Cross Country Differences in Productivity: The Role of Allocative Efficiency

Transcription

1 Cross Counry Differences in Produciviy: The Role of Allocaive Efficiency By Eric Barelsman, John Haliwanger and Sefano Scarpea 1 Draf: November 27, 2006 Absrac There is a growing body of evidence suggesing ha healhy, marke economies exhibi boh saic and dynamic allocaive efficiency, whereby more producive businesses have a larger marke share and reallocaion of oupus and inpus wihin secors shifs resources from he less o more producive businesses (e.g. (Barelsman, Haliwanger, and Scarpea 2004a);(Foser, Haliwanger, and Krizan 2001)). By conras, in a fricionless economy here are no welfare or produciviy gains o be obained from resource reallocaion (Levinsohn and Perin 2006). In his paper, we consider models ha feaure ase and echnology such ha a marke wih free enry susains firms operaing wihin a wide range of produciviy (e.g, (Resuccia and Rogerson 2004), and (Hsieh and Klenow 2006)). We hen invesigae o wha exen disorions in signals o decision makers generae model oucomes ha rack (may help explain?) paerns of resource allocaion and produciviy observed across counries, secors and over ime. 1. Respecively: Vrije Universiei Amserdam and Tinbergen Insiue; Universiy of Maryland, U.S. Census Bureau, and NBER; he World Bank and IZA, Bonn. We are graeful o he World Bank for financial suppor of his projec and o Karin Bouwmeeser, Helena Schweiger, and Vicor Sulla for excellen research assisance. The views expressed in his paper are hose of he auhors and should no be held o represen hose of he World Bank or is counries.

2 Inroducion There is a growing body of evidence suggesing ha healhy, marke economies exhibi boh saic and dynamic allocaive efficiency, whereby more producive businesses have a larger marke share and reallocaion of oupus and inpus wihin secors shifs resources from he less o more producive businesses (e.g.(barelsman, Haliwanger, and Scarpea 2004a);(Foser, Haliwanger, and Krizan 2001)). In an accouning sense, he conribuion of reallocaion o overall produciviy is sizeable, mainly because of he large differences in produciviy performance across firms even in narrowly defined indusries.. Empirical evidence suggess, for example, ha he difference in oal facor produciviy beween he 90 h and he 10 h perceniles wihin narrowly defined indusries is abou 99 log poins in he Unied Saes, and for labor produciviy is abou 140 log poins (see, e.g., (Syverson 2004a). This wide dispersion in produciviy provides considerable scope for reallocaion in promoing higher aggregae produciviy and growh. The evidence of large dispersion in firm produciviy and, in urn, of a sizeable conribuion of allocaive efficiency in marke economies raises a variey of quesions. The coexisence in narrowly-defined indusries of firms wih very differen produciviy performance even in healhy, marke economies is consisen wih he idea of subsanial fricions prevening resources from being immediaely allocaed o he highes valued use. Such fricions migh arise from many facors. I may be ha he fricions represen he coss of adjusmen eiher in he form of enry and exi coss, or adjusmen coss in reallocaing facors of producion such as capial and labor (World Bank, Doing Business, 2006). The laer migh involve a range of coss including he search and maching fricions ha have been he focus of much of he recen lieraure on sudying he dynamics of he labor marke (see e.g. (Davis, Haliwanger, and Schuh 1996);(Resuccia and Rogerson 2004); (Hsieh and Klenow 2000; Hsieh and Klenow 2006)). There may be saic fricions ha imply persisen produciviy differences across producers in seemingly narrowly defined secors. Produc differeniaion wihin secors due o local markes is one such facor. For example, (Syverson 2004a) explored he role of local markes for he U.S. concree indusry and found ha variaion in he marke densiy across local markes significanly accouns for variaion in he dispersion of produciviy and prices across hese markes. Saic and dynamic fricions parly depend on marke characerisics and echnological facors. However, fricions can also be driven by policy and insiuions ha affec marke compeiion and adjusmen coss. While hese issues are hardly resolved for advanced economies, hey assume a larger role in emerging and ransiion economies. These counries end o have insiuional facors ha creae addiional disorions o he allocaive process. There are many facors ha may underlie such addiional disorions. Implici or explici subsidies o favored incumbens migh be one source of disorion. Excessive regulaions ha make he cos of doing business inefficienly high (e.g., sar up coss) is anoher facor. The infrasrucure for doing business (e.g., communicaions and ransporaion infrasrucure) may weaken an efficien allocaion process. Moreover, graf and corrupion ha yields uneven

3 applicaion of rules and regulaions also huring allocaive efficiency. All of hese facors worsen differen aspecs of he invesmen climae and affec he performance of privae firms. In his paper, we emphasize heir effecs of dispersion in business climae condiions across producers wihin each counry, since we are especially ineresed in he role of allocaive efficiency. The basic working hypohesis is ha fricions play an imporan role in he allocaive efficiency and, hrough his channel, o produciviy and oupu growh. While some of hese fricions are inheren even in well-developed marke economies, ohers are affeced by policy and insiuional seings and hus could be ackled by appropriae reforms. The evidence of large dispersion in allocaive efficiency even wihin OECD counries provides suppor o his claim. This is no a new hypohesis, and ohers have explored is implicaions in a variey of heoreical and empirical conribuions (see e.g.,(banerjee and Duflo 2004);(Resuccia and Rogerson 2004); and (Hsieh and Klenow 2006)). Our conribuion is o quanify he magniude of allocaive efficiency in a sample of indusrial, and emerging economies drawing from a harmonized, cross counry daabase wih key measures and momens from firm-level daa. Using a simple decomposiion of indusry-level produciviy ((Olley and Pakes 1996)), we find evidence of considerable cross counry variaion in allocaive efficiency across counries. For example, we find ha in he Unied Saes shifing from he exising firm allocaion of marke shares o one where shares are allocaed randomly would yield a 60 percen loss in labor produciviy. Moreover, we find significan changes in allocaive efficiency over ime in he Easern European counries during heir ransiion o a marke economy: marke oriened reforms in hese counries have led o an increase in allocaive efficiency of abou log poins. However, hese empirical decomposiions are only suggesive and require some more heoreical underpinning. Accordingly, we develop a simple model ha draws heavily on he models of (Resuccia and Rogerson 2004); and (Hsieh and Klenow 2006). We calibrae his model o mach he momens from our harmonized daa and in paricular he momens from he Olley-Pakes (hereafer OP) empirical decomposiion. We show ha, consisen wih our working hypohesis, an increase cross-firm dispersion of disorions yields a decline in allocaive efficiency as measured by he OP decomposiion. We push his finding furher by quanifying he magniude of disorions ha are needed o accoun for he variaion in he observed OP measures, boh beween and wihin counries. By exploring he heoreical underpinnings of he empirical decomposiions ha have appeared in he lieraure, we also discuss how disorions can affec differen margins of he reallocaion process. Consisen wih he inuiive explanaions of he Olley-Pakes decomposiions, we find ha disorions end o reduce he OP cross erm ha is o say he covariance of marke share and produciviy of a given se of marke paricipans. However, here are addiional margins impaced by disorions ha are, poenially, a leas as imporan. For one, disorions may impac no only he covariance for a given se of firms bu also he selecion process of marke paricipans. The impac ha marke disorions can have on selecion is nonrivial and has a variey of relaed

4 implicaions. Firs, he impac of disorions on selecion ends o lower he average unweighed produciviy of marke paricipans. Second, in a model where poenial marke paricipans mus pay an enry fee before learning heir produciviy, disorions will affec he number of new firms ha pay enry fees. Bu disorions can also affec he mix of firms and he scale of aciviy; for example he average firm size and he capiallabor raio. All of hese effecs can have adverse impacs on consumpion (or more generally welfare). The implicaion of hese findings is ha while measured allocaive efficiency in erms of he OP cross erm is a useful indicaor i is by far no he only one for quanifying he impac of disorions on marke economies. The paper proceeds as follows. Secion 2 describes he harmonized firm level daabase. Secion 3 presens some basic facs abou produciviy dispersion wihin he counries in our daabase as well as he paerns of allocaive efficiency as measured by he OP decomposiion. Secion 4 develops he model of allocaive efficiency wih disorions in he allocaive process. Secion 5 calibraes he model numerically o explore he implicaions of he model in ligh of he empirical paerns in secion 3. Secion 6 presens concluding remarks. 2. The harmonized firm-level daabase and indicaors of dispersion To assess he degree of firm heerogeneiy, he magniude and characerisics of labor reallocaion and, ulimaely, heir impac on secoral and aggregae produciviy, we use a harmonized firm-level daabase ha covers 24 indusrial and emerging economies. 2 These daa have been colleced from micro daa from business regisers, census and enerprise surveys playing paricular aenion a harmonizing, o he exen possible, of key conceps (e.g. enry, exi, or he definiion of he uni of measuremen) as well as a using common mehods o compue he indicaors. A deailed echnical descripion of he daase may be found in (Barelsman, Haliwanger, and Scarpea 2004b). 3 The daabase conains informaion on firm demographics, such as enry and exi, jobs flows, size disribuion and firm survival, as well as on produciviy disribuions and correlaes of produciviy. In paricular, informaion was colleced on he disribuion of labour and/or oal facor produciviy by indusry and year, and on he decomposiion of produciviy growh ino wihin-firm and reallocaion componens. Furher, informaion is provided on he averages of firm-level variables by produciviy quarile, indusry, and 2 3 The daase includes 10 OECD counries: (Canada, Denmark, Germany, Finland, France, Ialy, he Neherlands, Porugal, Unied Kingdom and Unied Saes) and 14 ransiion and emerging economies (Esonia, Hungary, Lavia, Romania, Slovenia; Argenina, Brazil, Chile, Colombia, Mexico, Venezuela, Indonesia, Souh Korea and Taiwan.(China)). However, differen pars of he analysis uses a smaller sub-se of counries do o daa availabiliy. The analysis of firm demographics is based on business regisers, census, social securiy daabases, or employmen-based regiser conaining informaion on boh esablishmens and firms. Daa for he analysis of produciviy growh come more frequenly from business surveys.

5 year. The classificaion ino abou 40 secors (roughly he 2-digi level deail of ISIC Rev3) coincides wih he OECD Srucural Analysis (STAN) daabase. 4 For he purposes of his paper, we consider wo indicaors of dispersion ha are paricularly relevan. Firs, we look a he disribuion of firm by size across indusries and counries. The observed wide, and skewed owards small unis, disribuion of firm size shows ample room for reallocaion of labor. For such reallocaion of labor be influence aggregae produciviy -- even if only in an accouning sense -- here mus be differences in produciviy levels across firms. The daa indeed allow looking a his second dimension of firm dispersion. Heerogeneiy in firm size among incumbens: Table 1 presens he raio of he average size of he fourh o he firs quarile of he disribuion of firms by size in he oal economy and he manufacuring secor. The able suggess a wide dispersion in firm size in all counries for which daa are available. Moreover, in mos counries he dispersion is larger in he manufacuring secor han in he oal business secor and, wihin manufacuring in high-ech indusries compared wih secor average. I could be sressed, however, ha he cross-counry comparison of firm size dispersion may be influenced by he overall dimension of he inernal marke especially for non-radeables and by he differen secoral composiion of he economy. The daa indeed sugges wider dispersion in firm size in some of he large economies e.g. he Unied Saes bu also in some of he ransiion economies in Easern Europe, where policy-induced disorions have allowed he survival of very large (formerly or sill) sae-owned firms ogeher wih many new smaller privae unis. In Annex Table 1, we presen he indicaors of firm size dispersion afer conrolling for indusry composiion. Wihin 2-3 digi manufacuring indusries, he iner-quarile raio of average firm size is considerable in all indusries and counries. Dispersion in labor produciviy and MFP: There are also wide differences in firms produciviy performance wihin manufacuring indusries. Table 2 presens he difference beween he fourh and firs quarile of he log labor produciviy. The differences are large: in beween log poins in mos counries. Par of he significan cross-counry differences is due o he difference secoral composiion (see column 2 in he Table), bu even conrolling for ha, he gap in labor produciviy beween he mos and he leas producive firms is wide. And, as in he case of he dispersion in firm size, he high ech secors end o be characerized by an even higher dispersion in all counries, confirming ha where here is more room for innovaion and marke experimenaion here is more heerogeiy in firm characerisics and performance. To confirm his, Table 3 and Table 4 presen he sandard deviaion in labor and mulifacor produciviy respecively for he differen manufacuring indusries. Moreover, here is a wider dispersion in labor han in MFP for all he counries for which we have daa (Figure 1) 4. See

6 3. Empirical Measures of Allocaive efficiency How does his observed heerogeneiy in firm characerisics and performance affec aggregae produciviy? To address his quesion, we look a measures of allocaive efficiency by adoping an empirical procedure o decompose he level of produciviy wihin indusries originally proposed by (Olley and Pakes 1996). They noe ha in he cross secion, he level of produciviy for a secor a a poin in ime can be decomposed as follows: P P = (1/ N ) P + ( θ θ )( P ) i i i i i where P is secoral produciviy, P i is firm-level produciviy, θi is he share of aciviy for he firm, N is he number of businesses in he secor and a bar over a variable represens he unweighed indusry average of he firm-level measure. The simple inerpreaion of his decomposiion is ha aggregae produciviy can be decomposed ino wo erms involving he un-weighed average of firm-level produciviy plus a cross erm ha reflecs he cross-secional efficiency of he allocaion of aciviy. The cross erm capures allocaive efficiency since i reflecs he exen o which firms wih higher han average produciviy have a greaer marke share. This decomposiion is easy o implemen as i involves measures of he un-weighed average produciviy and of he weighed average produciviy. Measuremen problems make comparisons of he levels of eiher of hese measures across secors or counries poenially problemaic, bu aking he difference beween hese wo measures reflecs a form of a difference-in-difference approach. As such, in principle he OP cross erm is comparable across counries since any measuremen problem ha affecs produciviy levels is differenced ou by he indicaor. In mos of he analysis in his paper, we use log labor produciviy a he micro level as our measure of P i, and he firm s labor share in he indusry as our measure of θ i. We focus on labor produciviy because i is more readily available (and likely more accuraely measured) in our sample of counries. A number of sudies (e.g., (Eslava e al. 2004), and (Foser, Haliwanger, and Krizan 2002)) have shown ha he paerns of he OP decomposiion wihin a counry are similar for labor produciviy and oal facor produciviy when using he same marke shares. In he numerically simulaed model discussed in he nex secions, we also use labor produciviy o derive inferences on he impac of disorions on he paerns of allocaive efficiency. Figure 2 shows he resuls of applying he OP decomposiion a he indusry level and hen aking he weighed average resuls by indusry for he counries in he harmonized daabase. We focus our aenion on he cross erm -- or as we denoe allocaive efficiency erm. We find ha for virually all counries he OP cross erm is posiive, suggesing a posiive covariaion beween marke share and is produciviy a he micro level. For example, we find ha allocaive efficiency is slighly less han 0.50 in he U.S.. Since produciviy is measured in logs, his implies ha, wihin he average U.S. manufacuring, labor produciviy would be abou 50 logs poin smaller if labor were

7 allocaed randomly. The inernaional comparison also suggess ha he OP cross erm is subsanially higher in he U.S. han in mos European counries. By conras, here is evidence of high if no higher allocaive efficiency in some Eas Asian economies.. Lain American economies have lower cross erms han he U.S., bu higher han mos European economies and he ransiion economies have he lowes cross erms. While he OP cross erm avoids he sandard problems of cross counry comparisons of produciviy, i is no immune from measuremen problems. In paricular, measuremen in he second momen of produciviy wihin an indusry ha sysemaically varies by counry because he firm level daa is sysemaically noisier in one counry han anoher will impac he OP cross erm. For example, classical measuremen error will reduce he OP cross erm since i will mimic a more random allocaion of marke shares wih respec o produciviy. This consideraion suggess some cauion in assessing he observed cross-counry paerns of allocaive efficiency presened in Figure 1.. To ackle his issue, we also presen he evoluion of he OP cross erm over ime wihin a counry. To he exen measuremen error wihin a counry is relaively sable over ime, he wihin counry variaion over ime in he OP cross erm will difference ou he counry-specific second momen measuremen error. The ransiion economies offer a suiable framework for assessing he poenial links beween disorions and allocaive efficiency. Over he period observed by he available daa (he 1990s), hese counries underook sysemic reforms in heir ransiions from cenral-planning o a marke economy. Arguably, many disorions affeced he differen margins of resource allocaion, from barriers o enry, o disored allocaion of resources across firms, secors and geographical areas. These disorions were gradually reduced if no eliminaed in he ransiion o a marke economy. Figure 3 shows he wihin counry variaion over ime for he ransiion economies of he OP cross erm. Ineresingly, excep for Esonia which sars wih a relaively high cross erm, he OP cross erm increases in he ransiion economies and, in many cases, subsanially. For example, in boh Hungary and Slovenia, he OP cross erm increases by abou 20 log poins during he ransion. Anoher advanage of examining he OP cross erm over ime wihin counries is ha he variaion in he esimaed allocaive efficiency can be pu in he conex of he overall paerns of produciviy growh. In oher words, changes in allocaive efficiency may ake place in he conex of increasing or decreasing labor produciviy. Figure 4 shows he paerns of overall (indusry-level) produciviy, he un-weighed average produciviy and he OP cross erm for Hungary and Slovenia. Pu in his conex, he OP cross erm has increased subsanially bu he overall and un-weighed produciviy erm have increased as well, and a a even faser pace in he case of Slovenia. Thus, i would be misleading o draw inferences from Figures 2 and 3 ha he OP cross erm accouns for mos of he cross secional or ime series variaion in produciviy beween and wihin counries. The main poins from Figures 2-4 are ha he OP cross erm is large, i varies subsanially across counries (wih associaed concerns abou measuremen error) and wihin ransiion economies a leas i has increased subsanially over ime. In

8 comparing hese findings o he exising lieraure, i is worh noing ha Olley and Pakes (1996) found a posiive cross erm using TFP as he measure of produciviy in he U.S. elecommunicaions indusry and ha he cross erm increased subsanially following deregulaion in he U.S. elecommunicaions indusry. Eslava e. al. (2006) found (also using TFP) ha he OP cross erm rose subsanially wihin 3-digi Colombian indusries in he 1990s a period of subsanial marke reforms in Colombia. These findings in he lieraure along wih Figures 2-4 srongly sugges ha he OP cross erm is capuring allocaive efficiency. However, since i is a simple empirical decomposiion wihou srucure, his evidence is only suggesive. To help pu srucure on he resuls in Figures 2-4, we urn o a model of allocaive efficiency and disorions in he nex secion. 4. A Simple Model of Allocaive Efficiency To guide our analysis of disorions and allocaive efficency we develop a simple model drawing heavily from (Resuccia and Rogerson 2004) and (Hsieh and Klenow 2006). Key feaures of he model are diminishing reurns and heerogeneous producion unis (as in (Hopenhayn 1992) and (Hopenhayn and Rogerson 1993)) ha face idiosyncraic disorions ((Resuccia and Rogerson 2004)). Saring wih he behaviour of firms, we assume ha firms produce according a producion funcion given by: γ α (1) Y = A( n f) k α, γ < 1 i i i i where Yi is oupu for firm in period, Ai is he firm specific ime invarian produciviy for firm i, ki is he amoun of capial inpu of firm i a ime, ni is he employmen, and f is overhead labor. Decreasing reurns is assumed due o some unobserved fixed facor in a manner similar o (Lucas 1978) where like he laer paper i is useful o hink abou his facor as managerial abiliy. 5 The decreasing reurns insures (as in (Lucas 1978)) ha he mos producive firm/manager does no ake over he marke. The overhead labor implies ha he disribuion of labor produciviy is no degenerae even in an economy wihou disorions (i.e., while he marginal produc of labor will be se equal o he wage rae, he average produc of labor will vary wih scale given overhead labor). Firms maximize profis, wihin an environmen wih disorions o capial expendiures and nominal oupu, in each period given by: γ α α (2) π = A (1 τ )( n f) k wn (1 κ ) rk i i i i i i i i 5 In order o ge dispersion, we require some curvaure in profis. Here we assume decreasing reurns bu a reasonable alernaive is o assume some degree of marke power in, e.g., a differeniaed produc environmen. Hsieh and Klenow (2006) assume he laer. In fuure drafs we will add a differeniaed produc srucure o explore he role of such a srucure for our resuls.

9 whereτ i is he firm specific and ime invarian disorion o revenue for firm i, κ i is he firm specific disorion o capial allocaion, is he wage paid o homogenous workers, and r R w = + δ, is he user cos of capial which equals he ineres rae plus he rae of depreciaion. In considering hese disorions, τ i can be inerpreed broadly o include any disorion ha impacs he scale of a business, while κi represens any disorion ha impacs he facor mix of a business. In wha follows, we call hese disorions a "scale disorion" and a "facor mix disorion" respecively. To make he model and analysis racable, we assume a simple ex ane and ex pos iming of informaion and decisions in any given period. Ex ane, before a firm eners, we assume ha firms do no know heir producion and disorion draws bu hey know he disribuion of hese idiosyncraic variables. There is a fixed cos of enry, given by, ha firms mus pay o ener and o learn heir draws from he join ex ane disribuion c e of produciviy and disorions, G ( A, τ, κ). Once a firm learns heir draws of Ai, τ i, and κ i, heir values remain consan. Firms discoun he fuure a rae β = (1/(1 + R)) and face an exogenous probabiliy of exiing in each period given by λ. Given free enry and he assumpions abou he arrival of informaion, firms ener up o he poin where he expeced discouned value of profis is jus equal o he enry fee. Moreover, given ha he draws are ime invarian and in he seady sae, he presen discouned value for an incumben firm i ex pos is given simply by: (3) W( A, τ, κ ) = π( A, τ, κ )/(1 ρ) where i i i i i i ρ = (1 λ) /(1 + R) In urn, he free enry condiion is given by: (4) e W = max(0, W( A, τκ, )) dg( A, τκ, ) c e = 0 A, τκ, Firms wih a low produciviy and/or a high scale (or facor mix) disorion draw will exi upon learning heir draws if hey canno cover heir fixed operaing coss. The opimal capial/labor allocaion will depend on inpu prices and he idiosyncraic capial disorion. An operaing firm will have capial and employmen given by:

10 (5) k * i γ α ( γ α) r α(1 τ) Ai (1 + κ) αw = r(1 + κ) 1/(1 γ ) (6) * * ( γ α) r ni = f + ki (1 + κ) αw Oupu and profis for he operaing firm are given by he (1) and (2). To close he model we mus describe labor supply and he behaviour of households and workers. In his case, his is relaively sraighforward as a fixed number of households are assumed o supply labor inelasically so ha aggregae labor supply is s equal o N. Aggregae labor demand is given by: d * N = n ( A, τ, κ) dμ( A, τ, κ) A, τκ, where μ( A, τ, κ) is he ex pos join disribuion for operaing firms of produciviy and disorions. In equilibrium he number of firms and wages mus saisfy boh he free enry condiion and ha labor demand equals aggregae labor supply. e d W = 0, N = N s Aggregae consumpion plus resources spen on enry and depreciaion will equal aggregae oupu: C + E c + δ K = Y e Where K is he aggregae of capial of ex-pos operaing firms. Underlying his model is he sandard assumpion ha households maximize uiliy and given he assumpion of inelasic labor supply his is assumed o be given by for he represenaive household: =0 β U ( C ) Subjec o he budge consrain:

11 = 0 p ( C + K + 1 (1 δ ) K ) = p ( w N + r K + π ) = 0 Where p is he ime zero price of period consumpion, w and r are he period renal prices of labor and capial measured relaive o period oupu, and π is he oal profi from he operaions of all plans. A sandard resul emerges from he firs order condiions of his problem given by: R = δ = ( 1/ β ) 1 r So he real ineres rae is pinned down by he discoun facor for uiliy and he capial depreciaion rae. 5. Calibraion of Model We explore calibraions of he model in wo seps. Firs, we choose some reasonable parameers for he model and explore is numerical properies. This analysis helps us o undersand he ineracions in he model and he role of disorions. Second, we push his calibraion harder by aemping o mach key momens from he micro daa as discussed in Secion 3. For our iniial numerical exploraions, we selec: γ = 0.9, α =0.1, λ =.10, his is consisen wih evidence f exi raes in he Unied Saes and oher OECD counries (Barelsman e al. 2004) R=.03, and δ =.12, consisen wih long run real ineres raes in OECD counries and ypical depreciaion raes from naional accouns. f=0.2,log(c e )=11.92, Mean(log(A)) = Sd Dev (log(a)) = 0.34 The laer are momens associaed wih he ex ane disribuion G(A,τ,κ). These parameers are roughly consisen wih he micro evidence for he U.S. For purposes of calibraion and esimaion, we sar wih he U.S. as a base case and hen ask wheher we can accoun for he evidence across counries as presened in Secion 3 wih varying he degree of disorions. For disorions, we consider four differen cases: (i) A no disorion case where τ = 0, and κ = 0, (ii) A random ex ane scale disorion case wih he ex ane mean(τ)=0 and corr(a, τ)=0 (iii) A random ex ane facor mix disorion case wih he ex-ane mean(κ)=0 and corr(a, κ)=0 (iv) A correlaed ex ane scale disorion case ex ane mean(τ)=0 and corr(a, τ)>0.

12 Table 5 shows he resuls from hese simulaions. For he base case (op row) wihou disorions, he OP cross erm using labor produciviy (where weighs are employmen) and oal facor produciviy (where weighs are he composie inpu) is posiive reflecing posiive allocaive efficiency in a marke economy wihou disorions. Noe, however, ha here is a much smaller dispersion of LP relaive o TFP. Even wih overhead labor, here is a endency for he average labor produciviy o be equalized. In our scenario wih disorions ha have ex ane zero mean, or wih zero mean disorions ha are uncorrelaed wih produciviy, we obain ha, hrough he selecion process, here is a non random, non mean zero ex pos disorions. This makes sense as firms wih low τ draws are more likely o survive and hose wih high τ draws are less likely o survive. Given his paern, he average surviving firm acually faces a negaive disorion (he equivalen o a posiive subsidy) and he ex pos correlaion beween disorions and produciviy is posiive (his is because one needs o be high produciviy o survive wih a high disorion). 6 The adverse consequences of hese disorions can be seen on several margins. For he uncorrelaed scale disorion case we see lower average labor produciviy and average TFP and lower OP cross erm for boh labor produciviy and TFP. We also see much lower survival which implies oo much churning and paying of enry coss relaive o he non-disored economy. All of hese facors conribue o lower consumpion. For he uncorrelaed facor mix disorion case, we see relaively lile impac on OP erms and unweighed produciviy. Indeed unweighed labor produciviy increases. Insead wha we see is ha capial labor raio is far oo high relaive o non-disored economy. In his case, he economy is induced o hold oo much capial given implied equilibrium facor price raio. Survival is oo low relaive o he non-disored economy. In combinaion, he high use of capial and he low survival conribue o very low consumpion. Noe as suggesed above ha he facor mix disorion case should be more inerpreed as a capial-labor disorion. Thus, economies wih disorions o he capiallabor raio in favor of capial will have labor produciviy ha is disoredly oo high given high capial-labor raio. For he correlaed scale disorion case, we see large impacs on unweighed TFP and OP cross erm for TFP. We also see large impacs on unweighed labor produciviy and OP cross erm. In his case, boh who says in and allocaion of hose ha are paricipans is disored subsanially. Ineresingly, overall survival is roughly he same as in he non-disored economy bu he mix of who survives is highly disored. Given ha survival raes are abou he same as in he nondisored economy, he disorion shows up clearly in OP cross erms for boh TFP and labor produciviy. This makes sense as OP cross erm capures misallocaion for a given 6 One open issue boh concepually and empirically is ha we permi disorions o be boh posiive and negaive. A business wih a negaive disorion effecively has a subsidy for aciviy on some margin. A relaed issue is ha we, as in Hsieh and Klenow (2006) view hese disorions as disorions no axes while Resuccia and Rogerson (2003) model hese disorions as axes. The laer is relevan for he welfare/consumpion impac as Resuccia and Rogerson (2003) make a lump sum ransfer ha could be posiive or negaive depending on he ax revenue.

13 se of paricipans. In his correlaed case, all of hese facors (i.e., impac on survival, misallocaion of who survives, and misallocaion of aciviy amongs survivors) conribue o lower consumpion. To provide furher perspecive on hese simulaions, Figures 5-8 presen scaerplos of he relaionship beween labor produciviy and employmen for he alernaive simulaed disribuions of businesses. In Figure 5, we presen he nodisorion case and observe a srong posiive and sysemaic relaionship beween labor produciviy and size. Underlying his paern is ha firms wih higher TFP hire more workers and exhibi higher labor produciviy a a decreasing rae given he decreasing reurns and overhead labor. Figure 6 shows he paern for he uncorrelaed scale disorion case. We observe a much less sysemaic relaionship beween labor produciviy levels and employmen. Moreover, he range of labor produciviy and size is much larger han in Figure 5, so we see more dispersion in boh labor produciviy and size. Figure 7 shows he paern for he uncorrelaed facor mix case. Ineresingly in his case here is a srong posiive relaionship beween labor produciviy and size as in he non-disored case. However, he problem here is no so much his covariance bu raher he disoredly high capial-labor raio. Figure 8 shows he correlaed scale disorion case. Here he misallocaion for survivors is apparen as well as he greaer dispersion in labor produciviy and size relaive o he non-disored economy. These simulaions are suggesive bu sill here is sill a gap beween key feaures of he simulaions and momens from he acual daa. For one, while he simulaions produce subsanial and sysemaic variaion in OP cross erms due o disorions, he variaion in he above simulaions in he OP cross erms is subsanially lower han observed in he acual daa. Recall ha in counries like U.S. he OP cross erm is almos 0.6 while in some ransiion economies i is negaive. A number of facors may be a work here. Under he parameerizaions under consideraion, ex ane dispersion in TFP is subsanial (and similar o ha observed in say U.S.) bu ex pos dispersion in TFP in he non-disored economy is much lower given selecion. Geing a model o have a large dispersion in TFP ex pos requires more fricions. Candidae fricions here include: (i) produc differeniaion as in (Hsieh and Klenow 2006); (ii) learning so ha some of he dispersion reflecs young businesses ha have no ye learned hey have low produciviy; (iii) adjusmen coss. Noe ha all of hese facors can help conribue o dispersion in TFP and labor produciviy. We hink all of hese facors should be considered in his conex alhough we noe ha consideraion of any one of hem raises addiional issues. For example, wih differeniaed producs i is imporan o disinguish beween measured TFP and physical TFP since ypical measured TFP will include price variaion (e.g., (Syverson 2004b),and Foser, Haliwanger and Syverson (2006)). In considering hese issues and in inerpreing he resuls in he daa vs. he simulaions, i is imporan o emphasize ha dispersion in TFP and labor produciviy is criical for OP cross erms for boh disored and non-disored economies. For non-disored economies, dispersion is criical o obain posiive and large OP cross erms. Mechanically, he OP cross erm can only be large and posiive if here are large gaps in produciviy beween he large and small marke share plans. For disored

14 economies, i would seem o be relaively sraighforward o reduce he OP cross erm wih sufficien disorions. Tha is, one migh argue ha OP cross erm would seemingly be driven o zero in an environmen where disorions jus creae noise in allocaion. The fallacy in his argumen is ha disorions impac no only allocaion bu selecion as we found in our simulaions disorions impac boh he amoun of selecion and he mix of selecion. Thus, i may be ha he OP cross erm does no fall so much as insead he amoun of mix of selecion is impaced by disorions. 6. Concluding Remarks As augh in principles of economics classes, marke economies provide incenives for an efficien allocaion of resources. Empirically, we have learned ha well-developed marke economies exhibi paerns of allocaion consisen wih shifing resources o more producive businesses, bu he ongoing allocaive process is complex and dynamic. A growing body of empirical evidence suggess subsanial dispersion in measures of boh TFP and labor produciviy even wihin narrowly defined secors. There is also evidence ha more producive businesses end o be larger han less producive ones. Moreover, he laer is an ongoing process wih resources coninually being reallocaed a a high pace away from less producive o more producive businesses. The observed dispersion of produciviy wihin narrowly defined secors and he ongoing reallocaion dynamics imply ha even in healhy marke economies here are subsanial fricions in allocaing resources o heir highes valued use. Pu simply, here is no free lunch in he efficien allocaion of resources subsanial ime and resources are spen in marke economies in he allocaion process. In counries wih sringen regulaions in goods and labor markes and in paricular in he ransiion economies of Easern Europe -- hese allocaion dynamics are arguably furher complicaed. In paricular, he presence of scale or mix disorions prevens resources from being allocaed efficienly. In his paper, we explore he associaed working conjecure ha differences in he level of produciviy across counries and over ime can be accouned for by such differences in he disribuion of disorions. We follow he recen lieraure by emphasizing ha he disorions in indusrial, bu especially in emerging and ransiion economies may have an imporan idiosyncraic componen. Based upon a novel cross counry daase wih harmonized saisics on he produciviy and size disribuions of firms wihin indusries across counries, we presen evidence of significan sizeable disorions in allocaive efficiency across counries and indusries. Using an accouning measure of allocaive efficiency, we find for example ha ha average labor produciviy in he manufacuring secor of he Unied Saes would be approximaely 50 percen lower if labor was allocaed randomly as opposed is acual allocaion. In ransiion economies, we find ha he accouning measure of allocaive efficiency was much smaller (even negaive) in he 1990s, bu has been increasing rapidly during he ransiion o a marke economy.

15 Wih hese findings as a backdrop, we build a model drawing heavily from he lieraure on he role of disorions in he allocaive process. We inroduce wo possible disorions: a scale disorion ha affecs he size he business; and a facor mix disorion ha affecs is facor inpu composiion. In calibraing he model, we show ha boh disorions can qualiaively generae he paerns of allocaive efficiency observed in our accouning decomposiions. However, we also find ha disorions impac more margins han one migh capure on he basis of he accouning decomposiions. For example, we find ha disorions impac no only he allocaion amongs survivors bu also who survives and how much churning here is as firms learn abou heir produciviy. These oher margins are very imporan in quanifying he adverse impac of disorions in he economy. We also noe ha precisely because oher margins do maer in shaping he effecs of disorions, our simulaed model canno fully mach some key aspecs of he firm dispersion observed in he acual daa. Wih our simulaed disorions alone, for example, i is a challenge o mach he wide dispersion in produciviy observed in he daa. One of he reasons is indeed ha in he acual daa we only observe he allocaion among survivors, while disorions also affec heir selecion. We speculae ha o mach he acual momens we would require models wih a richer menu of fricions han he simple model we consider in his paper.

16 References 1. Banerjee, Abhiji and Duflo, Esher. Growh Theory hrough he Lens of Developmen Economics Ref Type: Unpublished Work 2. Barelsman, Eric J., Haliwanger, John, and Scarpea, Sefano. Microeconomic evidence of creaive desrucion in indusrial and developing counries a. World Bank Policy Research Paper. Ref Type: Repor 3. Barelsman, Eric J., Haliwanger, John C., and Scarpea, Sefano. Disribued Analysis of Firm-level daa from Indusrial and Developing Counries. mimeo. 2004b. Ref Type: Unpublished Work 4. Davis, Seven J., John C. Haliwanger, and Sco Schuh Job creaion and desrucion. Cambridge: MIT Press. 5. Eslava, Marcela e al., "The Effecs of Srucural Reforms on Produciviy and Profiabiliy Enhancing Reallocaion: Evidence from Colombia," Journal of Developmen Economics 75 (2): (2004). 6. Foser, Lucia, John C. Haliwanger, and C. J. Krizan Aggregae Produciviy Growh: Lessons from Microeconomic Evidence. Chicago: Universiy of Chicago Press. 7. Foser, Lucia, John C. Haliwanger, and C. J. Krizan, "The Link Beween Aggregae and Micro Produciviy Growh: Evidence from Reail Trade," NBER Working Paper (9120.) (2002). 8. Hopenhayn, Hugo and Richard Rogerson, "Job Turnover and Policy Evaluaion: A General Equilibrium Analysis," Journal of Poliical Economy 101 (5): (1993). 9. Hopenhayn, Hugo A., "Enry, Exi, and Firm Dynamics in Long Run Equilibrium," Economerica 60 (5, pag ): (1992). 10. Hsieh, Chang-Tai and Peer J. Klenow, "Misallocaion and Manufacuring Produciviy in China and India," (2000). 11. Hsieh, Chang-Tai and Klenow, Peer J. Misallocaion and Manufacuring Produciviy in China and India Ref Type: Unpublished Work 12. Levinsohn, James and Perin, Amil. Measuring Aggregae Produciviy Growh Using Plan-level Daa Ref Type: Unpublished Work

17 13. Lucas, Rober E., "On he Size Disribuion of Business Firms," The Bell Journal of Economics 9 (2): (1978). 14. Olley, G. Seven and Ariel Pakes, "The Dynamics of Produciviy in he Telecommunicaions Equipmen Indusry," Economerica 64 (6): (1996). 15. Resuccia, Diego and Rogerson, Richard. Policy Disorions and Aggregae Produciviy wih Heerogeneous Plans Sociey for Economic Dynamics, working paper. Ref Type: Repor 16. Syverson, Chad, "Marke Srucure and Produciviy: A Concree Example," Journal of Poliical Economy 112 (6): (2004a). 17. Syverson, Chad, "Subsiuibiliy and Produc Dispersion," Review of Economics and Saisics 86 (2): (2004b).