Committing to Grow: Privatizations and Firm Dynamics in East Germany

Board of Governors of the Federal Reserve System International Finance Discussion Papers ISSN 1073-2500 (Print) ISSN 2767-4509 (Online) Number 1382 November 2023 Committing to Grow: Privatizations and Firm Dynamics in East Germany Ufuk Akcigit, Harun Alp, Andr´e Diegmann, and Nicolas Serrano-Velarde Please cite this paper as: Akcigit,Ufuk,HarunAlp,Andr´eDiegmann,andNicolasSerrano-Velarde(2023). “Committing to Grow: Privatizations and Firm Dynamics in East Germany,” International Finance Discussion Papers 1382. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/IFDP.2023.1382. NOTE: International Finance Discussion Papers (IFDPs) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the International Finance Discussion Papers Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers. Recent IFDPs are available on the Web at www.federalreserve.gov/pubs/ifdp/. This paper can be downloaded without charge from the Social Science Research Network electronic library at www.ssrn.com.

Committing to Grow: Privatizations and Firm Dynamics in East Germany* Ufuk Akcigit Harun Alp André Diegmann Nicolas Serrano-Velarde October 16, 2023 Abstract Thispaperinvestigatesauniquepolicydesignedtomaintainemploymentduringtheprivatization ofEastGermanfirmsafterthefalloftheIronCurtain. Thepolicyrequirednewownersofthefirms tocommittoemploymenttargets,withpenaltiesfornon-compliance. Usingadynamicmodel,we highlight three channels through which employment targets impact firms: distorted employment decisions, increased productivity, and higher exit rates. Our empirical analysis, using a novel dataset and instrumental variable approach, confirms these findings. We estimate a 22% points higher annual employment growth rate, a 14% points higher annual productivity growth, and a 3.6% points higher probability of exit for firms with binding employment targets. Our calibrated model further demonstrates that without these targets, aggregate employment would have been 15% lower after 10 years. Additionally, an alternative policy of productivity investment subsidies provedcostlyandlesseffectiveintheshortterm. Keywords: industrialpolicy,privatizations,productivity,size-dependentregulations. JELclassification: D22,D24,J08,L25. *WethankourdiscussantsChang-TaiHsieh,BrianViard,StefanObernberger,RiccardoZago,andtheseminarandconference participants at Nova Business School, Bocconi University, Halle Institute for Economic Research, NBER Summer Institute, Centre for European Economic Research (ZEW), Federal Reserve Board, Richmond FED, The Society of Labor Economists, UVA, Barcelona School of Economics Summer Forum, CSEF-IGIER Symposium on Economics and Institutions, EDHEC, FEP School of Economics and Management University of Porto, Hitotsubashi University, Lisbon Macro Workshop,andAIEA-NBERConference. Forprovidingvaluablesupportfordataaccessanddataexpertise,wethankSandraGottschalk(MannheimEnterprisePanel),AlexanderGiebler(ISUDdata)andChrisBerthold,AntjeKlünderandJana Michaelis(GermanFederalArchives). AkcigitgratefullyacknowledgesfinancialsupportfromtheMaxPlanckHumboldt- Research Award 2019. The views in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Board of Governors of the Federal Reserve System or any other person associated with the FederalReserveSystem. Akcigit: University of Chicago, Halle Institute for Economic Research (IWH) (uakcigit@uchicago.edu). Alp: Federal Reserve Board (harun.alp@frb.gov). Diegmann: Halle Institute for Economic Research (IWH), Institute for Employment Research(IAB),CenterforEuropeanEconomicResearch(ZEW)(andre.diegmann@iwh-halle.de). Serrano-Velarde: Bocconi University,IGIER(nicolas.serranovelarde@unibocconi.it).

1 Introduction Industrial policy is often designed during times of significant structural change caused by shocks related to competition (e.g., China shock), disruptive innovations (e.g., IT revolution), or political turbulence (as seen in the post-Soviet era in Eastern Europe). During these periods, policymakers consider the immediate costs of labor market disruption and how to mitigate them by introducing policies centered around employment considerations. However, there remains a lack of evidence on how such interventions dynamically affect reallocation and firm behavior. In this paper, we study a novel policy designed to preserve employment during the privatization of East German firms following the fall of the Iron Curtain. The privatization process represented a period of intense structural change and raised significant concerns about the social costs associated withhighunemployment. Inresponse,policymakersrequiredthatnewownersofEastGermanfirms commit to employment targets, with penalties imposed for falling below the committed employment level. In total, these labor commitments were applied to over 18,000 privatization contracts, covering more than 900,000 workers in East Germany. Our analysis proceeds in three steps. First, we introduce a dynamic model where firms operate under employment targets. An important feature of the model is that firms invest resources to improvetheirproductivity,allowingustostudytheendogenousresponseofproductivitytosuchtargets. The model highlights three channels through which a firm is affected by an employment target that is binding i.e., in which the target is higher than the current employment level. The first channel stemsfromthefirm’sstaticlabordecision,whichinducesanupwardlydistortedemploymentchoice. The second channel arises dynamically as binding targets induce higher productivity growth. This is because more productive firms hire more workers in our model, and this structure creates additional incentives to invest in productivity improvements to avoid the penalties. These two channels imply thatfirmswithbindingemploymenttargetsexperiencehigheremploymentandproductivitygrowth. The third channel operates through the extensive margin choice of the firm to exit. Firms with binding employment targets are more likely to exit as binding targets introduce a fixed-cost-like structure in the cash flow of the firm. In the second step, we take these predictions to the data. The empirical analysis relies on a novel datasetfromtheGermanarchivesthatcontainsallthedocumentationproducedbytheTreuhandanstalt (THA), the government agency responsible for the privatization process. Our data contains detailed contract-level information on employment targets and deadlines, as well as the dates and results of eachon-siteauditoftheemploymentcommitment. Tomeasurefirm-levelproductivity,wemergeour contract-level information with data from the Mannheim Enterprise Panel (MUP) and the SOESTRA survey of East German firms. The empirical identification of the link between employment targets and firm dynamics is a challenging task. The reason is that employment targets are not randomly allocated and might thus bias our empirical estimates. In the spirit of the literature on judge leniency (Bhuller, Dahl, Løken, and Mogstad 2020; Dobbie and Song 2015; Bernstein, Colonnelli, Giroud, and Iverson 2019), we develop 1

an instrumental variable (IV) approach that exploits heterogeneous preferences of privatizers and their quasi-random assignment to firms. To do so, we estimate the propensity of a privatizer to require binding labor commitments. We show that: (i) the probability of receiving a binding contract increases continuously along the labor preference measure, (ii) these preferences are heterogeneous across privatizers, and (iii) they are persistent across time. Importantly, we also provide evidence consistent with the quasi-random assignment mechanism of firms to privatizers. To do so, we use information from the balance sheets of firms before their privatization. Consistent with anecdotal evidence about the organization of THA, we find no evidence of an economically or statistically significant correlation of our instrument with a wide range of sectoral characteristics, employment and revenue measures, and other individual characteristics of the privatizers. Consistent with the model’s predictions, we find that binding labor targets are associated with higher employment and productivity growth, as well as increased firm exit over the labor commitment period. Our IV estimates reveal a 22% points higher annual employment growth rate for firms withbindinglaborcontractscomparedtothosewithout. Bindinglaborcontractsalsoleadtoanadditional yearly productivity growth of approximately 14% points. Additional evidence based on firms’ patenting activity during the commitment period also supports these findings. Furthermore, firms with binding contracts exhibit, on average, a 3.6% points higher probability of exiting by the end of thecommitmentperiod. Relativetothebaselineexitrateof5.5%,thisrepresentsaneconomicallysizable increase in the exit margin. We show that these results are robust to alternative specifications in terms of the measurement of the dependent variables, the construction of the instrumental variable, and the inclusion of additional contractual characteristic as control variables. In the last step of our analysis, we calibrate our model to the data and run several counterfactual scenarios in order to quantitatively assess the importance of the different channels on firm behavior. Toidentifytheparametersofthemodel,especiallythepenaltyofnotmeetingthetarget,wematchthe effects of binding employment commitments on firm outcomes uncovered in our empirical analysis. The calibrated model is able to reproduce the main patterns in the data well. Importantly, the model replicates firm-level growth patterns across the employment commitment distribution as well as the post-commitment employment dynamics, which are not targeted in the calibration process. We study three counterfactual economies. We first simulate an economy without employment targets and find that aggregate employment would be 15% points lower permanently after 10 years. Next, we decompose the impact of employment targets on total employment into its “static” and “dynamic” components by shutting down its impact on productivity improvements. Our calibrated model attributes one-third of the employment growth to dynamic effects in the short run. In the long run, the entire permanent increase in employment is driven by the dynamic effects. Lastly, we consider an alternative policy of subsidizing investment into productivity. We calibrate the subsidy rate to achieve the aggregate employment growth in the data during the commitment period. The implied cost of such a policy is high, amounting to 5% of the output. While this policy results in higher permanent employment levels relative to the employment target policy, the increase is more gradual over the commitment period. In other words, the subsidy policy is less effective in the short 2

run to preserve employment. The paper contributes to the recent literature revisiting the merits and costs of industrial policies. Lane (2022) and Choi and Levchenko (2021) use historical data to study the dynamic impact of the South Korean heavy and chemical industry drive from 1973 to 1979. Lane (2022) shows that this temporary drive shifted Korean manufacturing into more advanced markets, creating durable industrial change. Choi and Levchenko (2021) link the associated firm-level subsidies to persistent effects on firm size due to a combination of learning-by-doing and financial frictions. Kalouptsidi (2018) and Barwick, Kalouptsidi, and Zahur (2021) study the Chinese intervention in the shipbuilding industry. Kalouptsidi (2018) estimates that policy interventions reduced shipyard costs by 13-20% and reallocated international market shares. Barwick, Kalouptsidi, and Zahur (2021) disentangle the various subsidies during the intervention and estimates their impact. Acemoglu, Akcigit, Alp, Bloom, and Kerr (2018) demonstrate that strategic industrial policies have significant influence over firms’ composition, with the potential to harness the economy’s firm selection process to amplify overall productivity gains. Liu (2019) embeds industrial policy in a production network setting and applies it to interventions in South Korea in the 1970s and modern-day China. Finally, Giorcelli and Li 2021 estimate the long-term effects of technology and know-how transfers on China’s structural transformation using data from the Sino-Soviet alliance in the 1950s.1 Similar to these studies, we leverage unique historical microdata to develop identification strategies and causally estimate the dynamic impact of an industrial policy. At the same time, the type of policy we analyze is unique, insofar as it primarily puts constraints on firms as opposed to using subsidies. By temporarily constraining firms to keep a larger size, the policy produces strong incentives to improve productivity.2 The paper contributes to the important and long-standing debate on state versus private ownership (Megginson and Netter 2001). From the point of view of economic efficiency, the case for privatization lies in the concentration of control rights and cash flow rights in the hands of outside investors. Inthisway,afirm’snewownersareprovidedwithoptimalincentivestodisciplinemanagement and restructure activities (Barberis, Boycko, Shleifer, and Tsukanova 1996; Djankov and Murrell 2002; Dyck 1997; López-de Silanes 1997). At the same time, the case for privatisation leaves open the possibility for a welfare-maximizing government to address social and strategic goals through contracting and regulation (Shleifer 1998; Besley and Ghatak 2001). Our paper provides novel evidence on this latter point, as we study how the German government used a contractual approach to force buyers to internalize concerns about employment. The paper also contributes to the recent literature on the consequences of size-dependent regu- 1GlitzandMeyersson(2020)studieshowindustrialespionagebyEastGermanyaffectedTFPgapswithWestGermany. See also Harrison and Zaksauskiene˙ (2016) for a study of the role of the secret police as a market regulator in Soviet Lithuania. 2Note that the objectives and tools of industrial policies can be wide-ranging. In the context of innovation policies, a combinationoftaxes, subsidies, andregulationcanbedirectedatpromotingtechnologicalchangeandspecificindustries (Aghion, Dechezleprêtre, Hemous, Martin, and Van Reenen 2016; Acemoglu, Akcigit, Hanley, and Kerr 2016). Similarly, persistent gaps in economic performance across regions have prompted governments to create a variety of place-based economic development policies (Criscuolo, Martin, Overman, and Van Reenen 2019; Greenstone, Hornbeck, and Moretti 2010; KlineandMoretti2014). SeeJuhász,Lane,andRodrik2023foracomprehensiveliteraturereviewondevelopments intheeconomicsofindustrialpolicy. 3

lations that frequently favor smaller firms. These policies can potentially create distortions in the economy that affect aggregate productivity by misallocating resources toward less-productive firms (RestucciaandRogerson2008;HsiehandKlenow2009;Bartelsman,Haltiwanger,andScarpetta2013; Syverson 2011). Garicano, Lelarge, and Van Reenen (2016) leverage size-contingent laws in France to identifytheequilibriumandwelfareeffectsoflaborregulation. Braguinsky,Branstetter,andRegateiro (2011) document how the entire Portuguese firm size distribution has shifted over time to the left. They attribute this process to strong protections for regular workers. Martin, Nataraj, and Harrison (2017) use the elimination of small-scale industry promotion in India to study firm dynamics. The dismantlingofthesepoliciesleadsnotonlytoincreasedentry,butalsohigheroutputgrowthinmore exposed districts. We also analyze a policy distorting the firms’ employment decisions, but focus on its dynamic implications, similar to recent papers by Aghion, Bergeaud, and Van Reenen 2023 and Akcigit, Alp, Akgunduz, Cilasun, and Quintero 2023.3 We contribute to this literature by analyzing a unique policy creating distortions that are not only firm-specific but also actively push firms to grow or operate at a larger size. Consequently, the trade-offs generated by this policy are distinct from those introducing barriers to growth. We exploit this unique institutional setting to provide a comprehensive evaluation of its dynamic impact on firm employment, productivity and exit. Finally, our paper contributes to the understanding of the transition of former Eastern Block economiesandtheprocessofeconomicconvergencewithWesterncountries(DeLoeckerandKonings 2006).4 Similar to other Eastern European countries, East Germany started out with a lower level of a variety of macroeconomic indicators such as economic freedom, GDP per worker, nominal wages, and labor productivity (Lipschitz and McDonald 1990; Akerlof, Rose, Yellen, Hessenius, Dornbusch, and Guitian 1991; Fuchs-Schündeln, Nicola and Schündeln, Matthias 2020). Over the years following reunification, the Eastern German economy started to converge in many dimensions. The process of economicconvergencestudiedintheliteraturerangesfromitsimplicationsformanagementpractices (Dyck 1997), to labor reallocation (Dauth, Lee, Findeisen, and Porzio 2021; Fuchs-Schündeln and Schündeln2005)andmigration(Uhlig2008;Hunt2006;Peters2022;ReddingandSturm2008),capital investmentsfromWestGermany(Sinn2002),collectivebargainingagreements(BurdaandHunt2001; Burda 2010), as well as social and cultural ties (Alesina and Fuchs-Schündeln 2007; Burchardi and Hassan2013).5 Weshowinthispaperthattheimplementationofstrategicgoals–intheformoflabor commitments – set by the government can contribute to higher productivity growth by generating a dynamic catch-up in productivity of firms. 3Thesepapersstudy,throughthelensofendogenousgrowthmodels,size-dependentregulationsthatintroducebarriers togrowth. WhilethefirstpaperfocusesonadevelopedcountrysettinglikeFrance,thesecondpaperstudiesadeveloping countrysettingandconcernshowregulationsinteractwithinformalityinlabormarkets. 4JohnsonandPapageorgiou(2020)providearecentsurveyaboutthelargeliteratureineconomicconvergencebetween countriesingeneral. 5Theconvergenceprocessstartedtoleveloffandstagnateattheendofthe1990s. Thisobservednon-convergenceover thelast20yearsreceivedincreasingattentionintheliterature. SnowerandMerkl(2006)emphasizetheroleofgovernment transfers in explaining persistent unemployment gaps between East and West Germany, while Burda (2006) argues for capital accumulation frictions as a driver of slow labor productivity convergence. More recently, Heise and Porzio (2021) provide evidence for low labor mobility between East and West Germany. Bachmann, Bayer, Stüber, and Wellschmied (2022)relatehighermonopsonypowertolowerproductivityconvergence. 4

The remainder of this paper is organized as follows. Section 2 provides an overview of the German institutional framework after reunification. Section 3 introduces a simple model of firm growth to study the dynamic implications of employment targets. In Section 4, we describe the data and provide descriptive statistics. Section 5 reports our empirical results. The structural model and the quantitative estimation are shown in Section 6. Section 7 concludes. 2 Institutional Background Established in March 1990 under the last Communist regime in East Germany, the THA was given greaterauthorityinJuly1990throughtheTreuhandgesetz.6 Taskedwithmanagingandprivatizingthe companiesthathadpreviouslybeenownedbythestateofEastGermany,theTHAbecamethelargest holding company in the world, overseeing a portfolio of around 12,000 companies and employing approximately 4.5 million people, which made up about 50% of the total workforce population. The THA officially commenced its duties on July 1, 1990. The THA management inherited a diverse and often disjointed portfolio of activities structured withinlargecentrallyplannedconglomerates(Kombinate). Theinitialsteptakentowardstransforming these companies involved the splitting up of these large conglomerates into firms organized under privatelaw(Entflechtung). Inasecondstep,theTHArequiredtheseenterprisestosubmitanopening balance sheet in Deutsche Mark (Eröffnungsbilanz) and a business plan for review. The privatization process further streamlined business activities, as firms divested through asset sales. THA itself built up rapidly from an initial staff of about 200 mostly East German employees to an institution of around 4,000 employees plus 800 full-time consultants. These people were divided approximatelyequallybetweenthecentralofficeinBerlinandthe15branchofficesdistributedamong themajorcitiesofthenewfederalstates. Figure1representstheformerGermanDemocraticRepublic (GDR)districtsandthelocationofTHAbranchoffices. Smallerfirmswithfewerthan1,500employees were assigned to local branch offices, while larger firms were assigned to industry-based divisions in the Berlin headquarters. In particular, the THA headquarters organized firms with more than 1,500 employees or with revenue or balance sheet values above 1.5 million Deutsche Mark (DM).7 TheTHAutilizeddirectcashsalestoprivatizeassets,whichincludedbothprivatizationsofentire firms and divestitures/spin-offs resulting from firm restructuring and liquidation. Sales contracts were structured to include the sales price as well as potential guarantees made by purchasers regarding minimum levels of employment and investment. The imposition of employment targets reflected the obligation placed on the THA to take account of the social costsof unemployment. Following the reunification, East Germany experienced a sharp increase in the unemployment rate, reaching 10.2% by 1991 and further rising to 15.7% in 1994, which led to significant social unrest. In April 1991, the first president of THA, Detlev K. Rohwedder, was assassinated. Similarly, after the THA closed the 6GesetzzurPrivatisierungundReorganisationdesvolkseigenenVermögensofJune17,1990. 7Exceptionsfromthecutoffrulesare(i)ifthetotalsumoffirmsubsidiariesisabove1,500employeesand(ii)ifthefirm belongstooneofthefollowingsectors: foreigncommercebusiness,financialinstitutions,printingandnewspaper,DEFA, hotelsandtravelagencies,circuses,waterandsewage,energyandmining,transportation. 5

Figure1: THAHeadquartersandSubsidiaries Notes: ThefigureshowsthelocationoftheTHAheadquartersandlocalsubsidiariesacrossEastGermany. IncludingEast Berlin,theformerGDRconsistedof15district. EachdistrictpossessesalocalTHAoffice. Theheadquartersislocatedin EastBerlin,whichisindicatedbytheredcross. former VEB Kaliwerk Bischofferode, the employees went on an 81-day-long hunger strike (Bernhard 2011). The employment commitment typically consisted of an agreed number of full-time equivalent jobs that should be maintained for an agreed period of time. This commitment was specific to the acquired establishment and could not be fulfilled by employing individuals in other establishments of the acquirer (Siebert 1991; Fischer, Hax, and Schneider 1993). While these targets could result in discounts on the sales price, the valuation process for these commitments varied for each case, and therewasnofixedformulafollowedbytheTHA(DoddsandWächter1993). Toensureenforceability, penalty clauses were included in the contracts, stipulating payments to the THA if the agreed-upon employment levels were not met. The penalties were designed to approximate the cost of retaining an employee and were proportional to the shortfall in employment and prevailing industry wages. The employment target was subsequently monitored through multiple audits organized within the contract management system of the THA organization. 3 A Model of Firms with Employment Commitments Inthissection,webuildasimplemodeloffirmgrowthtostudytheimplicationsofsuchemployment targets implemented by THA. We consider an economy in continuous time, populated by a large 6

number of heterogeneous firms in productivity producing a homogeneous good. One of the main features of our model is that firms grow through improving their productivity through investment, which allows us to study the dynamic consequences of operating under such targets. At any point in time,firmschoose(i)theamountoflabortohireforproduction,(ii)howmuchtoinvestinimproving firm productivity, and (iii) whether to exit the economy or not. We further assume that labor supply is perfectly elastic and wage growth is exogenous.8 3.1 Static Environment Firms are endowed with a production technology that features decreasing returns to scale with respect to labor: y = z1−αlα , 0 < α < 1 t,j t,j t,j where z denotes the level of productivity at firm j at time t, which is heterogeneous across firms, t,j and l is the amount of labor hired. Firms take the wage rate w as given. Firms operate under t,j t perfect competition and the price of the homogeneous good is normalized to be one, without loss of generality. In what follows, we drop the time subscript t whenever it does not cause any confusion. Firms operate under employment targets, l∗, which are heterogeneous across firms and exogej nously set by the policy makers. Consistent with the institutional framework, firms pay a penalty if they operate below the target level of employment and the penalty is proportional to the missing amount of employment: (cid:16) (cid:17)+ γ l ∗−l w j j where γ is a parameter that controls the amount of penalty per missing employee as a fraction of the wage rate. Given this structure, the firm’s static profit maximization problem is given by (cid:110) (cid:111) Π(z ,l ∗) = max z1−αlα−wl −γ(l ∗−l )+ w . j j j j j j j lj ≥0 The next lemma describes the solution to the static profit maximization problem. Lemma 1 The optimal labor decision for a firm with productivity level z and employment target l∗ is given by    α1− 1 αz˜ if z˜ > z˜∗ ≡ l∗ 1 (Undistorted)   α1−α (cid:16) (cid:17) 1 l(z˜,l ∗) = 1− α γ 1−α z˜ if z˜ < z˜∗∗ ≡ l∗ 1 , (Distorted, No Bunching) (1)    ( 1− α γ )1−α  l ∗ if z˜ ∗∗ ≤ z˜ ≤ z˜∗, (Distorted, Bunching) 8Wethinkthisisareasonableassumptiongiventhatemploymentcommitmentscoveredapproximately20%oftheEast German workforce and the high prevailing unemployment rates. Moreover, this is a period when the wage setting was mainlydrivenbypoliticalconsiderationsratherthanmarketforces(KruegerandPischke1995; Hunt2001), whichmakes thewageimpactoftheemploymenttargetpolicieslessrelevant. 7

and the implied profits are Π(z˜,l∗) = π(z˜,l∗)w where   α1− α α (1−α)z˜ if Undistorted  (cid:16) (cid:17) α π(z˜,l ∗) = 1− α γ 1−α (1−α)z˜−γl ∗ if Distorted,No Bunching (2)   z˜1−αl ∗ α−l ∗ if Distorted, Bunching where z˜ ≡ z/w1− 1 α is the normalized productivity level with respect to the wage rate. The above lemma suggests that firms’ optimal employment choice and the resulting profits are positively correlated with the level of (relative) productivity, z˜, and can be characterized based on three productivity regions. For high-productivity firms, z˜ > z˜∗, the labor choices are not distorted, since the employment target is not binding. Firms with intermediate levels of productivity, z˜ ≤ z˜ ≤ ∗∗ z˜∗,decidetobunchatthetargetemployment,whichishigherthantheiroptimallevelofemployment if there were no employment targets. Finally, low-productivity firms, z˜ < z˜∗∗, find it too costly to operate at the target level of employment, but still their labor choices are distorted towards the target level. These labor choices underline the first channel through which firms with binding contracts i.e., having an employment target larger than the optimal employment level under no target experience a higher employment growth through the contract period, as their employment is simply distorted upward towards the target. We refer to this channel as the "direct" effect of binding labor contracts on employment growth. Figure2illustrateskeyimplicationsoftheexistenceofemploymenttargetsonfirmprofits. Theleft panelplotstotalprofitswithrespecttofirm-levelproductivity. Theblacklineprovidesthebenchmark for firms with no commitment, while the dashed red line plots profits for firms under commitment. Dashedverticallinesshowthethresholdproductivitylevels,z ∗ andz ∗∗. Theplotshowsthatdistorted firmshavelowerprofits. AsEquation2clarifies,theselowerprofitsareattributabletothepresenceof penalties that introduce a fixed-cost-like structure. These penalties increase in magnitude as the level of the target rises. This will have important dynamic implications for the exit decision, which will be discussed in the next subsection. TherightpanelofFigure2plotsmarginalprofitsacrossfirmproductivityandshowsthatdistorted firms have a higher marginal profit with respect to productivity. This is intuitive: increasing productivity not only increases profits but also reduces the amount of distortion (if the firm is bunching) or penalty paid (if the firm is not bunching) for those firms with binding contracts. In a dynamic setting, which will be introduced later, this implies that the increase in profits from productivity improvements will be higher for distorted firms relative to undistorted ones i.e., distorted firms would bemorewillingtoinvestinproductivityimprovements. Sincethelaborchoiceispositivelycorrelated with the level of productivity, this higher productivity growth by binding-contract firms constitutes the second channel for higher employment growth through its dynamic implications on productivity growth. 8

Figure2: ProfitsacrossFirmProductivity PanelA:TotalProfits PanelB:MarginalProfits 0.13 No commitment 0.55 0.125 With commitment 0.5 0.12 0.45 0.115 0.4 s s tifo rP 0 0 . . 0 2 3 . 5 5 3 tifo rP la n ig ra 0. 0 1 . 0 0 1 . 5 1 1 M 0.2 0.095 0.15 0.09 0.1 0.085 No commitment 0.05 With commitment 0.08 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 Productivity Productivity Notes: Theleftandrightpanelsplotstotalandmarginalprofitsacrossfirm-levelproductivity,respectively. Theblackline providesthebenchmarkforfirmswithnocommitment,whilethedashedredlineplotsprofitsforfirmsundercommitment. Dashedverticallinesshowthethresholdproductivitylevels,z∗ andz∗∗. 3.2 Dynamics Next, we describe the dynamic decisions of the firms. At any point, the owner decides whether to stay in the economy or exit. If she decides to exit, she needs to pay an exit cost, net of outside option, which we parameterize with C .9 If she stays in the economy, she makes the optimal labor e choice, as described above, and decides how much to invest in productivity growth by choosing the Poissonarrivalrateofimprovingtheproductivity, x,withthefollowingcostfunction(intermsofthe homogeneous good) φ c(x|z˜) = x2z˜w 2 which is convex in the success probability x, and φ is the scale parameter for the cost. This cost function assumes that the higher the current level of productivity, the higher the cost of investment. Theparticularnormalizationofthecurrentlevelofproductivityimpliesthatfirmgrowthisconsistent with Gibrat’s law in the absence of employment targets: the growth rate of sufficiently large firms (high productive firms) is independent of their size. If the investment is successful, the productivity improves from z to (1+λ)z, where λ is the parameter that controls the step size in productivity improvement. Finally, we assume that the labor commitment contracts expire at the firm level at the rate µ i.e., the employment target becomes zero and no longer binding. Given this structure, the dynamic problem of the firm can be represented by the following value function: 9This cost reflects not only the penalties from missing the target as employment becomes zero upon exiting, but also anyother,implicitorexplicit,costsduetoimpairedrelationsbetweentheacquirerofthefirmandthegovernment. 9

∂V(z˜,l ∗ )    π(z˜,l ∗ )w− φ 2 x2z˜w   rV(z˜,l ∗ )− = max −C e w,max +x[V(z˜(1+λ),l ∗ )−V(z˜,l ∗ )]  (3) ∂t  x≥0 +µ[V(z˜,0)−V(z˜,l ∗ )]  where V(z˜,l ∗ ) is the firm value. The outer maximization problem determines the endogenous exit decision of the firm. The value of staying is determined in the second maximization problem where the firm chooses how much to invest on productivity growth.10 The first line includes the instantaneous profits, minus the cost of investment on productivity. The second line expresses the changeinfirmvaluewhenthefirmissuccessfulwithitsinvestmentinimprovingproductivityatthe rate x. The last line represents the change in value when the labor commitment contract expires at the rate µ. The extensive margin choice above gives rise to the standard optimal stopping problem. Firms follow a cutoff rule under which they choose to exit when their productivity falls below a certain threshold. The threshold productivity for exit is higher for firms with higher employment targets duetotheassociatedlargerfixedcosts. Inotherwords, conditionaloninitialproductivity, firmswith higher employment targets would be more likely to exit the economy. For firms that choose to stay in the economy, optimal level of investment in productivity is given by (the arrival rate of improving productivity): V(z˜(1+λ),l ∗ )−V(z˜,l ∗ ) x(z˜,l ∗ ) = (4) 1 φz˜1−α which depends on the increase in the value of the firm in the case of a successful improvement in productivity. Since the value function inherits the properties of the profit function, the investment rate on productivity mimics the pattern of marginal profits similar to the case illustrated in Panel B of Figure 2: it is higher for firms that are distorted by the binding employment targets. 3.3 Taking Stock We finish the model discussion by summarizing the main insights and predictions. Our model clarifies three channels through which binding employment commitments affects firm decision. The first channel stems from the firm’s static labor decision under binding contracts, which induces an upwardly biased employment choice (equation 1) i.e., labor hoarding towards the target employment level. This results in a transitory expansion in employment, with firms reverting to their undistorted size once the policy expires.11 The second channel arises dynamically as firms operating under these binding commitments witness higher productivity growth induced by higher marginal profits fostering investment in productivity improvements. This dynamic effect arises as firms seek to “escape” fromcontractuallyimposedpenalties. Unlikethetransitorynatureofthefirsteffect, theemployment gainsresultingfromthesedynamicimprovementsinproductivityarepersistentandcontinuebeyond 10Employmentchoicewascharacterizedabove,soitistakenasgivenhere. 11Forsimplicity,ourmodelabstractsfromlaboradjustmentcosts. Incorporatingsuchcostswouldnotaffectthetransitory natureofthischannelbutwouldimplysometransitionphaseoccurringwhenthecommitmentexpires. 10

the commitment period. The third channel operates through the extensive margin choice of the firm to exit. Firms with binding employment targets are more likely to exit over the commitment period as binding targets introduce a fixed-cost-like structure in the cash flow of the firm. Inthenextsection,wewillempiricallytestthesetheoreticalimplicationsofthemodelinthedata. In particular, we look at the impact of binding employment contracts on (i) employment growth, (ii) productivity growth, and (iii) exit decision at the firm level. 4 Data and Descriptive Statistics 4.1 Contract Data The analysis relies on a unique dataset from the German Federal Archives (Bundesarchiv) containing all contracts and documentation produced by the THA. These data are confidential, and, thanks to an institutional cooperation with the IWH Institute, we are among the first to gain access to them. Importantly, the agency digitally recorded all contracts for monitoring and enforcement purposes in morethan500datatables(ISUDSystem). Thesetablesprovidecomprehensivecontract-levelinformation on the privatization of assets including all the employment commitments that have been agreed upon as well as dates and audits associated with each commitment. Appendix Section B provides a detailed description of the explored ISUD data tables. The dataset contains 18,235 contracts with labor commitments. For each contract, we observe the contractdatee.g.,thedatethecontractissignedwithanotaryandthecommittedlevelofemployment along with the date of the final commitment. As shown in Panel A of Figure 3, 90% of employment commitments are signed between 1991 and 1994.12 In total, all labor contracts amount to more than 900,000committedworkers,representingabout20%oftheinitialworkforcepopulationofTHAfirms. ThesecontractsunderwentregularauditsconductedbycontractmanagersemployedbytheTHA. Theseauditsserveasthebasisfortherealizedemploymentlevelsanalyzedinthispaper. Thecontract managers would approach the contracting party and conduct audits either through physical visits to the firm or via documentation. On average, each contract was audited 6.3 times, with a minimum of 1 audit and a maximum of 84 audits. Figure 3, Panel B, illustrates that approximately 82% of all contracts were audited at least twice. We focus on these contracts to measure employment growth during the commitment period. While we always have data on employment levels at the final commitment date, we approximate the initial employment levels using the first labor audit, which typically took place between three to six months after the contract was signed with the notary. Panel A of Table 1 shows contract-level employment information at the start date of the contract, the final level, and the target level. We provide descriptive statistics for the 14,726 contracts with at least two audits. With, on average 66 employees, firms had been relatively sizable at the onset of privatization. Over the course of the commitment period of, on average, three years, firms decreased their size. Panel B relates the initial 12Lessthan2%ofalllaborcontractsarewrittenoutin1996orlater(15contractsareobservedin2002). For168contracts wedonotobservethedateofthecontract. 11

sizetothefinaltarget. Thefractionoffirmsinitiallybelowtheirtargetis22%,whileabout20%ofthe firms receive a target that is equal to their initial size. In 10% of cases the firm stays at the committed size in the first and last audits. Figure3: ContractsandLaborAudits A:Numberofcontractsovertime stcartnoc fo rebmuN 0006 0004 0002 0 0001 008 006 004 002 0 )0001 ni( tnemyolpme dettimmoc fo rebmuN B:Auditdistribution 1990 1992 1994 1996 1998 2000 Year Contracts Labor target tnecreP 2. 51. 1. 50. 0 1 5 10 15 20 25+ Total Number of Audits per Contract Notes: PanelAplotsthetotalnumberofsignedcontractswithlaborcommitmentsbetween1990and2000aswellasthe accumulatednumberofcommitmentemployment. PanelBplotsthedistributionoflaborauditspercontract. For a subset of 1,272 firms, we observe the total amount of penalties claimed by the THA due to violations of labor commitments as well as the total number of violations. Based on these numbers, wecalculatethepenaltypermissedemployeetakingintoaccounttheproratatemporiscondition. This meansthat,forexample,ifafirmismissingcontinuouslyoneemployeeoverthecourseofthreeyears, thefirmmisses, intotal, threecommitmentsandneedstopaythreetimesoneemployee. Conditional on having at least one labor violation, the average firm deviates 2.2 times. The cumulative number of missed employment over multiple violations corresponds, on average, to 111 workers.13 Finally, consistent with documentation on THA policy, our calculations suggest that the average penalty per missed employee amounts to 10,768 EUR.14 Figure 4 empirically assesses the importance of employment targets in affecting firm’s labor choices. The horizontal axis measures the difference between the realized employment measured at the last audit of a commitment and the final employment target. Firms below 0 are smaller in terms of their realized employment relative to the committed level, whereas firms above 0 have a larger employment with respect to their committed level. The figure plots the bin counts around the normalized target shown by the red vertical line at zero, with each bin representing a unit of 13Themaximumcumulativemissedemploymentamountsto8,567workers. ThisisabovethemaximumofthefinalemploymenttargetinPanelA,astherecanbemultipleviolations. Inaddition,themaximumnumberinPanelAcorresponds tothefinalcommitmentlevel. 14Theeffectivepenaltycanbelowerbecauseof“conditionsbeyondthepurchaser’scontrol”(DoddsandWächter1993), renegotiation,minorviolationsoftargets(Bagatellfall),judicialdecisions,andbankruptcy. 12

Table1: SummaryStatistics N Mean SD Minimum Maximum (1) (2) (3) (4) (5) A:Averagefirmsize Initialemployment 14,726 66.20 319.57 0.00 23,691 Finalemployment 14,726 60.67 194.26 0.00 8540 Finalemploymenttarget 14,726 52.98 183.25 1.00 6906 B:Initialsizerelativetotarget Fractioninitiallybelowtarget 14,726 0.22 0.42 0.00 1.00 Fractioninitiallyattarget 14,726 0.20 0.40 0.00 1.00 Fractioninitially&finallyattarget 14,726 0.10 0.30 0.00 1.00 C:Penalties Numberofobservedviolation 1,272 2.24 1.29 1.00 12.00 Totalnumberofviolatedlabor 1,272 111.58 393.22 0.24 8,567.47 Penaltypermissedemployee(in1000EUR) 1,272 10.77 10.67 0.10 58.52 D:Productivity Initialproductivity 3,387 9.99 1.43 3.51 16.27 Finalproductivity 3,584 12.02 0.786 10.46 14.81 InitialTFP 3,118 6.81 1.23 2.70 9.79 FinalTFP 2,219 7.32 1.08 3.53 10.31 E:Marketexit Exituntilfinalcommitmentyear 4,622 0.055 0.22 0.00 1.00 Notes: Thetableshowssummarystatisticsofprivatizationcontracts. InPanelA,initialemploymentleveliscalculatedfor contractswithatleasttwoobservations. Thiscorrespondsto14,726contracts. InPanelC,weobserve1,272contractswith atleastoneobservedlaborcommitmentviolation. PanelsDandEarebasedonthelinkagebetweencontractsandexternal firm-level data described in Appendices C and D. Panel D provides model-consistent productivity and TFP measured in logs. PanelEshowstheexitindicatorattheendofthelaborcommitmentperiod. employment deviation. A striking feature of the data is the large spike exactly at the committed level of employment, suggesting the importance of these constraints for firms’ labor choices. Following Chetty, Friedman, Olsen, and Pistaferri (2011), we estimate an excess mass around the threshold of 652% relative to the average height of the counterfactual distribution.15 4.2 Matching Contracts to Firms The audits conducted on the contract-level data do not provide information regarding firm-level sales and post-privatization market exit. To construct productivity measures, we utilize data from theMannheimEnterprisePanel(MUP).Bymergingthecontractpartners’nameswiththeownership information in the MUP, we can generate productivity measures starting from 1993. This merging process enables us to measure firm sales at the end of the commitment period for nearly all linked firms. For a detailed description of the data merge between the two datasets, please refer to Supple- 15TheredlineinFigure4plotstheestimatedcounterfactualdensitybasedonatwelve-degreepolynomial(p =12)and anasymmetricwindowaroundthethresholdR=[3,−1]. R=[3,−1]denotestheomittedbunchingrangeincludingfirms having up to three more employees than their committed target. The yellow shaded region depicts the estimated excess massaroundthethreshold. FiguresA.5toA.7providerobustnesscheckswithrespecttothedegreeofthepolynomial,the bunchingwindow,andthebindefinition. TableA.11showstheresultsbysub-samples. 13

Figure4: EmploymentDistributionaroundtheCommitmentLevel Employment < Commitment Employment > Commitment Excess mass (b) = 6.521 Standard error = .8195 Excess share (%) = 0.243 ycneuqerF 0003 0002 0001 0 -40 -20 0 20 40 Realized minus Committed Employment Counterfactual distribution Observed distribution Notes: The figure shows the employment distribution around the committed employment (demarcated by the vertical red line at 0) for contracts between 1990-1995. The blue line in dots is a histogram of actual employment relative to the commitment target in the final commitment year. Each point shows the number of observations in employment count bins (deviation between the target and the realized employment). The solid line beneath the empirical distribution is a twelve-degreepolynomialfittedtotheempiricaldistributionexcludingtheareaofmissingoneemployeeandhavingthree employeesmorethancommitted. Theshadedregioninyellowistheestimatedexcessmass,whichis652%oftheaverage height of the counterfactual distribution beneath. Standard error is calculated using a parametric bootstrap procedure. EstimationbasedonChetty,Friedman,Olsen,andPistaferri(2011). mentary Appendix C. Overall, we identify the respective legal unit behind 4,622 contracts. We compute two measures of productivity growth for firms under employment commitments. First, we consider a model-consistent productivity measure given by sales per worker adjusted by the labor share in the production function. To assess initial productivity, we use information from the opening balance sheets of THA firms with contract data regarding employment at the time of privatization. To measure productivity at the end of the employment commitment, we merge the sales information from MUP with the final employment audit. Second, we use the Soestra firmlevel survey of THA firms to calculate firm-level Total Factor Productivity (TFP) as described in Appendix D. Panel D of Table 1 reveals a substantial increase in productivity during the commitment period. Thisnoteworthyimprovementalignswiththedocumentedconvergenceprocessobservedintheyears following reunification. Within the first decade after reunification, approximately half of the measured labor productivity gap and over one-third of the GDP per capita gap between East and West havebeenclosed(Burda2006).16 Startingin1990,ourcalculationsuggestsanincreaseinproductivity of 2 log points. The calculated improvements in TFP between the initial contract year and the final commitment period amounts to 0.51 log points. Finally, Panel E of Table 1 provides information on 16Basedonaggregatestatistics, Bachmann, Bayer, Stüber, andWellschmied(2022)showthatGDPperworkerincreased byabout0.7logpointsbetween1991and2000. 14

market exit for the matched contracts with the MUP. The sample size for this analysis is larger compared to the productivity assessment, because of missing data in terms of the sales variable. At the end of the commitment period, an exit share of 5.5% is observed. 5 Empirical Analysis of Labor Commitments and Firm Dynamics 5.1 Identification Strategy Addressingtheempiricalchallengeofnon-randomallocationoflaborcommitmentstofirmsiscrucial when analyzing firm-level responses. For example, if high labor targets are assigned to low-growth firms, it may lead to an underestimation of the impact of employment commitments. To tackle this issue, we develop a framework for reduced-form identification inspired by methods used in studies onjudgeleniency(Bhuller,Dahl,Løken,andMogstad2020;DobbieandSong2015;Bernstein,Colonnelli, Giroud, and Iverson 2019) and patent evaluators (Sampat and Williams 2019). These studies typically estimate the fixed traits or preferences of decision makers regarding outcomes under their control, such as leniency or toughness. By combining this estimate of the fixed trait with the quasirandom allocation of decision makers, we obtain an exogenous shifter that helps mitigate potential biases in future cases. The proposed empirical framework for the analysis is well-suited to our institutional setting for severalreasons. ThenumberofprivatizationsinthoseyearsmeantthatTHAagentstypicallyworked on multiple cases. Importantly, the breakneck speed of privatizations generated, within offices, randomness in the assignment of these cases. A consultant with the THA in those years described the process as “an exceptional situation where there was a lot of improvisation.” Finally, at the moment of privatization, each THA agent possessed significant discretion in establishing the conditions for the firm to be privatized, thus leaving room for privatizer traits to matter in the process.17 Instrument Construction. In our setting, we observe the name of the privatizer for 11,194 signed contracts. These contracts are handled by 1,659 different individuals with an average of 6.7 cases per privatizer. FigureA.1inAppendixAshowsthedistributionofcasesperprivatizer. Weconditionour baseline sample on having at least five privatizations per privatizer. This generates a final sample of 9,363 privatizations as our baseline.18 The first step in the analysis is to construct a measure for privatizers’ stringency in assigning binding labor commitments i.e., labor targets that force the firm to grow. Our measure is the average propensity of privatizers to require binding labor commitments. To address the own-observation problem and control for the THA office level effects, we follow the literature by estimating the following leave-one-out measure of binding commitment: 17The organizational stress and complexity of privatizing the East German economy cannot be underestimated. The THAwasdescribedas“anadolescentbureaucracy,bornofchaosanddestinedtobephasedoutwithouteverfunctioning normally. Itisahumancreation,whippedtogetherquicklyandthenputunderextremepressurewithouttimetoprepare” (DoddsandWächter1993). Theagencyofficiallyterminateditsoperationattheendof1994anditsmissionwastakenover byasuccessoragencyentitledBundesanstaltfürvereinigungsbedingteSonderaufgaben(Böick2018). 18Weproviderobustnessestimatesusing10casesperprivatizerinAppendixA. 15

(cid:32) (cid:33) (cid:32) (cid:33) 1 ∑ noj 1 ∑ no Z = (Binding )−Binding − (Binding )−Binding , ioj n −1 k i n −1 k i oj k=1 o k=1 where i denotes the firm, o the THA office, and j the assigned privatizer.19 Binding is an indicator i variable equal to 1 if initial employment is smaller than the final employment target, i.e., the firm is constrained to grow. n is the number of cases handled by the privatizer in THA office o, and n oj o is the number of cases handled by the local THA office. Note that the second term in the formula normalizes the measure by taking into account the average office propensity of writing out binding commitments. This is important as the characteristics of firms to be privatized are different across THA offices. Z , therefore, measures the leave-one-out measure of binding labor requirements of ioj privatizer j assigned to firm i. Figure 5 plots the relationship between binding labor commitments and the estimated privatizer preferences. The density plot is accompanied by a local linear regression highlighting considerable variation in how privatizers impose labor commitments. The probability of receiving a binding contract increases continuously along the stringency measure e.g., moving from the lowest decile to the highest decile increases the probability of assigning a binding contract by 21% points.20 Random Assignment. We use this measure of privatizer labor preference to indirectly test the assignment mechanism of cases to privatizers. The test exploits pre-assignment information on 12,500 firms that submitted their opening balance sheets in July 1990. We link the labor contracts to these firms and test whether pre-privatization firm-level characteristics correlate with our measure of privatizer stringency. Table 2 tests the random assignment mechanism. Each coefficient in column (1) represents a single regression, with the independent variable being our measure of labor preferences (conditional on fully interacted year and local office fixed effects). Column (2) provides p-values with two-way clusteredstandarderrorsattheprivatizerandlocalofficelevel. Finally, weprovideadjustedp-values for multiple testing using the procedure proposed by Romano and Wolf (2005a) and Romano and Wolf (2005b) with 1,000 bootstrap replications. Estimates in Table 2 provide strong evidence that, conditional on fully interacted year and local officefixedeffects,casesarerandomlyassignedtoprivatizersinoursample. Forexample,theresults indicatethata1%pointincreaseinlaborpreferencesisassociatedwithaninsignificant0.1%increase in production workers. Similarly, we find no economically or statistically significant correlation of our instrument with a wide range of employment and revenue measures (including initial labor productivity). In terms of sectoral affiliation, only 2 out of 16 coefficients are individually significant 19WeobservetheassignmentoffirmstoTHAofficesandtheassignmentofcontractstofirms. Forasubsetofprivatizers weobserveseveraloffices. Inthesecases,weestimatethemodewithineachprivatizertoassignaTHAoffice,and,ifthe modeisadraw,weassigntherespectiveheadquarters. 20Figure A.2 shows that preferences for binding labor commitments are consistent within the individual privatizer i.e., the correlation coefficient between the leave-one-out measure in the previous case and the leave-one-out measure of the currentcase(theorderofthecasesisdefinedbythedatethecontractissigned)is0.91. 16

Figure5: First-stageanalysis .4 .3 .2 .1 )tcartnoc gnidniB(rP .05 .04 .03 .02 .01 0 ytisneD 0 10 20 30 40 50 Privatizer Stringency Notes: The figure plots the probability of having a binding contract (initial firm size < final committed size) against the leave-one-out mean privatizer stringency (× 100) on the right y-axis. The plotted solid line corresponds to a local linear regression of binding contracts on the privatizer stringency. The two dashed lines show the corresponding 95% CI. All plotted values in the local linear regression are mean-standardized residuals from regressions on THA subsidiary times year of privatization fixed effects. The histogram shows the density of privatizer stringency (left y-axis). The figure is constructedbyconditioningofhavinghandledatleastfiveprivatizationcontractsandexcludestopandbottom1%ofthe stringencymeasure. Totalnumberofcontractsis9,363. atthe5%level. Adjustingformultipletesting,however,thereisnostatisticallysignificantrelationship between the privatizer stringency measure and sector affiliation. The data also includes individual-level characteristics of the privatizer. We leverage this information to examine whether the stringency measure predicts the number of cases, the gender of the privatizer, and whether they hold a PhD degree. If there is systematic variation in the stringency measure based on academic qualifications or the ability to handle privatizations, as reflected in the number of cases, it could suggest that decisions regarding labor commitments are influenced by heterogeneity in skills rather than preferences (Chan, Gentzkow, and Yu 2022). However, our findings, presented in the table, indicate that privatizer characteristics do not have any predictive power over ourinstrument. Thissuggeststhatheterogeneityindecision-makingstemsfrompreferencesforstrict labor contracts. The last two rows of the table present regression results of our stringency measure on the probability of renegotiating contract conditions. Firms were able to renegotiate if they failed to meet their committed targets, which may indicate a potential reduction in the effective stringency of contracts. However, the table demonstrates that the instrument is not correlated with future renegotiations. This is consistent with the fact that the organization of privatizations and contract management were handled by different units within the THA. Anadditionalempiricalchallengerelatestohowlaborcommitmentscaninfluencebuyerselection. High-qualitybuyersmightfinditeasiertoagreetobindinglaborcommitments,therebygeneratinga 17

Table2: TestofRandomAssignmentofFirms/ContractstoPrivatizers Indep. variable: Stringency Dep. variables Coefficient p-value Adj. p-value Mean Standarddeviation (1) (2) (3) (4) (5) Employment Accounting -0.0039 0.2776 0.9830 2.2540 1.4640 Purchasing 0.0019 0.6773 1.0000 1.6380 1.5600 HR -0.0003 0.9397 1.0000 1.8840 1.7680 Production -0.0010 0.8725 1.0000 4.4340 2.4600 R&D 0.0004 0.8956 1.0000 1.2500 1.8300 Sales -0.0012 0.8321 1.0000 2.2520 1.8720 Administration -0.0039 0.3806 0.9970 3.2680 1.8720 Firmsizeabove2000 0.0005 0.4863 0.9990 0.1060 0.3100 Revenue Revenue -0.0110 0.1926 0.9211 8.0840 3.3820 Revenueupper80p -0.0007 0.4295 0.9980 0.1920 0.3940 ShareofrevenueWestEurope 0.0007 0.1548 1.0000 0.2560 0.4360 Productivity Laborproductivity -0.0013 0.5481 1.0000 3.3440 1.2640 Productivityupper80p -0.0006 0.3947 1.0000 0.1960 0.3960 Sectoraffiliation Agriculture,forestry,fishing -0.0001 0.6859 1.0000 0.0140 0.1140 Energyandwater 0.0000 0.9144 1.0000 0.0160 0.1280 Miningandquarrying 0.0001 0.4999 1.0000 0.0080 0.0840 Chemicalindustryandpetroleum 0.0005 0.1356 0.9970 0.0480 0.2160 Plasticsandrubber 0.0001 0.8000 1.0000 0.0100 0.1040 Extractionofcut-stoneandsand 0.0001 0.7342 1.0000 0.0240 0.1560 Iron,casting,steelforming 0.0000 0.9654 1.0000 0.0240 0.1520 Steelconstruction,mechanicalengineering 0.0021 0.0207 0.4635 0.1660 0.3720 Electricalengineering,optics 0.0008 0.2174 0.9600 0.0680 0.2520 Wood,paper,printindustry 0.0000 0.9417 1.0000 0.0440 0.2040 Textileandclothing 0.0001 0.7956 1.0000 0.0580 0.2340 Foodandbeverageindustry -0.0006 0.1089 0.9770 0.0460 0.2100 Constructionandbuildingstrades -0.0008 0.2166 0.9311 0.0580 0.2320 Wholesaleandforeigntrade -0.0003 0.5554 1.0000 0.0580 0.2320 Retailtrade -0.0007 0.2162 0.8821 0.0340 0.1780 Service -0.0006 0.0419 0.9311 0.0380 0.1900 Privatizercharacteristics Numberofcases 0.1280 0.2705 0.4635 30.2820 25.1260 Gender -0.0001 0.9338 1.0000 0.8640 0.3420 PhDdegree 0.0009 0.4499 0.9970 0.2660 0.4420 Renegotiationattempt Laborrenegotiation -0.0002 0.4142 1.0000 0.0640 0.2440 Anyrenegotiation -0.0009 0.1788 0.9980 0.3820 0.4860 Notes: Thesampleisbasedon7,152contractswithemployment,revenueandsectorinformationattheyearofreunification 1990. Employment, revenue, and productivity is measure in logs. All explanatory variables refer to the THA initial firm. Each line represents a single regression of the explanatory variable on the stringency measure that takes values between 0 (minimum) and 100 (maximum) controlling for THA office and year of privatization fixed effects. Standard errors are two-way clustered at privatizer and THA office level. p-values in column (2) correspond to the regression model and are two-wayclusteredattheprivatizerandTHAofficelevel. p-valuesincolumn(3)adjustformultipletestingusingRomano- Wolf procedure (Romano and Wolf 2005a; Romano and Wolf 2005b) with 1,000 bootstrap replications. **p<0.1, **p<0.05, ***p<0.01. correlation between buyer types and labor targets. To check whether this is the case, we link investor names from the contract data with external MUP data via record linkage. This allows us to obtain investor level information relative to size, credit rating, location (East/West), and industry. After the 18

record linkage, we observe investor characteristics for 4,993 privatization contracts. Similar to Table 2, Table A.1 shows the correlation between our instrumental variable and investor characteristics. Overall, we do not observe any systematic evidence for buyer selection. Monotonicity. Theinterpretationofourinstrumentalvariableestimatesreliesnotonlyonthevalidity of the exclusion restriction but also on the accompanying monotonicity condition. In our context, the monotonicity condition implies that firms with a strict labor commitment assigned to a lenient privatizer would have also received a strict commitment if they were assigned to a tough privatizer, and vice versa. Frandsen, Lefgren, and Leslie (2023) show that it is possible to relax the strict (pairwise) monotonicity assumption to an average monotonicity assumption and still recover a weighted average of individual treatment effects. This average monotonicity assumption requires that the data contain only complier groups where the covariance between privatizer stringency and binding labor commitments is positive. To test this condition, we first conduct an examination of the correlation between privatizer stringency and binding commitments across various observable subgroups (Bhuller, Dahl, Løken, and Mogstad 2018; Dobbie, Goldin, and Yang 2018). In Table A.2, we present the results of first-stage regressions for different firm size groups based on the 1990 measurements and also for sub-samples grouped by sector affiliation. We construct our instrument using the entire sample and perform thefirst-stageregressionsonthesub-samples.21 Asexpectedundertheassumptionofaveragemonotonicity,allfirst-stagecoefficientsarepositiveandstatisticallysignificant. Finally,followingFrandsen, Lefgren,andLeslie(2023),wealsoimplementandfailtorejectthejointnullhypothesisthatpairwise monotonicity and exclusion hold.22 IV Model. The next step of the analysis is to embed this instrument into a 2SLS equation relating THA’s employment targets to outcome variables such as employment growth, productivity growth, and exit. The regression model can be written as: y = β 1(Binding )+X (cid:48) θ+(cid:101), i i i i withy ,forexample,indicatingthegrowthrateofemploymentbetweenthefirstandthefinalaudits.23 i X includes log initial employment measured at the first audit to account for pure size effects of the i privatized firms as well as industry dummies. The empirical model further includes the number of monthsbetweenthetimethecontractissignedandthefirstauditandthenumberofmonthsbetween thefirstandfinalauditstocapturedifferencesincommitmentlengthacrosscontracts. Theparameter 21Toensureanadequatesamplesize,theseregressionsareconductedonsub-sampleswithaminimumoftwoprivatizationsperprivatizer. 22Table A.3 provides the test for the joint null hypothesis that the exclusion and monotonicity assumptions hold for differentnumbersofknotsandBonferroniweightsusingthesuggestedquadraticspline(controllingforTHAofficetimes yearfixedeffects). 23Thismeasureofemploymentgrowthboundsthegrowthratebetween-2and2andreducesthepossibilitythatresults aredrivenbyoutliers. Weproviderobustnesschecksusinglogdifferencesand(L it /L it−1 )1/#years−1asthegrowthmeasure. Notealsothatthefinalauditcorrespondscloselytothedateofthefinalcommitment. 19

of interest is β, which measures the effect of assigning a binding contract on the growth rate of the firm. Our research design exploits the quasi-random assignment of cases to THA privatizers with different preferences for binding labor commitments. We specify our first-stage equation for binding labor targets, Binding, as: i Binding = γZ +X (cid:48) λ+κ i i(j) i i where Z denotes labor preferences of privatizer j assigned to case i as defined above. The model i therefore estimates the local average treatment effect of THA labor requirements on firm outcomes. 5.2 Labor Commitments and Firm Growth Webeginbydocumentinghowemploymentcommitmentsdynamicallyaffectedfirmpoliciesinterms of labor. The description of employment dynamics by initial contract conditions uses the sample of 11,194 contracts for which we have multiple audits as well as information on privatizer names and offices. WefollowDavisandHaltiwanger(1999)andconstructthefirm-levelgrowthratebetweenthe firstauditandthefinalauditofthecommitmentas(L −L )/0.5(L +L ),where L denotesthe it it−1 it it−1 i levelofemploymentoffirmi. Subscripttreferstothedateofthefinalcommitmentand,consequently, t−1 refers to the first audit that approximates firm size when the contract is signed.24 Figure 6 provides descriptive evidence on the distribution of growth rates over the course of the contract period. In the left panel, the dashed line plots the distribution of growth rates for the entire sample. On average firm employment grows by 6% between the initial and final audits. In addition, the panel distinguishes between firms according to their initial size and final target. Firms initially at or above their committed target (grey bars) shrink, on average, by -6.8%, whereas firms initially below their target (empty bars) grow, on average, by 54%. The right panel of Figure 6 provides the full distribution of employment growth according to the ratio of initial size over final employment target. A striking negative relationship emerges between the distance to the final target and subsequent employment growth. Firms that have high targets relative to their initial size grow their workforce significantly more than firms with targets close to their initial size. Firms with lax targets relative to their initial employment had leeway to adjust and subsequently shrunk significantly. Overall, the figure suggests the importance of firm employment targets as a determinant of firm employment policies. Table 3 presents the estimates for the labor growth equation. Columns (1) to (3) provide OLS estimates, while columns (4) to (6) provide IV estimates. Standard errors are two-way clustered at the privatizer and office level. Columns (1) to (3) suggest that firm growth is positively correlated withbindinglaborcommitments. Conditionalonthesetofbaselinecontrols,industrydummies,and privatizer characteristics, the association between binding contracts and employment growth is on 24Thelengthoftheemploymentcommitmentisheterogeneousacrosscontracts. Onaveragecommitmentlengthisthree years. However, we observe commitments that can last for 11 months to 98 months at the 1st and 99th percentiles of the distribution. Intheempiricalspecificationwecontrolfordifferencesincommitmentlength. 20

average 49% points until the final commitment date. The estimated OLS coefficient is unaffected by the inclusion of additional control variables to those in the baseline specification. Figure6: EmploymentDynamicsbyInitialSizetoTarget A:EmploymentGrowth ytisneD 3 2 1 0 B:EmploymentGrowthbyInitialSizetoTarget -2 -1 0 1 2 Employment growth distribution Initial size above target Initial size below traget Growth distribution htworg tnemyolpmE 5. 0 5.- .2 .6 1 1.4 1.8 2.2 Ratio Initial Size to Final Target Average growth rate 90% Confidence Intervals Notes: PanelAshowstheoverallemploymentgrowthdistributionaswellastheemploymentgrowthdistributiondistinguishing by firms initially below or above (including firms initially at their target) their commitment employment level. PanelBshowsaveragegrowthratesbythedistanceoftheinitialsizetothefinaltarget. IV estimates in columns (4) to (6) suggest that the causal effect of binding labor targets is significantly larger with respect to OLS estimates. In these specifications, firms are estimated to grow their workforce by 68% points in the three years following the random assignment of a binding labor contract. The effect is not only economically large but also precisely estimated at the 1% significance level. Consistent with the previous evidence, the first-stage statistics for weak instruments is large(PanelB).TheKleinbergen-PaapF-testsrejectthehypothesisofweakinstrumentswithstatistics ranging between 15 and 17. Economically, the first-stage estimates imply that a 10% increase in labor preferences of privatizers result in a 2% points higher likelihood of a binding labor contract. Consistent with the quasi-random assignment mechanism, the inclusion of additional covariates does not affect the first-stage coefficients.25 The results are robust to a series of alternative specifications. A challenge to our identification strategy is that privatizer decisions are multidimensional. The exclusion restriction requires that pri- 25TableA.7providesawaytotestwhetherthedifferencesbetweentheOLSandtheIVestimatesaredrivenbytreatment effectheterogeneity. WefollowBhuller,Dahl,Løken,andMogstad(2020)andfirstperformaprincipalcomponentanalysis using one component based on pre-determined employment (see employment categories in Table 2) and revenue figures measurein1990, aswellasinitialemploymentatcontractdate, andthesectoraffiliation. Wethenseparatethepredicted componentintoquartilesandseparatelyestimatethecompliershareforeachquartilegroupusingthefirst-stageregression specification. Finally,were-weightthefullestimationsamplebyusingthesub-samplecompliersharesasweights. PanelB of Table A.7 shows that re-weighting based on observed characteristics increases the OLS estimate slightly from 0.49 to 0.52. The difference between the re-weighted OLS and IV estimate, however, remains stark. This suggests that effect heterogeneityisunlikelytoexplainthedifferences. 21

Table3: RegressionResults,EmploymentGrowth OLSModel IV-Model (1) (2) (3) (4) (5) (6) PanelA:Second-stageresults Bindingcontract 0.4992*** 0.4973*** 0.4975*** 0.7016*** 0.6740*** 0.6873*** (0.031) (0.030) (0.030) (0.219) (0.231) (0.235) PanelB:First-stageresults Privatizerstringency 0.0020*** 0.0018*** 0.0018*** (0.000) (0.000) (0.000) Observations 9,363 9,363 9,363 9,363 9,363 9,363 Averageemploymentatcontractdate 60.064 60.064 60.064 60.064 60.064 60.064 Averagegrowthrate(non-bindingcontracts) -0.063 -0.063 -0.063 -0.063 Sharewithbindingcontracts 0.207 0.207 0.207 0.207 0.207 0.207 F-Statistic 17.01 14.67 14.76 Samplecondition Baselinecontrols Yes Yes Yes Yes Yes Yes Industrycontrols No Yes Yes No Yes Yes Individualcontrols No No Yes No No Yes Notes: The table shows OLS and IV regression results of employment growth on binding contracts. Panel A shows the reducedformregressingthebindingcontractindicatoronthestringencymeasure. PanelBshowsthesecond-stageresults. AllspecificationscontrolforfullyinteractedTHAagencyandyearfixedeffectsandareconditionalonhavingatleastfive privatizations per privatizer. F-Statistic refers to the Kleibergen-Paap F-Statistic. Baseline controls are time between the first and last audits measured in months, time between contract date and first audit measured in months, and log initial employmentlevelmeasuredatthefirstaudit. Industrycontrolsare2-digitindustrydummies. Individualcontrolsreferto thegenderoftheprivatizerandadummyforaPhDdegree. Standarderrorsaretwo-wayclusteredatprivatizerandTHA officelevel. Instrumentreferstotheleave-one-outmeasureofassigningbindingcontracts. *p<0.1,**p<0.05,***p<0.01. vatizers affect firms’ outcomes only through binding labor commitments. THA privatizers, however, negotiated not only on labor commitments, but also on associated penalties, investment commitments, andsalesprice. FollowingBhuller, Dahl, Løken, andMogstad(2020), weaddressthisissueby augmenting our baseline model with controls for these dimensions of privatization contracts. Consistentwiththeexclusionrestriction,TableA.4showsthataddingextensiveandintensiveinvestment preferences does not affect our baseline results qualitatively and, further, does not have any explanatory power in the first and second stages. In Panel A of Table A.5, we also control for subsequent renegotiation attempts initiated by buyers. Panel B varies the sample according to the number of cases handled and the construction of the instrument. Panel C estimates our model using alternative measures of firm growth. Finally, Table A.6 addresses potential sample selection of GDR firms into labor commitments by including the estimated inverse mills ratio from a Heckman model. Again, estimates are unaffected.26 5.3 Labor Commitments and Productivity Growth Todisentanglethemechanismbehindthegrowthinemploymentweextendtheempiricalanalysisto productivitydynamicsinthematchedsampleof2,395privatizationcontractswithcompleteinforma- 26TheHeckmanselectionequationisbasedonaprobitregressionwiththeoutcomevariablebeingequalto1iftheinitial GDRfirmisobservedamongthecontractswithlaborcommitmentsandzerootherwise. Weuseasexplanatoryvariables loginitialfirmsizeandloginitialsalesoveremploymentmeasuredin1990aswellassector-andTHAofficefixedeffects. Resultsforemploymentgrowth,productivitygrowthandfirmexitinA.6. 22

tion. We construct the model-consistent measure of productivity as sales/employmentα, taking into account the labor share in the production. In the baseline analysis, we set α = 0.8, consistent with the aggregate labor share during this period. We also construct measures of productivity based on TFP by matching the contracts to a survey that contains information on THA firm investments. This allows us to obtain the associated capital stock of the firm and estimate a Cobb-Douglas production function. Figure 7 describes the relationship between productivity growth and the ratio of initial employment relative to the final target. The figure plots local linear regressions on both sides of the vertical line separating initially binding and non-binding contracts. The figure provides two major insights. First,growthinproductivityisrelativelyconstantforfirmsabovethethresholdforbindingcontracts. The average growth rate in the data amounts to 86.8% which indicates substantial improvements in productivity during the first years after reunification. Second, productivity growth is significantly higher for firms initially below their committed employment. Figure7: ProductivityGrowthandtheDegreeofBindingContracts projection linear fit ytivitcudorP ni htworG 2.1 1.1 1 9. 8. 7. 6. 5. .4 .5 .6 .7 .8 .9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 Ratio Initial Employment Size to Final Commitment Notes: The figure plots the growth in productivity between the initial year of the contract and the final commitment year against the ratio of initial employment relative to the final commitment level. Contracts below 1 have initially lower employmentthancommitted. Contractsabove1haveinitiallyhigheremploymentthancommitted. Theplottedvaluesin thelocallinearregressionaremean-standardizedresidualsfromaregressiononinitiallaborproductivity,employmentand industry-fixedeffects. Thetwogreydashedlinescorrespondtothe90%CI.Thebluelineshowsalinearfitofaregression ofproductivitygrowthontheratioofinitialsizetofinalcommitmentamongcontractstotheleftoforat1. Theredline projectsthelinearfitintotheareawheretheinitialsizeisbelowthecommittedlevel(totheleftof1). Thefigureexcludes topandbottom4%ofthetightnessmeasure. Totalnumberoffirmsis2,395. Table 4 provides OLS and IV estimates for productivity growth following the same growth rate formula as for employment. The specification controls again for fully interacted THA office and year fixed effects, industry dummies, log initial employment, log initial productivity, the purchasing price and the time between the first and last audits as well as between the contract date and the first audit. 23

Columns (1) and (2) of Table 4 provide OLS evidence that firms with binding labor contracts experience higher productivity growth of around 8 to 9% points. Column (2) adds controls for decile dummies of the purchasing price. Column (3) provides IV evidence on productivity growth. To do so, we implement a two-sample 2SLS estimation by using the predicted values from the firststage regression of the full sample in the second-stage regression of the sub-sample that includes information on productivity growth. To calculate the standard errors, we perform 2,500 bootstrap replications presented in parentheses. Over the course of five years measured between July 1990 and the end of the commitment period, on average, firms with binding labor contracts experience a total productivity growth of roughly 73% points. Columns (4) to (6) present our results based on the TFP growth measure. Again, we find that TFP increased by 66% points more for firms with binding contracts over the commitment period. As in the employment growth regressions, the OLS estimates for productivity display a downward bias with respect to the IV estimates. We provide extensive robustness checks for our productivity growth results in Appendix Tables A.9 and A.10. Amongst others, we demonstrate that the estimates remain robust when varying the labor share parameter, and when including additional controls such as purchasing price, the presence of investment targets, and the inclusion of exiting firms. Finally, TableA.8providessupportingevidencefortheproductivitychannelbyanalyzingpatenting activity. The outcome variable is equal to 1 if the firm has at least one patent during the commitment period and 0 otherwise. OLS results show a positive and significant association between bindingcontractsandpatenting. Althoughimpreciselyestimated,the2S2SLScoefficientshowsagain a downward bias of the OLS point estimate. Table4: RegressionResults,ProductivityGrowth Productivity, α =0.8 TFP OLS OLS 2S2SLS OLS OLS 2S2SLS (1) (2) (3) (4) (5) (6) Bindingcontract 0.0965*** 0.0835*** 0.7109** 0.1107*** 0.1239*** 0.6608*** (0.022) (0.023) (0.363) (0.039) (0.041) (0.213) Observations 2,395 2,395 1,612 1,825 1,825 1,825 Averageproductivityatcontractdate 10.599 10.599 10.633 6.813 6.813 6.813 Averageproductivitygrowth(non-binding) 0.851 0.851 0.854 0.332 0.332 0.332 Sharewithbindingcontracts 0.171 0.171 0.155 0.162 0.162 0.162 Baselinecontrols Yes Yes Yes Yes Yes Yes Industrycontrols Yes Yes Yes Yes Yes Yes Individualcontrols Yes Yes Yes Yes Yes Yes Purchasingprice No Yes Yes No Yes Yes Notes: The table shows OLS and 2S2SLS regression results of measures of productivity growth on binding contracts. All specifications control for fully interacted THA agency and year fixed effects. Binding contracts are defined as initial firm size below the committed target level. Controls are as in the baseline specification. Additional controls are log initial productivity and the purchasing price. Standard errors in columns (1)-(3) are two-way clustered at privatizer and THA officelevel. Thestandarderrorsincolumns(4)-(6)arebootstrappedusing2,500replications. *p<0.1,**p<0.05,***p<0.01. 24

5.4 Labor Commitments and Market Exit We now turn to the analysis between binding labor commitments and market exits. To measure firm exit we again use the merged sample of contracts to the MUP data. We also use a second measure of exit based on the final labor audit reporting 0 workers. Figure 8 describes the relationship between the exit probability of firms and the ratio of initial employment relative to the final target. Similar to the productivity growth patterns, the share of firms exiting the market is relatively constant for non-binding firms above the vertical line of 1. The Figure8: MarketExitandtheDegreeofBindingContracts projection linear fit ytilibaborP tixE tekraM 3. 52. 2. 51. 1. 50. 0 .4 .6 .8 1 1.2 1.4 1.6 1.8 2 Ratio Initial Employment Size to Final Commitment Notes: The figure plots market exit rates against the ratio of initial employment relative to the final commitment level. Contracts below 1 have initially lower employment than committed. Contracts above 1 have initially higher employment than committed. The plotted values in the local linear regression are mean-standardized residuals from a regression on initialemploymentandindustry-fixedeffects. Thetwogreydashedlinescorrespondtothe90%CI.Thebluelineshowsa linearfitofaregressionofmarketexitontheratioofinitialsizetofinalcommitmentamongcontractstotheleftoforat 1. Theredlineprojectsthelinearfitintotheareawheretheinitialsizeisbelowthecommittedlevel(totheleftof1). The figureexcludestopandbottom3%ofthetightnessmeasure. Totalnumberoffirmsis4,596. exit rate for these firms amounts to 4.8% on average. With an average exit rate of 8.4%, firms with binding contracts show a higher level of market exit that is increasing in the tightness measure. Table 5 provides a regression version of Figure 8, controlling again for the same variables as before. The first three columns provide the results using the MUP exit indicator, whereas the last three columns are based on zero employment in the ISUD data (conditional on the same sample). Both regression specifications generate similar OLS results. Binding contracts are associated with an increase in market exit of around 2.5% points. The IV estimation, although with lower precision, confirms the higher propensity to exit of firms with binding labor commitments. These results are consistent with the model’s prediction that lower profits associated with the penalties lead firms to exit at a higher rate. 25

Table5: RegressionResults,ExitProbabilityatFinalCommitment MUPexitindicator ISUD0employment OLS 2S2SLS OLS 2S2SLS (1) (2) (3) (4) (5) (6) Bindingcontract 0.0262** 0.0248** 0.1302 0.0216* 0.0193* 0.0358* (0.011) (0.012) (0.108) (0.011) (0.010) (0.021) Observations 4,563 4,563 2,804 4,563 4,563 2,804 Exitshare(non-binding) 0.049 0.049 0.047 0.035 0.035 0.030 Sharewithbindingcontracts 0.171 0.171 0.171 0.171 0.171 0.171 Samplecondition Baselinecontrols Yes Yes Yes Yes Yes Yes Industrycontrols Yes Yes Yes Yes Yes Yes Individualcontrols Yes Yes Yes Yes Yes Yes Purchasingprice No Yes Yes No Yes Yes Notes: The table shows OLS and 2S2SLS regression results of exiting probabilities at the end of the commitment period. The outcome variable takes the value of 1 if the firm is exiting by the end of the commitment period and 0 otherwise. All specifications control for fully interacted THA agency and year fixed effects. Binding contracts are defined as initial firmsizebelowthecommittedtargetlevel. Controlsareasinthebaselinespecification. Additionalcontrolvariableisthe purchasingprice. Standarderrorsincolumns(1),(2),(4)and(5)aretwo-wayclusteredatprivatizerandTHAofficelevel. Thestandarderrorsincolumns(3)and(6)arebootstrappedusing2,500replications. *p<0.1,**p<0.05,***p<0.01. 6 Quantitative Analysis In this section, we present the calibration of the model using firm-level data and provide several counterfactual analyses to quantify the various channels by which the binding employment targets impact firm dynamics. 6.1 Calibration We start by setting some of the parameter values externally. We choose the labor share parameter in theproductionfunction, α, equalto0.8tomatchthelaborearningshare. Consistentwiththeaverage contract length of three years in the data, we set the arrival rate of contract expiration, µ, to 1/3. Annual wage growth rate is set to 10% to match the average real wage growth rate over 1990 and 1996inEastGermany(Hunt2001). Therestoftheparametersarecalibratedinternallybyminimizing thedistancebetweenthemomentsfromthefirm-leveldataweusedintheempiricalpartofthepaper andtheirmodelimpliedcounterparts.27 Inparticular,let ME denotethevectorofempiricalmoments and let M(Ω) denote the vector of model-simulated moments and Ω is the set of parameters to be Ω calibratedinternally. Wethensearch tominimizetheabsoluterelativedeviationbetweenthemodel and data; that is, we solve min ∑ |M m E −M m (Ω)| . Ω |ME| m m We use the point estimates of the effect of binding contracts on employment growth and productivity growth presented in Section 5 to discipline the cost of not hitting the target, γ. We further use 27Inoursetting,wecannotseparatelyidentifythestepsizeandthecostscaleparameterinproductivityimprovements, λandφ,respectively. Therefore,wefixthevalueofthestepsizeat0.25andcalibratethecostscaleparameterinternally. 26

regression results on the impact of binding contracts on exit rates of firms to inform the exit cost parameter, C . Finally, we include the growth rate of total employment for firms with binding contracts e over the commitment period to pin down the investment cost parameter, φ. We use the following procedure to calibrate the model: For given values of parameters, we first solve the value function in equation 3 and use the implied optimal decisions to simulate a cohort of firms. We initialize the cohort by using the sample of firms used in the empirical part of the paper and take the employment target as given in the data. Crucially, each firm is simulated in line with the time span from its initial audit to its final audit. Finally, we use the simulated data to construct thetargetedmoments. Werepeatthisprocessandsearchovertheparameterspaceuntilweminimize the distance between model-implied moments and the data. 6.2 Calibration Results and Goodness of Fit Table6and7containthecalibratedparametersandthetargetedmoments,respectively. Asseenfrom Table 6, the model is able to replicate the targeted moments well. In particular, we were able to fit higher employment and productivity growth of firms with binding contracts with a relatively parsimonious model. Our calibration suggests that for every missing employee relative to the committed labor target, firms pay a fine that corresponds to 68% of the average wage, given by γ. Table6: MomentsUsedinCalibration # Description Model Data M Employmentgrowthregression 0.489 0.498 1 M Productivitygrowthregression 0.083 0.083 2 M Exitrateregression 0.030 0.027 3 M Totalempl. growthrateoffirmswithbindingcontracts 0.614 0.672 4 Table7: InternallyCalibratedParameters Description Model Estimate Penaltyfornothittingtargetemployment γ 0.676 Scaleforinvestmentcostparameter φ 0.030 Costofexit C 58.39 e Non-targeted Moments The calibrated model also performs well in matching some important patternsinthedatathatwerenottargeted. InFigure9,wedepicttheemploymentgrowthbytheratioof initialemploymentrelativetotargetemployment,analogoustoPanelBofFigure6. Theblackandred dotsshowthemodel-impliedemploymentgrowthratesandthedata,respectively. Althoughweonly target the average excess growth rate of binding-contract firms (employment regression coefficient) in the calibration, our model successfully matches employment growth rates across the entire range of employment-to-target ratios. The calibrated model also captures the post-commitment employment dynamics fairly well. Fig- 27

ure 10 illustrates the evolution of total employment during and after the commitment period both in the model and data. The results suggest that not only do these firms with binding contracts experience higher employment growth during the commitment period, but these employment gains are persistent at least six years subsequent to the commitment period. This persistent employment effect is consistent with the dynamic productivity gains implied by binding labor targets in the model. Figure9: EmploymentGrowth Notes: The figure depicts the employment growth at the firm level by the ratio of initial employment relative to target employment. Theblackandreddotsshowthemodel-impliedemploymentgrowthratesandthedata, respectively. Gray dots show the employment growth rates under the counterfactual economy where there are no employment targets. The x-axisisdividedinto20quantilebinsandeachdotrepresentsaveragevaluewithinthatbin. Figure10: TotalEmploymentinthePost-CommitmentPeriod Notes: The figure shows the evolution of total employment during and after the commitment period both in the model (black lines) and data (red lines). Dashed and solid lines show the total employment for binding and not binding firms, respectively. Allseriesarenormalizedto1atthebeginningofthepolicyperiod. 28

6.3 Counterfactuals To quantify the importance of the different channels through which employment targets affect firm dynamics, we start with a simple exercise where we simulate a counterfactual economy under which there are no employment targets. In particular, we keep all other parameters of the model as in the baseline values and set the cost of commitment parameter to zero, γ = 0. Gray dots in Figure 9 show the employment growth rate by the ratio of initial employment relative to target employment in this counterfactual economy. As seen from the figure, the employment growth rate is substantially reducedintheabsenceoftargets,especiallyforthosefirmsthathavemorebindingcontractsinitially. This counterfactual economy implies a 20% drop in the total employment, rather than the 5% drop we observe in the data, suggesting that employment targets had a non-trivial role in shaping the aggregate employment dynamics. Ournextexercisedecomposestheimpactofemploymenttargetsontotalemploymentinto"static" and "dynamic" effects. To isolate the static effect, we simulate a counterfactual economy where firms still operate under employment targets but we shut down the "escape" productivity growth effect by assuming that marginal profits are not affected by the employment targets and are set to the value under the case of no employment targets for all firms. In other words, the second counterfactual economy only includes the direct, static effect of employment targets on employment choices. Fig- Figure11: TotalEmploymentunderCounterfactualEconomies Notes: The figure shows total employment implied by the model under three different cases: (i) baseline economy (solid black line), (ii) counterfactual economy with no dynamic effect (red dashed line) and (iii) counterfactual economy with investmentsubsidies. "Nodynamiceffect"counterfactualisobtainedbysettingthemarginalprofitstothevalueunderthe caseofnoemploymenttargetsforallfirms. Theinvestmentsubsidylevelissettomatchthetotalemploymentgrowthin the data through the period the original policy was implemented. All series are normalized to 1 at the beginning of the period. ure 11 summarizes the results by plotting the total employment over time under commitment (black line) and under no dynamic effects (red dotted line), relative to the total employment under no com- 29

mitment counterfactual. The decomposition results suggest that the dynamic effect of employment targets on productivity growth contributed to the employment dynamics significantly. As can be seen from the opening distance between the black line and the red dotted line, the importance of the dynamic channel increases over time. Our calibrated model attributes one-third of the employment growth over three years to dynamic effects. After 10 years, the static effect of the policy completely disappears (red dotted line goes back to 1), as by that time there are no more firms with employment commitmentleft. Thatis,alltheemploymentgainsshownbytheblacklineareduetodynamiceffect after10years,implyinga15%permanentemploymentincreaserelativetoanocommitmentscenario. Lastly, we consider an alternative policy where firms are provided uniform investment subsidies insteadofemploymenttargets. Wechoosethesubsidylevelsuchthatthetotalemploymentgrowthis thesameasinthedatathroughtheperiodtheoriginalpolicywasimplemented. Thebluedottedline inFigure11depictsthetotalemploymentgrowthundersuchaninvestmentsubsidy. Thecomparison betweenthesubsidypolicyandthecommitmentpolicyintermsofemploymentdynamicsrevealstwo important distinctions. Firstly, the subsidy incurs a substantial cost to attain the same aggregate employment growth, equivalent to around 5 percent of total output. Secondly, while the subsidy policy resultsinhigherlevelsofpermanentemployment,itfallsshortinaddressingshort-termemployment concerns compared to the commitment policy. 7 Conclusion Inthispaper,westudytheimplicationsofapolicythatimposedemploymenttargetstopushfirmsto groworlimittheirdownsizing. Thetypeofpolicyinterventionweanalyzeisunique,insofarasitprimarilyputsconstraintsonfirms’laborchoicesasopposedtousingsubsidies. Ourthree-stepanalysis involved the construction of a dynamic model, an empirical assessment, and counterfactual simulations based on a calibrated model. The model highlighted the dual effect of binding employment targets, simultaneously inducing higher employment and productivity growth, yet increasing the probability of firm exit due to lower profitability. Empirical validation, using a rich German dataset and an instrumental variable approach, confirmed these theoretical predictions. We find that a 22% points higher annual employment growth rate was achieved under binding labor contracts, along with a substantial 14% points annual increase in productivity. However, these gains were somewhat counterbalanced by a 3.6% points higher firm exit probability. In a series of counterfactual scenarios, the absence of employment targets indicated a potential 15% drop in aggregate employment after 10 years, emphasizing the policy’s significant role in the labor market. We explore an alternative policy of productivity investment subsidies and find that it could potentially yield higher permanent employment, albeit at a high cost and a slower employment response. In this paper, our primary objectiveis to understand the positive implications of thepolicy, rather than exploring its detailed welfare implications. As such, the model we employ focuses on firm-level decisions, analyzing how the policy shapes and influences firm behavior statically and dynamically. It’s important to recognize that a comprehensive normative analysis would necessitate to specify 30

variousotherdetailsoftheeconomicenvironment. Thisencompassesfactorssuchastheexternalities present within the economy and the interactions between the policy and these externalities. We leave this interesting avenue for future research. References A cemoglu , D., U. A kcigit , H. A lp , N. B loom , and W. K err (2018): “Innovation, reallocation, and growth,” American Economic Review, 108(11), 3450–91. A cemoglu , D., U. A kcigit , D. H anley , and W. K err (2016): “Transition to clean technology,” Journal of Political Economy, 124(1), 52–104. A ghion , P., A. B ergeaud , and J. V an R eenen (2023): “The impact of regulation on innovation,” American Economic Review, forthcoming. A ghion , P., A. D echezleprêtre , D. H emous , R. M artin , and J. V an R eenen (2016): “Carbon taxes, path dependency, and directed technical change: Evidence from the auto industry,” JournalofPolitical Economy, 124(1), 1–51. A kcigit , U., H. A lp , Y. E. A kgunduz , S. M. C ilasun , and J. M. Q uintero (2023): “Cost of SizedependentRegulations: TheRoleofInformalityandFirmHeterogeneity,”Discussionpaper,Working Paper. A kerlof , G. A., A. K. R ose , J. L. Y ellen , H. H essenius , R. D ornbusch , and M. G uitian (1991): “East Germany in from the cold: the economic aftermath of currency union,” Brookings Papers on Economic Activity, 1991(1), 1–105. A lesina , A., and N. F uchs -S chündeln (2007): “Good-bye Lenin (or not?): The effect of communism on people’s preferences,” American Economic Review, 97(4), 1507–1528. B achmann ,R.,C.B ayer ,H.S tüber ,andF.W ellschmied (2022): “MonopsonyMakesFirmsnotonly Small but also Unproductive: Why East Germany has not Converged,” . B arberis , N., M. B oycko , A. S hleifer , and N. T sukanova (1996): “How does privatization work? Evidence from the Russian shops,” Journal of Political Economy, 104(4), 764–790. B artelsman , E. J., J. H altiwanger , and S. S carpetta (2013): “Cross-country differences in productivity: The role of allocation and selection,” American Economic Review, 103(1), 305–334. B arwick , P. J., M. K alouptsidi , and N. B. Z ahur (2021): Industrial Policy Implementation: Empirical Evidence from China’s Shipbuilding Industry. Cato Institute. ernhard B , H. (2011): “Bischofferode 1993: Hungerstreik im Kaliwerk,” https://www.mdr.de/geschichte/ddr/wirtschaft/hungerstreik-im-kaliwerk-100.html. 31

B ernstein , S., E. C olonnelli , X. G iroud , and B. I verson (2019): “Bankruptcy spillovers,” Journal of Financial Economics, 133(3), 608–633. B ersch , J., S. G ottschalk , B. M üller , and M. N iefert (2014): “The Mannheim Enterprise Panel (MUP) and Firm Statistics for Germany,” ZEW-Centre for European Economic Research Discussion Paper, (14-104). B esley , T., and M. G hatak (2001): “Government versus private ownership of public goods,” The Quarterly journal of economics, 116(4), 1343–1372. B huller ,M.,G.B.D ahl ,K.V.L øken ,andM.M ogstad (2018): “Intergenerationaleffectsofincarceration,” in AEA Papers and Proceedings, vol. 108, pp. 234–240. American Economic Association 2014 Broadway, Suite 305, Nashville, TN 37203. (2020): “Incarceration,recidivism,andemployment,”JournalofPoliticalEconomy,128(4),1269– 1324. B öick , M. (2018): Die Treuhand: Idee-Praxis-Erfahrung 1990-1994. Wallstein Verlag. B raguinsky , S., L. G. B ranstetter , and A. R egateiro (2011): “The incredible shrinking Portuguese firm,” Discussion paper, National Bureau of Economic Research. B urchardi , K. B., and T. A. H assan (2013): “The economic impact of social ties: Evidence from German reunification,” The Quarterly Journal of Economics, 128(3), 1219–1271. B urda , M. C. (2006): “Factor reallocation in eastern Germany after reunification,” American Economic Review, 96(2), 368–374. (2010): “The East German Economy in the Twenty-First Century,” in Conference Paper Washington. B urda , M. C., and J. H unt (2001): “From reunification to economic integration: Productivity and the labor market in Eastern Germany,” Brookings papers on economic activity, 2001(2), 1–92. C han , D. C., M. G entzkow , and C. Y u (2022): “Selection with variation in diagnostic skill: Evidence from radiologists,” The Quarterly Journal of Economics, 137(2), 729–783. C hetty , R., J. N. F riedman , T. O lsen , and L. P istaferri (2011): “Adjustment costs, firm responses, and micro vs. macro labor supply elasticities: Evidence from Danish tax records,” The Quarterly Journal of Economics, 126(2), 749–804. C hoi , J., and A. A. L evchenko (2021): “The long-term effects of industrial policy,” Discussion paper, National Bureau of Economic Research. C riscuolo , C., R. M artin , H. G. O verman , and J. V an R eenen (2019): “Some causal effects of an industrial policy,” American Economic Review, 109(1), 48–85. 32

D auth ,W.,S.L ee ,S.F indeisen ,andT.P orzio (2021): “TransformingInstitutions: LaborReallocation and Wage Growth in a Reunified Germany,” Discussion paper, mimeo. D avis , S. J., and J. H altiwanger (1999): “Gross job flows,” HandbookofLaborEconomics, 3, 2711–2805. D e L oecker , J., and J. K onings (2006): “Job reallocation and productivity growth in a post-socialist economy: Evidence from Slovenian manufacturing,” European Journal of Political Economy, 22(2), 388–408. D jankov , S., and P. M urrell (2002): “Enterprise restructuring in transition: A quantitative survey,” Journal of Economic Literature, 40(3), 739–792. D obbie , W., J. G oldin , and C. S. Y ang (2018): “The effects of pretrial detention on conviction, future crime, and employment: Evidence from randomly assigned judges,” American Economic Review, 108(2), 201–40. D obbie ,W.,andJ.S ong (2015): “Debtreliefanddebtoroutcomes: Measuringtheeffectsofconsumer bankruptcy protection,” American Economic Review, 105(3), 1272–1311. D odds , P., and G. W ächter (1993): “Privatization Contracts with the German Treuhandanstalt: An Insiders’ Guide,” The International Lawyer, pp. 65–90. yck D , I. A. (1997): “Privatization in Eastern Germany: management selection and economic transition,” The American Economic Review, pp. 565–597. F ischer ,W.,H.H ax ,andH.K.S chneider (1993): Treuhandanstalt: DasUnmöglichewagen: Forschungsberichte. De Gruyter Akademie Forschung. F randsen , B., L. L efgren , and E. L eslie (2023): “Judging Judge Fixed Effects,” American Economic Review, 113(1), 253–77. F uchs -S chündeln , N., and M. S chündeln (2005): “Precautionary savings and self-selection: evidence from the German reunification “experiment”,” The Quarterly Journal of Economics, 120(3), 1085–1120. uchs chündeln icola and chündeln atthias F -S , N S , M (2020): “The long-term effects of communism in Eastern Europe,” Journal of Economic Perspectives, 34(2), 172–91. G aricano , L., C. L elarge , and J. V an R eenen (2016): “Firm size distortions and the productivity distribution: Evidence from France,” American Economic Review, 106(11), 3439–79. G iorcelli , M., and B. L i (2021): “Technology transfer and early industrial development: evidence from the Sino-Soviet alliance,” Discussion paper, National Bureau of Economic Research. G litz , A., and E. M eyersson (2020): “Industrial espionage and productivity,” American Economic Review, 110(4), 1055–1103. 33

G reenstone , M., R. H ornbeck , and E. M oretti (2010): “Identifying agglomeration spillovers: Evidencefromwinnersandlosersoflargeplantopenings,”JournalofPoliticalEconomy,118(3),536–598. H arrison , M., and I. Z aksauskiene˙ (2016): “Counter-intelligence in a command economy,” The Economic History Review, 69(1), 131–158. H eise ,S.,andT.P orzio (2021): “Theaggregateanddistributionaleffectsofspatialfrictions,”Discussion paper, National Bureau of Economic Research. H sieh , C.-T., and P. J. K lenow (2009): “Misallocation and manufacturing TFP in China and India,” Quarterly Journal of Economics, 124(4), 1403–1448. H unt , J. (2001): “Post-unification wage growth in East Germany,” Review of Economics and Statistics, 83(1), 190–195. (2006): “StaunchingemigrationfromEastGermany: Ageandthedeterminantsofmigration,” Journal of the European Economic Association, 4(5), 1014–1037. J ohnson , P., and C. P apageorgiou (2020): “What remains of cross-country convergence?,” Journal of Economic Literature, 58(1), 129–75. J uhász , R., N. J. L ane , and D. R odrik (2023): “The New Economics of Industrial Policy,” Discussion paper, National Bureau of Economic Research. alouptsidi K , M. (2018): “Detection and impact of industrial subsidies: The case of Chinese shipbuilding,” The Review of Economic Studies, 85(2), 1111–1158. K line , P., and E. M oretti (2014): “Local economic development, agglomeration economies, and the big push: 100 years of evidence from the Tennessee Valley Authority,” The Quarterly Journal of Economics, 129(1), 275–331. K rueger , A. B., and J.-S. P ischke (1995): “A Comparative Analysis of East and West German Labor Markets: Before and After Unification,” in: Freeman und Katz: Differences and Changes in Wage Structures,, pp. 405–446. ane L ,N.(2022): “Manufacturingrevolutions: IndustrialpolicyandindustrializationinSouthKorea,” Available at SSRN 3890311. L ipschitz , L., and D. M c D onald (1990): “German unification: economic issues,” Discussion paper, International Monetary Fund Washington. L iu , E. (2019): “Industrial policies in production networks,” TheQuarterlyJournalofEconomics, 134(4), 1883–1948. L ópez - de S ilanes ,F.(1997): “Determinantsofprivatizationprices,”TheQuarterlyJournalofEconomics, 112(4), 965–1025. 34

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G M artin , L. A., S. N ataraj , and A. E. H arrison (2017): “In with the big, out with the small: Removing small-scale reservations in India,” American Economic Review, 107(2), 354–86. M egginson , W. L., and J. M. N etter (2001): “From state to market: A survey of empirical studies on privatization,” Journal of economic literature, 39(2), 321–389. M ergele , L., M. H ennicke , and M. L ubczyk (2020): “The big sell: privatizing East Germany’s economy,” . eters P , M. (2022): “Market size and spatial growth—evidence from Germany’s post-war population expulsions,” Econometrica, 90(5), 2357–2396. R edding , S. J., and D. M. S turm (2008): “The costs of remoteness: Evidence from German division and reunification,” American Economic Review, 98(5), 1766–1797. R estuccia , D., and R. R ogerson (2008): “Policy Distortions and Aggregate Productivity with Heterogeneous Establishments,” Review of Economic Dynamics, 11(4), 707–720. R omano , J. P., and M. W olf (2005a): “Exact and approximate stepdown methods for multiple hypothesis testing,” Journal of the American Statistical Association, 100(469), 94–108. (2005b): “Stepwise multiple testing as formalized data snooping,” Econometrica, 73(4), 1237– 1282. S ampat ,B.,andH.L.W illiams (2019): “Howdopatentsaffectfollow-oninnovation? Evidencefrom the human genome,” American Economic Review, 109(1), 203–36. S hleifer , A. (1998): “State versus private ownership,” Journal of Economic Perspectives, 12(4), 133–150. S iebert ,H.(1991): “Germanunification: theeconomicsoftransition,”EconomicPolicy,6(13),287–340. S inn , H.-W. (2002): “Germany’s economic unification: An assessment after ten years,” Review of international Economics, 10(1), 113–128. S nower , D. J., and C. M erkl (2006): “The caring hand that cripples: The East German labor market after reunification,” American Economic Review, 96(2), 375–382. S yverson , C. (2011): “What determines productivity?,” Journal of Economic literature, 49(2), 326–65. U hlig , H. (2008): “The slow decline of East Germany,” Journal of Comparative Economics, 36(4), 517– 541. W ahse ,J.,V.D ahms ,R.S chäfer ,andJ.K ühl (1996): “Beschäftigungsperspektivenvonprivatisierten Unternehmen,” Mitteilungen aus der Arbeitsmarkt-und Berufsforschung, 29(1), 106–116. 35

Supplementary Appendix A Further Empirical Results FigureA.1: PrivatizationsperPrivatizer Number of privatizers: 1659 Number of contracts: 11194 tnecreP 3. 2. 1. 0 1 5 10 15 20 25+ Number of observations per privatizer Notes: Thefigureplotsthenumberofprivatizationhandledperindividualprivatizer(winsorizedat25). Thetotalnumber of privatizations is 11,194. These cases are handled by 1,659 individuals. 5.04% of all cases are organized by privatizers onlyobservedonceinthesample. Thiscorrespondsto652individuals. 36

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G TableA.1: TestofRandomAssignmentofInvestorstoPrivatizers Indep. variable: Stringency Dep. variables Coefficient p-value Adj. p-value Mean Standarddeviation (1) (2) (3) (4) (5) Employment Loginvestorsize -0.0033 0.1228 0.9740 2.4600 1.8220 Investorsize>100employees -0.0008 0.1135 0.9181 0.1400 0.3480 Creditrating Creditworthinessinvestor 0.0768 0.3334 0.9990 284.38 101.58 Highrating -0.0004 0.1101 0.9800 0.0640 0.2460 Location WestGermaninvestor 0.0001 0.8356 0.9990 0.6780 0.4680 Sectoraffiliation Agriculture,forestry,fishing 0.0002 0.5666 0.9930 0.0120 0.1120 Miningandquarrying -0.0001 0.4517 0.9980 0.0040 0.0700 Manufacturing -0.0006 0.1441 0.9860 0.1740 0.3800 Energy -0.0003 0.0505 0.0859 0.0040 0.0680 Water 0.0001 0.7685 0.9990 0.0380 0.1940 Construction -0.0005 0.3395 0.9860 0.1260 0.3320 Retailtrade 0.0012 0.0113 0.5375 0.2000 0.4000 Transportation -0.0003 0.2162 0.9940 0.0540 0.2280 Hospitality 0.0001 0.3576 0.9990 0.0400 0.1940 ICT -0.0001 0.6517 0.9990 0.0420 0.2000 BaningandInsurance 0.0000 0.9975 1.0000 0.0320 0.1760 RealEstate 0.0001 0.7775 0.9990 0.0580 0.2340 Technicalservices -0.0003 0.5271 0.9980 0.1020 0.3040 Economicservices 0.0002 0.2320 0.9940 0.0340 0.1800 Other 0.0003 0.6932 0.9940 0.0760 0.2640 Notes: Thesampleisbasedon4,993contractsmatchedtoinvestorcharacteristics. Eachlinerepresentsasingleregressionof theexplanatoryvariableonthestringencymeasurethattakesvaluesbetween0(minimum)and100(maximum)controlling forTHAofficeandyearofprivatizationfixedeffects. Standarderrorsaretwo-wayclusteredatprivatizerandTHAoffice level. p-valuesincolumn(2)correspondtotheregressionmodelandaretwo-wayclusteredattheprivatizerandTHAoffice level. p-valuesincolumn(3)adjustformultipletestingusingRomano-Wolfprocedure(RomanoandWolf2005a;Romano andWolf2005b)with1,000bootstrapreplications. **p<0.1,**p<0.05,***p<0.01. TableA.2: First-StageRegressionResultsbySub-Samples Baseline Employmentin1990 Revenuein1990 Sectoraffiliation <p(75) <p(50) <p(75) <p(50) Tradeable Non-tradeable (1) (2) (3) (4) (5) (6) (7) Privatizerstringency 0.0015*** 0.0011** 0.0015*** 0.0012** 0.0012* 0.0016*** 0.0009* (0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) Observations 10,616 6,077 6,545 5,985 3,982 5,230 3,003 Averageemploymentatcontractdate 63.22 57.75 73.638 72.39 70.166 69.032 70.554 Averagegrowthrate .062 .028 .084 .038 .038 .048 .006 Sharewithbindingcontracts .208 .184 .232 .194 .194 .22 .146 Samplecondition Baselinecontrols Yes Yes Yes Yes Yes Yes Yes Individualcontrols Yes Yes Yes Yes Yes Yes Yes Industrycontrols Yes Yes Yes Yes Yes Yes Yes Notes:ThetableshowsIVregressionresults. AllspecificationscontrolforfullyinteractedTHAagencyandyearfixedeffects andareconditionalonhavingatleast2privatizationsperprivatizer. Allstratavariables(e.g., employmentin1990)refer totheinitialfirmfromwherethecontractwasgenerated. Thereare335contractsaffiliatedwiththeagriculturesectornot presentedinthetable. F-StatisticreferstotheKleibergen-PaapF-Statistic. Baselinecontrolsarethetimebetweenthefirst andlastauditsmeasuredinday,timebetweencontractdateandfirstauditmeasuredindays,andloginitialemployment level (+1) measured at the first audit. Individual controls are the gender of the privatizer and academic degree (PhD). Industrycontrolsare2-digitindustrydummies. Standarderrorsaretwo-wayclusteredatprivatizerandTHAofficelevel. Instrumentreferstotheleave-one-outmeasureofassigningbindingcontracts. **p<0.1,**p<0.05,***p<0.01. 37

TableA.3: TestofJointNullofMonotonicityandExclusion 10knots 15knots ω=1 ω=0.8 ω=0.5 ω=0.3 ω=1 ω=0.8 ω=0.5 ω=0.3 (1) (2) (3) (4) (5) (6) (7) Teststatistic 535 535 535 535 484 484 484 484 d.f. (509) (509) (509) (509) (504) (504) (504) (504) P-value [0.204] [0.255] [0.408] [0.680] [0.733] [0.916] [1.000] [1.000] Notes: ThetablepresentsresultsfromthetestproposedinFrandsen,Lefgren,andLeslie(2023)forthejointnullhypothesis thatthemonotonicityandexclusionrestrictionshold. WetestthisnullusingTHAofficetimesyear-of-privatizationeffects conditional on having handled at least 5 privatizations. Columns (1) to (4) provide the results imposing 10 knots in the quadraticsplinefunction. Columns(5)to(8)providetheresultsimposing15knotsinthequadraticsplinefunction. Each columnisassociatedwithdifferentweightingschemesbetweenthefitandslopecomponentsofthetest. Afailuretoreject thenullimpliesthatwecannotrejectthehypothesisthatthemonotonicityandexclusionrestrictionsjointlyhold. Thetest wasimplementedinStataviathepackagetestjfe. TableA.4: RegressionResultsAccountingforExtensive/IntensiveMarginPreferences IV-ModelResults Firststage (1) (2) (3) (4) Bindingcontract 0.7246*** 0.7317*** 0.7070*** (0.223) (0.225) (0.249) Investmentpreferences Extensivemargin -0.0005 -0.0005 -0.0004 -0.0005 (0.001) (0.001) (0.001) (0.000) Intensivemargin -0.0005 -0.0004 -0.0004 -0.0004 (0.001) (0.001) (0.001) (0.000) Privatizerstringency(instrument) 0.0018*** (0.000) Observations 9,363 9,363 9,363 9,363 Averageemploymentatcontractdate 60.064 60.064 60.064 60.064 Averagegrowthrate .064 .064 .064 .064 SharewithBindingcontracts 0.207 0.207 0.207 0.207 F-Statistic 17.67 17.03 14.47 Samplecondition Baselinecontrols Yes Yes Yes Yes Individualcontrols No Yes Yes Yes Industrycontrols No No Yes Yes Notes: The table shows IV regression results. All specifications control for fully interacted THA agency and year fixed effectsandareconditionalonhavingatleast5privatizationsperprivatizer. Extensivemargininvestmentpreferencerefer tocontractswithanyinvestment commitments. Intensivemarginpreferencerefertocontracts withtheinvestmenttarget over initial employment in upper decile of the distribution. F-Statistic refers to the Kleibergen-Paap F-Statistic. Baseline controls are the time between the first and last audits measured in days, the time between contract date and first audit measured in months, and initial employment level measured at the first audit. Individual controls are the gender of the privatizer and academic degree (PhD). Industry controls are 2-digit industry dummies. Standard errors are two-way clustered at privatizer and THA office level. Instrument refers to the leave-one-out measure of assigning tight contracts. **p<0.1,**p<0.05,***p<0.01. 38

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G FigureA.2: PersistenceofPrivatizerCharacteristics 100 80 60 40 20 0 -20 citsiretcarahc rezitavirp tuo-eno-evaeL coeff.: .914 (s.e.: .0071) -20 -10 0 10 20 30 40 50 60 70 80 90 100 Lag leave-one-out privatizer characteristic Notes: Thefigureplotstheleave-one-outrateofatightcontract(initialfirmsize<finalcommittedsize)inthepreviouscase againsttheleave-one-outrateofabindingcontractinthecurrentcase. Allplottedvaluesaremean-standardizedresiduals fromregressionsonfullyinteractedTHAofficeandyearofprivatizationfixedeffects. Thebluelinecorrespondstoalinear regression. Thefigureisconstructedbyconditioningofhavinghandledatleastfiveprivatizationcontracts. Totalnumber ofobservationsis8,759. FigureA.3: LaborProductivityacrossFirmSize 11.5 11 10.5 10 9.5 ytivitcudorp robal nL .3 .25 .2 .15 .1 .05 0 ytisneD 2 4 6 8 Log employment in 1990 Notes: The figure plots labor productivity across the firm size distribution among 7,620 initial GDR firms with sales and employmentinformationin1990. Thefigureexcludethetopandthebottom1%oftheproductivitymeasure. 39

TableA.5: RobustnessTests,EmploymentGrowth Dep. variable: Firmgrowth RandomAssignment Coefficient First-stage F-Statistic JointF-test(p-value) (1) (2) (3) (4) A:Instrumentconstruction Onlypastdecisionsforinstrument 1.1690*** 0.0008*** 9.505 0.226 (0.354) (0.000) Above10casesperprivatizer 0.9589** 0.0023*** 11.38 0.355 (0.346) (0.001) Verytightcontracts 1.0160** 0.0015*** 8.511 0.237 (0.424) (0.001) Fulltightnessdistribution -0.0097*** 0.1285*** 21.58 0.848 (0.003) (0.028) Contractwithzeroemploymentinfirst&last 0.6380** 0.0016*** 14.83 0.392 (0.254) (0.000) B:Controlvariables&sampleselection Controlforrenegotiationattempts 0.6588*** 0.0018*** 16.24 0.408 (0.222) (0.000) Controlforpenaltyclause 0.7621*** 0.0016*** 13.65 0.408 (0.264) (0.000) Controlforpurchasingprice&investmenttarget 0.5524** 0.0017*** 13.65 0.408 (0.222) (0.000) Yearsbetweencontractsigned&firstaudit<2 0.6558** 0.0020*** 16.86 0.617 (0.273) (0.001) Monthbetweenfirst&lastaudit>12 0.7387** 0.0018*** 14.26 0.695 (0.269) (0.000) MUPsubsample 0.5435** 0.0018*** 10.45 0.486 (0.256) (0.001) C:Manipulationoftheoutcomevariable Logemploymentdifferences 0.8953** 0.0018*** 12.33 0.408 (0.331) (0.000) Annualizedfirmgrowth,(Lt/L t−1 )1/#year−1 0.2822** 0.0018*** 14.76 0.408 (trimmedattheupperpercentile) (0.114) (0.000) Growthrate<2&>−2 0.7443** 0.0015*** 12.33 0.404 (0.314) (0.000) Notes: ThetableshowsIVregressionresults. AllspecificationscontrolforfullyinteractedTHAofficeandyearfixedeffects and are conditional on having at least five privatizations per privatizer. For sample size reasons, the MUP subsample is conditionalonhavingatleastthreeobservationsperprivatizer. Column(1)showsthepointestimateofthemainvariable ofinterest(exceptthespecificationwithatleast10observationsperprivatizer). Column(2)showsthecorrespondingfirststagecoefficient. F-Statisticincolumn(3)referstotheKleibergen-PaapF-Statistic(first-stage). Allspecificationscondition onthefullsetofcontrolvariablesincludingbaselinecontrols(log)timebetweenthefirstandlastaudits(+1)measuredin days,logtimebetweencontractdateandfirstaudit(+1)measuredindays,andloginitialemploymentlevel(+1)measured atthefirstaudit),individualcontrols(genderoftheprivatizerandacademicdegree(PhD)),and2-digitindustrycontrols. Column(4)showstheF-StatisticofajointF-testofrandomassignment. Thedependentvariableisalwaystheinstrument regressed on log initial employment variables (accounting, purchasing, HR, production, sales, administration, R&D), and loginitialrevenuemeasuredin1990(conditionalonindustry-fixedeffectsandfullyinteractedTHAofficeandtimefixed effects). Standarderrorsaretwo-wayclusteredatprivatizerandTHAofficelevel. ***p<0.1,**p<0.05,***p<0.01. 40

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G TableA.6: OLSRegressionResults,AdjustmentforSampleSelection Employment Productivity Exit (1) (2) (3) (4) (5) (6) Bindingcontract 0.4313*** 0.4302*** 0.0938*** 0.0884*** 0.0229* 0.0223* (0.025) (0.025) (0.022) (0.023) (0.011) (0.011) Millsratio -0.0444 -0.1695** -0.0194** (0.030) (0.076) (0.008) Observations 8,333 8,333 2,399 2,399 3,877 3,876 Averageemploymentatcontractdate 70.926 70.926 47.336 47.336 47.336 47.336 Meanoutcome(non-bindingcontracts) -0.062 -0.062 0.852 0.852 0.051 0.051 Sharewithbindingcontracts 0.192 0.192 0.188 0.188 0.188 0.188 Samplecondition Baselinecontrols Yes Yes Yes Yes Yes Yes Industrycontrols No No No No No No Individualcontrols No No No No No No Notes: The table shows OLS regression results of employment growth, productivity growth and firm exit on binding contracts with and without the inverse mills ratio. The inverse mills ratio is calculated based on a probit specification with the outcome variable being equal to 1 if the GDR initial firms is observed with privatizations contracts that include labor commitment. Explanatory variable in the selection equation are log employment measured in 1990, log sales over employment measured in 1990, THA office FE and industry FE. The selection equation controls for missing values in employmentandsalesoveremploymentbyintroducingdummyvariables. Allspecificationsinthesecondstagecontrolfor fullyinteractedTHAagencyandyearfixedeffects. Baselinecontrolsaretimebetweenthefirstandlastauditsmeasuredin months,timebetweencontractdateandfirstauditmeasuredinmonths,andloginitialemploymentlevelmeasuredatthe firstaudit. Standarderrorsaretwo-wayclusteredatprivatizerandTHAofficelevel. *p<0.1,**p<0.05,***p<0.01. TableA.7: OLSRegressionResultswithDifferentControlVariables&Weighting OLS-ModelResults (1) (2) (3) A:Baseline Bindingcontract 0.4992*** 0.4975*** 0.4975*** (0.031) (0.030) (0.030) B:Complierre-weighting Bindingcontract 0.5336*** 0.5341*** 0.5298*** (0.030) (0.030) (0.030) Observations 9,363 9,363 9,363 Averageemploymentatcontractdate 60.064 60.064 60.064 Averagegrowthrate .064 .064 .064 Sharewithbindingcontracts .207 .207 .207 Samplecondition Baselinecontrols Yes Yes Yes Individualcontrols No Yes Yes Industrycontrols No No Yes Notes: The table shows OLS regression results. All specifications control for fully interacted THA agency and year fixed effectsandareconditionalonhavingatleastfiveprivatizationsperprivatizer. Baselinecontrolsarelogtimebetweenthe firstandlastaudits(+1)measuredindays,logtimebetweencontractdateandfirstaudit(+1)measuredindays,andlog initialemploymentlevel(+1)measuredatthefirstaudit. Individualcontrolsarethegenderoftheprivatizerandacademic degree (PhD). Industry controls are 2-digit industry dummies. Standard errors are two-way clustered at privatizer and THAofficelevel. *p<0.1,**p<0.05,***p<0.01. 41

TableA.8: RegressionResults,CumulativePatentsduringCommitmentPeriod OLS 2S2SLS (1) (2) (3) (4) Bindingcontract 0.0050** 0.0049* 0.0061** 0.0081 (0.002) (0.002) (0.003) (0.057) Observations 4,563 4,563 4,563 1,430 MeanofYofbindingcontracts .012 .012 .012 .012 MeanofYofnon-bindingcontracts .008 .008 .008 .006 Samplecondition Baselinecontrols Yes Yes Yes Yes Industrycontrols Yes Yes Yes Yes Privatizercontrols No Yes Yes Yes Purchasingprice No No Yes Yes Notes: The table shows OLS regression results of patenting probabilities during the commitment period. The outcome variable takes the value of 1 if the firm has at least one patent during the period under commitment. All specifications controlforfullyinteractedTHAagencyandyearfixedeffects. Bindingcontractsaredefinedasinitialfirmsizebelowthe committed target level. Baseline controls are the timing variable as in the baseline specification, log initial firm size, and an indicator if the firm has at least one patent before the contract date. Industry controls are 2-digit industry dummies. Standarderrorsaretwo-wayclusteredatprivatizerandTHAofficelevel. Thestandarderrorincolumn(4)isbootstrapped using2,500replications. *p<0.1,**p<0.05,***p<0.01. TableA.9: RobustnessTests,ProductivityGrowth OLS 2S2SLS (1) (2) (3) (4) (5) (6) A:Baseline&industrycontrols α =1 α =0.7 α =0.8 α =1 α =0.7 α =0.8 Bindingcontracts 0.0753*** 0.1059*** 0.0965*** 0.6378* 0.6195* 0.6307* (0.025) (0.021) (0.022) (0.359) (0.368) (0.362) Averageproductivitygrowthrate 0.884 0.868 0.875 0.884 0.870 0.876 Observations 2,395 2,395 2,395 1,612 1,612 1,612 B:Investmentcommitmentcontrol α =1 α =0.7 α =0.8 α =1 α =0.7 α =0.8 Bindingcontracts 0.0647** 0.0909*** 0.0828*** 0.6793* 0.7004* 0.6989* (0.026) (0.022) (0.023) (0.370) (0.380) (0.374) Averageproductivitygrowthrate 0.884 0.868 0.875 0.884 0.870 0.876 Observations 2,395 2,395 2,395 1,612 1,612 1,612 C:Includingexits α =1 α =0.7 α =0.8 α =1 α =0.7 α =0.8 Bindingcontracts 0.0827** 0.1074*** 0.0999*** 0.5173 0.5650 0.5522 (0.032) (0.029) (0.030) (0.497) (0.509) (0.503) Averageproductivitygrowthrate 0.808 0.792 0.799 0.822 0.810 0.816 Observations 2,480 2,480 2,480 1,656 1,656 1,656 D:July1990productivity α =1 α =1 Bindingcontracts 0.0564** 0.6701* (0.026) (0.370) Averageproductivitygrowthrate 0.882 0.882 Observations 2,414 1,624 Notes: ThetableshowsOLSand2S2SLSregressionresultsofmeasuresofproductivitygrowthonbindingcontracts. Allspecifications control for fully interacted THA agency and year fixed effects. Binding contracts are defined as initial firm size below the committed targetlevel. Baselinecontrolsareasinthebaselinespecification. PanelAcontrolsforthebaselinecontrolvariables. PanelBincludesas furthercontrolsadummyifthecontracthasinvestmentcommitments. PanelCincludesexiterswhencalculatingproductivitygrowth rates. PanelCperformsa(cid:101)-transformation. PanelDusestheactualmeasurenumbersofsalesandemploymentin1990. Standarderrors incolumns(1)-(3)aretwo-wayclusteredatprivatizerandTHAofficelevel.Thestandarderrorsincolumns(4)-(6)arebootstrappedusing 2,500replications.*p<0.1,**p<0.05,***p<0.01. 42

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G TableA.10: RobustnessTests,TFPGrowth OLS 2S2SLS (1) (2) (3) (4) A:Baseline Bindingcontracts 0.1108** 0.1239*** 0.1266*** 0.6608*** (0.039) (0.041) (0.042) (0.213) Observations 1,825 1,825 1,825 1,825 B:Includingexits Bindingcontracts 0.0784 0.0931* 0.0978* 0.6159*** (0.052) (0.052) (0.054) (0.213) Observations 1,835 1,835 1,835 1,835 TFPgrowth .348 .348 .348 .348 Laborprod. growth .53 .53 .53 .53 Baselinecontrols Yes Yes Yes Yes Individualcontrols Yes Yes Yes Yes Industrycontrols Yes Yes Yes Yes Purchasingprice No Yes Yes Yes Investmenttarget No No Yes No Notes: The table shows OLS and 2S2SLS regression results of TFP growth on binding contracts. Panel A provides the baselineresultsconditionalonsurvivalinthefinalcommitmentyear. PanelBinducesa-2forfirmsthatexitintheyearof thefinalcommitment. AllspecificationscontrolforfullyinteractedTHAagencyandyearfixedeffects. bindingcontracts are defined as initial firm size below the committed target level. Baseline controls are as in the baseline specification. Column(1)controlsforthebaselinecontrolvariables. Column(2)includesthepurchasingprice(flexiblyintroducedusing deciledummies). Columns(3)and(4)includeadummyforinvestmenttargets. Standarderrorsincolumns(1)-(3)aretwowayclusteredatprivatizerandTHAofficelevel. Thestandarderrorincolumn(4)isbootstrappedusing2,500replications. *p<0.1,**p<0.05,***p<0.01. FigureA.4: CorrelationProductivityMeasures ProductivityinLevel 16 14 12 10 8 6 4 2 0 -2 PFT YearlyChangeinProductivity 12 coeff.: .8497 (s.e.: .0029) 10 8 6 4 2 0 -2 -4 -6 -8 2 4 6 8 10 12 14 16 18 20 Log Labor Productivity PFT atleD coeff.: .7791 (s.e.: .0035) -8 -6 -4 -2 0 2 4 6 8 10 Delta Log Labor Productivity Notes: The figure plots two measures of firm-level productivity using a matched sample of contracts with THA survey data(Soestra). TheleftpanelshowsthecorrelationbetweentheloglaborproductivityandlogTFP.Therightpanelshows the correlation between the yearly change in log labor productivity and the yearly change in log TFP. TFP is calculated basedonaCobbDouglasproductionfunctionwithlogrevenueasmeasuredoutputandlogemploymentandlogcapital asinputs. 43

FigureA.5: Bunchingwithdifferentpolynomials A:Polynomialdegree: 6 Employment < Commitment Employment > Commitment Excess mass (b) = 10.1 Standard error = 1.973 ycneuqerF 0004 0003 0002 0001 0 B:Polynomialdegree: 9 Employment < Commitment Employment > Commitment Excess mass (b) = 8.359 Standard error = 1.407 -40 -20 0 20 40 Committed - Realized Employment Counterfactual distribution Observed distribution ycneuqerF 0004 0003 0002 0001 0 C:Polynomialdegree: 18 Employment < Commitment Employment > Commitment Excess mass (b) = 5.795 Standard error = .663 -40 -20 0 20 40 Committed - Realized Employment Counterfactual distribution Observed distribution ycneuqerF 0004 0003 0002 0001 0 -40 -20 0 20 40 Committed - Realized Employment Counterfactual distribution Observed distribution Notes: The figures show the employment distribution around the committed employment (demarcated by the vertical red line at 0) for contracts between 1990-2002. The blue line in dots is a histogram of actual employment relative to the commitmenttargetinthefinalcommitmentyear. Eachpointshowsthenumberofobservationsinemploymentcountbin (deviationbetweenthetargetandtherealizedemployment). Thesolidlinebeneaththeempiricaldistributionisatwelvedegreepolynomialfittedtotheempiricaldistributionexcludingtheareaofmissingoneemployeeandhave4employees more than committed. The shaded region in yellow is the estimated excess mass. Standard error is calculated using a parametricbootstrapprocedure. EstimationbasedonChetty, Friedman, Olsen, andPistaferri(2011). PanelAshowsthe results using a six-degree polynomial order. Panel B shows the results using a ninth-degree polynomial order. Panel C showstheresultsusingaeighteenth-degreepolynomialorder. FigureA.6: BunchingwithsymmetricR A:R=[1,1] Employment < Commitment Employment > Commitment Excess mass (b) = 5.209 Standard error = .529 ycneuqerF 0004 0003 0002 0001 0 B:R=[2,2] Employment < Commitment Employment > Commitment Excess mass (b) = 6.341 Standard error = .757 -40 -20 0 20 40 Committed - Realized Employment Counterfactual distribution Observed distribution ycneuqerF 0004 0003 0002 0001 0 C:R=[4,4] Employment < Commitment Employment > Commitment Excess mass (b) = 7.806 Standard error = 2.293 -40 -20 0 20 40 Committed - Realized Employment Counterfactual distribution Observed distribution ycneuqerF 0004 0003 0002 0001 0 -40 -20 0 20 40 Committed - Realized Employment Counterfactual distribution Observed distribution Notes: The figures show the employment distribution around the committed employment (demarcated by the vertical red line at 0) for contracts between 1990-1995. The blue line in dots is a histogram of actual employment relative to the commitmenttargetinthefinalcommitmentyear. Eachpointshowsthenumberofobservationsinemploymentcountbin (deviationbetweenthetargetandtherealizedemployment). Thesolidlinebeneaththeempiricaldistributionisatwelvedegreepolynomialfittedtotheempiricaldistributionexcludingtheareaofmissingoneemployeeandhave4employees more than committed. The shaded region in yellow is the estimated excess mass. Standard error is calculated using a parametricbootstrapprocedure. EstimationbasedonChetty, Friedman, Olsen, andPistaferri(2011). PanelAshowsthe resultsexcluding-1and1. PanelBshowstheresultsexcluding-2and2. PanelCshowstheresultsexcluding-4and4. 44

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G FigureA.7: Bunchingwithpercentdeviationbin A:1percentbins Employment < Commitment Employment > Commitment Excess mass (b) = 23.64 Standard error = 7.275 ycneuqerF 0004 0003 0002 0001 0 B:2percentbins Employment < Commitment Employment > Commitment Excess mass (b) = 11.18 Standard error = 2.483 -40 -20 0 20 40 Committed over Realized Employment Counterfactual distribution Observed distribution ycneuqerF 0004 0003 0002 0001 0 C:5percentbins Employment > Comitment Employment < Commitment Excess mass (b) = 2.95 Standard error = .6922 -40 -20 0 20 40 Committed over Realized Employment Counterfactual distribution Observed distribution ycneuqerF 0004 0003 0002 0001 0 -50 -25 0 25 50 Committed over Realized Employment Counterfactual distribution Observed distribution Notes: The figures show the employment distribution around the committed employment (demarcated by the vertical red line at 0) for contracts between 1990-1995. The blue line in dots is a histogram of actual employment relative to the commitmenttargetinthefinalcommitmentyear. Eachpointshowsthenumberofobservationsinemploymentcountbin (deviationbetweenthetargetandtherealizedemployment). Thesolidlinebeneaththeempiricaldistributionisatwelvedegreepolynomialfittedtotheempiricaldistributionexcludingtheareaofmissingoneemployeeandhave4employees more than committed. The shaded region in yellow is the estimated excess mass. Standard error is calculated using a parametricbootstrapprocedure. EstimationbasedonChetty, Friedman, Olsen, andPistaferri(2011). PanelAshowsthe resultsbyconstructing1percentagebindeviations. PanelBshowstheresultsbyconstructing2percentagebindeviations. PanelCshowstheresultsbyconstructing5percentagebindeviations. 45

TableA.11: BunchingbySub-Samples Excessmass(b) Standarderror (1) (2) A:Industryaffiliation Agriculture,energy,mining 9.076 3.220 Chemistry,plastics 4.952 0.9231 Extractionofcut-stone,iron,casting,steelforming 7.842 3.351 Steelconstruction,mechanical&electricalengineering,automobile 6.699 1.023 Paper,print,textile,food 7.617 1.070 Constructionandbuildingstrades,wholesale,retail 7.257 1.227 Transportation,communication,insurance 5.799 1.325 B:Contractmaturity 16to31months 6.574 1.198 Below16months 8.697 3.010 Above31months 7.212 1.118 C:Numberofaudits Multipleaudits 6.521 0.773 D:Penaltycondition Excludecontractswithoutpenaltyclause 6.627 0.889 E:Initialsize Belowtarget 6.473 0.989 Abovetarget 5.151 0.540 Notes: Thetableshowsbunchingestimatesoftheemploymentdistributionaroundthecommittedemploymentforcontracts between1990-1995bydifferentgroups. Thecounterfactualdistributionisabasedonatwelve-degreepolynomialfittedto theempiricaldistributionexcludingtheareaofmissingoneemployeeandhavingthreeemployeesmorethancommitted. Standarderrorsarecalculatedusingaparametricbootstrapprocedurewith100replications. EstimationbasedonChetty, Friedman, Olsen, and Pistaferri (2011). Panel A shows the results by industry. Panel B shows the results by contract maturity cutting at 25th (16 months between contract date and final commitment) and 75th (31 months between contract dateandfinalcommitment)percentile. PanelsCandDselectonlycontractswithmultipleauditsandwithapenaltyclause, respectively. PanelEdistinguishesbyinitialcontractsize(measureatthefirstaudit)relativetothefinaltarget. 46

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G B Data Addendum - ISUD Data Environment This section provides an overview and a description of data used in the empirical analysis. The data were provided to the authors on the basis of an agreement between the IWH (Halle) and the German Federal Archives (Bundesarchiv). This agreement involved the transfer of more than 500 separate data tables in digitized format (csv) on activities of Treuhand. The timeline in Figure B.1 visualizes the level and timing of observations. The main identifiers in the ISUD environment are at the firm level and at the contract level. The former is constituted by information from firms submitting a balance sheet (DM Eröffnungsbilanz) and transitioning into the THA portfolio. The THA assigns initial IDs to each firm, and, in the case of restructurings and firm separations, new IDs are created. Once assets are sold out of the firms, we observe contract IDs. These contracts are organized and used by the contract management teams (VM) to follow up on payments and obligations of buyers.28 FigureB.1: Timelinefromreunificationtothemarketperiod Eröffnungsbilanz THAPeriod Privatizations MarketPeriod (01.07.1990) Firms FirmID ContractID Liquidation t BalanceSheets THAPlanning VMControlManagement MUP Baselinenumbers Growth (Basiskennziffern) Guarantee Payments Employment Financials (Bürgschaften) (Raten) Exit (Finanzierungsdaten) LegacyDebt Outlays&Inflows Liquidation (Altkredit) (Budgetverfolgung) (Abwicklungsdaten) Liabilities (Verbindlichkeiten) Commitments Audits (Zusicherungen) (Überprüfung) Twotablesareusedtomeasurefirm-levelinformation: basis_kennziffernandbasis_kennziffern_ 91. The table basis_kennziffern_91 comprises most of the information and, therefore, is the main table. Incaseofmissingvalues, wesearchforinformationinbasis_kennzifferntocomplementand to construct a comprehensive cross-section of firm information for the year 1990.29 The information 28SectionCdescribesthemergebetweencontractsandexternalfirm-leveldata,theMannheimEnterprisePanel(MUP), tostudydynamicsbeyondthecommitmentperiod. 29Theinformationcanbecombinedtoconstructayearlypanelwithinformationatthefirmlevelbetween1989and1994. Thisdatasetcannotbeusedtostudytheevolutionoffirmsovertimebecausethefirmdisappearsfromthedatasetoncethe firmtransitionsoutoftheTHAportfolioeitherbecauseofaprivatizationorliquidation. 47

relates to employment (including a breakdown into production workers, HR, and administration), revenues (including a breakdown of revenues in East and West Europe), and the assignment of firms to THA offices (headquarters or local subsidiary). The data contains a total of 13,552 legal firm entities, out of which 93.3% are observed for the first time in 1990.30 We complement the data with additional industry information from the SOESTRA survey (see Mergele, Hennicke, and Lubczyk 2020). The final data set is used in the analysis to study random assignment of firms to privatizers in Table 2, to calculate labor productivity growth between 1990 and the final commitment year in Table 4, market exit effects in Table 5, and to construct Figure A.3. A second set of data tables provides information on ownership changes of firms: besitz_91 and besitz. Similarly, besitz_91 comprises most information, and besitz is used to fill missing values. Combining the two tables generates a dataset with information on 13,051 firms about partial sales, privatizationsandliquidationdecisions. Thesedataallowsustonotonlytrackchangesinownership, but also to calculate the share of firms privatized or liquidated. We refer to this estimated share in Section 2. One of the main challenges of the ISUD data environment is to link information at the firm level with contract-level information. This link is important for two reasons. First, it allows us to study random assignment, productivity growth, and market exit. Second, it provides us information on which THA division handles the privatization of the firm. We first describe the data tables used to construct the link between firms and contracts. Table B.1 provides an overview of the data tables and a short description. The data table ASVA01T forms the main source of information for contracts. It provides us with information on the contract ID and the contract date. It does not, however, provide information on the link between the contracts and the firm. For this reason, we search for this information across the ISUDsystem. ThetablesASVA02T,VATVT,ASVA22T,ASVA50T,andFE3_VTareidentifiedtobecandidates that possess the link. Due to the degree of non-missing information, the two most important tables are ASVA02T and VATVT. The search process generates 48,086 unique contracts with a firm link. Another advantageous feature of ASVA01T is that it contains not only the contract ID but also the string names of privatizers who handle the contracts and communicate/negotiate with potential investors. We clean the variable “PNAME” which is labeled as “Name d. zuständigen Privatisierers”. In the overall file, we generate 3,521 unique names for 58,544 contracts after name cleaning. The main reason for losing contracts is missing values in this name variable. Out of the 256,842 contracts in the data table, 147,060 do not have information on the name of the privatizer. The reason why most of the contracts do not possess a name of a privatizer is because the contracts are not related to firms but represent estate, machinery or land deals. Therefore, these contracts are not related to firmsandconsequentlydonothaveaprivatizerattachedtoit. Linkingcontractstocontainprivatizer information, labor commitment contracts, and firm links generates a sample of 11,194 contracts as shown in Section 5. 30THAcreatedlegalentitiesovertime,and,asaresult,5.1%offirmsareobservedforthefirsttimein1991,and1.2%in 1992,and0.48%in1993. 48

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G TableB.1: Contract-LevelDataTables Tablenames Description A:Baselinetables ThetablescontainsmasterdataandstatusinformationforcontractssignedwiththeTHA.Itcombines ASVA01T manyvariablesfromdifferenttables.ThetablecontainsthecontractID(sysnr),thedateofthecontract signedwiththenotary,andthenameoftheprivatizer.Totalnumberofuniquecontracts:256,842. Thetableprovidesinformationonpartialcontracts. Itcontainsthelinkbetweenthecontractsandthe ASVA02T firms,thefixedpricepayedbythecontractpartner,andtheassignmenttoTHAoffices.Totalnumberof uniquecontracts:213,052.Uniquecontractswithanon-missingcontract-firmlink:22,837. Thetableprovidesinformationonpartialcontracts. Itcontainsthelinkbetweenthecontractsandthe VATVT firms. Total number of unique contracts: 37,967. Unique contracts with a non-missing contract-firm link:30,745. Thistableprovidesinformationonmappings. Itcontainsthelinkbetweenthecontractsandthefirms. ASVA22T Totalnumberofuniquecontracts:40,036.Uniquecontractswithanon-missingcontract-firmlink:9,784. Thistableprovidesheaderdataforconcertedaction. Itcontainsthelinkbetweenthecontractsandthe ASVA50T firms.Totalnumberofuniquecontracts:82.Uniquecontractswithanon-missingcontract-firmlink:82. Thistableprovidesinformationonprocesses/operationsofmaintablesrelatedtofinancials.Itcontains FE3_VT thelinkbetweenthecontractsandthefirms.Totalnumberofuniquecontracts:1,723.Uniquecontracts withanon-missingcontract-firmlink:1,710. B:LaborCommitments&Audits Thistableprovidesinformationonlaborcommitmentsofthecontractpartner. Totalnumberofunique VAPST contracts:17,753.Totalnumberofobservations:52,438. Thistableprovidesinformationonlaboraudits.Totalnumberofuniquecontracts:16,583.Totalnumber VAPIT ofobservations:116,619. Thistableprovidesinformationonlaborauditsandislabeledashistoryinthedocumentation. Total VAPITH numberofuniquecontracts:19,052.Totalnumberofobservations:102,933. Thistable,amongothers,providesinformationonlaborcommitments. Totalnumberofuniqueoverall ASVA12T contracts: 275,054. Totalnumberofuniquecontractswithpositivenumberofcommittedlabor: 22,535. Totalnumberofobservations:322,829. Thistable, amongothers, providesinformationonlaboraudits. Totalnumberofuniqueoverallcon- ASVA13T tracts: 47,111. Totalnumberofuniquecontractswithpositivenumberofauditedlabor: 15,702. Total numberofobservations:153,155. C:InvestmentCommitments&Audits Thistableprovidesinformationoninvestmentcommitmentsofthecontractpartner. Totalnumberof VAZST uniquecontracts:18,120.Totalnumberofobservations:20,366. Thistableprovidesinformationoninvestmentaudits. Totalnumberofuniquecontracts: 16,806. Total VAZIT numberofobservations:32,096. Thistableprovidesinformationoninvestmentauditsandislabeledashistoryinthedocumentation. VAZITH Totalnumberofuniquecontracts:26,195.Totalnumberofobservations:60,159. Thistable,amongothers,providesinformationoninvestmentcommitments. Totalnumberofunique ASVA15T overallcontracts:274,375. Totalnumberofuniquecontractswithpositivenumberofcommittedinvestment:24,220.Totalnumberofobservations:280,370. Thistable,amongothers,providesinformationoninvestmentaudits. Totalnumberofuniqueoverall ASVA16T contracts:47,111.Totalnumberofuniquecontractswithpositivenumberofauditedinvestment:15,619. Totalnumberofobservations:64,725. After this preparation of baseline tables, we obtain information on labor commitments and labor audits. We start with the original files that are called VAPST for commitment information and VAPIT for information on audits (see Panel B of Table B.1). These two tables can be seen as the original tables as suggested from the delivered pdf documentation by the German Federal Archives. The 49

pdf file for labor commitments is shown in Figure B.2. It shows the template how the data was collected in the first place by THA employees. The top right corner corresponds to the tables VAPST and VAPIT, respectively. In these two data tables we observe 17,753 unique contracts with labor commitments and 16,583 contracts with at least one audit. As presented in Panel B, the total number ofobservationsinbothtablesishigherbecausetherecanbemultiplecommitmentsfordifferentyears of the commitment period as well as several audits per commitment. FigureB.2: PaperFile: LaborCommitment Notes: ThefiguresshowtheoriginaltemplateusedbytheTHAtodocumentlaborcommitments. We perform the following steps to clean the data. First, we drop observations without date information in both tables and select the first contract within the contract ID in case there are several partial contracts per ID. Out of the 116,619 contract-audit observations, these selection steps reduce the sample by 36 and 674, respectively. Out of the 52,438 contract-commitment observations, these selection steps reduce the sample by 1,414 and 367, respectively. Within the VAPIT file we also drop observationswherethenumberofemployeesattheauditiszero, butthevariablethatstateswhether employeeinformationisreportedissettozero. Thisreducesthesamplefurtherby2,536observations. In order to obtain an initial firm size measure at the contract level, we select the first audit. The last audited labor information provides a measure of the size at the final commitment time. We further perform basic data cleaning steps: (i) we drop contracts if the date of the last commitment is before thedateofthecontractwiththenotary(7observations),(ii)ifthetimebetweentwoconsecutivecommitments is negative, and (iii) if the final employment commitment is zero (224 observations). This 50

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G generates a sample with 15,538 labor commitment contracts with at least one matched employment audit. The ISUD environment further contains a table called ASVA12T with labor commitment contracts. The original table has 322,829 observations. The majority of these observations are labeled as having no labor commitments. We compare this data table with the original VAPST table. Conditional on observing one contract ID in both tables (VAPST and ASVA12T) shows that the information is identical. However, ASVA12T has 5,125 additional contracts with labor commitments that are not included in VAPST. These additional contracts are, on average, later written out and are entered into the ISUD data system mainly in 2003 and 2004. After following the same data cleaning steps, we end up with 3,385additionalcontracts. Intermsoflaboraudits,however,thesecontractsarenotobservedinVAPIT. There exists another data table that is a natural suspect and is called ASVA13T. But again, this table does not contain audit information for the additional contracts with observed labor commitments.31 After searching for possible contracts with additional audit information, we found that the history versionofVAPIT,calledVAPITH,issuitabletofillpartsofthemissingauditsfromASVA12T. Amongthe 3,385 additional contracts after basic data cleaning steps, we are able to merge the audit information for2,702contracts. Together,thesedatatablesgenerateourfinalsampleof18,235contractswithlabor commitments. For the empirical specifications accounting for extensive/intensive margin privatizer preferences presented in Table A.4, we make further use of investment commitment contracts. The logic and steps in the data cleaning process apply similarly to investment commitment contracts. Figure B.3 shows the template used for the documentation of investment commitments. The baseline data table for investment with information on investment commitments is called VAZST, whereas the table for investment audits is called VAZIT. Panel C of Table B.1 provides a list and short description of the investment commitment related data tables. AfterbasicdatacleaningstepsandcombiningcommitmentinformationinVAZSTwithauditinformation in VAZIT, we obtain a dataset with 15,086 investment commitments. The data table ASVA15T has 7,127 additional contracts that are not observed in the baseline files. Similar to the additional employment contracts, ASVA16T does not contain audits to these additional contracts. Again, exploiting VAZITH, the history file of VAZIT, we are able to add 4,978 contracts. Together, these data tables generate our final sample of 20,062 contracts with investment commitments. One remarkable difference between investment and labor commitment contracts is the number of audits. While the share of contracts with only one audit is about 17% among the labor commitment contracts, this share is 65.2%. Due to the flow nature of investment commitment, there are fewer auditsduringthecommitmentperiod. Combininglaborwithinvestmentcontractsresultsinasample of 23,662 unique contract-level observation. Among them, 14,635 contracts have both, labor and investment commitments, 5,427 only have investment commitments, and 3,600 contracts only have labor commitments. In order to calculate extensive margin preferences i.e., writing contracts with 31Outofthe5,125additionalcontractswithlaborcommitmentsASVA12T,17contractsarefoundinVAPITand22contracts arefoundinASVA13T. 51

FigureB.3: PaperFile: InvestmentCommitment Notes: ThefiguresshowtheoriginaltemplateusedbytheTHAtodocumentinvestmentcommitments. any labor commitment condition we merge this combined dataset with the 58,544 contracts with cleaned privatizer names. 52

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G C Data Addendum - Merging Contracts to Mannheim Enterprise Panel Data This section describes the merge between our baseline contract-level data and the Mannheim Enterprise Panel data, which cover firms in East Germany starting from 1993 to 2019 (the most recent wave). TheMannheimEnterprisePanel(MUP),isthemostcomprehensivemicrodatabaseofcompanies in Germany outside of administrative data. Official administrative data is usually not accessible to the public. The data contains detailed information on the firm-level that is often hard to come by in administrative records such as, for instance, the date of creation and closure of a company, ownership structures, and credit rating scores. Besides that, the dataset comprises employment, sales, and industry affiliation information. The MUP is based on the firm data pool of Creditreform e.V., which is the largest credit rating agency in Germany. While it has broad overall coverage it does not offer 100% coverage (for further details, see Bersch, Gottschalk, Müller, and Niefert (2014)). At the level of the contracts, we do not observe firm names that would allow a string matching based on these names. Instead, we explore the ownership information in both datasets. In the MUP data,weobserveforeachfirmowner. Inthecontract-leveldata,wehaveaccesstothecontractpartner, who ususally becomes the new owner of the company after the contract is signed with the notary. Among the 18,235 contracts in the baseline data, we start off with 9,538 that can be linked via name matching between the owners in the MUP and contract partners in the contract data. These observationscorrespond to11,199 contractpartners. Theseindividualsusually havemultiple linksto firms at different points in time and across space. In order to select the correct firm to the contract, we perform the following pre-selection: • Drop if firm is located in West Germany • Drop if original firm under Treuhand is located in different Federal State than MUP firm • Drop if firm/contract location, date of incorporation, contract date is missing • Drop if date of incorporation/ownership start is after 2000 • Drop if contract date is five years after date of incorporation The first two selection criteria are based on regional information. We assume that the contract does not belong to the privatized eastern firm or asset if the firm in the MUP dataset is located in West Germany. We also drop observations if the former GDR firm and the MUP firm are located in different Federal States (within East Germany). Moreover, if we do not observe the region, the contract date or the date of incorporation, we drop the entire observation. We also drop observations if the date of incorporation or the start of the ownership period is post 2000. As a last step of the pre-selection procedure, we drop contracts if the contract date is more than five years after the date of incorporation. The reason behind this is that the contract date should mark the creation of a firm and therefore should be close to the date of incorporation. This leaves us with 7,415 contracts and 53

8,952 contract partners. Per contract partner, we find about two owners with the same name in the MUP data at the median. The 99th percentile corresponds to 51 potential matches, which are rather common names that are matched several times in the MUP data. We therefore exclude the upper decile (more than nine different IDs in the MUP data) of the matches as a further pre-selection step. With these potential matches at hand, we need to select the firm that matches best. In order to perform the selection and exclude MUP firms that are likely not behind the privatization contract, we construct three indicator variables based on the region (county, state), the dates (incorporation, contract), and the employment deviation. At the regional level, we construct an indicator equal to 1 if the regional information in both dataset coincide. The date indicator is equal to 1 if the absolute difference between the two available dates is at most three years. For the indicator for employment deviation, we first calculate observed employment deviations for all the year where we observe employmentnumbersinbothdatasets. Itispossibletohavemorethanoneobservationperfirmbecause audits happen at different points in time. The employment deviation measure, naturally, can only be calculated among contracts with labor commitments. The employment deviation indicator is set to be 1 for the match with the smallest difference. We then drop potential matches if regional and date information do not coincide with each other. In cases where we only observe date information, we select the MUP firm with the closest date of incorporationtothecontractdate. If,forexample,therearetwopossiblematchesofMUPfirmsinthe sameregionandincorporatedinthesameyear,weneedtodropthecontractentirelyfromthesample as we cannot select the bast match. Our final match consists of 4,805 firms with labor commitment contracts that are observed in a panel structure. Table C.1 provides an overview on the selection criteria. It states that 38% of our matches are based on the exact county, date (date of incorporation and contract date) and audit information. Another15.5%ofthematchesareselectedbasedontheFederalStateinformation, thedateandaudit information. This indicates that slightly more than 50% are based on region, date and employment information available in both dataset. Then, there are some few matches of around 10% that are only based on region and date or region and audit information. About a quarter of the matches are based only on the information of the contract date and the date of incorporation, whereas 2.6% are only basedonauditinformation. Finally,8.1%oftheselectedMUPfirmsareselectedbecausethereisonly one possible match, i.e., the matched owner has only one firm ID attached. Given the pre-selection criteria, all observed matches are in the same state. Conditional on nonmissing county information, our final matched MUP firms to contracts that come out of former GDR firms are in 73% of all cases located in the same county as the MUP firms. Moreover, the average absolute difference between the date of the contract and the date of the incorporation in the MUP data is 1.12 years (median is equal to 1 year). Based on the firm-year observation, we are able to merge employment audits from the contract management system of the ISUD environment. Note that in the selection procedure, we have used the match with the smallest deviation. We will now be able to justify the match by studying employment number differences between the two datasets. For 3,609 firms, we observe at least one audit 54

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G TableC.1: SourcesofSelectedMatches Share Selectedbasedoncounty,date&auditinformation 0.378 Selectedbasedonstate,date&auditinformation 0.155 Selectedbasedoncounty&date 0.016 Selectedbasedonstate&date 0.009 Selectedbasedoncounty&auditinformation 0.061 Selectedbasedonstate&auditinformation 0.024 Selectedbasedononlystate 0.006 Selectedbasedononlydate 0.247 Selectedbasedononlyauditinformation 0.026 Selectedbasedonly1possiblemerge 0.081 Notes: ThetableshowsthesourceofselectedmatchesbetweentheISUDdataandtheMUPdataset. Themajorityofselected matchesarebasedoncounty,dateandauditinformation. About25%areonlyselectedbaseondateinformationand16% arebasedonthesamestate,dateandauditinformation. (with positive employment information) which allows us to calculate employment differences. The median (mean) estimated difference in employment is 0 (2.67). However, we observe large tails in the distribution of employment differences. For this reason, we further drop matches with absolute employment differences above 500. After this adjustment, our sample consists of 4,735 firms. At this stage, we do not drop firms if precise calculations of employment differences are not possible, which means that we rely on date and regional information for the merge. Figure C.1 provides a comparison between the contract-level employment information and the MUPdata. PanelAprovidesavisualizationofcountdifferenceswithamedianofzero(maximumof 500 by construction). Panel B shows the inverse hyperbolic sine transformed employment numbers between the MUP firm-level data in black and the contract-level data in grey. These results suggests that the MUP firm-level data shows slightly more mass among smaller firms. FigureC.1: ComparisonofEmploymentFiguresbetweenContractsandMUP A:EmploymentDifferences ytisneD 80. 60. 40. 20. 0 B:EmploymentDistributions Number of firms: 3609 Median: 0 -400 -200 0 200 400 600 Absolute employment difference ytisneD 3. 2. 1. 0 Number of firms: 3609 0 2 4 6 8 Number of employees (inverse hyperbolic sine) Firm-level data Contract-level data Notes: Panel A shows employment differences between matched contracts and firms in the MUP data that is centered around0. PanelBshowsthelogemploymentdistributionofmatchedcontractsandtheemploymentdistributioninthe MUPdataset. Numberofobservationswithemploymentinformationinbothdatasetsis3,609. 55

To evaluate the quality of the match, we calculate the share of firms that are “close” to each other in terms of employment figures. To arrive to such a statement, we first calculate the relative employment differences as: (empl −empl ) employment = MUP ISUD , diff (empl +empl ) MUP ISUD where empl and empl refer to the respective employment figures in both datasets. We MUP ISUD then define a match to be close or acceptable if the employment difference is smaller or equal to following threshold value: 1 abs(employment ) ≤ . diff (cid:112) (min[empl ,empl ]+1) MUP ISUD Thisequationtakesintoaccountthelevelofemploymentandallowsforhigherrelativedeviations among small firms. To provide an example, consider the following case with empl = 1 and ISUD empl = 3. Thisgeneratesarelativeemploymentdifference, employment , equalto0.5, whichis MUP diff smallerthanthethresholdvalueof0.707andthereforeconsideredtobecloseenoughtobeacceptable. The case where, for example, empl = 100 and empl = 300 also provides a measure of ISUD MUP employment equal to 0.5. However, the threshold value becomes 0.099 and therefore labels this diff merge as not close enough to be acceptable. Figure C.2 shows the same distributions among firms that are considered to be close i.e., have employment differences below the defined threshold value. At the firm level, 2,894 out of 3,609 firms are below the defined threshold value, which corresponds FigureC.2: CloseMatchesbetweenContractsandMUP A:EmploymentDifferences ytisneD 51. 1. 50. 0 B:EmploymentDistributions Number of firms: 2894 -80 -60 -40 -20 0 20 40 60 80 Absolute employment difference ytisneD 4. 3. 2. 1. 0 Number of firms: 2894 0 2 4 6 8 Number of employees (inverse hyperbolic sine) Firm-level data Contract-level data Notes: Panel A shows employment differences between matched contracts and firms in the MUP data that is centered around 0. Panel B shows the log employment distribution of matched contracts and the employment distribution in the MUP dataset. The sample is conditional on fulfilling the threshold rule. Number of observations with employment informationinbothdatasetsis2,894. to an acceptance rate of 80%. We therefore judge the success of the merge to be relatively high. 56

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G Based on this sample, we can re-calculate the employment distribution around the commitment level as shown in Figure 4. Figure C.3 shows the bunching estimate using the employment information in the MUP for the year of the final commitment. We adjust the bunching window slightly by excluding the area of 6 and less above the committed level as well as 1 and 2 employment below the committed level. FigureC.3: EmploymentDistributionaroundtheCommitmentLevelusingFirm-LevelData Employment < Commitment Employment > Commitment Excess mass (b) = 6.312 Standard error = .882 Excess share (%) = 0.187 ycneuqerF 004 003 002 001 0 50 70 90 110 130 150 Committed - Realized Employment Counterfactual distribution Observed distribution Notes: Thefigureshowstheemploymentdistributionaroundthecommittedemployment(demarcatedbytheverticalred line at 0) for firms with matched contracts. The blue line in dots is a histogram of actual employment relative to the commitment target in the final commitment year. Each point shows the number of observations in employment count bin (deviation between the target and the realized employment). The solid line beneath the empirical distribution is a twelve-degreepolynomialfittedtotheempiricaldistributionexcludingtheareaofmissingtwoemployeeandhavingsix employeesmorethancommitted. Theshadedregioninyellowistheestimatedexcessmass,whichis631%oftheaverage height of the counterfactual distribution beneath. Standard error is calculated using a parametric bootstrap procedure. EstimationbasedonChetty,Friedman,Olsen,andPistaferri(2011). Similar to the baseline bunching estimates, bunching with the matched sample between the contracts and the MUP is estimated to be 6.312. The estimated standard error is 0.88, indicating a significance level of 1%. This value is, furthermore, relatively close to 6.52 presented in Figure 4. Overall, these results suggest that the merge between the two datasets can be considered highly reliable. 57

D Data Addendum - Treuhand Firm Survey Data This section describes how we construct firm-level capital stock and TFP estimates and the merge between our baseline contract-level data and the THA firm survey data. The bi-annual survey was conducted by the the SOESTRA institute with its first wave in April 1991. The survey data has been used and analyzed, among others, by Wahse, Dahms, Schäfer, and Kühl (1996) and Mergele, Hennicke, and Lubczyk (2020). The focus of the questionnaire was on employment and most of the survey waves also contain questions on firm revenue. Important for our purpose to construct the firm-level capital stock is the fact that some waves also contain information on investments. Apart from these main variables, the survey contains baseline information on the sector affiliation, the location of the firm, and end dates of THA ownership and labor commitments (if any). Out of these waves, we first construct an (unbalanced) monthly firm panel between 1991 and 2000. This initial panel contains 11,105 Treuhand firms. D.1 Constructing Firm-Level Capital Stock and TFP Measures The first aim is to convert the monthly panel into a yearly panel. Out of 36,735 revenue observations over the years between 1991 and 2000 and belonging to 9,596, 69% of the information belongs to an end-of-year question. Thus, more of the revenue information is related to a full calendar year. Further,15%oftherevenuequestionsaskforrevenuenumbersduringthefirsthalfoftheyear,andthe remainingbelongseithertothefirstquarteroftheyear(9.4%)ortothethirdquarteroftheyear(6.6%). Likewise, the survey covers 17,896 investment information belonging to 6,743 firms. The majority of 95.3% of the investment numbers are related to the full calendar year, and the remaining 4.6% relate to the first six months of the year. Therefore, we harmonize the data to the yearly level by assuming linearitye.g.,ifweonlyobserverevenue/investmentinformationforthefirstsixmonthoftheyear,we multiply by 2 to construct the number for the year. In most cases, however, information are typically available for the full year and for a fraction of the year. We finally impute for 652 firms revenue informationandfor834firmsinvestmentinformationtotheendoftheyear. Regardingemployment, we construct the average employment level out of the monthly information. We complement the surveydataonyearlyemploymentandrevenuewithbasis_kennziffernasdescribedinAppendixB. The initial capital stock is constructed using balance sheet information submitted by the firms for the year 1990. The data table is called DM_BIL_N. The initial capital stock consists of tangible assets, including mainly properties, (technical) equipment, and machinery. These tangible assets represent 97% of the initial capital stock. The remaining fraction comes from breeding stock, concessions, and soil improvement. Initial capital stock information is available for 7,182 firms. We then clean the dataset and drop firms entirely if the firm does not have a single employment or sales information, whichdropstheinitialsampleof11,105firmsto10,390firms. Intheoccurrencethatemploymentand revenue information within the firm contain gaps, we linearly impute these gaps of up to two years. In order to calculate the yearly capital stock at the firm level, we start with the initial capital stock 58

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G measured in 1990, add investments, and assume a 10% depreciation rate. All Deutsch Mark (DM) values are deflated by the CPI measuredin 2016 prices. The capital stock can only be estimated if investment information is available. Table D.1 shows in column (4) that the question on investment primarily exists for the years between 1992 and 1995. Coverage is particularly low towards the end of the sample period and in 1991. For example, there are only 560 firms with full investment information between 1991 and 1994, and only 160 always have investment numbers. Likewise, but to a lower extent, column (3) shows the number of firms with revenue information. In the first two years, around 98% of all firms do have information on revenue, whereas this share decreases to 65% in 2000. TableD.1: ActualandImputedInvestmentInformation Year N Nwithinvestment Nwithimputedinvestment (1) (2) (3) (4) 1991 6,764 682 5,767 1992 6,764 3,572 3,130 1993 6,707 2,428 3,694 1994 6,583 1,364 4,198 1995 6,003 1,145 2,962 1996 5,187 500 2,383 1997 4,369 535 1,963 1998 3,633 555 1,447 1999 2,711 515 1,293 Notes: Thetableshowsthenumberoffirmsinthefinalsurveydataaswellasthenumberoffirmswithactualandimputed revenueandinvestmentinformation. To construct the capital stock, we first employ a machine-learning assisted imputation approach by predicting investment numbers and use the predicted values in case actual numbers are missing. We employ a standard least absolute shrinkage and selection operator (lasso) with an optimal tuning parameter using a 10-fold cross-validation. The covariates used in the baseline lasso regression include revenue and employment, both measured in size bins and 259 4-digit sector dummies. We perform the prediction exercise separately for every year. We provide the results for the investment imputation also using ln(employment) and ln(revenue) as well as these variables introduced with a seconddegreepolynomial. DuetothefactthatTreuhandfirmsgotrestructured(todifferentdegrees) until privatization, we also use a proportional imputation approach. For this approach, we approximate the initial capital stock by mimicking the faction of employment at privatization relative to the initial firm size. For example, if a firm gets privatized with 50 employees and the initial firm size in 1990 was 500 employees, we assume the initial capital stock to be 10% of the actual capital stock measured in 1990. Table D.2 provides baseline information for each lasso specification measuring employment and revenue in bins. Specifically, we introduce 11 employment size bins [1-4; 5-19; 50-99; 100-149; 150- 249; 250-499; 500-749; 740-1449; 1450-2999; 3000+] and 9 (ln) revenue size bins [<12.51356; 12.51356- 13.26366;13.26366-14.36855;14.36855-15.50374;15.50374-16.67438;16.67438-17.80855;17.80855-18.57818; 18.57818-20.10738; 20.10738+]. The number of non-zero covariates decreases as the sample size de- 59

TableD.2: LassoResults: ln(investment) N Optimal Numberofnon-zero Cross-validatedminimum lambda coefficients predictionerror (1) (2) (3) (4) 1991 4,908 0.015 163 2.131 1992 3,665 0.021 142 2.174 1993 2,112 0.032 99 2.041 1994 1,864 0.035 95 2.294 1995 750 0.057 65 2.322 1996 825 0.038 78 2.077 1997 851 0.039 94 2.046 1998 833 0.046 69 2.099 1999 739 0.033 86 1.848 Notes: Thetableshowssummaryresultsfromyearlylassoregressionswithln(investment)astheoutcomevariable. creases, indicated by a higher optimal cross-validated penalty parameter. Figure D.1 shows actual vs predicted investment numbers pooled over the whole time period. On average, actual and predicted numbers line up at the 45 degree line. Based on these predictions, we impute investment information in case actual investment information is missing and the selected covariates are not missing. Column (4) of Table D.1 shows the number of imputed observations over time. FigureD.1: CorrelationActualandPredictedValues Investment tnemtsevnI 81 61 41 21 01 10 12 14 16 18 Predicted investment Notes: Thefigureplotsactualvs. predictedinvestmentnumberspoolingallyearsbetween1991and1999withthecrossvalidatedlambda. In a next step, we construct the capital stock at the firm level starting with the initial capital stock in 1990 and add these (actual and imputed) investment numbers and subtract a 10% depreciation 60

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G rate. Figure D.2 provide firm-level averages over the period between 1990 and 1999. Although these numbers might not be representative for the East Germany economy due to selectivity and panel attrition, the panels A and C of the figure show an increasing trend in the constructed capital stock measure and firm revenue. Average investment amounts decrease over time ,indicting a disproportional high investment need. Average firm-level employment decreases over time. The drop in firm level employment is consistent with total employment in the economy, with the largest decreased happening between 1990 and 1991. FigureD.2: MainvariablesusedfromSoeastrafirmsurvey A:Capitalstock )MD 0001 ni( kcots latipaC 00004 00053 00003 00052 00002 B:Investment 1990 1992 1994 1996 1998 2000 Year )MD 0001 ni( tnemtsevnI 0006 0005 0004 0003 1990 1992 1994 1996 1998 2000 Year C:Revenue )MD 0001 ni( euneveR 00006 00005 00004 00003 00002 D:Employment 1990 1992 1994 1996 1998 2000 Year tnemyolpmE 004 003 002 001 1990 1992 1994 1996 1998 2000 Year Notes: Thefiguresplotaveragefirm-levelcapitalstock,investment,revenue,andemploymentnumbersbetween1991and 1999. In a next step, we aim to construct a measure of total factor productivity (TFP). Due to the fact that we have no information on intermediate inputs such as material, we run a simple Cobb-Douglas regression specification for each year, with input factors being firm-level employment and the constructed measure of capital. Output is measured by revenue. All variables are deflated by the CPI. 61

Specifically, we estimate y = α+β l +β k +(cid:101) i l i k i i where y is the logarithm of the firm’s output, in our case, revenue. l and k are the logarithm of i i i the firm inputs, in our case, the number of employees and the capital stock. We construct TFP as ω = exp(y −βˆ l −βˆ k ). Table D.3 provides the regression results separately for each year between i i l i k i 1991 and 1999. In Panel A, we provide the results using the baseline imputation approach of firm- TableD.3: RegressionResults: ln(revenue) 1991 1992 1993 1994 1995 1996 1997 1998 1999 (1) (2) (3) (4) (5) (6) (7) (8) (9) PanelA:baseline,covariates:employment&revenuedummies Ln(Empl.) 0.4899*** 0.6679*** 0.6937*** 0.6482*** 0.6683*** 0.5895*** 0.6070*** 0.6887*** 0.7290*** (0.020) (0.015) (0.013) (0.012) (0.014) (0.019) (0.019) (0.023) (0.025) Ln(Capital) 0.4582*** 0.3497*** 0.3454*** 0.3859*** 0.3654*** 0.4019*** 0.4407*** 0.4093*** 0.4002*** (0.018) (0.015) (0.014) (0.014) (0.014) (0.018) (0.019) (0.021) (0.023) N 6,448 6,449 5,823 5,251 3,847 2,677 2,296 1,771 1,502 R2 0.560 0.538 0.577 0.583 0.623 0.597 0.634 0.672 0.699 PanelB:covariates:ln(employment)&ln(revenue) Ln(Empl.) 0.4413*** 0.6505*** 0.6832*** 0.6378*** 0.6622*** 0.5794*** 0.5977*** 0.6796*** 0.7251*** (0.020) (0.016) (0.013) (0.012) (0.014) (0.019) (0.020) (0.023) (0.026) Ln(Capital) 0.4927*** 0.3613*** 0.3502*** 0.3897*** 0.3634*** 0.3988*** 0.4347*** 0.4056*** 0.3927*** (0.018) (0.015) (0.014) (0.013) (0.014) (0.018) (0.018) (0.021) (0.022) N 6,448 6,449 5,823 5,251 3,847 2,677 2,296 1,771 1,503 R2 0.566 0.542 0.580 0.587 0.625 0.599 0.635 0.673 0.699 PanelC:covariates:ln(employment)&ln(revenue)withsecondpolynomialorder Ln(Empl.) 0.4293*** 0.6509*** 0.6843*** 0.6383*** 0.6628*** 0.5805*** 0.5996*** 0.6763*** 0.7214*** (0.020) (0.016) (0.013) (0.012) (0.014) (0.019) (0.019) (0.023) (0.026) Ln(Capital) 0.5007*** 0.3603*** 0.3499*** 0.3909*** 0.3633*** 0.3947*** 0.4300*** 0.4092*** 0.3984*** (0.018) (0.015) (0.014) (0.013) (0.014) (0.018) (0.018) (0.021) (0.023) N 6,439 6,440 5,816 5,244 3,840 2,669 2,290 1,766 1,499 R2 0.563 0.538 0.577 0.584 0.621 0.592 0.629 0.671 0.699 PanelD:baselinewithproportionalinitialcapitalstock Ln(Empl.) 0.3012*** 0.5456*** 0.6002*** 0.5460*** 0.5764*** 0.4876*** 0.4872*** 0.5961*** 0.6467*** (0.022) (0.020) (0.017) (0.015) (0.017) (0.024) (0.024) (0.029) (0.033) Ln(Capital) 0.6537*** 0.4454*** 0.4109*** 0.4728*** 0.4569*** 0.4836*** 0.5351*** 0.4759*** 0.4508*** (0.020) (0.021) (0.018) (0.018) (0.018) (0.023) (0.024) (0.027) (0.029) N 4,279 4,280 4,004 3,683 2,748 1,919 1,649 1,283 1,063 R2 0.600 0.528 0.569 0.572 0.622 0.579 0.618 0.653 0.671 Meanrevenue 15.33 15.474 15.434 15.55 15.66 15.744 15.684 15.602 15.556 Meanemployment 4.636 3.966 3.676 3.484 3.394 3.338 3.436 3.526 3.714 Meancapital 15.674 15.748 15.81 15.878 15.95 15.948 15.922 15.906 15.85 Notes: The table shows production function estimation results of ln(revenue) for each year between 1991 and 1999 with inputs ln(employment) and ln(capital). Different panels indicate different lasso specifications to impute investment for constructing firm-level capital stock. Panel A (baseline) uses as covariates group size bins in employment and revenue. Panel B uses as covariates ln(revenue) and ln(employment). Panel C uses as covariates ln(revenue) and ln(employment) with a polynomial degree of order 2. Panel D uses as covariates the baseline revenue and employment introduced with sizedummies. Alllassospecificationinclude2594-digitsectordummies. level investment. Except for the first and the last year of the sample, we estimate β to be around 0.65 l and β to be around 0.35. Towards the end of the sample, both coefficients increase with a significant k decrease of the size of the sample. The estimates’ elasticities in the year 1991 are rather of equal size, and both are below 0.5. This might be the results of distorted firm sizes under socialism. Panels B and C provide the estimation results for the different lasso specifications. Panel D 62

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G providestheresultswiththebaselineimputationprocedureusingtheproportionalityapproximation oftheinitialcapitalstock. WhilePanelsBandCshowrathersimilarresults, PanelDshowsthat β is k higher by a magnitude of around 0.1, whereas β is lower by about the same magnitude. The reason l might be that the imputed investment numbers are relatively large, relative to the approximated initial capital stock, which increases the elasticity of capital in the production function estimation. D.2 Merging Contracts to Treuhand Firm Survey Data The section describes the linkage between the contracts and the survey data. This combined dataset allows us to estimate the effects of binding labor commitment contracts on TFP growth. The main challenge of linking the two datasets come from the fact that the survey data covers initial firm units, whereas the contracts might belong to only part of the firm assets. This becomes apparent because we observe multiple contracts within initial Treuhand firms. The initial firm survey sample covers 11,105 Treuhand firms with information on employment, revenue, and investments measured at different points in time at the monthly level. The ISUD data environmentcontains47,322contractsmergedto10,023TreuhandfirmIDs. Inordertoselectthecontracts that belong to the legal unit of the Treuhand firm, we merge contracts with labor commitments at the level of the Treuhand firm ID and month of the year. For example, in the case of two labor commitment contracts belonging to the same initial Treuhand firm, we can compare employment information from the survey and the audits and select the best match. Similar to Appendix Section C, we calculate the relative employment differences as: (empl −empl ) employment = survey ISUD , diff (empl +empl ) survey ISUD where empl and empl refer to the respective employment figures in both datasets and keep survey ISUD the contract with the smallest absolute deviation. In addition, we drop matched pairs if the absolute difference in both employment numbers is above 1000 employees (30 observations) and also drop 71 observations because two or more contracts generate the same deviation in employment, making it impossible to select the correct one. This generates a sample of 5,221 Treuhand firms with selected labor commitment contracts. Tojudgethesuccessofthelinkage, wedefineamatchtobecloseoracceptableiftheemployment difference is smaller or equal to the following threshold value: 1 abs(employment ) ≤ . diff (cid:113) (min[empl ,empl ]+1) survey ISUD Out of the 5,221 linked contracts, 73.07% fulfill this condition. We combine this dataset with the TFP measure at the firm level calculated and described in Section D.1. At the contract level, we merge information related to the labor commitment (first and last labor audit information including the timing, the final commitment level, the date of the contract signed with the notary) and related to the contract in general (privatizer information, THA office 63

information, sales price, investment target). This generates a sample of 2,185 firms with information on the change in TFP between the initial contract year and the final year of the labor commitment. We follow the empirical specification from the baseline model – including THA office times year fixed effects, initial firm size, time between the first and last audits, and industry and privatizerlevel contracts – and show results of binding contracts on TFP growth with and without including purchasing price and investment targets as control variables. We deviate from the baseline model by conditioning the sample on observing three or more privatizations per privatizer (instead of five) for sample size reasons. This defines the final sample of 1,962 firm-contract observations. Panel A of Table A.10 shows the results of binding labor contracts on firm-level TFP growth following the capital stock imputation of Panel A in Table D.3. OLS results show that binding labor commitments are associated with and increase in TFP growth of about 15% points. This point estimate is rather stable across different empirical specifications and close to the (cid:101) transformed labor productivity results of 0.145 presented in Tables 4 and A.9. The point estimates decrease by about 6% points to 10% points when including firm exits in Panel B of Table A.10. Column (4) provides the 2S2SLS results with a highly significant (bootstrapped) point estimate of around 0.93 (Panel A). Compared to Tables 4 and A.9 results are highly in line with each other. When including firm exits, 2S2SLS estimates decrease to 0.58 (significant at the 1% level). Again, compared to Panel C of Table A.9 with documented point estimates of around 0.55, these results are very consistent. TableD.4: OLSRobustnessResults: TFPGrowth Covariates: Covariates: Covariates: No ln(rev)&ln(empl) ln(rev)&ln(empl),poly2 Baseline/proportionalinitialcapital imputation (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Bindingcontracts 0.097*** 0.112*** 0.114*** 0.092*** 0.107*** 0.109*** 0.063 0.066 0.068* 0.1019 (0.034) (0.035) (0.036) (0.032) (0.034) (0.034) (0.037) (0.038) (0.039) (0.251) Observations 1,962 1,962 1,962 1,961 1,961 1,961 1,962 1,962 1,962 91 TFPgrowth .474 .474 .474 .486 .486 .486 .556 .556 .556 .386 Baselinecontrols Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Individualcontrols Yes Yes Yes Yes Yes Yes Yes Yes Yes No Industrycontrols Yes Yes Yes Yes Yes Yes Yes Yes Yes No Purchasingprice No Yes Yes No Yes Yes No Yes Yes No Investmenttarget No No Yes No No Yes No No Yes No Notes:ThetableshowsOLSregressionresultsofTFPgrowthonbindingcontractsfordifferentlassospecificationstoimpute investment for constructing firm-level capital stock. Columns (1)-(3) use as covariates ln(revenue) and ln(employment). Columns (4)-(5) use as covariates ln(revenue) and ln(employment) with a polynomial degree of order 2. Columns (7)-(8) useascovariatesthebaselinerevenueandemploymentintroducedwithsizedummies. Alllassospecificationinclude259 4-digit sector dummies. Column (10) provides the results without investment imputation. All regression specifications controlforfullyinteractedTHAagencyandyearfixedeffects. Bindingcontractsaredefinedasinitialfirmsizebelowthe committed target level. Baseline controls are as in the baseline specification. The purchasing price is flexibly introduced usingdeciledummies. Investmenttargetsisadummyisthecontractcontainsinvestmentcommitments. Standarderrors aretwo-wayclusteredatprivatizerandTHAofficelevel. *p<0.1,**p<0.05,***p<0.01. Table D.4 shows OLS estimation results with different specifications of the imputation of the capital stock variables (conditional on survival). The first six columns use revenue and employment in different combinations, whereas columns (7) to (9) approximate the initial capital stock measured in 1990 proportional to the employment share in the year of the contract. All specifications provide 64

ommitting to row rivatizations and irm ynamics in ast ermany C G : P F D E G positive point estimates between 0.07 and 0.12% points. The final column (10) provides the results without the imputation of the capital stock. Following the description in Section D.1, this results in a sample size of 91 observations. Although insignificant due to the sample size, the point estimate is with 0.102, rather close to the specifications with imputed capital stock. 65