Enterprise Restructuring in Transition：A Quantitative Survey

Enterprise Restructuring in Transition：A Quantitative Survey Enterprise Restructuring in Transition:A Quantitative Survey Article 1 8/12/8><#004699'>02 9:13 AM Page 739Journal of Economic Literature Vol. XL (September 2<#004699'>0<#004699'>02) pp. 739–792Enterprise Restructuring in Transition:A Quantitative Survey1SIMEONDJANKOVandPETERMURRELLtransition countries have been matched by1.Introductionenormous variance in the degree to whichVERTHELASTdecade, more thanenterprises have restructured their opera-O15<#004699'>0,<#004699'>0<#004699'>0<#004699'>0 large enterprises in 27 transi-tions and responded successfully to events.tion countries have encountered revolution-With changes in the institutional and policyary changes in every aspect of their politicalenvironment much faster and more encom-and economic environments. Some enter-passing than in virtually any other historicalprises have responded to the challenge, en-episode, this is as close to a policy laboratorytering world markets with great dynamismas economics gets.and becoming indistinguishable from theirThis mammoth quasi-experiment offerscompetitors in mature market economies.lessons of profound importance for eco-Many others remain mired in their past, un-nomic studies and economic policy. Sincedergoing protracted deaths, delayed at timesthe pace at which firms restructure is a fun-by their slippage into a world of barter anddamental determinant of economic growth,subsidies. Thus the revolutionary changes inanalysis of the determinants of restructuringin formerly socialist countries sheds light on1Djankov is senior economist at the World Bank.the bases of economic progress. Such analy-Murrell is professor of economics and chair of thesis addresses age-old questions and posesAcademic Council of the IRIS Center, University ofMaryland. We thank Erik Berglof, Bernard Black,new ones. What are the relative productivi-Olivier Blanchard, Morris Bornstein, Harry Broadman,ties of state and private enterprises? DoesDavid Brown, Wendy Carlin, Stijn Claessens,mass privatization work? What is the effi-Alexander Dyck, William Evans, John Earle, RomanFrydman, Cheryl Gray, Irena Grosfeld, Laszlociency cost of diffuse share ownership rela-Halpern, Judy Hellerstein, Janos Kornai, Johntive to block-holder -McMillan, John Nellis, Tatiana Nenova, Guyvate ownership? Which pri owners are most effective—managers,Pfeffermann, Yingyi Qian, Mark Schaffer, PaulSeabright, Marcelo Selowsky, Andrei Shleifer, Janworkers, banks, or investment funds? DoesSvejnar, and three anonymous referees for helpful ad-competition promote productivity change?vice. This research was made possible through supportDoes it matter whether competitive pres-provided by the World Bank and by the U.S. Agencyfor International Development under Cooperativesure comes from foreign or domestic firms?Agreement DHR-<#004699'>0<#004699'>015-A-<#004699'>0<#004699'>0–<#004699'>0<#0 04699'>031–<#004699'>0<#004699'>0 to the CenterTo what degree do soft budgets dull enter-for Institutional Reform and the Informal Sector(IRIS). The findings, interpretations, and conclusionsprise performance? Is a strengthening ofexpressed in this paper are entirely those of the au-managerial incentives sufficient to inspirethors. They do not necessarily represent the views ofturnaround, or is replacement of managersthe IRIS Center, US AID, the World Bank, its execu-tive directors, or the countries they represent.necessary for revitalization?739 Article 1 8/12/<#004699'>02 9:13 AM Page 74<#004699'>074<#004699'>0Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)Answers to these questions are obviouslystatements, although the reader might be re-of vital significance for economic delibera-assured to note that our priors to some ex-tions in general. But beyond this, the transi-tent cancel (Murrell 1992; Gerhard Pohl ettion process is important in and of itself be-al. 1997).cause of its geographical scope, the largeIn 6><#00aa00'>view of these factors, we adopt morechanges in levels of economic well-being inroutinized methods of synthesizing the evi-the last decade, and the ions fordence, drawing on insights from meta-the world economy ramificat and polity. Analysis ofanalysis, which has long been in use in otherenterprise restructuring is central in any ef-disciplines, particularly bio-medicine, psy-fort to understand the effects of the mostchology, and education (Morton Hunt3important reform measures adopted in1997).Apart from making our methods oftransition countries. With enterprise re-synthesis transparent, application of meta-structuring apparently more successful inanalysis has several other advantages. First,some countries than others, the naturalit provides tools to aggregate the results ofquestion that arises is whether such differ-many studies on a similar topic, combiningences relate systematically to policy. In thismany tests with weak power to produce apaper, we address this question by examiningsingle one with larger power. Second, thesehow the effects of policy have varied betweenmethods allow one to test hypotheses acrosstransition countries.groups of studies. For example, we examineOur objective is to survey the evidence onwhether the replacement of managers isthe determinants of enterprise restructuringmore effective than the addition of incen-2in transition.We provide such a synthesis,tives, and we test whether privatization hassummarizing the composite conclusionsstronger results in Eastern Europe than infrom more than one hundred empiricalthe former Soviet Union. Third, the synthe-studies. Where possible, we compare the re-sis of results can address the thorny issue ofsults from the transition literature with thosedifferences in the quality of studies, allowingfrom studies of market economies.one to gauge the extent to which the conclu-With such a large body of literature undersions change when one gives greater weightreview, it is necessary to pay special atten-to those studies that are methodologicallytion to the methodology of synthesis.sounder.Because there are so many results, verbalWe find that privatization is strongly asso-description alone would result in a hard-to-ciated with more enterprise restructuring.remember list. An interpretative summaryEconomic effects are quite often very large,presents its own dangers. Bayesian priorsfor example adding several percentagemight come to weigh too heavily in the syn-points to enterprise growth rates. The pri-thesis, a danger that is all too great in thevatization effect is, however, statisticallytransition arena, where the contentiousnessinsignificant in the Commonwealth of In-of the subject has encouraged forthrightdependent States (CIS). These results arestatements. Indeed, we have made suchrobust. They hold when we vary the empha-sis assigned to the results of different studies2The previous survey papers in this area, Josefby using weights that reflect the differing at-Brada (1996) and Oleh Havrylyshyn and Donaltention paid to bias or the overallMcGettigan selection (2<#004699'>0<#004699'>0<#004699'>0), used quite limited empirical evi-dence, which came almost exclusively from Centralquality of analysis.Europe and China. Now studies of other countries (theformer Soviet Union, Mongolia, and Vietnam) are be-3Examplecoming numerous, providing a much wider variety ofs of recent use of meta-analysis in eco-evidence. John Nellis (1999) covers the full range ofnomics are Kerry Smith and Ju Chin Huang (1995),Smith and Yoshiaki Kaoru (199<#004699'>0), David Neumark andcountries; here we provide a more systematic summaryof the evidence and focus on a wider set of determi-William Wascher (1998), Joseph Phillips and Ernestnants of enterprise restructuring.Goss (1995), and Thomas Stanley (1998). Article 1 8/12/<#004699'>02 9:13 AM Page 741Djankov and Murrell: Enterprise Restructuring in Transition741The survey also documents the effects ofnew human capital—is associated with im-different types of owners on enterprise re-proved enterprise performance. Statisticalstructuring. We find that state ownershipanalyses show that this is the case in bothwithin traditional state firms is less effectiveEastern European and CIS countries. Wethan all other ownership types, except forfind evidence that the strengthening ofworker-owners, who have a negative effect.managerial incentives leads to a greaterPrivatization to outsiders is associated withamount of restructuring, in economic5<#004699'>0 percent more restructuring than privati-terms.zation to insiders (managers and workers).Our review of existing research thenInvestment funds, foreigners, and otherleads to suggestions on three priority areasblock-holders produce more than ten timesfor future research. The design of theas much restructuring as diffuse individualstudies we survey was not driven by at-ownership. State ownership within partiallytempts to address difficult methodologicalprivatized firms is surprisingly effective, pro-problems like selection bias or simultane-ducing more restructuring than enterpriseity, or by efforts to examine the effects ofinsiders and non-block-holder outsiders.less easily quantifiable aspects of policy,The effects of different owners vary be-such as the quality of institutional con-tween regions. Workers are better owners instruction, or by considerations that viewedEastern Europe than in the CIS, whiletransition in the wider context of develop-banks and concentrated individual owner-ment. In the future, research should focusship are significantly more effective in theon these issues.CIS than elsewhere.We conclude by comparing the strengthProduct market competition has a signifi-of the effects of the different policy reformscant effect in improving enterprise perfor-examined in this paper. Privatization to out-mance. The economic effects can be large,siders is associated with the largest restruc-with a typical study indicating that enter-turing gains, while privatization to workersprises in highly competitive sectors arehas no effect in Eastern Europe and is detri-2<#004699'>0–3<#004699'>0 percent more productive than mo-mental in the CIS. Hardened budget con-nopolies. The sources of improvement differstraints are the second most importantbetween regions, however. In Easterndeterminant in the CIS, while the establish-Europe, the beneficial effect comes mainlyment of competition is second in Easternthrough import competition but is also evi-Europe.dent through domestic competition. In con-The paper is organized as follows.trast, in the CIS domestic competition isSection 2 lays out the nature of the studiessometimes statistically insignificant whileunder review, in general terms. Section 3import competition generally has a negativeinvestigates the empirical evidence oneffect on enterprise restructuring.whether state-owned or enter-undertake more privatized firmsWe next explore the link between economic restructuring.prise restructuring and the hardening ofSection 4 studies the effects of differentbudget constraints. The evidence is con-types of owners. Section 5 links productsistent with the <#00aa00'>view that hardened budgetmarket competition and enterprise restruc-constraints have had a beneficial effectturing efforts. Section 6 analyzes the role ofon enterprise restructuring in Easternsoft-budget constraints in limiting produc-Europe and the CIS. The effect in the CIStivity enhancements. Section 7 documentsis economically larger than that in Easternthe role of managers, focusing on manage-Europe.ment turnover and manager incentives.We also examine whether managementSection 8 develops the future researchturnover—or more broadly, bringing inagenda. Section 9 concludes. Article 1 8/12/<#004699'>02 9:13 AM Page 742742Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)2.The Nature of Reviewed Studies:Profits and efficiency were much less impor-Enterprise Restructuring and Itstant than they are under capitalism.DeterminantsThe enterprise was organized along veryhierarchical lines. Workers had virtually noWhat is enterprise restructuring and whatrole in enterprise decision making, except inchanges might induce it in transition coun-personnel policy, leading to firing rates thattries? The perspective taken here is found atwere extremely low by any standard (Davidthe core of every paper that we survey. TheGranick 1987). The enterprise was not onlyenterprises initially functioned in socialista producer of goods, but also played an im-economies, and their behavior was a productportant role in implementing the state’s so-of the institutions and policies of thosecial welfare policies. For these reasons, effi-economies. In the 199<#004699'>0s, those institutionsciency considerations were often secondaryand policies changed radically. Thesein determining the size of the workforce.changes compelled enterprises to adaptA labyrinthine bureaucracy replaced thetheir behavior in order to survive, and per-institutions and markets of capitalism.haps to succeed, in a new, liberalized, mar-Bureaucratic pressure substituted for 5><#880000'>com-ket environment. The term enterprise re-petition. The bureaucracy found customersstructuring has come to denote the wholeand determined prices, interceding betweenprocess undertaken by enterprises as theyproducer and buyer. It generated contractsadapt for survival and success in a marketand enforced them. Its one-year plans guar-4economy.anteed short-term working capital, and itsThe search for possible determinants ofinvestment projects automatically receivedenterprise restructuring therefore lies inlong-term credits. Given the ubiquitous rolethose institutions and policies that changedof the state, much was decided by negotia-most rapidly in the early years of transition.tion. One consequence of the frequency ofTo identify these, it is useful to reflect on thethese negotiations was the universal pres-central characteristics of the socialist econ-ence of soft budgets, which further turnedomy and its enterprises. These have beenattention away from profits and efficiency.widely discussed in the literature and wePre-transition reforms did change this stan-5only need to reiterate a few central issues.dard picture in a few countries, notablyThe classical socialist enterprise was by def-Yugoslavia (which was centralized only for ainition state-owned and was oriented to an few years), Hungary, and Poland (Leszekinput-output plan rather than any market.Balcerowicz 1995, and Kornai 1986). De-Meeting the plan was of prime importancecentralizing reforms reduced the scope of and the plan was normally very ambitious.bureaucratic decision making. Markets andTherefore, production issues dominated en-competition increased in importance.trepreneurship, marketing, and cost minimiza-Enterprises came closer to ultimate con-tion in managerial concerns. Consistently, thesumers. Paradoxically, however, abandonmenttypical manager was a production engineerof formal planning led to increased bargainingand not a businessman. Managers faced a mixbetween bureaucracy and enterprise, furtherof monetary and career-based incentives,softening budgets. Notably also, workerswhich were a function of plan fulfillment, en-gained more power within enterprises, acquir-terprise performance, and loyalty.ing experience at being informal owners.4In sum, the political pre-transition enterprisesSee, for example, Phillippe Aghion and MarkSchankerman (1999), Roman Frydman et al. (2<#004699'>0<#004699'>0<#004699'>0),were state-owned, protected from competi-and Lubomir Lizal, Miroslav Singer, and Jan Svejnartion, shielded from failure by soft budgets,(2<#004699'>0<#004699'>01).5and managed by production engineers withSee Joseph Berliner (1976), Murrell (199<#004699'>0), andJanos Kornai (199<#004699'>0) for details.incentives oriented toward the plan or politics. Article 1 8/12/<#004699'>02 9:13 AM Page 743Djankov and Murrell: Enterprise Restructuring in Transition743The enterprises were embedded in a set ofthe literature concerning which variables areinstitutions, for contracting, for financing,the best measures of restructuring, apartfor governance, that were far different fromfrom a greater preference for measures ofthose of a market economy. Transition poli-performance than for indexes of internal7cies aimed to change all of this. Therefore,decisions.the literature with which we are concernedThe typical study relates these restructur-emphasizes the following determinants ofing measures to their determinants, estimat-restructuring: ownership, competition, softing an equation of the form:budgets, managerial incentives and charac-Y = α +X β + γP + ε (1)teristics, and broader institutional changes.The widespread liberalization meant thatwhere the enterprise is the unit of observa-enterprises had to adapt in order to survivetion, Yis the restructuring measure, Pcap-and prosper markets that were increas-tures some aspect of the reforms to in whichingly contestable, the process of change thatthe enterprise is subject, e.g., ownershipthe literature conventionally defines as en-change, degree of hardness of the budget,terprise restructuring. In studies, theetc., Xis a vector of other enterprise some charac-measurement of restructuring focuses di-teristics, and εis an error term. γis the pa-rectly on enterprise decisions, for example,rameter of direct interest. The overwhelm-changes in the of corporate gover-ing majority of papers justifies their structure exactnance and management (Saul Estrin andchoice for (1) pragmatically, rather than pre-Adam Rosevear 1999a) or renovation of fac-senting a structural derivation.tories (Djankov 1999a), or investment ratesThe studies we examine use data from(Irena Grosfeld and Jean-Francois Nivetmedium-large and large enterprises. These1999). But more usually, the degree of en-enterprises were the core of the socialistterprise restructuring has been captured byeconomy, and when liberalization came theyperformance, with performance measuredremained going concerns, leading to someby variables that are objectives of companiescontinuity in operations and personnel. Thisin market economies and that were less im-continuity facilitated data collection. In con-portant for socialist enterprises. Thus, pro-trast, smaller enterprises were notoriouslyductivity (e.g., Stephen Smith, Beom-Cheolweak under socialism and soon wereCin, and Milan Vodopivec 1997; and Youngswamped by new entrants. They often van-Lee 1999) or profits (e.g., Stijn Claessensished, with their assets resurfacing in newand Simeon Djankov 1999a; and Estrin andactivities, with new personnel. Data collec-Rosevear 1999a) are often used. But sales ortion for smaller enterprises, therefore, facedrevenue have also been used extensivelyenormous difficulties, leading to few studies(e.g., Frydman et al. 1999; and Derek Jonesexamining the progress of the small enter-1998) under the premise that the ability toprises established before transition.hold on to customers or to find new ones isThe paragraphs above emphasize thean indicator of successful change within thecommon elements of the studies under re-enterprise, especially when accomplished inview. We now turn to important ways inthe face of steep recessions (Frydman et al.which studies differ. The cataloging of these61999).There seems to be no consensus indifferences and the determination of theireffect on results are important features6There is wide variation in whether levels or growthof our survey. We focus on three such vari-rates are used. This decision often depends on theations in the characteristics of studies:availability of data and also reflects whether lagged de-pendent variables are included as explanatory variables7in the equation. The differences between levels andThe lack of consensus probably results from thedifficulties of obtaining data, with researchers using thegrowth specifications diminish greatly with the inclu-best measure available.sion of lagged dependent variables. Article 1 8/12/<#004699'>02 9:13 AM Page 744744Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)9measurement of Y, selection bias, and over-selection effects or simultaneous causation.all quality.Typical cases of selection bias occur when ei-Many varieties of Yappear in the studies,ther the level of ownership or the use ofbut one distinction, already suggestedmanagerial incentives is systematically re-above, is worth emphasizing. One categorylated to some unobserved enterprise charac-of Ycomprises quantitative indicators of en-teristic that also affects Y. Examples ofterprise performance, which are based onsimultaneous causation are when an enter-accounting information. Other are some-prise’s market share is being used to meas-what softer, Y’s perhaps derived from surveyure competition or subsidies are being usedquestions on economic performance posedto measure soft budgets: in each case, theto managers, e.g., forecasts of sales in themeasured determinant of restructuring re-surveyed year as in John Earle (1998), or us-flects enterprise performance. These prob-ing information collected about reorganiza-lems have been thoroughly recognized intion, e.g., the introduction of new productsthe literature, but solutions are not alwaysas in Wendy Carlin et al. (2<#004699'>0<#004699'>01), or reflect-easy to obtain. Thus, for example, in theing operational factors only indirectly re-studies examining private versus state own-lated to current performance, e.g., the ex-ership in section 3, only 53 percent of the es-tent of wage arrears, as in Hartmuttimates of γemploy methods that mightLehmann, Jonathan Wadsworth, andserve to counter selection bias.Alessandro Acquisti (1999). We will refer toThe prevailing evidence suggests biasesthese two types of indicators as quantitativedue to selection effects or simultaneous8and qualitative.causation are a real possibility. For exam-The prevailing sentiment in the literatureple, Sweder van Wijnbergen and Antonis that the quantitative variables are to beMarcinin (1997) show that selection intotrusted more (even with the misreportingCzechoslovakia’s voucher program was non-and accounting difficulties that are rife inrandom and that this must be taken into ac-transition countries). They measure directlycount in ascertaining the effects of thethe prime output of enterprise restructur-voucher program on outcomes. Moreover,ing: economic performance. On the otherwhen there are OLS estimates and estimateshand, there is also the <#00aa00'>view that quantitativeemploying some technique to counter biasperformance might suffer when an enter-in the same paper, they quite often differprise is investing in large-scale reorganiza-considerably, suggesting the presence oftion and that the results of this processbias. (See, for example, David Brown andmight be observed earliest in qualitativeJohn Earle 2<#004699'>0<#004699'>0<#004699'>0; Rachel Glennerster 2<#004699'>0<#004699'>0<#004699'>0;variables. We focus primarily on the quanti-James Anderson, Lee, and Murrell 2<#004699'>0<#004699'>0<#004699'>0;tative indicators in this paper, deeming themand Earle 1998.)more reliable. Where a sufficient number ofThe fact that some studies have identifiedstudies is available, however, we examinenontrivial selection or simultaneity bias sug-both types.gests that we must be sensitive to its pres-A second crucial variation between papersence. Hence, in every section of this paper,lies in the degree of attention paid to possi-we present conclusions that allow the readerble biases in the estimates of γdue either toto understand how the synthesis of resultsmight have been affected by these biases.8WWe present sufficient information to allowe do not use indicators for which there is sub-stantial disagreement in the literature on whether theeach reader to judge which of the broadsign of γshould be positive or negative. The most perti-9nent example is employment, whose direction ofOur use of the term selection bias might not bechange would depend very much on the extent of ex-wholly consistent with the precise terminology ofcess labor under the old regime.econometrics. We simply follow the literature. Article 1 8/12/<#004699'>02 9:13 AM Page 745Djankov and Murrell: Enterprise Restructuring in Transition745conclusions emanating from this literatureBoycko, Andrei Shleifer, and Robert Vishnyfall under the shadow of selection bias. How1995) or through stimulation of a nascentwe do this is best explained in a practicalprivate sector (Kornai 199<#004699'>0; Murrell 1992,context, in the next section. But it must be1995). The relative emphasis on the differingemphasized that we have directly con-strategies waxed and waned with events.fronted issues of selection or simultaneity bi-With deep crisis in Eastern Europe in theases in reaching our conclusions, that weearly 199<#004699'>0’s, fast privatization gained empha-draw back from stating strong conclusionssis. However, after the recovery of Poland, awhere we feel the possibility of such biasrelatively slow privatizer, that emphasis de-leaves some residual doubt, and that never-clined (Brian Pinto, Marek Belka, and Stefantheless there are many strong conclusionsKrajewski 1993; Phillippe Aghion, Olivierthat we can reach.Blanchard, and Robin Burgess 1994; andA third variation between papers onBrada 1996). But Poland is only one of manywhich we focus is quality of analysis. One oftransition countries, an outlier at that. Nowthe primary objections to the application ofthere are fast privatizers performing wellmeta-analysis hinges on the fact that empirical work the(Estonia) and fast privatizers performingquality of varies greatlybadly (Russia), with similar variation acrossacross papers, meaning that a simple aggre-slow privatizers (Poland versus Romania),gation of results might give undue weight togiving sustenance to a variety of opinionslesser quality research. Therefore, someabout the results of privatization (Pohl et al.scholars prefer to focus reviews of empirical1997; Havrylyshyn and McGettigan 2<#004699'>0<#004699'>0<#004699'>0;literature on the high points, ignoring pa-Joseph Stiglitz 1999; Bernard Black, Reinierpers that are less convincing from a method-Kraakman, and Anna Tarassova 2<#004699'>0<#004699'>0<#004699'>0). Toological standpoint. However, it is also possi-referee this debate, it is necessary to turn to11ble to take a middle road, one that examinesthe microeconomic empirical literature.whether the composite results change whenStudies examining whether privatized en-considerations of quality are taken into ac-terprises perform better than state-owned en-count. This is our approach. How we do thisterprises use equation (1), with Pa measure ofis, again, best left until our methods can bethe degree of private ownership of an enter-1<#004699'>0explained in context in the next section.prise of pre-1989 vintage. A large variety ofvariables takes the place of Y, X, or Pwith no3.State versus Private Ownershipconsensus on the nature of the equation to beestimated. For Y, a significant number of pa-In early transition debates, there waspers use measures of output levels, captured,agreement on the goal of an economy domi-for example, by sales (e.g., Jozef Koningsnated by private ownership, but conflicting1997; and Earle 1998) or value added (e.g.,views on how best to attain this, through fastAnderson, Georges Korsun, and Murrellprivatization (Jeffrey Sachs 1992; Maxim2<#004699'>0<#004699'>0<#004699'>0; and Smith, Cin, and Vodopivec1997).1<#004699'>0An earlier version of this paper (Djankov andOther papers use growth rates of the sameMurrell 2<#004699'>0<#004699'>0<#004699'>0) examined other variations in thevariables (e.g., Frydman et al. 1999; and Jonesmethodological characteristics of studies. It applied the1998). Qualitative versions of Yare also exten-methods that appear below (on quality and selectionbias) also to the length of time (for the pertinent re-sively used, especially in studies of the formerform) that is embodied in the estimates and the num-Soviet Union, where there are greater difficul-ber and appropriateness of the variables used in X. Inties in obtaining accurate accounting data.the present paper, these dimensions of quality are sim-ply absorbed into the overall measure of quality. The11reader can refer to that earlier paper for the more de-William Megginson and Jeffry Netter (2<#004699'>0<#004699'>01) un-tailed results. The data compiled for that study, exceptdertook an extensive survey of the effects of privatiza-for our assessments of the overall quality of papers, aretion worldwide. Their results are primarily germane toavailable from the authors on request.countries with long-established market economies. Article 1 8/12/<#004699'>02 9:13 AM Page 746746Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)One common form of Pis the percentageThe similarities and differences betweenof shares owned privately (e.g., Brown andtwo papers (Frydman et al. 1999; andEarle 2<#004699'>0<#004699'>0<#004699'>0). Other papers use a dummyAnderson, Lee, and Murrell 2<#004699'>0<#004699'>0<#004699'>0) exhibitvariable capturing whether private owner-the methodological decisions to be madeship is above some threshold value, for ex-when conducting such studies and the varia-ample <#004699'>0 percent in Estrin and Roseveartions in results that can be obtained.(1999a) and 5<#004699'>0 percent in Derek Jones andFrydman et al. examine the performanceNiels Mygind (2<#004699'>0<#004699'>02). Frydman et al. (1999)from 199<#004699'>0 to 1993 of a panel of 218 pri-and Claessens and Djankov (2<#004699'>0<#004699'>0<#004699'>0) formvatized and state firms from the Czechtheir dummy variables on the basis of corpo-Republic, Poland, and Hungary, whilerate laws, the former defining a privatizedAnderson et al. focus on 1995 data for 211enterprise as one in which private ownershipprivatized (including partially state-owned)is above the 33.3-percent threshold neededMongolian enterprises. Data collection, in-to block strategic company decisions, whilecluding sample design, was carried outthe latter also uses the 66.7-percent levelspecifically for each of these studies, raisingthat guarantees that a state minority ownerthe quality, extensiveness, and appropriate-cannot veto such decisions.ness of the information collected but result-Variety also present in the X’s. Sectoral,ing in small sample sizes. Each study is exam-regional, and (in the case of panels) time andines the effects of privatization as a wholefirm dummies are common. Among the X’sand the effects of different owners. We dis-used in various studies are measures <#880000'>com-cuss the latter in the next section.petition, levels of of soft budgets, size of firms,Both studies wrestle with the choice ofaccess to finance, year of privatization, andthe dependent variable, how to measuremany more. Pre-privatization levels of de-ownership, which X’s to include, and how topendent variables are sometimes includedcounter selection bias. Their decisions differunder the premise that these might dampena great deal. Frydman et al. use four differ-the effects of selection bias. One commonent dependent variables, rates of growth offormulation includes capital and labor in therevenues, employment, revenues per em-13X’s and uses output for Y, with the basic equa-ployee, and costs per unit of revenue.tion then representing a production functionTheir ownership variable is a dummy. Asand the estimate of γcapturing the effect ofcontrols, they use initial levels (not growthownership on total factor productivity.rates) of the four performance measures, ac-The variety in the formulation of estimat-companied by sectoral, country, and timeing equations is a reflection of two factors.dummies. Selection bias is parried in a num-First, there is the absence of a single <#880000'>com-ber of ways, primarily by methods devel-pelling theory that models the process ofoped for the analysis of treatment effects12change within an enterprise.Without suchin panels. In separate analyses, the authorstheory, specifications for estimating equa-use a dummy variable (in X) capturing pre-tions rely on ad hoc formulations. Second,privatization differences between state firmsthe set of variables for which data are avail-and privatized firms; they employ a firmable greatly, with every study havingfixed-effects model, and they varies verify that per-deficiencies in some respect. Given theseformance of those firms slated for privat-reasons for the variety of approaches, withization, but not yet privatized, is closer tonone obviously superior to all others, itthat of state firms than privatized ones. Theseems judicious to include a wide range ofstudies in drawing general conclusions.13A major aspect of Frydman et al. (1999) is the ef-fect of privatization on different dependent variables.12Additionally, there is also no agreed theory on howFor reasons of space, we cannot examine this interest-to model the selection of firms for privatization.ing issue here. Article 1 8/12/<#004699'>02 9:13 AM Page 747Djankov and Murrell: Enterprise Restructuring in Transition747cumulative effect of these analyses is to con-differences in results is the countries stud-vince the reader that the privatization ef-ied, the most advanced transition countriesfects are real, rather than an artifact of selec-versus one of the most backward. We exam-tion for privatization.ine this issue later in this section. AnotherEnterprise record-keeping during Mon-possibility is the types of owners generatedgolia’s early transition was so poor thatduring privatization, which is the subject ofAnderson et al. could not obtain panel data.section 4.Hence, they focus on performance in oneOne respect in which the two papers areyear, using three different dependent vari-similar is in their use of quantitative depen-ables, total factor productivity (within adent variables derived from accounting infor-Cobb-Douglas production function), salesmation. In this respect, Carlin et al. (2<#004699'>0<#004699'>01)per employee, and value-added per em-provide an interesting contrast. They useployee. The equations for the latter two vari-data from a survey of over three thousandables include lagged dependent variable,firms in 25 countries. Because of the a logisti-thus nesting a specification in which growthcal and compatibility issues involved in col-is the dependent variable. Since the esti-lecting accounting data from such a wide va-mated coefficients on the lagged dependentriety of countries, they rely for theirvariables are significantly different fromdependent variables on responses to ques-one, this suggests that growth measures aretionnaires. One set of questions focused onnot suitable dependent variables. The own-changes in real sales and employment, lead-ership variable is the percentage of enter-ing to a measure of productivity growth.prise shares held by non-state owners.Another set focused on internal restructur-Controls include regional and sectoral dum-ing, asking about upgrading of existing prod-mies, levels of competition, and the pres-ucts, introduction of new products, openingence of soft budgets. Selection bias is coun-of new plants, and quality accreditation.tered through the use of instrumentalUsing principal components to combine thevariables. Suitable instruments were avail-answers on this latter set, the paper derives aable because of idiosyncratic features of themeasure of new product restructuring.privatization program and due to the varyingCarlin et al. do not find a direct effect ofincentives of different types of owners dur-change in ownership on productivity or salesing privatization. Comparison of OLS andgrowth, but they do find a significant effectinstrumental variables results suggests thatof privatization on new product restructur-OLS estimates of the effect of privatizationing. Moreover, new product restructuringare upwardly biased.directly increases sales and productivityFrydman et al. and Anderson et al. obtaingrowth, implying an indirect effect of own-strikingly different results. In Centralership on enterprise growth. These resultsEurope, privatization improves revenuegive some sustenance to the argument thatgrowth by approximately 7 percent a year; inqualitative variables provide useful informa-Mongolia, wholly private firms are 3<#004699'>0 per-tion where accounting ones are not avail-cent to 7<#004699'>0 percent less efficient than <#880000'>com-able. The authors nevertheless caution thepletely state-owned firms. Although therereader that they were not able to implementare many differences between the analysesprocedures to counter selection bias on theof these two papers, it is quite unlikely thatownership variable.methodological differences can explain theThe three papers discussed above gener-divergence in results, since both papers useate highly varying results, which is a charac-conventional methods to solve the usualteristic of the literature as a whole. In tableproblems and pay close attention to possible1,we focus on the economic size of effects.biases. An obvious candidate to explain theSuch information is usually not emphasized Article 1 8/12/<#004699'>02 9:13 AM Page 748748Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)TABLE1GAINSORLOSSESDUETOOWNERSHIPCHANGE: A SAMPLINGOFESTIMATESEstimates of the performance of a 1<#004699'>0<#004699'>0% private firm relative to a 1<#004699'>0<#004699'>0% state firm. Positive signs indicate superiorprivate performance. For the growth measures, relative performance is private-firm yearly growth rate minus state-firm yearly growth rate. For the levels measures, relative performance is private-firm productivity minus state-firmproductivity expressed as a % of state productivity.Output or Sales GrowthProductivity GrowthLevels of Productivity% private % privateprivate minusminus %minus %state as %statecountrystatecountryofstatecountry12118.7Poland4.3Central Europe??43Russia23127.3Central Europe3.5Russia62Russia3613??9.7Russia1<#004699'>0.6Kyrgyz14Estonia47141<#004699'>0.9Bulgaria3.6Georgia/Moldova??3<#004699'>0Lithuania(MBOs)(diffuse owners)47142.5Czech Republic<#004699'>0.9Georgia/Moldova49Latvia(voucher privatized)(individual owners)48152.1Hungary16.<#004699'>0Romania??65Mongolia598??4.5Russia1.<#004699'>025 countries36Romania61<#004699'>01618.<#004699'>0Kyrgyz3.1Eastern Europe14<#004699'>0SloveniaNotes:These estimates are based on the authors’ calculations, derived from information presented in the following pa-12345pers: Grosfeld and Nivet (1999); Frydman et al. (1999); Jones (1998); Pohl et al. (1997); Perevalov, Gimadi, and6789Dobrodey (2<#004699'>0<#004699'>0<#004699'>0); Roberts, Gorkov, and Madigan (1999); Djankov (1999a); Earle and Telgedy (2<#004699'>0<#004699'>01); Carlin et1<#004699'>0111213al. (2<#004699'>0<#004699'>01); Claessens and Djankov (2<#004699'>0<#004699'>0<#004699'>0); Brown and Earle (2<#004699'>0<#004699'>0<#004699'>0); Earle (1998); Jones and Mygind (2<#004699'>0<#004699'>02);141516Jones and Mygind (2<#004699'>0<#004699'>0<#004699'>0); Anderson, Lee, and Murrell (2<#004699'>0<#004699'>0<#004699'>0); Smith, Cin, and Vodopivec (1997).in the papers under review. Indeed, it is re-ization can only be reached by employingmarkable that in most cases we had to calcu-methods that aggregate the results of thelate the numbers in table 1 ourselves, some-many papers under consideration. Moreover,times using a considerable degree ofto understand the reasons for variations be-interpretation. Moreover, a significant num-tween estimates, the aggregation methodsber of papers did not offer enough informa-should offer the possibility of assessing howtion for analogous estimates to be calcu-methodological factors influence estimateslated. This suggests that the surveyed papersand should allow for comparisons betweendo not emphasize the economic interpreta-sets of studies. It is to this aggregate analysistion of their estimates sufficiently.that we now turn.The estimates in table 1 are sampled to3.1Statistical Significance: Combining reflect a variety of contexts and results. Thet-Statisticsfirst from the table is that privat-ization has economically conclusion significant effects.We are interested in the size and statistical^A second conclusion is that these effects aresignificance of the estimate of γ() in equa-γ^far from uniform. Privatization done in thetion (1). The ’s of different studies are notγright way, or under the right circumstances,directly comparable because Yand Parecan have huge positive effects, but privatiza-measured in many different ways. Therefore,tion can also be hugely detrimental.we seek methods of combining the results ofThis large variation in results suggests thatstudies with different characteristics. We firstan overall judgment on the effects of privat-begin by combining t-statistics. Article 1 8/12/<#004699'>02 9:13 AM Page 749Djankov and Murrell: Enterprise Restructuring in Transition749In this section, we combine the results ofing conclusions, since it combines probability37 distinct studies. Table 2 lists these stud-information erroneously, a point that is best il-lustrated with a simple example.ies, the countries included in them, their Y’s,Researcher A obtains a t-statistic of 2 in aand their P’s. Within these studies, we havestudy, pronouncing significance for the effect.identified 93 distinct empirical analyses and^Researcextracted 93 ’s together with their corre-her B, not favorable to A’s conclusions,γconducts two separate studies of two separatesponding t-statistics. We use more than oneanalysis from a single paper only in cases incountries and obtains t-statistics of 1 in eachwhich the estimates are derived from con-study. Researcher B triumphantly announcesthat A has been mistaken, for the vote is now 2ceptually distinct analyses, e.g., from <#880000'>com-pletely different forms of the dependentstudies to 1 for nonsignificance. But the <#880000'>com-bined statistic obtained by applying (2) to allvariable or from different countries.The theory justifying the aggregation ofthree studies is 4/?3 = 2.31. B has actuallyt-statistics is analogous to that used whenstrengthened support for A’s conclusions.conducting tests on the mean of a sample.Column (1) of table 3 contains the resultsCollect several independent observationsobtained by applying equation (2). The rowsfrom the same distribution, find their mean,reflect two different ways of grouping esti-and deduce the distribution of the mean.mates. First, there is the quantitative-qualita-The variance of the sample mean will betive division of dependent variables, dis-smaller than that of individual observations,cussed above. Second, there are regionalimplying that statistical tests based on thegroupings. Corresponding to much of themean will be more powerful than tests on in-rest of the literature (e.g., EBRD 1999), thedividual observations.basic split is between the non-Baltic former^15The data comprise Mt-statistics on ’s,γSoviet Union (the CIS) and the rest.Thet, ... , t. The following is normally dis-countries outside the CIS comprise Eastern1M14tributed:Europe and the Baltics (with one study ofΜChina). For brevity, but at the cost of sometk(2)?? Minaccuracy, we will refer to the two groups ask=1Eastern Europe (EE) and the CIS in the restM, which is the number of analyses, playsof the paper. Interestingly, once we seek athe role of sample size. It is readily apparentcriterion that corresponds to our split ofthat a set of analyses with small positivecountries, we find that it is the length of timet-statistics could be significant in the aggre-that the countries labored under commu-gate even with nonsignificance in each indi-nism, seventy years for each CIS country and16vidual analysis. As it happens, less than one-halfless than fifty years in EE.The readerof the t-statistics examined in this section aretherefore might like to think of our regionalstatistically significant, but collectively theygroups as “two generations” and “three gen-are highly significant.erations,” indicating the length of time underOur synthesis of results relies on tests of (2),communism.whereas the most common method of combin-The significant effects of privatizationing results in literature reviews is the methodshow clearly in all but one of the statisticsof vote-counting (John Hunter and FrankSchmidt 199<#004699'>0). Vote-counting concludes that15In the set of papers under consideration, therethere is nonsignificance in the aggregate whenare two studies of Mongolia. Since this countrya set of studies has a median t-statistic that islooks like a typical member of the CIS (Korsun andinsignificant. This method produces mislead-Murrell 1995), Mongolia is included in the CISgrouping.1416Assuming sample sizes in all analyses are suffi-Moldova is an exception to this rule. This countryciently large, which is the case here.contributes only a small amount of data to this paper. Article 1 8/12/<#004699'>02 9:13 AM Page 75<#004699'>0TABLE2STUDIESOFPRIVATIZATIONLevels or Privatization PaperCountriesDependent variablesGrowth Measure Anderson, Korsun, Murrell 2<#004699'>0<#004699'>0<#004699'>0Mongoliasoftness of budgetslevels% ownershipAnderson, Lee, Murrell 2<#004699'>0<#004699'>0<#004699'>0MongoliaTFP, labor productivitylevels% ownershipBrown and Earle 2<#004699'>0<#004699'>0<#004699'>0RussiaTFPlevels% ownershipBrown and Brown 1999Russiasectoral profitabilitylevels% of sectoral output in private firmsCarlin et al. 2<#004699'>0<#004699'>0125 countriesproductivity, creation growth,dummy if privatizedof new productslevelsClaessens and Djankov 2<#004699'>0<#004699'>0<#004699'>07 EE combinedTFP, labor productivitygrowthdummy if 33% privateDjankov 1999aGeorgia, Moldovasales, renovationsgrowthdummy if privatizedDjankov 1999cGeorgia, Moldova, labor productivity,Kazakstan, Kyrgyz, renovationsgrowth,% ownershipRussia, UkrainelevelsEarle 1998Russiamid-year projection levels% ownershipof labor productivityEarle and Estrin 1997Russiaqualitative restructuringlevels% ownershipEarle and Estrin 1998Russiamid-year projection of levels% ownershiplabor productivityEarle and Rose 1997Russianew goods productionlevelsdummy if privatizedEarle and Sabirainova 1999Russiawage arrearslevelsdummy if privatizedEarle, Estrin, Leshchenko 1996Russialinks to government, saleslevelsdummy if privatizedEarle and Telgedy 2<#004699'>0<#004699'>01RomaniaTFPlevels% ownershipEstrin and Rosevear 1999aUkraineprofits, saleslevelsdummy if privatizedEstrin and Rosevear 1999bUkrainequalitative restructuringlevelsdummy if privatizedEvans-Klock and Samorodov1998Kyrgyzexcess employmentlevelsdummy if privatizedFrydman, Gray, Hessel, Hungary, Poland,revenues, laborgrowthdummy if 33%Rapaczynski 1999Czech RepublicproductivityprivateFrydman, Gray, Hessel, Hungary, Poland,default on debtslevelsdummy if 33%Rapaczynski 2<#004699'>0<#004699'>0<#004699'>0Czech RepublicprivateGlennerster 2<#004699'>0<#004699'>0<#004699'>0Macedoniaprofits per workergrowthdummy if privatizedGrigorian 2<#004699'>0<#004699'>0<#004699'>0LithuaniaTFP, labor productivitylevels% ownershipGrosfeld and Nivet 1999Polandoutputgrowthdummy if privatizedHendley, Murrell, Ryterman2<#004699'>0<#004699'>01Russiasuccess in transactionslevels% ownershipJones 1998Russiarevenue, labor productivitygrowthdummy if 5<#004699'>0% privateJones and Mygind 2<#004699'>0<#004699'>02EstoniaTFPlevelsdummy if 5<#004699'>0% privateKonings 1997Hungary, Romaniarevenueslevels% ownershipSloveniaLehmann, Wadsworth, Acquisti 1999Russiawage arrearslevels% ownershipLinz and Krueger 1998Russialabor productivitygrowthdummy if privatizedMajor 1999Hungarycapital productivitylevelsownership scalePerevalov, Gimadi, Dobrodey 2<#004699'>0<#004699'>0<#004699'>0Russiarevenues, labor growth,dummy if privatizedproductivitylevelsPohl, Anderson, Claessens, Bulgaria, CzechTFPgrowthdummy if privatizedDjankov 1997Republic, Hungary, Poland, Romania, Slovenia, SlovakiaRoberts, Gorkov, Madigan 1999Kyrgyz Republicsales, labor productivity, growth,dummy if privatizedqualitative restructuringlevelsSmith, Cin, Vodopivec 1997SloveniaTFPlevels% ownershipWarzynski 2<#004699'>0<#004699'>01bUkraineproductivitygrowthdummy if privatizedXu and Wang 1999Chinareturn on equity andassetslevels% ownershipZemplinerova, Lastovicka, Czech Republicrestructuring indexlevelsdummy if stateMarcincin 1995majority owner Article 1 8/12/<#004699'>02 9:13 AM Page 751Djankov and Murrell: Enterprise Restructuring in Transition751TABLE 3EFFECTSOFPRIVATEVERSUSSTATEOWNERSHIPONENTERPRISERESTRUCTURINGComposite statistics derived from 93 analyses appearing in 37 studiesWeighting Method Used(1)(2)(3)RegionsType of Dependent Number of Analysesattempts Variable in the ContributingAnalysesto the Compositeto handle overall ratingTest Statisticnoneselection biasof qualitynormally distributed test statisticsAll countriesboth9313.4813.3411.19EEboth3715.7115.4915.18CISboth533.952.84<#004699'>0.<#004699'>07All countriesquantitative6411.7811.<#004699'>079.54EEquantitative3415.3715.1414.92CISquantitative28<#004699'>0.37??2.28??3.73All countriesqualitative296.647.585.99EEqualitative33.713.713.49CISqualitative255.356.765.<#004699'>06appearing in column (1). Thus, the first con-This weighting procedure discounts studiesclusion from this table is that the aggregatewith smaller weights. For example, if thereeffects of privatization are positive. This alsoare particular concerns about selection bias,applies when both quantitative and qualita-the weights would reflect attention paid totive indicators are examined separately. Thethis issue.one case where the effects of privatizationWe have constructed two sets of weights.are not significantly positive is for quantita-First, we rated papers on a scale of one totive indicators for the CIS.three, reflecting the amount of attentionHow robust are these conclusions? Thepaid to the problem of selection bias. Apapers that contribute analyses to our datascore of one indicates no attempt at coun-vary a great deal, in the quality of analysis intering such bias, as in Djankov (1999a),general, and the amount of attention paid towhile two suggests a less-than-complete selection bias in particular. Therefore, dueattempt, for example, using fixed effectscaution suggests that we examine whetherwithout any strong justification that thisthe above conclusions could reflect method-approach addresses the problem of biasological deficiencies in some of the surveyed(David Grigorian 2<#004699'>0<#004699'>0<#004699'>0). A three involves apapers, rather than real economic phenom-convincing attack on the problem, as inena.Brown and Earle (2<#004699'>0<#004699'>0<#004699'>0), which uses bothA straightforward way to undertake such anfixed effects and instrumental variables.examination is to weight the various t-statis-Obviously the choice of scale is ad hoc, buttics, using weights reflecting methodologicalit does fulfill our basic purpose, which is todifferences. Suppose that there are weights,check whether conclusions change notice-w, ... , w, for each t-statistic. Then the fol-ably when discounting results derived from1Mlowing statistic has a normal distribution:weaker methodologies. This weightingplays a role analogous to the procedure ofΜΜ2deemphasizing some papers in the stan-wtw(3)kkk??k=1k=1dard literature review. But it approaches Article 1 8/12/<#004699'>02 9:13 AM Page 752752Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)the issue in a somewhat more transparent3.2Strength of Effect: Comparisons manner, since the reader can directly observeacross Regionsthe consequences of the deemphasis.Although it is tempting to do so, one can-We also constructed a rating of the over-not immediately conclude from table 3 thatall quality of the papers. This rating reflectsthe effect of privatization on the quantitativesystematic information collected on spe-variables in EE is greater than the effect incific issues such as selection bias, numberthe CIS. Table 3 provides information onlyof reform years reflected in the data, theon the statistical significance of an effect rel-nature of Y, and appropriateness of X, asative to zero. It is quite possible that an ef-well as our own subjective assessment offect can be weaker in statistical terms but17the strength of the analysis.It also re-stronger in economic terms. To compare theflects the relative standing of the journal insize of the two economic effects, it is firstwhich the paper is published, if it has beennecessary to identify a statistic that summa-published. The rating is on a scale of one torizes aggregate effect size for a heteroge-ten. Again, the use of this rating is to exam-neous set of studies. In order to focus onine whether conclusions change noticeablysubstantive results, the more formal discus-if the results from weaker methodologiession of that statistic is relegated to an appen-are discounted.dix. We concentrate on intuition across the columns of tableThe t-statistic is not below.Comparisons suitable because its3 suggest robustness in the conclusions formagnitude reflects sample size, which hasqualitative variables in both regions and fornothing to do with the economic effect ofquantitative variables in EE. For example,^privatization. is not suitable because it isγthe evidence in table 3 suggests quite clearlydependent on the units of measurement of Ythat the overall conclusions for EE are notand P, which vary greatly between studies.vitiated by the possibility of selection bias.On the positive side, the t-statistic is invari-Table 3 serves to blunt the perception held^ant to units of measurement, while doesγby some that selection bias presents an in-not depend on sample size. Thus, we seek asurmountable problem in assessing the ef-measure that combines the positive proper-fects of privatization.ties of the t-statistic and of the estimated pa-There is one case where weighting byrameter, while eliminating their negativequality does change the aggregate result: thecharacteristics.statistic for quantitative variables in the CISThe standard procedure in meta-analysischanges from nonsignificance to signifi-is to use the partial correlation coefficientcantly negative. Given this sensitivity to the(Robert Rosenthal 1984). Partial correlationemphasis placed on different results andcoefficients are not usually published, butgiven the differences in the results for quali-they are easily calculated from regressiontative and quantitative measures for the CIS,18statistics that appear in all papers.Sinceit is difficult to draw an overall conclusion oncorrelation coefficients are unit-free, theythe effects of privatization in this region. Ifare comparable across a heterogeneous setthe results for the quantitative variables areof studies and therefore can be aggregateddeemed most reliable, then our analysis sug-18gests that privatization has not had positiveWilliam Greene (2<#004699'>0<#004699'>0<#004699'>0, pp. 233–36) proves this re-sult within the classical regression framework, to whicheffects.a majority of the studies reviewed conform. When ap-plying this result outside this framework, we simply ap-17An earlier version of this paper presented resultspeal to analogy. This suggests that more effort shouldfor many sets of weights, reflecting each methodologi-be made in including information in published paperscal variable on which we collected systematic informa-that allows the estimates to be more readily comparedtion. The reader can examine these results in Djankovto estimates in other papers that examine the same eco-and Murrell (2<#004699'>0<#004699'>0<#004699'>0).nomic phenomena. Article 1 8/12/<#004699'>02 9:13 AM Page 753Djankov and Murrell: Enterprise Restructuring in Transition753to form a summary statistic on effect size forgether or when the quantitative variables arethe whole set of studies.examined alone, the test statistics affirm (atThe first two rows of column (1) of table 4the 1-percent significance level) that the pri-show the simple unweighted means of thevatization effect is stronger in EE than in theestimates of the partial correlations derivedCIS. For the quantitative variables, the usefrom two groups of studies, those on EE andof the weights adjusting for methodologicalthose on the CIS. As expected, the EE esti-quality actually strengthens this conclusion,mate is larger. As the appendix shows, it issuggesting that selection bias, or otherstraightforward to construct a normally dis-methodological deficiencies, do not play anytributed test statistic that reflects the nullrole in leading us to this conclusion. The re-hypothesis that these means are equal.sults are similar, but weaker statistically, forThis two test statistic appears in the third row ofthe qualitative variables, perhaps becausetable 4, providing evidence that the privat-there are so few studies for EE that fit intoization effect is significantly larger in Easternthis category. In sum, there is strong evi-Europe than in the CIS.dence that the effect of privatization inOne disadvantage of working with partialEastern Europe has been much greater incorrelation coefficients is that their size ismagnitude than in the CIS.not obviously interpretable in terms of realeconomic phenomena. However, a few facts4.The Effects of Different Types of Ownersmight make the numbers in table 4 moreeasily interpretable. For example, if a studyOne reason that changes of ownershiphad the same sample size and standard er-might have had varying effects across re-rors as the Frydman et al. result in table 1gions is that different privatization processes(second row, first column), then a partialled to different mixes of owners. In thecorrelation coefficient of <#004699'>0.<#004699'>05 would corre-absence of quick re-trading of sharesspond to a percentage growth of sales of pri-(Anderson, Korsun, and Murrell 1999;vatized enterprises that is 3.51 higher thanJoseph Blasi and Andrei Shleifer 1996), thethat of state enterprises. Using the Brownowners created by the privatization processand Earle (2<#004699'>0<#004699'>0<#004699'>0) result (table 1, first row,have had more than a temporary effect onlast column) in a similar fashion, a partialenterprise behavior. This is important, ofcorrelation coefficient of ??<#004699'>0.<#004699'>05 would corre-course, only if ownership makes a differencespond to privatized enterprises being 38in restructuring. As it happens, transition ex-percent less efficient than state enterprises.perience offers unusually good evidence onWeighted means of the partial correlationthis score.coefficients can also be calculated and a nor-The pre-transition literature offers a vari-mally distributed test statistic can be derivedety of disparate empirical results. There hasto examine the significance of the differencebeen general support for Adolf Berle andbetween the two weighted means. ColumnsGardiner Means’ (1933) contention that dif-(2) and (3) of table 4 present the results forfuse ownership yields power to managersthe aggregate partial correlation coefficientswith less interest than shareholders in re-calculated using the weights that were de-structuring (Bruce Johnson et al. 1985;scribed in the context of table 3. Thus, dif-Richard Roll 1986). Randall Morck, Andreiferences between the columns of the tableShleifer, and Robert Vishny (1988) found thatreflect the sensitivity of the overall results tosome managerial ownership can amelioratevarying judgments on the importance ofsuch problems, while Shleifer and Vishnymethodological quality.(1986) argue that block-holders have an in-When examining the test statistics forcentive to monitor management. However,qualitative and quantitative variables to-block-holders also have opportunities to Article 1 8/12/<#004699'>02 9:13 AM Page 754754Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)TABLE4COMPARINGTHESIZEOFPRIVATIZATIONEFFECTSACROSSREGIONSPartial correlation coefficients and test statistics derived from 9<#004699'>0 analyses from 36 studiesWeighting Method UsedRegionsType of Dependent Number of Analyses(1)(2)(3)Variable in the ContributingAnalysesto the Compositeattempts Statisticto handle overall ratingnoneselection biasof qualitypartial correlation coefficients and test statistics on their differencesEEboth37<#004699'>0.<#004699'>083<#004699'>0.<#004699'>073<#004699'>0.<#004699'>075CISboth53<#004699'>0.<#004699'>038<#004699'>0.<#004699'>041<#004699'>0.<#004699'>016Test statistic for both3.892.744.82differenceEEquantitative34<#004699'>0.<#004699'>078<#004699'>0.<#004699'>07<#004699'>0<#004699'>0.<#004699'>073CISquantitative28<#004699'>0.<#004699'>016??<#004699'>0.<#004699'>0<#004699'>01??<#004699'>0.<#004699'>019Test statistic for quantitative4.144.635.41differenceEEqualitative3<#004699'>0.146<#004699'>0.146<#004699'>0.145CISqualitative25<#004699'>0.<#004699'>063<#004699'>0.<#004699'>089<#004699'>0.<#004699'>064Test statistic for qualitative1.991.361.84differenceexpropriate value from minority sharehold-Therefore, evidence from transition coun-ers. Quite separately, the comparative sys-tries has a vital contribution to make. Intems literature pursued the question oftransition economies, ownership was largelywhether the pathologies predicted by thedetermined through political and adminis-neoclassical model of the labor-managed processes rather than endogenouslyoutweigh the motivational firmtrative effects of workerin markets with low transaction costs. Inownership. The evidence is mixed (Alanmany cases, ownership structure is exoge-Blinder 199<#004699'>0; John Bonin, Derek Jones, andnous. In others, it is relatively easy to obtainLouis Putterman 1993; Douglas Kruse andreliable instruments to counter endogeneityJoseph Blasi 1995).bias. The evidence from transition shouldThe existing literature on marketoffer much better information on the effectseconomies does not provide a clear pictureof different owners than has been generated19of which type of owner is most advantageouspreviously.for restructuring. Ronald Coase (1988) andThe results for transition countries areHarold Demsetz and Kenneth Lehn (1985)unique in another respect. Many differentsuggest a reason for the lack of clarity. Theytypes of owners were created. At the time ofargue that the evidence from mature marketprivatization, government agencies andeconomies might be spurious. If the transac-firms knew the structure of ownership andtion costs of taking value-maximizing posi-could impart this information to researchers.tions in firms are low, each firm would have19For example, Megginson and Netter’s (2<#004699'>0<#004699'>01) <#880000'>com-the “right” ownership structure: there is noprehensive survey of the effects of privatization in allrelationship between ownership type and re-types of economies does not reflect on the relative ef-fectiveness of different types of private owners.structuring. Article 1 8/12/<#004699'>02 9:13 AM Page 755Djankov and Murrell: Enterprise Restructuring in Transition755The resultant ownership information was atafter privatization is narrower and the de-a level of detail that could not be obtained ingree of ownership concentration less.mature market economies. Hence, there is aManagerial and worker ownership are highlylarge set of studies examining the effects ofcorrelated due to the nature of the privatiza-different owners. By aggregating the results,tion scheme and there is no value in differ-we are able to compare the effects of a <#880000'>com-entiating between them. Individuals withprehensive set of owners, something notsmall stakes are dominant among outsiderpreviously accomplished in the literature.owners. Hence, analysis of the effects of dif-We begin by examining the characteristicsferent owners in Anderson et al. only con-of the pertinent papers. The methodologiestrasts the state (in a privatized firm) versusused by the papers are essentially the sameoutsiders versus insiders. In contrast toas those discussed in section 3. In fact, manyFrydman et al.’s dummy variables, owner-papers begin with an analysis of state versusship is measured by the percentage of sharesprivate and then turn to a disaggregation ofheld by each type of owner, perhaps theprivate ownership, e.g., Brown and Earlemost usual approach, e.g., Brown and Earle(2<#004699'>0<#004699'>0<#004699'>0), Earle (1998), Frydman et. al. (1999),(2<#004699'>0<#004699'>0<#004699'>0), Claessens and Djankov (1999a,b),and Jones (1998). Such papers use the sameEarle and Estrin (1997), and Smith, Cin,methodology and independent variables inand Vodopivec (1997).both parts of their analyses, simply making PFrydman et al. find that foreign and do-in (1) a vector that captures the ownershipmestic financial owners produce the largestheld by different entities.positive effects, while outsider owners out-A further focus on Frydman et al. (1999)perform insiders, results echoed later in ourand Anderson, Lee, and Murrell (2<#004699'>0<#004699'>0<#004699'>0) ex-synthesis of many papers. Their results arehibits the methodological decisions to bemixed when comparing insider ownership tomade when choosing the elements of Pandstate ownership in enterprises with 1<#004699'>0<#004699'>0-the variations in results that can be obtained.percent state ownership. Anderson et al.Frydman et al. examine outsiders, insiders,find that outsiders and insiders perform lessand the state in one analysis, and foreigners,effectively than the state in privatized enter-domestic financial firms, domestic nonfinan-prises, while there is no significant differ-cial firms, domestic individuals, the state (in aence between insiders and outsiders.privatized firm), the state (in a nonprivatizedWhat explains the differences in thesefirm), managers, and workers in anothertwo sets of results? Selection bias or omit-analysis. This is an unusually comprehensiveted variables seem unlikely candidates,list of owners to appear in one paper, and mostgiven the attention paid by both studies topapers use a much smaller set, e.g., Smith,these issues. Rather, the outsider ownersCin, and Vodopivec (1997) using foreigners,are very different in Central Europe, whereinsiders, and the state, and Claessens andthey are block-holders, than in CentralDjankov (1999a) using managers and out-Asia, where they are individuals, usuallysiders. The ownership variables in Frydman etwith tiny amounts of ownership. Andersonal. are dummies capturing whether the givenet al. examine whether diffusion of owner-owner is the largest shareholder, a commonship can explain their results and do findapproach, e.g., Jones and Mygind (2<#004699'>0<#004699'>0<#004699'>0,better performance in those few enter-2<#004699'>0<#004699'>02). Measurement of in this wayprises in which block-holders have ownership ob-2<#004699'>0reflects a somewhat unusual feature of thetained seats on the boards of directors.Frydman et al. data, the highly concentratedThe positive relation between ownershipownership in all privatized firms.In Mongolia, examined by Anderson, Lee,2<#004699'>0However, because of lack of instruments, they areand Murrell (2<#004699'>0<#004699'>0<#004699'>0), the variety of owners not able to control for selection bias on this variable. Article 1 8/12/<#004699'>02 9:13 AM Page 756756Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)concentration and performance is a direct4.1Quantifying the Differences focus of Claessens and Djankov’s (1999b)between Ownersstudy on the Czech Republic.We now turn to the quantitative synthesisIn reconciling the findings of Frydman etof results. The objective of this synthesis isal. and Anderson et al., the results on stateto find the relative sizes of the effects of dif-ownership warrant particularly close exami-ferent owners and to test which owners havenation. Anderson et al. were able to samplesignificantly different effects from others.only privatized enterprises (including onesWe follow the literature in identifying thewith large state ownership). Thus their studyownership categories that we analyze, iden-includes no firms that remained under tradi-tifying those ownership types that recur con-tional state ownership. Frydman et al. sam-22sistently in a number of studies.This re-pled both privatized and nonprivatized firms,sulted in categories, some of whichthus having two state ownership eleven variables.23overlap:The Anderson et al. results on state owner-1) traditional state ownership: enterprisesship are comparable only to those ofthat are 1<#004699'>0<#004699'>0-percent state-owned and thatFrydman al. on state ownership in privat-have not been part of a privatization et program;ized firms. When this is understood, the two2) the state in commercialized (or corpo-studies’ results appear much more consis-ratized) enterprises: state ownership in en-tent, since Frydman et al. find that stateterprises that have been legally separatedownership in a privatized firm performs atfrom the state bureaucracy, that are treatedleast as well as the median type of privateas private under the corporate governance21owner.However, the two studies providelaws, and, usually, that have been part of avery different interpretations of their results.24privatization program;Frydman et al. <#00aa00'>view the state as passive, withprivate owners dominating the decisions of22Since there is no source that provides a definitionpartially state-owned privatized firms.of a standard set of ownership categories, there are in-Anderson et al. <#00aa00'>view the Mongolian state asevitable variations across the empirical papers in howdifferent owners are defined. Where papers used own-active and enormously pressured by eco-ership types that differed too much from the types thatnomic necessity. The government was moreappeared in other papers, we did not use the results.willing than insiders to push for efficiencyThis happened in only a small number of cases.23For example, workers and managers together areand more able to do so than outsiders.insiders. One should not assume that an aggregate cat-In a reconciliation of the resultsegory necessarily produces the providing same result as the sumof the two studies, the above paragraphsof its parts. For example, estimates of the effects ofmanagers and workers usually appear in different stud-provide an important warning to the harriedies than those for insiders as a whole. Which owners arereader trying to absorb a large number of re-included in a study, however, is not random. Re-sults very quickly. Details matter crucially,searchers are more likely to collect data on a specificowner when the privatization process gives scope forfor example which study used the privatiza-that owner to be active in privatization procedures.tion agency to obtain its sample or whichHence, the results for insiders probably reflect some-study’s data reflect concentrated ownership.what different circumstances than those for managersand workers on their own. We should not expect to seeChoice of estimation method can even influ-the insider effect simply equal to the weighted sum ofence the interpretation of which owners arethe worker and managerial effect.24reflected in results, fixed effects estimatesIn most countries commercialization (or corpora-tization) occurred as preparation for privatization.reflecting only ownership that changes overTherefore, our data on this ownership type is domi-the sample period. These facts are often notnated by results for enterprises that have less than 1<#004699'>0<#004699'>0-emphasized in papers.percent state ownership. Where a study did not provideenough information for us to know which of the twocategories of state ownership applied or where the21As is clearly shown when we aggregate results be-study mixed the two types, we assigned the results tolow, this result is echoed in many other studies.the traditional state ownership category. Article 1 8/12/<#004699'>02 9:13 AM Page 757Djankov and Murrell: Enterprise Restructuring in Transition7573) enterprise insiders: a composite group,Second, there is no longer a binary compari-including both workers and managers;son between state and private, but a multi-4) outsiders: a composite group consistinglateral comparison between many types ofof all nonemployee, nonstate owners;owners. Therefore, a dummy-variable re-5) workers (nonmanagerial employees);gression framework is required to combine6) managers (managerial employees);results, rather than simply using differences7) foreign owners of all types;between means. Third, to take advantage of8) banks, except those included in 7;information on the precision of estimates9) investment funds, except if ownershipfrom different papers, it is necessary to useof the fund is such that 2, 7, or 8 applies;generalized least squares. Fourth, the1<#004699'>0) block-holders: outsider ownershipweighting of results according to method-concentrated in the hands of large individualological quality also requires generalizedowners, except where the block-holderleast squares. A full consideration of thesecould be classified in 2, 6, 7, 8, or 9;complications entails an extended discussion11) diffuse outsider: the residual outsiderof methodology, in which the average readerownership category, when outsider ownersis probably not interested. The discussion isare not identified as belonging to 7, 8, 9, ortherefore relegated to an appendix.1<#004699'>0; a category dominated by individual out-From 24 on twenty countries, wesider owners, each of whom owns a studies minis-have compiled all usable information on the25cule portion of any specific enterprise.effects of different types of owners on quan-All papers that we review examine a sub-titative enterprise outcomes. The ertinentset of these eleven types of owners. For ex-papers, together with p the countries andample, a paper might focus on the state ver-owners in each study, are listed in table 5.sus insiders and outsiders and estimate:From these studies, we derive the t-statis-tics and then the partial correlation coeffi-Y=α+Xβ + δO+ θI+ ε (4)cients corresponding to ownership effects.Oand Iare measures of the amount ofThe data are then combined in a generalizedownership held by outsiders and insiders,least squares framework to obtain a compos-state ownership is the omitted variable, δite estimate of the effect of each of the newand θare the parameters of interest, and allowners (2–11 above) relative to traditionalother variables are as defined before. Thestate ownership. These estimates of owner-information of prime interest is the compar-ship effects are in the form of partial correla-ison of δto θ(outsiders versus insiders) andtions, as in section 3.each of these to <#004699'>0 (insiders or outsiders ver-The estimates of the partial correlationssus the state). We examine both statisticalappear in figure 1. These estimates reflect asignificance and the magnitude of the own-process that weights more heavily thoseership effect.studies that have paid more attention to the26The methods used here are extensions ofissue of selection bias.We focus on thisthose already introduced in section 3.methodological issue because readers ofHowever, a number of complications arise inprevious versions of this paper have indi-27developing the results. First, important in-cated that this is of most concern.By <#880000'>com-formation is often missing from papers (rele-paring the results that are obtained when allvant to the statistical precision of the <#880000'>com-26The weighting scale is the same as in the previousparison between δand θ). Therefore, it issection.27necessary to investigate how to fill this gap.Suspicions about selection bias arise naturally.Perhaps the state kept the best enterprises during pri-25This category is used only when the study differ-vatization, or managers fought harder to retain controlentiates between various types of outsiders. When allwhen prospects were good, or foreigners were willingoutsiders are treated as one, owner group 4 applies.to pay for efficient enterprises only. Article 1 8/12/<#004699'>02 9:13 AM Page 758758Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)TABLE5STUDIESOFPRIVATIZATIONBYOWNERSHIPTYPEPaperCountriesOwnership TypesAnderson, Lee, Murrell 2<#004699'>0<#004699'>0<#004699'>0Mongolia2, 3, and 4Brown and Earle 1999Russia2, 3, 4, 7, and 11Claessens and Djankov 1999aCzech Republic4 and 6Claessens and Djankov 1999bCzech Republic2 and 7–1<#004699'>0Claessens, Djankov, Pohl 1997Czech Republic2 and 7–9Cull, Matesova, Shirley 2<#004699'>0<#004699'>02Czech Republic1 and 7–9Djankov 1999bGeorgia and Moldova1, 6, and 11Djankov 1999cSix CIS states2, 5–7, 1<#004699'>0, and 11Earle 1998Russia1, 4–11Earle and Estrin 1997Russia1, 4–11Earle and Telgedy 2<#004699'>0<#004699'>01Romania2, 3, 1<#004699'>0, and 11Estrin and Rosevear 1999aUkraine1–6Frydman, Gray, Hessel, Rapaczynski 1999Central Europe1–1<#004699'>0Grosfeld and Nivet 1999Poland1 and 2Jones 1998Russia1, 4–6, and 8Jones, Klinedinst, Rock 1998Bulgaria1 and 2Jones and Mygind 2<#004699'>0<#004699'>0<#004699'>0Baltics1, 5–7, and 1<#004699'>0Jones and Mygind 2<#004699'>0<#004699'>02Estonia2, 5–7, and 1<#004699'>0Konings 1997Slovenia, Hungary, Romania1 and 5Lee 1999China1 and 2Lehmann, Wadsworth, Acquisti 1999Russia1 and 7Roberts, Gorkov, Madigan 1999Kyrgyz Republic2–4Smith, Cin, Vodopivec 1997Slovenia1, 3, and 7Weiss and Nikitin 1998Czech Republic2 and 8–1<#004699'>0Ownership types are coded as follows (see text for details):1) traditional state ownership2) the state in commercialized (or corporatized) enterprises3) enterprise insiders4) outsiders5) workers (nonmanagerial employees)6) managers (managerial employees)7) foreign owners of all types8) banks9) investment funds1<#004699'>0) block-holders11) diffuse outsiderpapers are treated equally (see appendix) toconsiderable degree of sensitivity to how se-those in figure 1, one can examine whetherlection bias is handled. Thus, our assessmentthe results seem particularly sensitive to se-of the results is that a high degree of confi-lection bias. Three conclusions follow fromdence can be in the relative rankingthis comparison. First, the relative placed positionthat appears in figure 1, except for someof most owners remains the same whateverresidual doubt about the positions of man-the treatment of selection bias. Second, se-agers and workers, especially the former.lection bias reduces the estimated magni-Figure 1 suggests that differences be-tudes of most ownership effects. Third, onlytween owners are of great economic impor-the results for managers and workers show atance. Privatization to workers is detrimental; Article 1 8/12/<#004699'>02 9:13 AM Page 759Djankov and Murrell: Enterprise Restructuring in Transition759<#004699'>0.1<#004699'>0<#004699'>0<#004699'>0.<#004699'>079<#004699'>0.<#004699'>08<#004699'>0<#004699'>0.<#004699'>06<#004699'>0<#004699'>0.<#004699'>049<#004699'>0.<#004699'>048<#004699'>0.<#004699'>043<#004699'>0.<#004699'>04<#004699'>0<#004699'>0.<#004699'>035<#004699'>0.<#004699'>025<#004699'>0.<#004699'>02<#004699'>0<#004699'>0.<#004699'>017<#004699'>0.<#004699'>011<#004699'>0.<#004699'>0<#004699'>04<#004699'>0.<#004699'>0<#004699'>0<#004699'>0diobcmbifnnoiuoalfsaovrnmftineceudskimiakssedsgtg-erwmneehe sirorroenscrslndkid–<#004699'>0.<#004699'>02<#004699'>0atiee vl–<#004699'>0.<#004699'>018rirfizusdsenudda ssltate–<#004699'>0.<#004699'>04<#004699'>0Category of OwnersFigure 1. How Ownership Affects Firm Performance after Privatization.Estimated Effects of Changing from Traditional State Ownership to Different Private Ownersprivatization to diffuse individual owners hasbeen developed organically for decades, butno effect, and privatization to investmentrather where ownership has been artificiallyfunds or to foreigners has a large positive ef-transferred, sometimes to private ownersfect. Loosely speaking, privatization to fundswho are creatures of the state. Then, if cor-is five times as productive as privatization toporate governance laws are weak, share re-insiders, while privatization to foreigners ortrading is sluggish, and the state is focusedblock-holders is three times as productive ason solving economic state owner-privatization to insiders. Foreigners were ex-ship problems, could easily be superior to some typespected to make productive changes, but it isof ownership (Anderson, Lee, and Murrellnotable that investment funds are signifi-2<#004699'>0<#004699'>0<#004699'>0). The superiority of state ownership incantly better than foreigners, and that bankscommercialized enterprises over traditionaland block-holders are close to foreigners.state ownership might arise because theSimilarly, diffuse individual ownership waspart-owners who are private are playing annot expected to be very effective, but it isimportant role in enterprise affairssurprising that it is indistinguishable from(Frydman et al. 1999) or because the verytraditional state ownership.act of commercialization changes the incen-A notable result is the difference betweentives facing the state when it intervenes intotraditional state ownership and state owner-enterprise affairs (Shleifer and Vishny 1994).ship in commercialized enterprises, a resultA conclusion implicit in the results of fig-that recurs from Central Europe to Chinaure 1 is that concentrated shareholding pro-(Claessens, Djankov, and Pohl 1997; Leeduces larger effects than diffuse sharehold-1999). Remember, of course, that this resulting. This is seen most clearly in theis not for economies in which ownership hasdifference between the effects of individualPartial Correlation Coefficients Article 1 8/12/<#004699'>02 9:13 AM Page 76<#004699'>076<#004699'>0Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)TABLE6TESTINGDIFFERENCESBETWEENTHEEFFECTSOFAVARIETYOFOWNERSStatistics derived from 341 pair-wise comparisons of owners taken from 24 studiesTraditional DiffuseWorkersStateIndividualManagerst-statistic for {(effect of owner listed on row) minus (effect of owner listed on column)}Traditional State1.36Diffuse Individual1.78<#004699'>0.5<#004699'>0Managers1.93<#004699'>0.9<#004699'>0<#004699'>0.63Insiders2.772.613.12<#004699'>0.45Outsiders3.2<#004699'>03.453.541.<#004699'>09Commercialized State4.125.697.211.99Banks3.533.423.<#004699'>041.93Foreign4.827.827.672.98Block-holders4.9<#004699'>07.737.743.<#004699'>09Investment Funds5.557.<#004699'>055.984.28Notes:To compare owners A and B: If the cell located at the intersection of A’s row and B’s column is blank, thenB is more productive than A. If the cell corresponding to A’s column and B’s row is nonempty, then the number inthat cell is the t-statistic for a test of the null hypothesis that B’s effect minus A’s effect is equal to zero.owners and those of block-holders, but it is6 immediately reveals an unusually large29also implicit in the effects of foreigners,number of significant t-statistics.This isfunds, and banks since these entities usuallypartially a consequence of employing aconcentrate their shareholdings. Claessensmethodology that combines results and em-and Djankov (1999b) focus directly on thebodies the precision of estimates from manyeffects of concentration in their study of thestudies. Just as in the previous section, weCzech Republic. They show that a 1<#004699'>0-percenthave been able to generate an unusualincrease in the percentage of shares held byamount of statistical power. However, it isthe largest five shareholders will increase la-not the case that all owners have signifi-bor productivity by 5 percent. They also findcantly different effects from all others. Fordiminishing returns to concentration. Theseexample, workers, diffuse individual owners,results are echoed in studies of Russia, e.g.,and traditional state ownership cluster at theEarle and Estrin (1997) and Brown andbottom, with effects that do not differ signif-Earle (2<#004699'>0<#004699'>0<#004699'>0).icantly. The effects of banks are significantlyTable 6 examines statistical significance ofdifferent from the effects of less than halfthe differences in figure 1, reporting the val-the other owners, partially reflecting the factues of t-statistics for standard tests of nullthat banks are included in few studies.hypotheses that one owner has the same ef-fect as another, for all pair-wise combina-4.2Comparing Owners across Regions28tions of owners.A first inspection of tableWe have found that privatization hasstronger effects in Eastern Europe than in28These test statistics are easily calculated using29standard generalized least squares procedures and im-The appendix suggests using a conservativeplicitly use the information on the variances of esti-benchmark for significance, but this statement holdsmates that is extracted from the original studies.even under this standard. Article 1 8/12/<#004699'>02 9:13 AM Page 761Djankov and Murrell: Enterprise Restructuring in Transition761TABLE6 (Cont.)Statistics derived from 341 pair-wise comparisons of owners taken from 24 studiesCommercializedInsidersOutsidersStateBanksForeignBlock-holderst-statistic for {(effect of owner listed on row) minus (effect of owner listed on column)}1.675.282.282.<#004699'>091.39<#004699'>0.636.664.183.44<#004699'>0.446.714.363.71<#004699'>0.56<#004699'>0.385.3<#004699'>04.6<#004699'>03.972.472.962.84the CIS and that different types of ownersThe data reject the null hypothesis that own-have different effects. This immediatelyership effects in the CIS are all the same asthose in EE, at the 1-percent significanceraises the question of whether the latterlevel. The next step then is to redo figure 1,could explain the former. This questionfor each region separately.would be best addressed using data on theIn all cases except one (workers), the own-distribution of ownership in different coun-ership effects for the CIS in figure 2 aretries, but there is no systematic collection of3<#004699'>0greater than for EE.How is this consistentsuch data. The available evidence does sug-with section 3, which showed that privategest that worker and diffuse individual own-ownership had stronger effects in EE than inership are more prevalent in the CIS than inthe CIS? There are two obvious reasons. First,EE, while foreign, investment fund, concen-the effect of workers is strongly negative andtrated individual, and bank ownership arethey own a much larger share of privatized en-less prevalent. (See, for example, Anderson,terprises in the CIS than elsewhere. (InLee, and Murrell 2<#004699'>0<#004699'>0<#004699'>0; Frydman et al. 1999;Frydman et al.’s 1999 Central European sam-Brown and Earle 2<#004699'>0<#004699'>0<#004699'>0; and Claessens andple, workers are the dominant owner in 9 per-Djankov 1999b.) Thus, since the CIS owner-cent of firms, while they are dominant in 5<#004699'>0ship portfolio contains a greater share of lesspercent of Earle and Estrin’s 1997 Russianeffective owners, structure of ownership is asample.) Second, commercialized state own-strong candidate to explain regional differ-ences in the effects of privatization.3<#004699'>0The effects of different types of ownersIt is worth emphasizing that these results do notshow that the various owners are more productive incould also vary between regions. To investi-the CIS than Eastern Europe in absolute terms.gate this, the natural first step is to seeRather, these results could just as easily be due to thewhether the vectors of estimated ownershiprelative unproductiveness of traditional state enter-effects differ significantly between regions.prises in the CIS, which are the basis for comparison. Article 1 8/12/<#004699'>02 9:13 AM Page 762762Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)<#004699'>0.2<#004699'>0<#004699'>0<#004699'>0.162<#004699'>0.15<#004699'>0<#004699'>0.126<#004699'>0.12<#004699'>0<#004699'>0.115<#004699'>0.1<#004699'>02<#004699'>0.<#004699'>094<#004699'>0.1<#004699'>0<#004699'>0<#004699'>0.<#004699'>087<#004699'>0.<#004699'>079<#004699'>0.<#004699'>069<#004699'>0.<#004699'>068<#004699'>0.<#004699'>063<#004699'>0.<#004699'>05<#004699'>0<#004699'>0.<#004699'>04<#004699'>0<#004699'>0.<#004699'>028<#004699'>0.<#004699'>028<#004699'>0.<#004699'>017<#004699'>0.<#004699'>011<#004699'>0.<#004699'>0<#004699'>05<#004699'>0.<#004699'>0<#004699'>0<#004699'>0<#004699'>0.<#004699'>0<#004699'>0<#004699'>0diobcmbif–<#004699'>0.<#00469 9'>0<#004699'>03nnoiuoalfsaovrnmftineudscekimiakssedsgtg-erwmneehe sirorroenscrslndkidatieevl rirfizusdse–<#004699'>0.<#004699'>05<#004699'>0nudda –<#004699'>0.<#004699'>053ssltate–<#004699'>0.1<#004699'>0<#004699' >0EECISCategory of OwnersFigure 2. Regional Variations in the Effects of Different Types of Owners.Comparing Ownership Effects in Eastern Europe to Those in the CISership is separated from traditional state own-While we have treated the two regions asership in figure 1 but not in table 4.individually homogenous in deriving figure 2,Commercialized state ownership has aboveit should be noted that there are also largedifferences between the estimates for differ-average effects (relative to traditional stateent countries within the same region. For ex-ownership) in the CIS and below average inample, Romania seems to be an outlier inEE. Thus, if one conceives of the privatizationEastern Europe. John Earle and Almosprocess as two steps, first the move away fromTelgedy (2<#004699'>0<#004699'>01) find that diffuse individualtraditionalstate ownership and then the moveand insider ownership are surprisingly pro-away from state ownership of all forms, figureductive in Romania. the studies for2 and table 1 together suggest that the firstthe Indeed, CIS are somewhat more homogenousstep alone was a important contribution of thethan those outside the CIS, as evidenced byprivatization process in the CIS, while this was31the R-squares for the regressions that providenot so in Eastern Europe.the data for figure 2: <#004699'>0.73 in the CIS and <#004699'>0.31outside that region. This is perhaps a reflec-tion of the greater diversity, in history and in31Lest some readers find this statement confusing, itreform programs, within EE. Hence, figure 2should be emphasized that it was sometimes possiblefor an enterprise to pass through a privatization pro-is hardly a last answer, but rather a beginning,gram and still end up with 1<#004699'>0<#004699'>0-percent state owner-in a search for the reasons why different own-ship! Going through a privatization program and reduc-ers are more effective in some settings thantion of state ownership were not synonymous, althoughhighly correlated.others. In section 8, we provide some discus-Partial Correlation Coefficients Article 1 8/12/<#004699'>02 9:13 AM Page 763Djankov and Murrell: Enterprise Restructuring in Transition763sion of this issue, examining an institutionalespecially from abroad, destroys networkinterpretation of the patterns in figure 2.capital and harms enterprise performance inthe early transition period. The former5.Product Market CompetitionSoviet Union experienced significant lossesin markets with the disintegration of theThere is a substantial theoretical literaturetrade relations between Russia and the otheranalyzing the relationship between competi-former republics (Simeon Djankov andtion and enterprise efficiency. The generalCaroline Freund 2<#004699'>0<#004699'>02). As traditional mar-hypothesis is that increased competitionkets collapsed and as countries were openedstimulates improvements in productivity.up to foreign competition, many firms foundTwo lines of argument support this hypothe-it difficult to restructure their product linessis. The first is derived from the literature onand at the same time reorient their sales to-X-inefficiency, while the second centers onwards new markets.industry rationalization. The X-inefficiencyThe empirical literature outside of transi-literature argues that managerial effort is un-tion economies yields a mixed picture on theder-supplied in the absence of vigorous <#880000'>com-relationship between competition and enter-petition. Henrik Horn, Harold Lang, andprise efficiency. Few empirical investigationsStefan Lundgren (1995) show that greaterusing firm-level data have established acompetition induces an expansion of outputstrong link between greater competition andby incumbent firms through improved inter-subsequent changes in enterprise perfor-nal technical efficiency without any realloca-32mance.Two studies of British manufactur-tion of resources across firms. Earlier studiesing firms (Stephen Nickell, Sushil Wadhwani,(Bengt Holmstr?#004699'>0?m 1982; Barry Nalebuff andand Martin Wall 1992; Stephen Nickell 1996)Joseph Stiglitz 1983) argue that incentiveuse a panel to show that market concentra-schemes for managers will generate bettertion has an adverse effect on total factor pro-results the greater the number of playersductivity. In contrast, David Blanchflower(firms) involved because of greater opportu-and Stephen (1996) find no effectnities for performance comparison.of changes Machin in domestic market structure onA second line of argument is that in-the productivity of British plants. Silkecreased competition may lead to a rationali-Januszewski, Jens Koke, and Joakim Winterzation of oligopolistic industries as firms are(2<#004699'>0<#004699'>01) use a sample of publicly listed Germanforced to compete for market share (Klauscompanies to show that neither import <#880000'>com-Schmidt 1997). Resource reallocation oc-petition nor domestic market competition hascurs across firms, within and between sec-beneficial effects on TFP growth. Studies ontors. Although the shake-out may result in adeveloping countries also find ambiguous re-transitional decline in productivity as firmssults. Ann Harrison (1994) finds that a reduc-with increasing returns to scale lose domes-tion of tariffs in Cote d’Ivoire and the subse-tic market share, over time this may be re-quent increase in import penetration had anversed as market share expands due to exitinsignificant effect on TFP growth. In con-of competitors and access to export markets.trast, Sweder van Wijnbergen and AnthonyMuch depends on the existence of scaleVenables (1993) find a strong positive effecteconomies and the ease of entry and exit.of increased import penetration on labor pro-In transition economies, increased compe-ductivity in Mexican manufacturing firms.tition may have negative effects on efficiency.Competition creates incentives for breaking32contracts when institutions are weak (OlivierOn the macro level, Francisco Rodriguez andDani Rodrik (1999) survey the literature on open tradeBlanchard and Michael Kremer 1997). Barrypolicies and economic growth and conclude that theIckes, Randi Ryterman, and Stoyan Tenevevidence that openness is associated with growth is(1995) suggest that excessive competition,very weak. Article 1 8/12/<#004699'>02 9:13 AM Page 764764Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)The initial period of transition provides aThe analyses are fairly homogeneous. Inunique opportunity to test the importance ofmost cases, the dependent variable is quanti-product market competition. A majority oftative: 55 analyses use TFP as the indicator oftransition economies rapidly liberalizedenterprise restructuring, nineteen use labortheir trade regimes. Many went on to demo-productivity, and two use sales growth. Sixnopolize their industrial sectors by breakinganalyses, all derived from Kreacic (1998), useup conglomerates, spinning off individualqualitative indicators of restructuring (facili-production units, and allowing entry of newties renovation, establishment of a marketingprivate firms (Simeon Djankov and Bernarddepartment, and computerization of the ac-Hoekman 2<#004699'>0<#004699'>0<#004699'>0; Lizal, Singer, and Svejnarcounting function). Import competition is2<#004699'>0<#004699'>01). The short period in which theseproxied by the import penetration ratiochanges took place allows the researcher to(twenty analyses), or the industry-level tariffidentify the timing of the policy change andrate (eight analyses). Domestic market <#880000'>com-to control for other economy-wide and firm-petition is measured by either the Herfindahlspecific factors that influence the change inindex (25 analyses), the percentage of salesproductivity.revenues of the top two, three, or four firmsWe find 23 studies that explicitly investi-in the respective industry (fifteen analyses),gate the effect of product market competi-the number of local that enter-tion on enterprise restructuring. Amongprise competitors managers perceive as rivals (twelvethose, thirteen focus on EE countriesanalyses), or the change in the enterprise’s(Claessens and Djankov 2<#004699'>0<#004699'>0<#004699'>0; Rumenown market share (two analyses).Dobrinsky et al. 2<#004699'>0<#004699'>01; Djankov andSome explanatory variables raise the issueHoekman 1998, 2<#004699'>0<#004699'>0<#004699'>0; Grigorian 2<#004699'>0<#004699'>0<#004699'>0; Irenaof endogeneity. In proxying for the level ofGrosfeld and Thierry Tressel 2<#004699'>0<#004699'>01; Philipdomestic competition, Hersch, Kemme, andHersch, David Kemme, and SharadBhandari (1994) asked enterprise managersBhandari 1994; Laszlo Halpern and Gaborto “categorize the number of other firms nowK?#004699'>0?r?#004699'>0?si 1998; K?#004699'>0?nings 1997, 1998; Nikolayproducing in your main domestic market asMarkov et al. 2<#004699'>0<#004699'>0<#004699'>0; Mary Shirley and Lixineither none, 1 to 1<#004699'>0, 11 to 5<#004699'>0, 51 to 1<#004699'>0<#004699'>0, orColin Xu 2<#004699'>0<#004699'>01; Frederic Warzynski 2<#004699'>0<#004699'>01a)more than 1<#004699'>0<#004699'>0” (p. 358). Dobrinsky, Dochev,and ten use data for Russia (Manuellaand Markov (2<#004699'>0<#004699'>01) use lagged own marketAngelucci et al. 2<#004699'>0<#004699'>01; Annette Brown andshare as one of their proxies. Both variablesDavid Brown 1999; Earle and Estrin 1998;are problematic since the number of <#880000'>com-Brown and Earle 2<#004699'>0<#004699'>0<#004699'>0; Boris Kuznetsov etpetitors or one’s market share may be en-al. 2<#004699'>0<#004699'>02; Yuri Perevalov et al. 2<#004699'>0<#004699'>0<#004699'>0), Georgiadogenous to the performance of the enter-(Vladimir-Goran Kreacic 1998), Ukraineprise. Accordingly, when we aggregate the(Warzynski 2<#004699'>0<#004699'>01b), Mongolia (Anderson,empirical analyses using weights, we assign aLee, and Murrell 2<#004699'>0<#004699'>0<#004699'>0), or all transitionweight of 1 to both studies for “attempts toeconomies (Carlin et al. 2<#004699'>0<#004699'>01). We are ablehandle endogeneity.” Studies using theto distinguish 82 separate analyses, whereHerfindahl index, e.g., K?#004699'>0?nings (1998), avoidthe authors use either different measures ofthis problem and are given a rating of three.enterprise restructuring (total factor produc-The endogeneity problem is also presenttivity levels or growth rate, labor productivityin studies that use manager perceptions oflevel or growth, sales growth, and qualitativeimport competition or the import penetra-variables like renovation of facilities) or dif-tion ratio as a measure of competition fromferent indicators of competitive pressures.abroad. The former suffers from obvious en-Twenty-eight analyses use import -dogeneity, while the second depends on thetion as the main competi explanatory variable, whilevolume of imports, which is endogenous to52 focus on domestic market structure.the performance of the producers in the Article 1 8/12/<#004699'>02 9:13 AM Page 765Djankov and Murrell: Enterprise Restructuring in Transition765TABLE7GAINSORLOSSESDUETOPRODUCTMARKETCOMPETITION: A SAMPLINGOFESTIMATES1StudyCountryDependent VariableExplanatory VariableEconomic EffectAngelucci et al. 2<#004699'>0<#004699'>01RomaniaLog TFP levelHerfindahl level19.1%Perevalov et al. 2<#004699'>0<#004699'>0<#004699'>0RussiaLog TFP levelNumber of competitors2.4%Grosfeld and Tressel 2<#004699'>0<#004699'>01PolandTFP growthOwn market share1.4 ppDjankov and Hoekman 2<#004699'>0<#004699'>0<#004699'>0BulgariaTFP growthHerfindahl level1.4 ppWarzynski 2<#004699'>0<#004699'>01bUkraineTFP growthNumber of competitors??3.8 ppBrown and Earle 2<#004699'>0<#004699'>0<#004699'>0RussiaLog TFP levelHerfindahl level1.4%Konings 1998BulgariaTFP growthHerfindahl level4.4 ppKonings 1998EstoniaTFP growthHerfindahl level??1.<#004699'>0 ppHalpern and Korosi 1998HungaryLog TFP levelHerfindahl level??2.6%Anderson, Lee, Murrell 2<#004699'>0<#004699'>0<#004699'>0MongoliaLog TFP levelOwn market share9.2%Claessens and Djankov 2<#004699'>0<#004699'>0<#004699'>0HungaryLabor prod. growth3-firm concentration7.1 ppMarkov et al. 2<#004699'>0<#004699'>0<#004699'>0BulgariaLog TFP levelOwn market share??23.7%Markov et al. 2<#004699'>0<#004699'>0<#004699'>0BulgariaLog TFP level3-firm concentration5.9%Angelucci et al. 2<#004699'>0<#004699'>01RomaniaLog TFP levelImport penetration??6.<#004699'>0%Markov et al. 2<#004699'>0<#004699'>0<#004699'>0BulgariaLog TFP levelImport penetration??2.1%Brown and Earle 2<#004699'>0<#004699'>0<#004699'>0RussiaLog TFP levelImport penetration1.4%Djankov and Hoekman 2<#004699'>0<#004699'>0<#004699'>0BulgariaTFP growthImport penetration1.1 ppHalpern and Korosi 1998HungaryLog TFP levelImport penetration3.<#004699'>0%Dobrinsky et al. 2<#004699'>0<#004699'>01BulgariaLog TFP levelImport penetration??<#004699'>0.7%Kreacic 1998GeorgiaLog TFP levelImport penetration??5.8%Claessens and Djankov 2<#004699'>0<#004699'>0<#004699'>0Czech Rep.Labor prod. growthTariff levels4.3 pp1The economic effect is calculated from the coefficient estimates assuming a 1<#004699'>0-percentage-points change in theexplanatory variable. pp means percentage points.domestic sector. Accordingly, we assign aage-points reduction in the Herfindahl indexvalue of 1 for handling endogeneity to stud-results in an increase in TFP growth of 19.1ies that use perception of import competi-percent; a 1<#004699'>0-percentage-points increase intion, e.g., Konings (1997); a 2 to all analysesimport penetration results in a TFP growththat use import penetration, e.g., Angelucciloss of 6 percent. The negative effect ofet al. (2<#004699'>0<#004699'>01); and a 3 for all analyses that usecompetition is particularly pronounced forchanges in the statutory tariff rate, e.g.,state-owned enterprises, and is insignificantDjankov and Hoekman (2<#004699'>0<#004699'>0<#004699'>0). The endo-for privatized firms. The authors concludegeneity ratings are also reflected in the over-that the technology gap between domesticall quality rating.firms and importing rivals may have beenTable 7 illustrates the variety of analysestoo wide and thus acted to discourage enter-in terms of country samples, the choice ofprise restructuring. In effect, domesticdependent and explanatory variables, and(state-owned) firms are priced out of theirthe estimated economic effects of competi-markets.tion on enterprise efficiency. For example,Grosfeld and Tressel (2<#004699'>0<#004699'>01), using a panelAngelucci et al. (2<#004699'>0<#004699'>01) study the effect ofof 153 Polish firms, find that the reductioncompetitive pressure on firm performancein own market share by 1<#004699'>0 percentage pointsin Romania, using panel data from 1997 toresults in 1.4-percentage-points increase in331998.The study finds that a 1<#004699'>0-percent-productivity growth. Again, this effect isstronger for privatized firms. The results are33The study also uses a large panel of Polish firms.robust to the use of alternative estimationAs the data cover both small and large companies, theyare not used in our analysis.techniques, and the inclusion of controls for Article 1 8/12/<#004699'>02 9:13 AM Page 766766Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)ownership concentration and ownershipChanges in domestic market structure aretypes. The evidence supports the <#00aa00'>view thatimportant in explaining enterprise restruc-competition-enhancing policies and privat-turing in both the CIS and EE samples.ization should be pursued simultaneously.(The exception is in the CIS when we weightThe analyses indicate that product mar-according to attention paid to endogeneityket competition has been a major force be-bias.) These results are upheld in a study ofhind improvements in enterprise productiv-over 3,3<#004699'>0<#004699'>0 enterprises in 25 transitionity in transition economies as a whole, aseconomies (Carlin et al. 2<#004699'>0<#004699'>01) that showsshown in table 8A. When we divide thestrong positive effects of the reduction ofsample into analyses based on import <#880000'>com-market concentration on firm efficiency.petition versus domestic market structureThe significant effect of changes in domestic(for all countries together), we find thatmarket structure on enterprise restructuringeach is generally significant in explainingin the CIS resonates with recent evidenceenterprise performance. The effect of do-on high barriers to entry in transitionmestic market structure is always signifi-economies. Djankov et al. (2<#004699'>0<#004699'>02a) documentcant. Import competition is insignificant inthe number of procedures and the associ-explaining enterprise restructuring whenated time and cost for starting a new busi-giving less weight to studies that do not con-ness in 85 countries around the world, in-front endogeneity problems. thecluding 22 transition economies, to explainresults by region, Examining the effects are strong forpoor aggregate growth performance in theEE countries, where in all cases the t-statis-CIS. Establishing a new business in Russiatics are significant at conventional levels.or Ukraine takes twice as much effort, time,For CIS countries, increased competitiveand money than a start-up in Easternpressures are not associated with enhancedEurope, and three times longer than estab-restructuring.lishing a new business in Latvia orA further subdivision of the analysesLithuania. The authors argue that entry bar-shows an interesting pattern: import compe-riers serve to impede product market <#880000'>com-tition in the CIS countries does not have apetition. Similarly, Andrei Shleifer andsignificant effect on enterprise restructuringDaniel Treisman (2<#004699'>0<#004699'>0<#004699'>0) document the pres-and has a negative sign. In contrast, importence of significant geographic segmentationcompetition is significant in explaining en-in Russia and lackluster competition at theterprise restructuring in the EE sample,regional level.with t-statistics varying between 3.37 andTable 8B examines the statistical signifi-4.15. Since all the Russian studies use im-cance of the variation in the effect of prod-port penetration variables that are regionuct market competition across regions andspecific, it would not be correct to concludeacross types of competition. In the firstthat this difference is due to the fact thatpanel, we show competition from local pro-Russia has a larger internal market and theducers has a stronger effect than importeffect of import competition may be mutedcompetition, but the difference is not statis-or imprecisely measured. What explains thistically significant. The second panel showsdifference then? The evidence is consistentthat competition has a stronger effect in ex-with the theoretical predictions in Blanchardplaining enterprise restructuring in EEand Kremer (1997) and Ickes, Ryterman,countries than in CIS countries and that thisand Tenev (1995). In the CIS, the rapiddifference is statistically significant whenopening to competition from abroad actedweighting for attempts to handle endogene-to deter domestic restructuring. Perhaps aity, and when controlling for the overallgradual approach to trade liberalization isquality of the paper. The third panel <#880000'>com-the preferred policy choice.pares the relative importance of foreign Article 1 8/12/<#004699'>02 9:13 AM Page 767Djankov and Murrell: Enterprise Restructuring in Transition767TABLE8IMPORTANDDOMESTICCOMPETITIONANDENTERPRISERESTRUCTURINGComposite statistics derived from 82 analyses appearing in 23 studiesWeighting Method UsedCategory Number of Analyses(1)(2)(3)Contributing toattempts the Composite Testto handle Statisticendogeneityoverall ratingnonebiasof qualityA. Testing the Effects of Competitionnormally distributed test statisticsAll countries825.134.145.<#004699'>01Import competition282.861.793.3<#004699'>0Domestic competition544.273.783.79EE, all competition585.<#004699'>094.9<#004699'>05.<#004699'>08CIS, all competition241.58<#004699'>0.681.22EE, import competition2<#004699'>03.973.374.15CIS, import competition8??<#004699'>0.93??1.12??1.55EE, domestic competition383.413.653.16CIS, domestic competition162.591.562.1<#004699'>0B. Comparing Types of Competition and Regionspartial correlation coefficients and test statistics on their differenceImport competition28<#004699'>0.<#004699'>0<#004699'>03??<#004699'>0.<#004699'>013<#004699'>0.<#004699'>0<#004699'>02Domestic competition54<#004699'>0.<#004699'>025<#004699'>0.<#004699'>015<#004699'>0.<#004699'>021Test statistic for difference??1.43??1.8<#004699'>0??1.53EE countries58<#004699'>0.<#004699'>032<#004699'>0.<#004699'>026<#004699'>0.<#004699'>025CIS countries24??<#004699'>0.<#004699'>018??<#004699'>0.<#004699'>03<#004699'>0??<#004699'>0.<#004699'>014Test statistic for difference2.683.<#004699'>052.<#004699'>08EE, import competition2<#004699'>0<#004699'>0.<#004699'>036<#004699'>0.<#004699'>029<#004699'>0.<#004699'>028CIS, import competition8??<#004699'>0.<#004699'>08<#004699'>0??<#004699'>0.<#0046 99'>087??<#004699'>0.<#004699'>096Test statistic for difference3.413.523.59EE, domestic competition38<#004699'>0.<#004699'>03<#004699'>0<#004699'>0.<#004699'>024<#004699'>0.<#004699'>023CIS, domestic competition16<#004699'>0.<#004699'>014??<#004699'>0.<#004699'>0<#004699'>02<#004699'>0.<#004699'>017Test statistic for difference<#004699'>0.461.21<#004699'>0.3<#004699'>0competition in the CIS and EE countries,The findings on the effect of import <#880000'>com-suggesting that foreign competition has apetition deserve special attention. In the CIS,large positive effect in Eastern Europe andimport competition has a large negative effectthe Baltics and a large negative effect in thein economic terms, although this effect is sta-CIS countries. The last panel shows thattistically not robust. In Eastern Europe, im-there are no discernible patterns in the wayport competition has a positive effect in eco-in which the effects of domestic marketnomic terms, but the results of the individualstructure differ between the CIS and EEstudies are mixed, consistent with the litera-countries.ture on other developing economies. Article 1 8/12/<#004699'>02 9:13 AM Page 768768Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)6.Soft Budgets(1994) and Boycko, Shleifer, and Vishny(1996). These papers model bargaining be-The concept of soft budget constraintstween politicians and managers, which leadswas coined by Kornai (1979, 198<#004699'>0) to explainto equilibria with subsidies to firms and ex-shortages in socialist economies, especiallycess employment. Politicians use govern-in the market socialist system. Kornai de-ment subsidies to induce firms to maintain ascribes the soft budget phenomenon ashigher-than-efficient level of employment in“firms are bailed out persistently by stateorder to enlarge their own political con-agencies revenues do not cover costs”stituency. Since the model predicts that when sub-(Kornai 1998, p. 12) and defines soft bud-sidies increase when there is a decline in en-gets as “the expectation of the decision-terprise profits, these political decisionsmaker as to whether the firm will receive35constitute a soft-budget policy regime.help in time of trouble ... ” (Kornai 1998,These two theories of the causes of softp. 14). Stiglitz further narrows the definitionbudgets differ significantly. The first explainsto situations when “enterprises believe thataccommodating lending behavior deter-any losses they incur will be made good bymined by concerns of social stability, whilethe government” (Stiglitz 1994, p. 184).the second suggests that soft budgets ariseTwo alternative theories explore the causesfrom politicians’ narrower self-interest. In34of soft budgets.Kornai (1979, 198<#004699'>0) relatesboth, soft budgets compensate the enterprisethe softness of budget constraints to the pa-for keeping surplus employment. The pre-ternalistic attitude of the government, whichdicted effect on enterprise restructuringresults in the accommodation of enterprisefrom soft budgets is lack of productivity im-requests for new finance. Firms are financedprovements and continuation of unprofitableeven when the expected new project’s returnproduction (and nonproduction) activities.is below the real interest rate. The govern-The theoretical literature suggests that anment’s goal is to provide economic securityempirical measure of the incidence of softfor enterprise employees and supply socialbudgets needs to satisfy two criteria. First, itservices (kindergartens, schools, hospitals,has to capture the expectations of enterpriserecreation facilities) in the enterprise.managers of a future government bail-out,A different reason for the existence of softnot the current policy of the government.budgets is advanced in Shleifer and VishnyImprudent behavior of managers that re-sults in poor enterprise performance can oc-34A third theory, relevant for both transition and de-veloped market economies, views soft budgets as thecur in enterprises that have not yet receivedcontinued extension of credit even when the substan-government subsidies but expect to do performance of an already-financed investmentwhen they are in sodard trouble. Also, enterprisesproject has been revealed (Michel Dewatripont andEric Maskin 1995; Maskin 1999). Because of asymmet-that have received subsidies in the past mayric information, even poor projects are initially fi-not expect those to continue in the future,nanced. By the time creditors can observe project qual-perhaps as a result of new government poli-ity they will continue to lend, because refinancing maystill maximize the expected value of funds that can becies. Second, the expectation of governmentrecovered. A fourth theory, advanced in Stiglitz (1994),support is contingent on the enterprise be-suggests that soft budgets arise when financing institu-ing in financial distress. Both Kornai’s theorytions have an incentive to make large gambles. In thismodel, an insolvent bank is willing to invest in a riskyand the Shleifer-Vishny-Boycko model showproject with low or even negative expected payoff, be-that without the fear of imminent job losses,cause the bank will become solvent if the gamble paysoff; and if the gamble doesn’t pay off, the bank will be35no worse off than it was before it made the loan, i.e., itIn Shleifer and Vishny (1994) there are severalwill still be insolvent. We do not discuss these two the-types of equilibria, depending on whether bribes occurories in the context of soft budget constraints in transi-and on the nature of property rights. The effect a de-tion, because the source of soft budgets here is a (pri-cline in of profits is either an increase or no change invate) financial institution, not the government.subsidies, depending on which equilibrium pertains. Article 1 8/12/<#004699'>02 9:13 AM Page 769Djankov and Murrell: Enterprise Restructuring in Transition769i.e., if the enterprise is solvent, there is nois used in Anderson, Lee, and Murrellreason for the government to extend new (2<#004699'>0<#004699'>0<#004699'>0) to study the effect of soft budgets onfinancing to the enterprise.enterprise performance.Few empirical papers meet these criteriaWhile theoretically most appropriate, thefor measuring soft budgets. The study thatexpectational measure used in Anderson etcomes closest to the prescribed measure isal. is not easily replicable in other studiesAnderson, Korsun, and Murrell (2<#004699'>0<#004699'>0<#004699'>0).since it relies on a survey instrument de-They survey 249 enterprises that underwentsigned for studying the soft budgets phe-Mongolia’s mass-privatization program bynomenon. In the absence of such data, othermid-1995, asking enterprise managers whatauthors use quantitative variables that proxyproportion of lost revenues the state wouldfor the expectations of enterprise managers,make up if losses threatened the enterprise’swhile different variables are used to control36ability to maintain its employment level.for the financial distress of the enterprise.The resulting measure of soft budgets is ex-The best example of this approach is Davidpectational in nature, and the expectationsLi and Minsong Liang (1998), who constructare limited to the situation where the enter-indicators of soft budgets from survey dataprise has lost revenues. Almost three-quarterson employment of nonproduction workers,of the surveyed enterprise managers, 73.1investment with below-average rates of re-percent, thought that the government wouldturn, and distribution of bonuses in excess ofdo nothing to help out, 2.4 percent of man-these regulated by the Chinese government.agers thought that the government wouldAll three factors are shown to contribute tobail them out completely, and 13.6 percentthe financial distress of the enterprise. Usingof managers thought that the governmenta sample of 681 enterprises for the periodwould make up at least half of the decline in198<#004699'>0 to 1994, Li and Liang find that if all ofrevenues. The perceived degree of budgetthe non-production workers, defined assoftness was shown to be strongly correlatedworkers with zero marginal product (such aswith the amount of central governmentcleaners, political counselors, etc.), wereownership. Almost half of the enterpriseseliminated, 38 percent of financial losseswith majority central government ownershipwould be avoided. As another example, ex-expected some financial assistance, whilecessive bonuses accounted for 39 percent of31.3 percent of the managers of these enter-total enterprise losses. The empirical ques-prises expected that the government wouldtion that the authors ask is whether enter-make up least half the decline in rev-prises respond to financial losses, at defined asenues. The data on managers’ expectationsnegative cash flows, by reducing any of thefactors that cause these losses initially. The36answer is No. All enterprises in financialThe question was phrased as follows: “Suppose thatunfortunate market conditions resulted in a sudden dropdistress expected to be bailed out by thein your enterprise’s revenues, so that you might have togovernment.lay off workers. How likely is it that the government (ei-Several other papers illustrate the diffi-ther national or local) would help your enterprise out, sothat it would not be forced by its financial situation to layculties in constructing a soft budget proxy inoff workers? Please indicate your expectation of thethe absence of survey data. These studieslikely government reaction by choosing a point on thescale from <#004699'>0 to 1<#004699'>0—a “<#004699'>0” means that you think the gov-rely on data from the financial statements ofernment would do absolutely nothing to help out and aenterprises. Wafa Abdelati and Stijn“1<#004699'>0” means that you think the government would <#880000'>com-Claessens (1996) and Fabrizio Coricelli andpletely make up for the decline in revenues in some way,Simeon Djankov (2<#004699'>0<#004699'>02) use samples of overand a “5” means that the government would make uphalf the decline in revenues. Choose any number be-four thousand Romanian companies and dis-tween <#004699'>0 and 1<#004699'>0, indicating your expectation concerningtinguish among chronic loss-makers and ad-the extent to which the government would help out”justed enterprises. Chronic loss-makers areAnderson, Korsun, and Murrell (2<#004699'>0<#004699'>0<#004699'>0, p. 222). Article 1 8/12/<#004699'>02 9:13 AM Page 77<#004699'>077<#004699'>0Journal of Economic Literature, Vol. XL (September 2<#004699'>0<#004699'>02)defined as enterprises with negative cashanalysis does not capture the financial dis-flows during the 1991 to 1994 period.tress of the enterprise, making the resultsAdjusted enterprises are defined as enter-difficult to compare with the theories softprises that had negative cash flows in 1991budgets. For example, of product specific sub-or 1992 or both, but then recorded positivesidies, often the result of price controls, cancash flows in 1993 and 1994. The proxy forbe given to both profitable and loss-makingsoft budgets is the amount of new bank fi-enterprises. Studies using such data docu-nancing, net of accrued interest expenses.ment the effect of distortions due to priceSince the banking system was fully in statecontrols as much as the incentives of softhands in 1994, this proxy measure capturesbudgets.the major channel of government funding,Earle and Estrin are in good company.as tax arrears and direct government subsi-Claessens and Djankov (1998), Djankov anddies are shown to be small relative to theHoekman (2<#004699'>0<#004699'>0<#004699'>0), Grigorian (2<#004699'>0<#004699'>0<#004699'>0), andflow of new credit. The authors seek to an-Carlin et al. (2<#004699'>0<#004699'>01) use similar variables asswer the question whether chronically loss-proxies for soft budgets. This pattern is inmaking enterprises received financial sup-part explained by the fact that all five stud-port in order to keep employment high asies focus on the effects of ownershipthat of profitable as enterprises. The answer ischanges and use the soft budgets proxy onlyYes. In fact, during the sample period, loss-as a control variable. We include these stud-making enterprises received about 5<#004699'>0-per-ies in the a share of survey since they provide valuablecent more bank credit as revenuesinformation on the incidence of govern-and shed less labor than did adjusted firms.ment financing of loss-making enterprisesAssuming that the government would con-in transition economies, and some of the ef-tinue its policy of keeping loss-making en-fects that this financing has. If anything, ourterprises afloat, these studies provide empir-criticism of their approach reflects theical support for the presence of soft budgets.dearth of research focused on the soft bud-Stijn Claessens and Kyle Peters (1997) use agets issue.similar approach to study the effect of softThe studies discussed so far document nu-budget constraints in Bulgaria. They usemerous channels of government support fortrade arrears to other state enterprises, andailing enterprises. While Abdelati andthe share of external financing in total assets,Claessens (1996) and Coricelli and Djankovas alternative proxies for soft budgets. The(2<#004699'>0<#004699'>02) identify new credit from the state-dependent variable is the change in valueowned banking sector as the main conduit ofadded.soft budgets in Romania during the earlyEarle and Estrin (1998) use data fortransition period, and Claessens and Petersabout three hundred Russian enterprises(1997) finds similar results for Bulgaria,during 199<#004699'>0 to 1994 and construct a mea-Mark Schaffer (1998) shows that bank lend-sure of government assistance to enter-ing was not a major source of soft budgets inprises, which they interpret as a proxy forHungary during the early transition period,soft budgets. The measure relies on flow-of-even before the banking sector was privat-funds data and includes all possible chan-ized. Once the Hungarian banking systemnels of government support, such as federalwas largely privatized, by 1995, bank creditsubsidies and investments, tax benefits,was restricted to profitable projectspreferential credits, extra-budgetary funds,(Claessens and Djankov 1998). Tax and socialtax exemptions and other benefits associ-security arrears to the central governmentated with foreign trade, and local govern-were the main source of soft financing inment subsidies and tax arrears. In additionPoland in 1993, while trade credit was not ato relying on a current policy proxy, thesource of soft budgets in either Hungary or Article 1 8/12/<#004699'>02 9:13 AM Page 771Djankov and Murrell: Enterprise Restructuring in Transition77137European countries. Djankov and HoekmanPoland.These results are reflected also in(2<#004699'>0<#004699'>0<#004699'>0) find an unrealized annual growth of 3Claessens and Djankov (1998) who find that,percentage points in Bulgaria over theby the mid-199<#004699'>0s, financing of loss-making1992–95 period. Earle and Estrin (1998) findenterprises by the still state-owned bankinga 5.7 percent unrealized labor productivitysector was the primary channel of soft bud-growth. Alfandari, Fan, and Freinkmangets in Bulgaria, the Czech Republic,(1996) show that recipients of soft financingRomania, Slovakia, and Slovenia, but not inrecord labor productivity growth that is 6 per-Hungary or Poland. Not coincidentally, bycent less than that of non-recipients. Finally,the end of the 199<#004699'>0s all of these countriesAbdelati and Claessens (1996) find that ahad experienced severe (Bulgaria andone-standard-deviation increase in the flow ofRomania) or at least significant (the Czechfinancing from state banks in Romania is as-Republic, Slovakia, and Slovenia) bankingsociated with a 14.2 percent unrealized laborcrises.productivity growth per annum. This evi-Shleifer and Treisman (2<#004699'>0<#004699'>0<#004699'>0) show thatdence suggests that soft budgets are a majortax exemptions b