, columns 2, 3 and 4, respecively. These can be compared o he Olley-Paes esimaors, where expor saus is reaed as an exogenous shoc o invesmen and exi decisions, column 5, and where expor saus is considered o be a conrol variable ha leads o a second-order Marov process in produciviy - column 6. 11 In columns 7 and 8 we repea he previous wo esimaions where we use insrumened probabiliies of being an exporer insead of expor saus. Finally, we spli exporers and non-exporers ino sub-samples wihin indusries, allowing for differenial echnology and facor shares, reaing he probabiliy of being an exporer 10 Inroducing he exporing informaion ino he producion funcion can be reaed as inroducing an addiional inpu in producion. If one has o idenify he coefficien on he exporing inpu in he hird sep, i would imply ha exporing only affecs he average of he fuure produciviy disribuion and hence leaves no scope for learning by exporing o be a heerogeneous process across companies. In addiion i would imply ha he effec is ime-invarian, i.e. every year exporing raises oupu (condiioned on labour and capial) by he coefficien esimaed on exporing. 11 The sandard errors of all Olley-Paes esimaion rouines are boosrapped using 1,000 replicaions. 13
(condiional on company characerisics) as an exogenous shoc o invesmen and exi decisions and alernaively, as an endogenous conrol variable leading o a second-order Marov process in produciviy - columns 9 o 12, respecively. By comparing resuls from OLS and GLS wih he Olley-Paes esimaors we see plausible (and expeced) changes in he esimaes of he parameric bea s and in produciviy, ω. The R 2 on he explained movemens in value added progressively increases as we incorporae a richer model of he unobservable. In columns (6) and (8) we allow expor saus and he proabiliy of being an exporer o be considered a conrol variable ha leads o a second-order Marov process in produciviy. This changes he bea s and he average and variance in our company level produciviy, ω, esimaes. In columns (11) and (12) we go furher by allowing he parameric echnologies o be differen across subsamples of non-exporers and exporers, where we allow he proabiliy of being an exporer o be considered a conrol variable ha leads o a second-order Marov process in produciviy Armed wih hese various esimaes of our company level produciviy, ω, we analyse he conribuions of exporers and non-exporers o aggregae produciviy. In he UK manufacuring here is a srong posiive correlaion (correlaion coefficiens ranging around 80%) beween expor inensiy he raio of expors o oal sales - and aggregae produciviy, measured on he basis of various company TFP measures, over he period of analysis as illusraed in Figure 2. OP is he benchmar TFP measure derived from Olley-Paes esimaor where no expor-saus informaion is used; OPex1 is he TFP measure derived from an esimaion over companies where we allow for exogenous rade orienaion shocs and where produciviy is firs-order Marov process; OPex2 is he TFP measure derived from an esimaion over companies where endogenous exporing decisions creae a second-order Marov process. Nex, xopex1 denoes a TFP measure esimaed separaely over sub-samples of exporers and non-exporer where ω was modelled as a firs-order endogenous (insrumening he expor variable wih prediced value) Marov process. Analogously, xopex2 is a TFP measure esimaed under he assumpion ha endogenous exporing leads o a second-order endogenous Marov process. The relaionship beween expor inensiy and TFP depiced in Figure 2 may lead one o hin ha recen improvemens in produciviy are expor lead (Becerman, 1965; Kaldor, 1970). Ye, micro-daa sudies such as Barnes and Hasel (2000; 2001) and Disney e al. (2003) indicae ha he expansion of more efficien companies accouned for beween one hird and a half of he labour produciviy growh in he UK during he 1990s and even for a larger share of TFP growh. To relae indusry-level produciviy o rade orienaion, we sar by defining indusry produciviy, P, as a mare-share weighed sum of he company produciviy levels: P = ω, s i i where ω i is company produciviy as defined in previous secions and s i is he value of company i's i 14
real revenue relaive o oal indusry revenue in year. Wih his formulaion, shifs of oupu from low produciviy o high produciviy companies will conribue posiively o indusry produciviy growh, even if no individual company experiences a produciviy increase. This analysis is appropriae because our ulimae ineres is in he abiliy of he companies in he indusry o conver he se of inpus used in he indusry ino oupu, and movemens of resources from low o high produciviy companies can be jus as effecive in increasing indusry oupu as are produciviy improvemens wihin individual companies. As shown by Olley and Paes (1996), aggregae produciviy can be rewrien as: P P+ Δ Δω, where P is he unweighed mean = s i i produciviy over all companies in a paricular indusry, in year and he Δ is used o represen a deviaion from he un-weighed mean in year. The second erm in he equaion is he sample covariance beween company produciviy and mare share in year, and summed up over he number of companies in he year. The larger his covariance, he higher he share of oupu ha is allocaed o more producive companies and he higher is indusry produciviy. Table 4 repors he changes in aggregae produciviy for each of he eleven aggregae (2- digi) indusries over he 1997-2001 period as he decomposiions of aggregae produciviy change is repored separaely for exporers and non-exporers. We repor decomposiions of aggregae produciviy based on five produciviy measures corresponding o he original Olley-Paes algorihm (OP) and o our modified esimaion algorihm where we use exporing informaion and model he exporing saus as a firs- or second-order Marov process, while esimaing he oal sample (OPex1 and OPex2) and he exporer and non-exporer sub-samples separaely (xopex1 and xopex2). Furhermore, we repor separae decomposiions of aggregae produciviy for wo disinc periods, before (for 1997-98) and afer (for 2000-01) he implemenaion of he Euro, in he beginning of 1999. By looing in changes of aggregae produciviy in he wo periods wih disinc exchange rae regimes and inernaional rade condiions we are able o derive imporan resuls concerning he impac of foreign rade and macroeconomic condiions on produciviy. Ineresingly, we find a dual paern, which cerainly is no expor driven. For he exporer sample (conaining more producive companies), mare share expansion drives aggregae produciviy, raher han produciviy improvemens wihin companies as shown in Table 4, columns 4 o 7. Such aggregae oucomes can be explained by mechanisms oulined in he Meliz (2003) model, driven by micro selecion and mare share reallocaion effecs. For he non-exporer sample (conaining less producive companies), he paern is quie he opposie wihin company produciviy change is larger han he produciviy change due o mare share reallocaions. This evidence suggess ha here is no much learning by exporing bu raher less producive companies i 15
end o improve heir produciviy in an aemp o converge o he more producive exporing companies. One would be wrong o assume ha TFP is expor lead in he radiional sense. Looing ino he paerns of specific indusries, we noice ha here are ineresing differences beween exporer and non-exporer samples as well as wih respec o ime periods, before and afer he implemenaion of he Euro. For exporers he relocaion of mare shares (column 6 vs. column 7) plays more imporan role in he changes of aggregae produciviy compared o non-exporers. Furhermore, a general paern across he indusries is ha aggregae produciviy, driven mosly by mare share relocaion increases more significanly in he pos-euro period. Considering non-exporers, in general, i is eviden ha he non-exporing companies were loosing produciviy in he pre-euro period while in he period afer 1999 here is significan increase in wihin company produciviy (column 10 vs. column 11), which seems o play imporan role in he changes of he aggregae produciviy as well. While increasing heir individual produciviies, non-exporing companies remained relaively small as here are no significan mare share reallocaions in he sample. The decomposiions of he oal manufacuring sample for exporers and non-exporers by ime periods confirm he paers observed in individual indusries. Noeworhy is also he fac ha he produciviy measure, based on he original OP esimaion algorihm, wihou aing ino accoun he company exporing saus, leads ofen o differen inerpreaions abou he facors affecing aggregae produciviy compared o measures derived on he basis of our modified esimaion algorihm. Furhermore, he produciviy measures calculaed by using exporing informaion and esimaes of he oal sample (OPex1 and OPex2) show similar decomposiion paern o he one depiced by produciviy measures calculaed on he basis of esimaes of separae exporer and non-exporer samples (xopex1 and xopex2). Imporanly, all repored produciviy measures where exporing informaion is incorporaed ino he esimaion algorihm exhibi similar paerns and confirm he robusness of our conclusions concerning he changes in aggregae produciviy. The observed general paern indicaes ha in he exporer sample a larger share of indusry oupu is being reallocaed o he more producive companies, and hus, indusry produciviy is higher han he unweighed company mean. Unlie he unweighed mean produciviy, he magniude of covariance erm does vary grealy over ime and more so for exporers, alhough he non-exporer sample also exhibis significan reallocaions in some indusries and periods. This variaion in he magniude of he covariance erms indicaes ha mare share reallocaions raher han shifs of he company produciviy disribuion are he main source of aggregae indusry produciviy growh observed in Figure 2. I is also imporan o sress ha in he sample of nonexporers, which are he less producive companies in every indusry, here is a endency of wihin 16
company produciviy improvemens. This paern suggess a rend oward convergence beween exporer and non-exporer samples as he rend srenghens afer he implemenaion of he Euro. 5 CONCLUSION In his paper we ouline a mehodology for esimaing he parameers of a producion funcion while lining he unobservable produciviy o an endogenous company level rade orienaion choice, amongs oher facors. Following Acerberg e al. (2005) we adjus he algorihm in Olley and Paes (1996) by augmening invesmen and exi decisions o allow for exogenous demand shocs by rade orienaion, assuming ha labour and capial are sae variables, and produciviy follows a firs-order Marov process. We exend he framewor furher by allowing exporing o be an addiional conrol variable ha is driven by lagged produciviy as in Meliz (2003), leading produciviy o follow a second-order Marov process. We esimae he parameers of producion funcions for exporing and non-exporing samples of companies wihin he 4-digi UK manufacuring indusries, for he period 1997-2001. Allowing for rade-orienaion grealy enhances our abiliy o obain consisen and unbiased esimaes of he parameers of he producion funcion. We find ha over he period of inroducion of he Euro improvemens in aggregae produciviy were driven by exporers - mainly by mare share reallocaions away from inefficien and owards efficien expor companies. Aggregae produciviy also benefied from improvemens in produciviy of non-exporers bu was driven by improvemens wihin companies raher han by mare share reallocaions. These findings show a dual paern in aggregae produciviy changes and seem o suppor he idea ha a susained real exchange rae appreciaion can induced expor (company) cleansing as well as compeiive pressure on non-exporers o increase produciviy in he UK manufacuring. 17
REFERENCES Acerberg, D., Benard, L., Berry, S., and Paes, 2005. A. Economeric Tools for Analyzing Mare Oucomes, Forhcoming chaper in Handboo of Economerics, Volume 6. Aw, B.Y. and Hwang, A.R., 1995. Produciviy and he expor mare: A firm-level analysis. Journal of Developmen Economics 47, 313 332. Aw, B.Y., Chen, X., and Robers, M.J., 2001. Firm-level evidence on produciviy differenials and urnover in Taiwanese manufacuring, Journal of Developmen Economics 66, 51-86. Barnes, M., and Hasel, J., 2000. Produciviy growh in he 1990s: Evidence from Briish plans, Queen Mary, Universiy of London Draf Paper, available a hp://www.qmw.ac.u/~uge193. Barnes, M., and Hasel, J., 2001. Produciviy, compeiion and downsizing. In Summary Volume of Treasury Growh Seminar, presened a No.11 Downing Sree, Ocober 2000, HM Treasury, London, March, available a hp://www.qmw.ac.u/~uge193 and hp://www.hm-reasury.gov.u. Becerman, W., 1965. Demand, expors and growh. In W. Becerman and Associaes, The Briish economy in 1965, pp. 44 72. The Naional Insiue of Economic and Social Research. Bernard, A., Eaon, J., Jensen, J.B., and Korum, S., 2003. Plans and produciviy in inernaional rade, American Economic Review 93(4), 1268-1290. Bernard, A. and Jensen, J.B., 1995. Exporers, jobs and wages in U.S. manufacuring, 1976 1987. The Brooing Papers on Economic Aciviy. Microeconomics 1995, 67 112. Bernard, A. and Jensen, J.B., 1999. Exporing and produciviy, NBER Woring Paper No. 7135. Bernard, A. and Jensen, J.B., 2001. Why some firms expor, NBER Woring Paper No. 8349. Bernard, A. and Wagner, J., 1997. Expors and success in German manufacuring, Welwirschafliches Archiv 133, 134-157. Bernard, A. and Wagner, J., 2001. Expor enry and exi by German firms, Welwirschafliches Archiv 137, 105-123. Berry, S., Levinsohn, J., and Paes, A., 1995. Auomobile process in mare equilibrium, Economerica 63(4), 841-890. Casellini, D., 2001. Expor behaviour and produciviy growh: Evidence from Ialian manufacuring firms, Universia di Urbino, mimeo. Clerides, S.K., Lach, S., and Tybou, J.R., 1998. Is learning-by-exporing imporan? Microdynamic evidence from Colombia, Mexico and Morocco. Quarerly Journal of Economics CXIII, 903 947. Delgado, M., Farinas, J., and Ruano, S., 2002. Firm produciviy and expor mares: A nonparameric approach, Journal of Inernaional Economics 57, 397 422. 18
De Loecer, J., 2004. Do expors generae higher produciviy? Evidence from Slovenia, LICOS Discussion Paper 151/2004, K.U. Leuven. Disney, R., Hasel, J., and Heden, Y., 2003. Resrucuring and produciviy growh in UK manufacuring, Economic Journal 113 (July), 666-694. Ericson, R. and Paes, A., 1995. Marov-perfec indusry dynamics: A framewor for empirical wor. Review of Economic Sudies 62, 53 82. Fernandes, A., 2001. Trade policy, rade volumes and plan-level produciviy in Colombian manufacuring indusries, Yale Universiy, mimeo. Girma, S., Greenaway, D., and Kneller, R., 2004. Does exporing lead o beer performance? A microeconomeric analysis of mached firms. Review of Inernaional Economics 12(5), 855-866. Harris, Richard. G., 2001 Is There a Case For Exchange-Rae-Induced Produciviy Change? In Revisiing he Case for Flexible Exchange Raes Proceedings of a Conference, November 2000. (Oawa: Ban of Canada) Hopenhayn, H., 1992. Enry, exi, and firm dynamics in long-run equilibrium. Economerica 60, 1127 1150. Kaldor, N., 1970. The case for regional policies. Scoish Journal of Poliical Economy 17, 337 348. Kaayama, H., Lu, S., and Tybou, J., 2003.Why plan-level produciviy sudies are ofen misleading, and an alernaive approach o inerference, NBER WP 9617. Kraay, A., 1999. Expors and economic performance: Evidence from a panel of Chinese enerprises, World Ban, mimeo. Levinsohn, J. and Perin, A., 2003. Esimaing producion funcions using inpus o conrol for unobservables, Review of Economic Sudies 70, 317-341. Lopez-Cordova, J.E., 2002. NAFTA and Mexico s manufacuring produciviy: An empirical invesigaion using mirco-level daa 1993-1999, Iner-American Developmen Ban, mimeo. Masin, E. and Tirole, J., 1988. A heory of dynamic oligopoly: I&II, Economerica 56, 549-600. Meliz, M.J., 2003. Esimaing produciviy in differeniaed produc indusries, Economerica 71(6), 1695-1725. Olley, S. and Paes, A., 1996. The dynamics of produciviy in he elecommunicaions equipmen indusry, Economerica 64(6), 1263-1297. Paes, A., 1994. Dynamic srucural models, problems and prospecs: mixed coninuous discree conrols and mare ineracions. In: Laffon, J.-J., Sims, C. (Eds.), Advances in Economerics: The Sixh World Congress of he Economeric Sociey, volume II, Chaper 5. Cambridge Universiy Press, New Yor, pp. 171 260. 19
Pavcni, N., 2002. Trade liberalizaion, exi, and produciviy improvemens: Evidence from Chilean plans, Review of Economic Sudies 69, 245-276. Robers, M.J. and Tybou, J.R., 1997. The decision o expor in Colombia: An empirical model of enry wih sun coss. American Economic Review 87 (4), 545 564. Van Biesebroec, J., 2003. Exporing raises produciviy in sub-saharan African manufacuring firms, NBER WP 10020. 20
Figure 1: Indusry exporing inensiy 40 35 Number of indusries 30 25 20 15 10 5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Indusry exporing inensiy (share of exporers) 21
Figure 2: Aggregae produciviy and expor inensiy of he UK manufacuring 1.2 Index 1.15 1.1 1.05 1 0.95 0.9 1997 1998 1999 2000 2001 Expor inensiy (expors/oal revenue) Aggregae produciviy (OP, 0.41) Aggregae produciviy (OPex1, 0.81) Aggregae produciviy (OPex2, 0.83) Aggregare produciviy (xopex1, 0.79) Aggregae produciviy (xopex2, 0.79) Year Noe: Indexes are normalised a 1 for 1997; correlaion coefficiens beween aggregae produciviy, for each produciviy measure, and expor inensiy are repored in he parenheses. 22
Table 1: Company level facs on exporing Exporer share Percenage of all companies Percenage of oal oupu 15.6 74.4 Produciviy Sandard deviaion of log produciviy (%) Exporer less non-exporer average log produciviy (%) Labour produciviy (LP), wihin oal 90.2 16.8 manufacuring Labour produciviy (LP), wihin 4-digi 85.2 13.3 indusries Exporer size advanage Raio of average UK sales Raio of average oal sales 3.8 6.5 Expor inensiy (%) Percenage of all exporers Percenage of oal oupu of exporers 0 o 30 66.7 41.7 30 o 70 25.8 32.4 70 o 100 7.5 25.8 Noe: The saisics are calculaed from average company characerisics over he 1994-2003 period. Labour produciviy (LP) is measured as value added per worer. Heerogeneiy is he sandard deviaion of he logarihm of LP, muliplied by 100. The produciviy advanage of exporers is he difference (muliplied by 100) in he mean logarihms of produciviy beween exporing and non-exporing companies. Wihin indusry indicaes ha we subrac (from he log of produciviy for each company) average log produciviy of he appropriae 4-digi indusry. The size advanage of exporers is he average shipmens of exporing companies relaive o he average for non-exporing companies, presened as a simple raio. 23
Table 2: Summary saisics Variables Age Value added Toal fixed asses Employmen Invesmen E N T E N T E N T E N T E N T 1997 28.9 (24.0) 25.3 (21.6) 27.4 (23.1) 24.8 (272.8) 8.5 (46.3) 18.1 (211.0) 25.6 (403.6) 9.0 (54.1) 18.7 (310.8) 626 (3197) 266 (1263) 477 (2584) 5.4 (75.2) 2.4 (17.6) 4.2 (58.7) 1998 28.5 (23.7) 25.0 (21.4) 27.0 (22.8) 24.0 (277.5) 8.7 (48.8) 17.4 (212.0) 28.9 (629.7) 8.0 (46.6) 19.9 (476.4) 596 (3823) 263 (1177) 453 (2991) 11.6 (360.9) 2.2 (14.0) 7.5 (272.6) 1999 28.4 (23.6) 24.5 (20.9) 26.7 (22.5) 25.6 (319.7) 8.9 (50.0) 18.3 (242.4) 27.3 (616.5) 7.8 (36.2) 18.8 (463.5) 557 (3615) 258 (1066) 427 (2807) 5.7 (98.4) 2.2 (18.9) 4.2 (74.9) 2000 28.6 (23.5) 24.9 (21.1) 27.0 (22.6) 25.2 (355.6) 10.7 (59.9) 18.9 (270.2) 33.8 (948.0) 8.5 (44.6) 22.8 (713.2) 518 (3214) 272 (1114) 411 (2528) 10.7 (411.8) 2.2 (30.2) 7.0 (310.1) 2001 28.7 (23.6) 24.8 (21.4) 27.0 (22.7) 27.4 (416.1) 11.2 (63.6) 20.2 (314.2) 36.5 (1013.5) 9.4 (55.6) 24.6 (759.2) 528 (3262) 286 (1318) 422 (2595) 6.0 (145.3) 2.0 (23.0) 4.3 (109.8) Average 28.6 (23.7) 24.9 (21.2) 27.0 (22.7) 25.4 (332.9) 9.6 (54.4) 18.6 (253.6) 30.4 (760.0) 8.5 (47.6) 21.0 (573.7) 564 (3437) 269 (1187) 437 (2710) 7.9 (262.4) 2.2 (21.6) 5.5 (198.3) Noe: Toal number of observaions, afer applying lags and deleing observaions wih missing values, for he smalles esimaed sample covering five years and he 1997-2001 period is 24,338 for he oal sample (T); for exporers (E) observaions are 13,831 and for non-exporers (N) - 10,507. Moneary values are in millions of consan (wih respec o year 2000) pounds serling. Sandard deviaions are in parenheses. 24
Table 3: Weighed average coefficien esimaes for he oal sample of UK manufacuring companies Parameers Esimaion mehod Expor saus no considered Expor saus considered Exogenous Endogenous OLS GLS fe OP OP 1 s OP 2 nd OP 1 s OP 2 nd OP 1 s order MP OP 2 nd order MP order MP order MP order MP order MP E NE E NE (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) b_l s.e 0.70 0.05 0.47 0.10 0.61 0.08 0.61 0.09 0.66 0.06 0.61 0.08 0.66 0.05 0.56 0.12 0.65 0.11 0.59 0.07 0.68 0.08 b_ s.e. 0.21 0.04 0.13 0.06 0.29 0.05 0.29 0.05 0.24 0.04 0.28 0.05 0.26 0.04 0.28 0.07 0.26 0.06 0.28 0.05 0.26 0.05 b_a s.e 0.02 0.06-0.08 0.42-0.02 0.11-0.12 0.15 0.11 0.10 0.01 0.11 0.03 0.07 0.16 0.16 0.18 0.13-0.02 0.08 0.07 0.09 log ω 3.24 4.94 3.14 3.24 3.19 3.29 3.32 3.57 3.18 3.59 3.06 s.d. 0.75 1.06 0.81 0.82 0.79 0.82 0.79 1.08 1.08 0.98 0.98 log ω* 3.26 5.02 3.16 3.25 3.18 3.31 3.29 3.55 3.09 3.57 3.04 s.d. 0.11 0.31 0.53 0.57 0.54 0.59 0.57 0.90 0.90 0.83 0.83 R 2 0.77 0.73 0.97 0.98 0.99 0.98 0.99 0.99 0.99 0.99 0.99 No obs. 24,338 24,338 24,338 24,338 24,338 24,338 24,338 13,831 10,507 13,831 10,507 Noe: Coefficien esimaes repored here are weighed averages of coefficiens esimaed wihin each 4-digi indusry in he sample. ω is produciviy measure wih las sage esimaion error and ω* is produciviy measure ne of he las sage esimaion error. Coefficiens repored in bold are significan a he 1% level or beer. 25
Table 4 Decomposiions of produciviy change over he 1997-2001 period, by indusry and expor saus Indusry (Obs.) Period Esimaion mehod Aggregae produciviy in 1997or 2000 (logω) Aggregae produciviy change Exporers Wihin company produciviy change Share reallocaion produciviy change Aggregae produciviy in 1997 or 2001 (logω) Non-exporers Aggregae produciviy change Wihin company produciviy change Share reallocaion produciviy change (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) 97-98 xopex2 4.30-5.8 0.4-6.2 3.01-3.6 1.6-5.2 xopex1 4.83-7.9 0.3-8.2 3.17-2.0 1.7-3.7 OPex2 2.91-3.8 0.7-4.5 2.25-1.2 0.6-1.8 OPex1 2.88-2.8 0.4-3.2 2.18-1.3 0.7-2.0 15 Food and beverages (1394) 18 Wearing apparel (534) OP 3.33-1.0 2.2-3.2 2.96-4.2-0.8-3.4 00-01 xopex2 4.56 5.6 1.5 4.1 2.62 13.2 1.6 11.6 xopex1 4.21 9.5 1.8 7.7 2.79 11.8 1.2 10.6 OPex2 2.46 5.1 0.8 4.3 2.17 7.6 2.6 5.0 OPex1 2.63 7.9 0.8 7.1 2.50 8.0 2.0 6.0 OP 3.05 10.1 1.6 8.5 2.43 17.8 2.5 15.3 97-98 xopex2 5.65-3.5-0.2-3.3 3.82-0.1 0.2-0.3 xopex1 5.24-3.7-0.1-3.6 4.05-0.1 0.3-0.4 OPex2 3.95-6.8-0.6-6.2 3.29-0.4 0.4-0.8 OPex1 4.07-5.1-0.5-4.6 3.40-0.7 0.3-1.0 OP 2.62-10.5-1.8-8.7 2.46 0.1 2.8-2.9 00-01 xopex2 5.86 4.6 0.9 3.7 3.91 2.5 2.8-0.3 xopex1 5.42 5.1 1.1 4.0 4.17 2.4 2.5-0.1 OPex2 3.99 7.4 2.3 5.1 3.49 3.9 3.6 0.3 OPex1 4.17 6.6 2.3 4.3 3.54 2.9 2.4 0.5 OP 2.91 15.5 7.2 8.3 2.72 5.5 5.3 0.2 26
Table 4 coninued (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) 21 97-98 xopex2 3.86 7.1-4.4 11.5 2.81-9.4-6.9-2.5 Pulp and xopex1 3.22 11.7-4.2 15.9 2.43-12.4-6.9-5.5 paper OPex2 3.48 5.7-1.5 7.2 3.33-5.6-4.5-1.1 (894) OPex1 3.70 6.0-1.8 7.8 3.60-6.5-4.3-2.2 22 Publishing and prining OP 3.56 2.2-2.8 5.0 3.41-5.0-4.2-0.8 00-01 xopex2 3.68 4.7 2.0 2.7 2.21-2.9-1.1-1.8 xopex1 3.24 4.6 2.7 1.9 2.45-3.4-1.4-2.0 OPex2 3.62 2.5 2.4 0.1 3.40-2.2-1.3-0.9 OPex1 4.07 2.3 2.2 0.1 3.67-3.3-1.2-2.1 OP 3.77 2.4 1.7 0.7 3.46-2.1-0.9-1.2 97-98 xopex2 3.99 4.3-0.9 5.2 3.68-8.8 0.2-9.0 xopex1 3.77 6.1-0.1 6.2 3.42-6.7 0.4-7.1 OPex2 3.97 4.9-1.1 6.0 3.80-5.0 0.1-5.1 OPex1 4.26 7.9-0.6 8.5 4.23-7.2 0.6-7.8 (4636) OP 4.69-1.6-0.6-1.0 4.24-10.6-1.0-9.6 00-01 xopex2 4.47 6.2 2.1 4.1 4.03-3.1-0.6-2.5 xopex1 4.10 4.0 1.2 2.8 3.97-2.8-0.1-2.7 OPex2 4.27 5.4 0.9 4.5 3.77-2.1-0.4-1.7 OPex1 4.34 7.7 0.7 7.0 3.96-2.3-0.1-2.2 23 o 26 Chemicals and fuel (4352) OP 4.06-1.2 0.9-2.1 3.92-1.4-0.1-1.3 97-98 xopex2 3.71-9.4-1.0-8.4 3.41-12.0 1.3-13.3 xopex1 3.80-5.6-1.0-4.6 3.31-14.0 0.8-14.8 OPex2 4.20-5.4-1.3-4.1 4.02-7.1 0.7-7.8 Opex1 3.81-7.9-2.1-5.8 3.75-6.0 1.0-7.0 OP 4.55-4.7-0.1-4.6 4.16-18.0 1.3-19.3 00-01 xopex2 3.82 6.8 0.8 6.0 3.53-2.6-0.5-2.1 xopex1 4.00 4.4 0.4 4.0 3.66-1.5-0.3-1.2 OPex2 4.29 3.0 0.1 2.9 4.17-2.6-0.7-1.9 Opex1 4.22 3.9 0.1 3.8 3.93-2.9-0.9-2.0 OP 4.58 2.4 0.0 2.4 3.60 1.6 1.1 0.5 27
Table 4 coninued (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) 27 and 28 97-98 xopex2 3.20-2.6-2.3-0.3 2.74-1.9-2.3 0.4 Basic and xopex1 3.55-3.4-2.6-0.8 3.19-2.0-2.8 0.8 fabricaed Opex2 3.43-4.0-3.6-0.4 3.34-1.5-2.2 0.7 meals Opex1 3.68-4.2-3.4-0.8 3.55-1.7-2,1 0.4 (2985) OP 1.67-4.1-6.6 2.5 1.62-1.6-1.8 0.2 00-01 xopex2 3.30 5.8 1.1 4.7 2.62 1.3-1.3 2.6 xopex1 3.62 4.9 1.2 3.7 3.00 2.4-1.5 3.9 Opex2 3.30 8.0 0.8 7.2 3.21 1.1-0.9 2.0 Opex1 3.52 9.3 0.6 8.7 3.39 1.4-0.8 2.2 29 Nonelecrical machinery OP 1.71-5.4 3.3-8.7 1.57 1.0-1.9 2.9 97-98 xopex2 3,81 4.4-0.3 4.7 2.77 5.2 0.6 4.6 xopex1 3.68 2.0-0.5 2.5 3.27 4.5 0.8 3.7 Opex2 3.13 4.0-1.2 5.2 3.01 5.2 0.3 4.8 Opex1 3.67 3.5-0.1 3.6 3.51 4.7 0.4 4.3 (1036) OP 2.42 9.8 1.3 8.5 1.65 9.6 1.5 8.1 00-01 xopex2 3,68 13.8 4.6 9.2 2.83-4.5 2.3-6.8 xopex1 3.58 7.3 3.3 4.0 3.37-3.2 1.8-5.0 Opex2 2.98 14.4 4.1 10.3 2.79-2.9 0.6-3.5 Opex1 3.59 9.0 3.4 5.6 3.52-3.5 1.7-5.2 30 o 32 Elecrical machinery (3054) OP 2.49 12.6 4.6 8.0 1.68-13.6 3.7-17.3 xopex2 3.86 8.8 2.2 6.6 3.21-10.7 2.3-13.0 xopex1 3.40 3.8 0.9 2.9 2.95-10.1 2.8-12.9 Opex2 3.67 10.3 1.7 8.6 2.80-8.9 2.7-11.6 Opex1 3.25 8.2 1.0 7.2 3.07-8.8 2.6-11.4 OP 3.11 7.6-0.2 7.8 2.97-8.1 4.5-12.6 xopex2 3.66-2.4 4.5-6.9 3.08 7.5 1.9 5.6 xopex1 3.76-3.4 4.9-8.3 3.22 6.2 0.4 5.8 Opex2 3.07-4.7 5.9-10.6 2.88 8.4 1.9 6.5 Opex1 3.56-6.0 4.8-10.8 3.33 9.7 1.8 7.9 OP 3.09-9.2 4.7-13.9 2.79 13.9 1.9 12.0 28
Table 4 coninued (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) 33 xopex2 3.95-1.4 0.2-1.6 3.69-1.6 0.6-2.2 Precision xopex1 4.56-1.4 0.4-1.8 3.87-1.7 0.7-2.4 insrumens Opex2 3.97-1.7 0.4-2.1 3.20-1.9 0.0-1.9 (1155) Opex1 3.88-1.5 0.3-1.8 3.70-1.4 0.7-2.1 34 and 35 Transporaion equipmen (976) 36 Furniure and oher manufacuring OP 2.87-1.9 0.3-2.2 2.70-2.1 0.0-2.1 xopex2 3.88 3.9 1.5 2.4 3.61 4.5 3.8 0.7 xopex1 4.47 3.6 1.5 2.1 3.91 3.8 3.4 0.4 Opex2 3.93 4.8 2.0 2.8 3.51 4.7 3.6 1.1 Opex1 3.81 4.0 1.6 2.4 3.52 4.4 3.5 0.9 OP 2.83 5.0 2.2 2.8 2.67 4.8 4.0 0.8 xopex2 4.00 3.7-2.8 6.5 2.16 4.6 5.6-1.0 xopex1 3.88 4.3-3.3 7.6 2.61 2.7 4.6-1.9 OPex2 4.01 5.4-2.9 8.3 2.59 3.7 3.8-0.1 OPex1 3.99 9.4-2.1 11.5 3.31 5.8 4.0 1.8 OP 3.02 13.6-3.0 16.6 2.47 6.8 3.3 3.5 xopex2 4.26 4.6 2.0 2.6 2.91-1.4 7.9-9.3 xopex1 4.24 4.8 2.3 2.5 3.78-1.4 6.0-7.6 OPex2 4.19 2.0 0.6 1.4 3.33-1.2 6.5-7.8 OPex1 4.17 4.3 1.0 3.3 3.66-0.9 4.8-5.7 OP 3.24 4.9 0.8 4.1 2.96 2.3 6.5-4.2 97-98 xopex2 3.24-1.5-0.5-1.0 2.75 0.6-1.0 1.6 xopex1 2.64-1.8-0.1-1.7 2.44 1.2-0.9 2.1 OPex2 2.79-1.3-0.3-1.0 2.67 1.4-0.7 2.1 OPex1 2.82-1.5-0.4-1.1 2.36 1.3-1.3 2.6 (2737) OP 2.76-1.2-0.8-0.4 2.45 0.6-1.6 2.2 00-01 xopex2 3.33 2.9 0.4 2.5 2.85 6.9 3.7 3.2 xopex1 2.73 5.5 1.1 4.4 2.56 6.0 3.0 3.3 OPex2 2.68 4.0 0.7 3.3 2.56 6.4 4.9 1.5 OPex1 2.60 3.9 0.8 3.1 2.55 9.3 5.4 3.9 OP 2.84 0.6-0.4 1.0 2.66 5.8 3.9 1.9 29
Table 4 coninued (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Toal 97-98 xopex2 3.83 2.6-0.6 3.2 3.14-7.4-0.3-7.1 manufacuring xopex1 3.88 2.6-0.3 2.9 3.51-6.0-0.4-5.6 (24338) OPex2 3.58 4.9-0.6 5.5 3.30-3.7-0.1-3.6 OPex1 3.62 5.0-0.3 5.3 3.40-3.4-0.1-3.3 OP 3.87-1.2-0.5-0.7 3.28-9.2-0.2-9.0 00-01 xopex2 3.66 5.0 1.8 3.2 3.28 1.0 3.9-4.9 xopex1 3.81 6.0 1.5 4.5 3.37 1.3 2.6-1.3 OPex2 3.76 3.5 1.3 2.2 3.45 1.0 3.5-2.5 OPex1 3.98 3.7 1.3 2.4 3.56 0.4 3.1-2.7 OP 3.89 1.1 1.6-0.5 3.20 3.2 1.0 2.2 Noe: xopex2 denoes a produciviy measure of produciviy calculaed separaely for exporer and non-exporer sub-samples and modelling omega as 2 nd order endogenous (insrumening expor variable wih prediced value) Marov process. Analogously, xopex1 is a produciviy measure where 1 s order endogenous Marov process is modelled; OPex2 is a measure where 2 nd order endogenous Marov process is modelled and he esimaion is done for he pooled sample of exporers and non-exporers; OPex1 is a measure where 1 s order endogenous Marov process is modelled and he pooled sample is esimaed; OP is he benchmar sandard Olley-Paes esimaor where no expor-saus informaion is used. Resuls repored for oal manufacuring are no weighed by indusry. 30