Committing to grow: Privatizations and firm dynamics in East Germany

Akcigit, Ufuk; Alp, Harun; Diegmann, André; Serrano-Velarde, Nicolas Working Paper Committing to grow: Privatizations and firm dynamics in East Germany IAB-Discussion Paper, No. 01/2024 Provided in Cooperation with: Institute for Employment Research (IAB) Suggested Citation: Akcigit, Ufuk; Alp, Harun; Diegmann, André; Serrano-Velarde, Nicolas (2024) : Committing to grow: Privatizations and firm dynamics in East Germany, IAB-Discussion Paper, No. 01/2024, Institut für Arbeitsmarktund Berufsforschung (IAB), Nürnberg, https://doi.org/10.48720/IAB.DP.2401 This Version is available at: https://hdl.handle.net/10419/283019 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-sa/4.0/deed.de IAB‑DISCUSSIONPAPER Articles on labour market issues 01|2024 Committing to Grow: Privatizations and Firm Dynamics in East Germany Ufuk Akcigit, Harun Alp, André Diegmann, Nicolas Serrano‑Velarde ISSN 2195‑2663 Committing to Grow: Privatizations and Firm Dynamics in East Germany Ufuk Akcigit (University of Chicago, Halle Institute for Economic Research (IWH)), Harun Alp (Federal Reserve Board), André Diegmann (Halle Institute for Economic Research (IWH), Institute for Employment Research (IAB), Center for European Economic Research (ZEW)), Nicolas Serrano‑Velarde (Bocconi University, IGIER) Mit der Reihe „IAB‑Discussion Paper“ will das Forschungsinstitut der Bundesagentur für Arbeit den Dialog mit der externen Wissenschaft intensivieren. Durch die rasche Verbreitung von Forschungsergebnissen über das Internet soll noch vor Drucklegung Kritik angeregt und Qualität gesichert werden. The “IAB‑Discussion Paper” is published by the research institute of the German Federal Employment Agency in order to intensify the dialogue with the scientific community. The prompt publication of the latest research results via the internet intends to stimulate criticism and to ensure research quality at an early stage before printing. Contents 1 Introduction...............................................................................6 2 Institutional Background .............................................................. 11 3 A Model of Firms with Employment Commitments ................................. 13 3.1 Static Environment ..................................................................... 13 3.2 Dynamics ................................................................................ 15 3.3 Taking Stock............................................................................. 18 4 Data and Descriptive Statistics........................................................ 19 4.1 Contract Data ........................................................................... 19 4.2 Matching Contracts to Firms .......................................................... 23 5 Empirical Analysis of Labor Commitments and Firm Dynamics ................... 24 5.1 Identification Strategy.................................................................. 24 5.2 Labor Commitments and Firm Growth .............................................. 30 5.3 Labor Commitments and Productivity Growth ..................................... 34 5.4 Labor Commitments and Market Exit................................................ 35 6 Quantitative Analysis................................................................... 38 6.1 Calibration ............................................................................... 38 6.2 Calibration Results and Goodness of Fit ............................................ 39 6.3 Counterfactuals ......................................................................... 41 7 Conclusion .............................................................................. 42 References .................................................................................. 44 Supplementary Appendix.................................................................. 50 A Further Empirical Results.............................................................. 50 B Data Addendum – ISUD Data Environment .......................................... 65 C Data Addendum – Merging Contracts to Mannheim Enterprise Panel Data ....... 72 D Data Addendum – Treuhand Firm Survey Data...................................... 79 D.1 Constructing Firm‑Level Capital Stock and TFP Measures ........................ 79 D.2 Merging Contracts to Treuhand Firm Survey Data ................................. 84 List of Figures .............................................................................. 88 List of Tables ............................................................................... 89 IAB‑Discussion Paper 01|2024 3 Abstract We develop a labor demand model that encompasses pre‑match hiring cost arising from tight labor markets. Through the lens of the model, we study the effect of labor market tightness on firms’ labor demand by applying novel Bartik instruments to the universe of administrative employment data on Germany. In line with theory, the IV results suggest that a 10 percent increase in labor market tightness reduces firms’ employment by 0.5 percent. When accounting for search externalities, we find that the individual‑firm wage elasticity of labor demand reduces from ‑0.7 to ‑0.5 at the aggregate level. For the 2015 minimum wage introduction, the elasticities imply only modest disemployment effects mirroring empirical ex‑post evaluations. Moreover, the doubling of tightness between 2012 and 2019 led to a significant slowdown in employment growth by 1.1 million jobs. Zusammenfassung Dieses Papier untersucht eine wirtschaftspolitische Maßnahme, die darauf abzielt, die Beschäftigung während der Privatisierung ostdeutscher Unternehmen nach dem Fall des Eisernen Vorhangs zu sichern. Die neuen Eigentümern der Unternehmen verpflichten sich zu Arbeitsplatzzusagen, wobei Sanktionen bei Nichteinhaltung vertraglich implementiert waren. Mit einem dynamischen endogenen Wachstumsmodell zeigen wir drei Wege auf, wie sich Arbeitsplatzzusagen auf Unternehmen auswirken. Arbeitsplatzzusagen führen (i) zu einer verzerrte Unternehmensgröße, (ii) steigern die Produktivität und (iii) erhöhen die Marktaustrittswahrscheinlichkeit. Anhand eines Instrumentalvariablenansatzes und unter Verwendung von einzigartigen Vertragsdaten werden die Modellaussagen bestätigt. Das Instrument der Arbeitsplatzzusagen führt zu einer um 22 Prozentpunkte höheren jährliche Wachstumsrate der Beschäftigung, einer um 14 Prozentpunkte höheren jährliche Wachstumsrate der Produktivität und eine um 3,6 Prozentpunkte höhere Wahrscheinlichkeit des Marktaustritts. Das kalibrierte Wachstumsmodell zeigt, dass ohne diese Zusagen die Gesamtbeschäftigung nach 10 Jahren um 15 Prozent niedriger gewesen wäre. Darüber hinaus erweist sich eine alternative wirtschaftspolitische Maßnahme der Investitionssubventionen zur Steigerung der Produktivität als teuer und weniger effektiv in der kurzen Frist. 4 IAB‑Discussion Paper 01|2024 JEL D22, D24, J08, L25 Keywords industrial policy, privatizations, productivity, size‑dependent regulations Acknowledgements We thank our discussants Chang‑Tai Hsieh, Brian Viard, Stefan Obernberger, Riccardo Zago, and the seminar and conference 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, and AIEA‑NBER Conference. For providing valuable support for data access and data expertise, we thank Sandra Gottschalk (Mannheim Enterprise Panel), Alexander Giebler (ISUD data) and Chris Berthold, Antje Klünder and Jana Michaelis (German Federal Archives). Akcigit gratefully acknowledges financial support from the Max Planck Humboldt‑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 Federal Reserve System. 5 IAB‑Discussion Paper 01|2024 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 with high unemployment. In response, policymakers required that new owners of East German firms 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 improve their productivity, allowing us to study the endogenous response of productivity to such targets. 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 stems from the firm’s static labor decision, which induces an upwardly distorted employment choice. 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 that firms with binding employment targets experience higher employment and productivity growth. 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 dataset from the German archives that contains all the documentation produced by the Treuhandanstalt (THA), the government agency responsible for the privatization process. Our data contains detailed contract‑level information on 6 IAB‑Discussion Paper 01|2024 employment targets and deadlines, as well as the dates and results of each on‑site audit of the employment commitment. To measure firm‑level productivity, we merge our 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 et al., 2020; Dobbie/Song, 2015; Bernstein et al., 2019), we develop 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 percent points higher annual employment growth rate for firms with binding labor contracts compared to those without. Binding labor contracts also lead to an additional yearly productivity growth of approximately 14 percent 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 percent points higher probability of exiting by the end of the commitment period. Relative to the baseline exit rate of 5.5 percent, this represents an economically sizable 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. To identify the parameters of the model, especially the penalty of not meeting the target, we match the effects of binding employment 7 IAB‑Discussion Paper 01|2024 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 percent 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 percent 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 run to preserve employment. The paper contributes to the recent literature revisiting the merits and costs of industrial policies. Lane (2022) and Choi/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/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/Zahur (2021) study the Chinese intervention in the shipbuilding industry. Kalouptsidi (2018) estimates that policy interventions reduced shipyard costs by 13‑20 percent and reallocated international market shares. Barwick/Kalouptsidi/Zahur (2021) disentangle the various subsidies during the intervention and estimates their impact. Acemoglu et al. (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/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 8 IAB‑Discussion Paper 01|2024 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. Figure 2 illustrates key implications of the existence of employment targets on firm profits. The left panel plots total profits with respect to firm‑level productivity. The black line provides the benchmark for firms with no commitment, while the dashed red line plots profits for firms under commitment. Dashed vertical lines show the threshold productivity levels, z∗ and z∗∗. The plot shows that distorted firms have lower profits. As Equation 2 clarifies, these lower profits are attributable to the presence of 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. The right panel of Figure 2 plots marginal profits across firm productivity and shows that distorted 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 be more willing to invest in productivity improvements. Since the labor choice is positively correlated 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. 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 Ce.9 If she stays in the 9 This cost reflects not only the penalties from missing the target as employment becomes zero upon exiting, but also any other, implicit or explicit, costs due to impaired relations between the acquirer of the firm and the government. 15 IAB‑Discussion Paper 01|2024 Figure 2: Profits across Firm Productivity 12345678 Productivity 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 0.55 Profits No commitment With commitment 12345678 Productivity 0.08 0.085 0.09 0.095 0.1 0.105 0.11 0.115 0.12 0.125 0.13 Marginal Profits No commitment With commitment Notes: The left and right panels plots total and marginal profits across firm‑level productivity, respectively. The black line provides the benchmark for firms with no commitment, while the dashed red line plots profits for firms under commitment. Dashed vertical lines show the threshold productivity levels, z∗ and z∗∗ . economy, she makes the optimal labor choice, as described above, and decides how much to invest in productivity growth by choosing the Poisson arrival rate of improving 16 Panel A: Total Profits Panel B: Marginal Profits IAB‑Discussion Paper 01|2024 the productivity, x, with the following cost function (in terms of the homogeneous good) ϕ 2 ˜c(x|z˜) = x zw 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. The particular normalization of the current level of productivity implies that firm growth is consistent 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:        2 ˜π(˜z, l∗)w − ϕ x zw 2 ∂V (˜z, l∗)   (3) rV (˜z, l∗) − = max −Cew, max +x [V (˜z(1 + λ), l∗) − V (˜z, l∗)]    ∂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 change in firm value when the firm is successful with its investment in improving productivity at the 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 due to the associated larger fixed costs. In other words, conditional on initial productivity, firms with higher employment targets would be more likely to exit the economy. 10 Employment choice was characterized above, so it is taken as given here. 17 IAB‑Discussion Paper 01|2024 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” from contractually imposed penalties. Unlike the transitory nature of the first effect, the employment gains resulting from these dynamic improvements in productivity are persistent and continue beyond 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. In the next section, we will empirically test these theoretical implications of the model in the data. 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. 11 For simplicity, our model abstracts from labor adjustment costs. Incorporating such costs would not affect the transitory nature of this channel but would imply some transition phase occurring when the commitment expires. 18 IAB‑Discussion Paper 01|2024 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 more than 500 data tables (ISUD System). These tables provide comprehensive contract‑level information 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 contract date e.g., the date the contract is signed with a notary and the committed level of employment along with the date of the final commitment. As shown in Panel A of Figure 3, 90 percent of employment commitments are signed between 1991 and 1994.12 In total, all labor contracts amount to more than 900,000 committed workers, representing about 20 percent of the initial workforce population of THA firms. These contracts underwent regular audits conducted by contract managers employed by the THA. These audits serve as the basis for the realized employment levels analyzed in this paper. The contract 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 percent 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 12 Less than 2 percent of all labor contracts are written out in 1996 or later (15 contracts are observed in 2002). For 168 contracts we do not observe the date of the contract. 19 IAB‑Discussion Paper 01|2024 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 size to the final target. The fraction of firms initially below their target is 22 percent, while about 20 percent of the firms receive a target that is equal to their initial size. In 10 percent of cases the firm stays at the committed size in the first and last audits. Figure 3: Contracts and Labor Audits A: Number of contracts over time 0 200 400 600 800 1000 Number of committed employment (in 1000) 0 2000 4000 6000 Number of contracts 1990 1992 1994 1996 1998 2000 Year Contracts Labor target 0 .05 .1 .15 .2 Percent 1 5 10 2015 25+ Total Number of Audits per Contract Notes: Panel A plots the total number of signed contracts with labor commitments between 1990 and 2000 as well as the accumulated number of commitment employment. Panel B plots the distribution of labor audits per contract. Source: ISUD. 20 B: Audit distribution IAB‑Discussion Paper 01|2024 Table 1: Summary Statistics N Mean SD Minimum Maximum (1) (2) (3) (4) (5) A: Average firm size Initial employment 14,726 66.20 319.57 0.00 23,691 Final employment 14,726 60.67 194.26 0.00 8540 Final employment target 14,726 52.98 183.25 1.00 6906 B: Initial size relative to target Fraction initially below target 14,726 0.22 0.42 0.00 1.00 Fraction initially at target 14,726 0.20 0.40 0.00 1.00 Fraction initially & finally at target 14,726 0.10 0.30 0.00 1.00 C: Penalties Number of observed violation 1,272 2.24 1.29 1.00 12.00 Total number of violated labor 1,272 111.58 393.22 0.24 8,567.47 Penalty per missed employee (in 1000 EUR) 1,272 10.77 10.67 0.10 58.52 D: Productivity Initial productivity 3,387 9.99 1.43 3.51 16.27 Final productivity 3,584 12.02 0.786 10.46 14.81 Initial TFP 3,118 6.81 1.23 2.70 9.79 Final TFP 2,219 7.32 1.08 3.53 10.31 E: Market exit Exit until final commitment year 4,622 0.055 0.22 0.00 1.00 Notes: The table shows summary statistics of privatization contracts. In Panel A, initial employment level is calculated for contracts with at least two observations. This corresponds to 14,726 contracts. In Panel C, we observe 1,272 contracts with at least one observed labor commitment violation. Panels D and E are based on the linkage between contracts and external firm‑level data described in Appendices C and D. Panel D provides model‑consistent productivity and TFP measured in logs. Panel E shows the exit indicator at the end of the labor commitment period. Source: ISUD, MUP, SOESTRA. 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, we calculate the penalty per missed employee taking into account the pro rata temporis condition. This means that, for example, if a firm is missing continuously one employee over the course of three years, the firm misses, in total, three commitments and needs to pay three times one employee. 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 21 IAB‑Discussion Paper 01|2024 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 Figure 4: Employment Distribution around the Commitment Level Employment > CommitmentEmployment < Commitment Excess mass (b) = 6.521 Standard error = .8195 Excess share (%) = 0.243 0 1000 2000 3000 Frequency -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‑degree polynomial fitted to the empirical distribution excluding the area of missing one employee and having three employees more than committed. The shaded region in yellow is the estimated excess mass, which is 652 percent of the average height of the counterfactual distribution beneath. Standard error is calculated using a parametric bootstrap procedure. Estimation based on Chetty et al. (2011). Source: ISUD. 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 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. 13 The maximum cumulative missed employment amounts to 8,567 workers. This is above the maximum of the final employment target in Panel A, as there can be multiple violations. In addition, the maximum number in Panel A corresponds to the final commitment level. 14 The effective penalty can be lower because of “conditions beyond the purchaser’s control” (Dodds/Wächter, 1993), renegotiation, minor violations of targets (Bagatellfall), judicial decisions, and bankruptcy. 22 IAB‑Discussion Paper 01|2024 Following Chetty et al. (2011), we estimate an excess mass around the threshold of 652 percent 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 the Mannheim Enterprise Panel (MUP). By merging the contract partners’ names with the ownership 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 Supplementary 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 firm‑level 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. This noteworthy improvement aligns with the documented convergence process observed in the years 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 have been closed (Burda, 2006).16 Starting in 1990, our calculation suggests an increase in productivity 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 15 The red line in Figure 4 plots the estimated counterfactual density based on a twelve‑degree polynomial (p = 12) and an asymmetric window around the threshold R = [3, −1]. R = [3, −1] denotes the omitted bunching range including firms having up to three more employees than their committed target. The yellow shaded region depicts the estimated excess mass around the threshold. Figures A.5 to A.7 provide robustness checks with respect to the degree of the polynomial, the bunching window, and the bin definition. Table A.11 shows the results by sub‑samples. 16 Based on aggregate statistics, Bachmann et al. (2022) show that GDP per worker increased by about 0.7 log points between 1991 and 2000. 23 IAB‑Discussion Paper 01|2024 information on 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 percent is observed. 5 Empirical Analysis of Labor Commitments and Firm Dynamics 5.1 Identification Strategy Addressing the empirical challenge of non‑random allocation of labor commitments to firms is crucial 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 on judge leniency (Bhuller et al., 2020; Dobbie/Song, 2015; Bernstein et al., 2019) and patent evaluators (Sampat/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 quasi‑random 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 several reasons. The number of privatizations in those years meant that THA agents typically worked 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 17 The organizational stress and complexity of privatizing the East German economy cannot be underestimated. The THA was described as “an adolescent bureaucracy, born of chaos and destined to be phased out without ever functioning normally. It is a human creation, whipped together quickly and then put under extreme pressure without time to prepare” (Dodds/Wächter, 1993). The agency officially terminated its operation at the end of 1994 and its mission was taken over by a successor agency entitled Bundesanstalt für vereinigungsbedingte Sonderaufgaben (Böick, 2018). 24 IAB‑Discussion Paper 01|2024 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 with binding labor commitments. Conditional on the set of baseline controls, industry dummies, and privatizer characteristics, the association between binding contracts and employment growth is on average 49 percent 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. 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 percent 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 percent significance level. Consistent with the previous evidence, the first‑stage statistics for weak instruments is large (Panel B). The Kleinbergen‑Paap F‑tests reject the hypothesis of weak instruments with statistics ranging between 15 and 17. Economically, the first‑stage estimates imply that a 10 percent increase in labor preferences of privatizers result in a 2 percent 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 25 Table A.7 provides a way to test whether the differences between the OLS and the IV estimates are driven by treatment effect heterogeneity. We follow Bhuller et al. (2020) and first perform a principal component analysis using one component based on pre‑determined employment (see employment categories in Table 2) and revenue figures measure in 1990, as well as initial employment at contract date, and the sector affiliation. We then separate the predicted component into quartiles and separately estimate the complier share for each quartile group using the first‑stage regression specification. Finally, we re‑weight the full estimation sample by using the sub‑sample complier shares as weights. Panel B 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 heterogeneity is unlikely to explain the differences. 31 IAB‑Discussion Paper 01|2024 Figure 6: Employment Dynamics by Initial Size to Target A: Employment Growth 0 1 2 3 Density -2 -1 0 1 2 Employment growth distribution Initial size above target Initial size below traget Growth distribution -.5 0 .5 Employment growth .2 .6 1 1.4 1.8 2.2 Ratio Initial Size to Final Target Average growth rate 90% Confidence Intervals Notes: Panel A shows the overall employment growth distribution as well as the employment growth distribution distinguishing by firms initially below or above (including firms initially at their target) their commitment employment level. Panel B shows average growth rates by the distance of the initial size to the final target. Source: ISUD. 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 privatizers affect firms’ outcomes only through binding labor commitments. THA privatizers, however, negotiated not only on labor commitments, but also on associated penalties, investment commitments, and sales price. Following Bhuller et al. (2020), we address this issue by augmenting our baseline model with controls for these dimensions of privatization contracts. Consistent with the exclusion 32 B: Employment Growth by Initial Size to Target IAB‑Discussion Paper 01|2024 Table 3: Regression Results, Employment Growth OLS Model IV‑Model (1) (2) (3) (4) (5) (6) Panel A: Second‑stage results Binding contract 0.4992*** 0.4973*** 0.4975*** 0.7016*** 0.6740*** 0.6873*** (0.031) (0.030) (0.030) (0.219) (0.231) (0.235) Panel B: First‑stage results Privatizer stringency 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 Average employment at contract date 60.064 60.064 60.064 60.064 60.064 60.064 Average growth rate (non‑binding contracts) ‑0.063 ‑0.063 ‑0.063 ‑0.063 Share with binding contracts 0.207 0.207 0.207 0.207 0.207 0.207 F‑Statistic 17.01 14.67 14.76 Sample condition Baseline controls Yes Yes Yes Yes Yes Yes Industry controls No Yes Yes No Yes Yes Individual controls 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 reduced form regressing the binding contract indicator on the stringency measure. Panel B shows the second‑stage results. All specifications control for fully interacted THA agency and year fixed effects and are conditional on having at least five 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 employment level measured at the first audit. Industry controls are 2‑digit industry dummies. Individual controls refer to the gender of the privatizer and a dummy for a PhD degree. Standard errors are two‑way clustered at privatizer and THA office level. Instrument refers to the leave‑one‑out measure of assigning binding contracts. *p<0.1, **p<0.05, ***p<0.01. Source: ISUD. restriction, Table A.4 shows that adding extensive and intensive investment 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 26 The Heckman selection equation is based on a probit regression with the outcome variable being equal to 1 if the initial GDR firm is observed among the contracts with labor commitments and zero otherwise. We use as explanatory variables log initial firm size and log initial sales over employment measured in 1990 as well as sector‑ and THA office fixed effects. Results for employment growth, productivity growth and firm exit in A.6. 33 IAB‑Discussion Paper 01|2024 5.3 Labor Commitments and Productivity Growth To disentangle the mechanism behind the growth in employment we extend the empirical analysis to productivity dynamics in the matched sample of 2,395 privatization contracts with complete information. 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, growth in productivity is relatively constant for firms above the threshold for binding contracts. The average growth rate in the data amounts to 86.8 percent 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. 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. 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 percent 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 first‑stage 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 percent points. Columns (4) to (6) present our results based on the TFP growth measure. Again, we find that TFP increased by 66 percent 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 34 IAB‑Discussion Paper 01|2024 Figure 7: Productivity Growth and the Degree of Binding Contracts linear fitprojection .5 .6 .7 .8 .9 1 1.1 1.2 Growth in Productivity .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 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 initial labor productivity, employment and industry‑fixed effects. The two grey dashed lines correspond to the 90 percent CI. The blue line shows a linear fit of a regression of productivity growth on the ratio of initial size to final commitment among contracts to the left of or at 1. The red line projects the linear fit into the area where the initial size is below the committed level (to the left of 1). The figure excludes top and bottom 4 percent of the tightness measure. Total number of firms is 2,395. Source: ISUD, MUP. 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, Table A.8 provides supporting evidence for the productivity channel by analyzing patenting 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 binding contracts and patenting. Although imprecisely estimated, the 2S2SLS coefficient shows again a downward bias of the OLS point estimate. 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. 35 IAB‑Discussion Paper 01|2024 Table 4: Regression Results, Productivity Growth Productivity, α = 0.8 TFP OLS OLS 2S2SLS OLS OLS 2S2SLS (1) (2) (3) (4) (5) (6) Binding contract 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 Average productivity at contract date 10.599 10.599 10.633 6.813 6.813 6.813 Average productivity growth (non‑binding) 0.851 0.851 0.854 0.332 0.332 0.332 Share with binding contracts 0.171 0.171 0.155 0.162 0.162 0.162 Baseline controls Yes Yes Yes Yes Yes Yes Industry controls Yes Yes Yes Yes Yes Yes Individual controls Yes Yes Yes Yes Yes Yes Purchasing price 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 office level. The standard errors in columns (4)‑(6) are bootstrapped using 2,500 replications. *p<0.1, **p<0.05, ***p<0.01. Source: ISUD, MUP. 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 exit rate for these firms amounts to 4.8 percent on average. With an average exit rate of 8.4 percent, 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 percent 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. 36 IAB‑Discussion Paper 01|2024 Figure 8: Market Exit and the Degree of Binding Contracts linear fitprojection 0 .05 .1 .15 .2 .25 .3 Market Exit Probability .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 initial employment and industry‑fixed effects. The two grey dashed lines correspond to the 90 percent CI. The blue line shows a linear fit of a regression of market exit on the ratio of initial size to final commitment among contracts to the left of or at 1. The red line projects the linear fit into the area where the initial size is below the committed level (to the left of 1). The figure excludes top and bottom 3 percent of the tightness measure. Total number of firms is 4,596. Source: ISUD, MUP. Table 5: Regression Results, Exit Probability at Final Commitment MUP exit indicator ISUD 0 employment OLS 2S2SLS OLS 2S2SLS (1) (2) (3) (4) (5) (6) Binding contract 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 Exit share (non‑binding) 0.049 0.049 0.047 0.035 0.035 0.030 Share with binding contracts 0.171 0.171 0.171 0.171 0.171 0.171 Sample condition Baseline controls Yes Yes Yes Yes Yes Yes Industry controls Yes Yes Yes Yes Yes Yes Individual controls Yes Yes Yes Yes Yes Yes Purchasing price 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 firm size below the committed target level. Controls are as in the baseline specification. Additional control variable is the purchasing price. Standard errors in columns (1), (2), (4) and (5) are two‑way clustered at privatizer and THA office level. The standard errors in columns (3) and (6) are bootstrapped using 2,500 replications. *p<0.1, **p<0.05, ***p<0.01. Source: ISUD, MUP. 37 IAB‑Discussion Paper 01|2024 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 the production function, α, equal to 0.8 to match the labor earning share. Consistent with the average 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 percent to match the average real wage growth rate over 1990 and 1996 in East Germany (Hunt, 2001). The rest of the parameters are calibrated internally by minimizing the distance between the moments from the firm‑level data we used in the empirical part of the paper and their model implied counterparts.27 In particular, let ME denote the vector of empirical moments and let M(Ω) denote the vector of model‑simulated moments and Ω is the set of parameters to be calibrated internally. We then search Ω to minimize the absolute relative deviation between the model and data; that is, we solve ∑ |ME − Mm(Ω)| m min . 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 regression results on the impact of binding contracts on exit rates of firms to inform the exit cost parameter, Ce. Finally, we include the growth rate of total employment for firms with binding contracts 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 27 In our setting, we cannot separately identify the step size and the cost scale parameter in productivity improvements, λ and ϕ, respectively. Therefore, we fix the value of the step size at 0.25 and calibrate the cost scale parameter internally. 38 IAB‑Discussion Paper 01|2024 final audit. Finally, we use the simulated data to construct the targeted moments. We repeat this process and search over the parameter space until we minimize the distance between model‑implied moments and the data. 6.2 Calibration Results and Goodness of Fit Table 6 and 7 contain the calibrated parameters and the targeted moments, respectively. As seen from 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 percent of the average wage, given by γ. Table 6: Moments Used in Calibration # Description Model Data M1 Employment growth regression 0.489 0.498 M2 Productivity growth regression 0.083 0.083 M3 Exit rate regression 0.030 0.027 M4 Total empl. growth rate of firms with binding contracts 0.614 0.672 Table 7: Internally Calibrated Parameters Description Model Estimate Penalty for not hitting target employment γ 0.676 Scale for investment cost parameter ϕ 0.030 Cost of exit Ce 58.39 Non‑targeted Moments The calibrated model also performs well in matching some important patterns in the data that were not targeted. In Figure 9, we depict the employment growth by the ratio of initial employment relative to target employment, analogous to Panel B of Figure 6. The black and red dots show the model‑implied employment growth rates and the data, respectively. Although we only 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. Figure 10 illustrates the evolution of total employment during and after the 39 IAB‑Discussion Paper 01|2024 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. Figure 9: Employment Growth Notes: The figure depicts the employment growth at the firm level by the ratio of initial employment relative to target employment. The black and red dots show the model‑implied employment growth rates and the data, respectively. Gray dots show the employment growth rates under the counterfactual economy where there are no employment targets. The x‑axis is divided into 20 quantile bins and each dot represents average value within that bin. Figure 10: Total Employment in the Post‑Commitment Period 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. All series are normalized to 1 at the beginning of the policy period. 40 IAB‑Discussion Paper 01|2024 Giorcelli, Michela; Li, Bo (2021): Technology transfer and early industrial development: evidence from the Sino‑Soviet alliance. Tech. Rep., National Bureau of Economic Research. Glitz, Albrecht; Meyersson, Erik (2020): Industrial espionage and productivity. In: American Economic Review, Vol. 110, No. 4, p. 1055–1103. Greenstone, Michael; Hornbeck, Richard; Moretti, Enrico (2010): Identifying agglomeration spillovers: Evidence from winners and losers of large plant openings. In: Journal of Political Economy, Vol. 118, No. 3, p. 536–598. Harrison, Mark; Zaksauskienė, Inga (2016): Counter‑intelligence in a command economy. In: The Economic History Review, Vol. 69, No. 1, p. 131–158. Heise, Sebastian; Porzio, Tommaso (2021): The aggregate and distributional effects of spatial frictions. Tech. Rep., National Bureau of Economic Research. Hsieh, Chang‑Tai; Klenow, Pete J (2009): Misallocation and manufacturing TFP in China and India. In: Quarterly Journal of Economics, Vol. 124, No. 4, p. 1403–1448. Hunt, Jennifer (2006): Staunching emigration from East Germany: Age and the determinants of migration. In: Journal of the European Economic Association, Vol. 4, No. 5, p. 1014–1037. Hunt, Jennifer (2001): Post‑unification wage growth in East Germany. In: Review of Economics and Statistics, Vol. 83, No. 1, p. 190–195. Johnson, Paul; Papageorgiou, Chris (2020): What remains of cross‑country convergence? In: Journal of Economic Literature, Vol. 58, No. 1, p. 129–75. Juhász, Réka; Lane, Nathan J; Rodrik, Dani (2023): The New Economics of Industrial Policy. Tech. Rep., National Bureau of Economic Research. Kalouptsidi, Myrto (2018): Detection and impact of industrial subsidies: The case of Chinese shipbuilding. In: The Review of Economic Studies, Vol. 85, No. 2, p. 1111–1158. Kline, Patrick; Moretti, Enrico (2014): Local economic development, agglomeration economies, and the big push: 100 years of evidence from the Tennessee Valley Authority. In: The Quarterly Journal of Economics, Vol. 129, No. 1, p. 275–331. Krueger, Alan B; Pischke, Joern‑Steffen (1995): A Comparative Analysis of East and West German Labor Markets: Before and After Unification. In: in: Freeman und Katz: Differences and Changes in Wage Structures„ p. 405–446. Lane, Nathan (2022): Manufacturing revolutions: Industrial policy and industrialization in South Korea. In: Available at SSRN 3890311. Lipschitz, Leslie; McDonald, Donogh (1990): German unification: economic issues. Tech. Rep., International Monetary Fund Washington. 47 IAB‑Discussion Paper 01|2024 Liu, Ernest (2019): Industrial policies in production networks. In: The Quarterly Journal of Economics, Vol. 134, No. 4, p. 1883–1948. López‑de Silanes, Florencio (1997): Determinants of privatization prices. In: The Quarterly Journal of Economics, Vol. 112, No. 4, p. 965–1025. Martin, Leslie A; Nataraj, Shanthi; Harrison, Ann E (2017): In with the big, out with the small: Removing small‑scale reservations in India. In: American Economic Review, Vol. 107, No. 2, p. 354–86. Megginson, William L; Netter, Jeffry M (2001): From state to market: A survey of empirical studies on privatization. In: Journal of economic literature, Vol. 39, No. 2, p. 321–389. Mergele, Lukas; Hennicke, Moritz; Lubczyk, Moritz (2020): The big sell: privatizing East Germany’s economy. In: . Peters, Michael (2022): Market size and spatial growth—evidence from Germany’s post‑war population expulsions. In: Econometrica, Vol. 90, No. 5, p. 2357–2396. Redding, Stephen J; Sturm, Daniel M (2008): The costs of remoteness: Evidence from German division and reunification. In: American Economic Review, Vol. 98, No. 5, p. 1766–1797. Restuccia, Diego; Rogerson, Richard (2008): Policy Distortions and Aggregate Productivity with Heterogeneous Establishments. In: Review of Economic Dynamics, Vol. 11, No. 4, p. 707–720. Romano, Joseph P; Wolf, Michael (2005a): Exact and approximate stepdown methods for multiple hypothesis testing. In: Journal of the American Statistical Association, Vol. 100, No. 469, p. 94–108. Romano, Joseph P; Wolf, Michael (2005b): Stepwise multiple testing as formalized data snooping. In: Econometrica, Vol. 73, No. 4, p. 1237–1282. Sampat, Bhaven; Williams, Heidi L (2019): How do patents affect follow‑on innovation? Evidence from the human genome. In: American Economic Review, Vol. 109, No. 1, p. 203–36. Shleifer, Andrei (1998): State versus private ownership. In: Journal of Economic Perspectives, Vol. 12, No. 4, p. 133–150. Siebert, Horst (1991): German unification: the economics of transition. In: Economic Policy, Vol. 6, No. 13, p. 287–340. Sinn, Hans‑Werner (2002): Germany’s economic unification: An assessment after ten years. In: Review of international Economics, Vol. 10, No. 1, p. 113–128. Snower, Dennis J; Merkl, Christian (2006): The caring hand that cripples: The East German labor market after reunification. In: American Economic Review, Vol. 96, No. 2, p. 375–382. 48 IAB‑Discussion Paper 01|2024 Syverson, Chad (2011): What determines productivity? In: Journal of Economic literature, Vol. 49, No. 2, p. 326–65. Uhlig, Harald (2008): The slow decline of East Germany. In: Journal of Comparative Economics, Vol. 36, No. 4, p. 517–541. Wahse, Jürgen; Dahms, Vera; Schäfer, Reinhard; Kühl, Jürgen (1996): Beschäftigungsperspektiven von privatisierten Unternehmen. In: Mitteilungen aus der Arbeitsmarkt‑und Berufsforschung, Vol. 29, No. 1, p. 106–116. 49 IAB‑Discussion Paper 01|2024 Supplementary Appendix A Further Empirical Results Figure A.1: Privatizations per Privatizer Number of privatizers: 1659 Number of contracts: 11194 0 .1 .2 .3 Percent 1 5 10 15 20 25+ Number of observations per privatizer Notes: The figure plots the number of privatization handled per individual privatizer (winsorized at 25). The total number of privatizations is 11,194. These cases are handled by 1,659 individuals. 5.04 percent of all cases are organized by privatizers only observed once in the sample. This corresponds to 652 individuals. Source: ISUD. 50 IAB‑Discussion Paper 01|2024 Table A.1: Test of Random Assignment of Investors to Privatizers Indep. variable: Stringency Dep. variables Coefficient p‑value Adj. p‑value Mean Standard deviation (1) (2) (3) (4) (5) Employment Log investor size ‑0.0033 0.1228 0.9740 2.4600 1.8220 Investor size > 100 employees ‑0.0008 0.1135 0.9181 0.1400 0.3480 Credit rating Creditworthiness investor 0.0768 0.3334 0.9990 284.38 101.58 High rating ‑0.0004 0.1101 0.9800 0.0640 0.2460 Location West German investor 0.0001 0.8356 0.9990 0.6780 0.4680 Sector affiliation Agriculture, forestry, fishing 0.0002 0.5666 0.9930 0.0120 0.1120 Mining and quarrying ‑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 Retail trade 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 Baning and Insurance 0.0000 0.9975 1.0000 0.0320 0.1760 Real Estate 0.0001 0.7775 0.9990 0.0580 0.2340 Technical services ‑0.0003 0.5271 0.9980 0.1020 0.3040 Economic services 0.0002 0.2320 0.9940 0.0340 0.1800 Other 0.0003 0.6932 0.9940 0.0760 0.2640 Notes: The sample is based on 4,993 contracts matched to investor characteristics. 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‑way clustered at the privatizer and THA office level. p‑values in column (3) adjust for multiple testing using Romano‑Wolf procedure (Romano/Wolf, 2005a; Romano/Wolf, 2005b) with 1,000 bootstrap replications. **p<0.1, **p<0.05, ***p<0.01. Source: ISUD. 51 IAB‑Discussion Paper 01|2024 Table A.2: First‑Stage Regression Results by Sub‑Samples Baseline Employment in 1990 Revenue in 1990 Sector affiliation < p(75) < p(50) < p(75) < p(50) Trade‑ able Non‑ tradeable (1) (2) (3) (4) (5) (6) (7) Privatizer stringency 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 Average employment at contract date 63.22 57.75 73.638 72.39 70.166 69.032 70.554 Average growth rate .062 .028 .084 .038 .038 .048 .006 Share with binding contracts .208 .184 .232 .194 .194 .22 .146 Sample condition Baseline controls Yes Yes Yes Yes Yes Yes Yes Individual controls Yes Yes Yes Yes Yes Yes Yes Industry controls Yes Yes Yes Yes Yes Yes Yes Notes: The table shows IV regression results. All specifications control for fully interacted THA agency and year fixed effects and are conditional on having at least 2 privatizations per privatizer. All strata variables (e.g., employment in 1990) refer to the initial firm from where the contract was generated. There are 335 contracts affiliated with the agriculture sector not presented in the table. F‑Statistic refers to the Kleibergen‑Paap F‑Statistic. Baseline controls are the time between the first and last audits measured in day, time between contract date and first audit measured in days, and log initial employment level (+1) 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 binding contracts. **p<0.1, **p<0.05, ***p<0.01. Source: ISUD. 52 IAB‑Discussion Paper 01|2024 Table A.3: Test of Joint Null of Monotonicity and Exclusion 10 knots 15 knots 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) Test statistic 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: The table presents results from the test proposed in Frandsen/Lefgren/Leslie (2023) for the joint null hypothesis that the monotonicity and exclusion restrictions hold. We test this null using THA office times year‑of‑privatization effects conditional on having handled at least 5 privatizations. Columns (1) to (4) provide the results imposing 10 knots in the quadratic spline function. Columns (5) to (8) provide the results imposing 15 knots in the quadratic spline function. Each column is associated with different weighting schemes between the fit and slope components of the test. A failure to reject the null implies that we cannot reject the hypothesis that the monotonicity and exclusion restrictions jointly hold. The test was implemented in Stata via the package testjfe. Source: ISUD. Table A.4: Regression Results Accounting for Extensive/Intensive Margin Preferences IV‑Model Results First stage (1) (2) (3) (4) Binding contract 0.7246*** 0.7317*** 0.7070*** (0.223) (0.225) (0.249) Investment preferences Extensive margin ‑0.0005 ‑0.0005 ‑0.0004 ‑0.0005 (0.001) (0.001) (0.001) (0.000) Intensive margin ‑0.0005 ‑0.0004 ‑0.0004 ‑0.0004 (0.001) (0.001) (0.001) (0.000) Privatizer stringency (instrument) 0.0018*** (0.000) Observations 9,363 9,363 9,363 9,363 Average employment at contract date 60.064 60.064 60.064 60.064 Average growth rate .064 .064 .064 .064 Share with Binding contracts 0.207 0.207 0.207 0.207 F‑Statistic 17.67 17.03 14.47 Sample condition Baseline controls Yes Yes Yes Yes Individual controls No Yes Yes Yes Industry controls No No Yes Yes Notes: The table shows IV regression results. All specifications control for fully interacted THA agency and year fixed effects and are conditional on having at least 5 privatizations per privatizer. Extensive margin investment preference refer to contracts with any investment commitments. Intensive margin preference refer to contracts with the investment target 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. Source: ISUD. 53 IAB‑Discussion Paper 01|2024 Figure A.2: Persistence of Privatizer Characteristics -20 0 20 40 60 80 100 Leave-one-out privatizer characteristic -20 -10 0 10 20 30 40 50 60 70 80 90 100 Lag leave-one-out privatizer characteristic coeff.: .914 (s.e.: .0071) Notes: The figure plots the leave‑one‑out rate of a tight contract (initial firm size < final committed size) in the previous case against the leave‑one‑out rate of a binding contract in the current case. All plotted values are mean‑standardized residuals from regressions on fully interacted THA office and year of privatization fixed effects. The blue line corresponds to a linear regression. The figure is constructed by conditioning of having handled at least five privatization contracts. Total number of observations is 8,759. Source: ISUD. Figure A.3: Labor Productivity across Firm Size 9.5 10 10.5 11 11.5 Ln labor productivity 0 .05 .1 .15 .2 .25 .3 Density 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 employment information in 1990. The figure exclude the top and the bottom 1 percent of the productivity measure. Source: ISUD. 54 IAB‑Discussion Paper 01|2024 Table A.5: Robustness Tests, Employment Growth Dep. variable: Firm growth Random Assignment Coefficient First‑stage F‑Statistic Joint F‑test (p‑value) (1) (2) (3) (4) A: Instrument construction Only past decisions for instrument 1.1690*** 0.0008*** 9.505 0.226 (0.354) (0.000) Above 10 cases per privatizer 0.9589** 0.0023*** 11.38 0.355 (0.346) (0.001) Very tight contracts 1.0160** 0.0015*** 8.511 0.237 (0.424) (0.001) Full tightness distribution ‑0.0097*** 0.1285*** 21.58 0.848 (0.003) (0.028) Contract with zero employment in first & last 0.6380** 0.0016*** 14.83 0.392 (0.254) (0.000) B: Control variables & sample selection Control for renegotiation attempts 0.6588*** 0.0018*** 16.24 0.408 (0.222) (0.000) Control for penalty clause 0.7621*** 0.0016*** 13.65 0.408 (0.264) (0.000) Control for purchasing price & investment target 0.5524** 0.0017*** 13.65 0.408 (0.222) (0.000) Years between contract signed & first audit < 2 0.6558** 0.0020*** 16.86 0.617 (0.273) (0.001) Month between first & last audit > 12 0.7387** 0.0018*** 14.26 0.695 (0.269) (0.000) MUP subsample 0.5435** 0.0018*** 10.45 0.486 (0.256) (0.001) C: Manipulation of the outcome variable Log employment differences 0.8953** 0.0018*** 12.33 0.408 (0.331) (0.000) Annualized firm growth, (Lt/Lt−1)1/#year − 1 (trimmed at the upper percentile) 0.2822** 0.0018*** 14.76 0.408 (0.114) (0.000) Growth rate < 2 & > −2 0.7443** 0.0015*** 12.33 0.404 (0.314) (0.000) Notes: The table shows IV regression results. All specifications control for fully interacted THA office and year fixed effects and are conditional on having at least five privatizations per privatizer. For sample size reasons, the MUP subsample is conditional on having at least three observations per privatizer. Column (1) shows the point estimate of the main variable of interest (except the specification with at least 10 observations per privatizer). Column (2) shows the corresponding first‑stage coefficient. F‑Statistic in column (3) refers to the Kleibergen‑Paap F‑Statistic (first‑stage). All specifications condition on the full set of control variables including baseline controls (log) time between the first and last audits (+1) measured in days, log time between contract date and first audit (+1) measured in days, and log initial employment level (+1) measured at the first audit), individual controls (gender of the privatizer and academic degree (PhD)), and 2‑digit industry controls. Column (4) shows the F‑Statistic of a joint F‑test of random assignment. The dependent variable is always the instrument regressed on log initial employment variables (accounting, purchasing, HR, production, sales, administration, R&D), and log initial revenue measured in 1990 (conditional on industry‑fixed effects and fully interacted THA office and time fixed effects). Standard errors are two‑way clustered at privatizer and THA office level. ***p<0.1, **p<0.05, ***p<0.01. Source: ISUD. 55 IAB‑Discussion Paper 01|2024 Table A.6: OLS Regression Results, Adjustment for Sample Selection Employment Productivity Exit (1) (2) (3) (4) (5) (6) Binding contract 0.4313*** 0.4302*** 0.0938*** 0.0884*** 0.0229* 0.0223* (0.025) (0.025) (0.022) (0.023) (0.011) (0.011) Mills ratio ‑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 Average employment at contract date 70.926 70.926 47.336 47.336 47.336 47.336 Mean outcome (non‑binding contracts) ‑0.062 ‑0.062 0.852 0.852 0.051 0.051 Share with binding contracts 0.192 0.192 0.188 0.188 0.188 0.188 Sample condition Baseline controls Yes Yes Yes Yes Yes Yes Industry controls No No No No No No Individual controls 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 employment and sales over employment by introducing dummy variables. All specifications in the second stage control for fully interacted THA agency and year fixed effects. 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 employment level measured at the first audit. Standard errors are two‑way clustered at privatizer and THA office level. *p<0.1, **p<0.05, ***p<0.01. Source: ISUD, MUP, SOESTRA. 56 IAB‑Discussion Paper 01|2024 Figure A.7: Bunching with percent deviation bin A: 1 percent bins B: 2 percent bins Employment > CommitmentEmployment < Commitment Excess mass (b) = 23.64 Standard error = 7.275 0 1000 2000 3000 4000 Frequency -40 -20 0 20 40 Committed over Realized Employment Counterfactual distribution Observed distribution C: 5 percent bins Employment > CommitmentEmployment < Commitment Excess mass (b) = 11.18 Standard error = 2.483 0 1000 2000 3000 4000 Frequency -40 -20 0 20 40 Committed over Realized Employment Counterfactual distribution Observed distribution Employment < CommitmentEmployment > Comitment Excess mass (b) = 2.95 Standard error = .6922 0 1000 2000 3000 4000 Frequency -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 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‑degree polynomial fitted to the empirical distribution excluding the area of missing one employee and have 4 employees more than committed. The shaded region in yellow is the estimated excess mass. Standard error is calculated using a parametric bootstrap procedure. Estimation based on Chetty et al. (2011). Panel A shows the results by constructing 1 percentage bin deviations. Panel B shows the results by constructing 2 percentage bin deviations. Panel C shows the results by constructing 5 percentage bin deviations. Source: ISUD. 63 IAB‑Discussion Paper 01|2024 Table A.11: Bunching by Sub‑Samples Excess mass (b) Standard error (1) (2) A: Industry affiliation Agriculture, energy, mining 9.076 3.220 Chemistry, plastics 4.952 0.9231 Extraction of cut‑stone, iron, casting, steel forming 7.842 3.351 Steel construction, mechanical & electrical engineering, automobile 6.699 1.023 Paper, print, textile, food 7.617 1.070 Construction and buildings trades, wholesale, retail 7.257 1.227 Transportation, communication, insurance 5.799 1.325 B: Contract maturity 16 to 31 months 6.574 1.198 Below 16 months 8.697 3.010 Above 31 months 7.212 1.118 C: Number of audits Multiple audits 6.521 0.773 D: Penalty condition Exclude contracts without penalty clause 6.627 0.889 E: Initial size Below target 6.473 0.989 Above target 5.151 0.540 Notes: The table shows bunching estimates of the employment distribution around the committed employment for contracts between 1990‑1995 by different groups. The counterfactual distribution is a based on a twelve‑degree polynomial fitted to the empirical distribution excluding the area of missing one employee and having three employees more than committed. Standard errors are calculated using a parametric bootstrap procedure with 100 replications. Estimation based on Chetty et al. (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 date and final commitment) percentile. Panels C and D select only contracts with multiple audits and with a penalty clause, respectively. Panel E distinguishes by initial contract size (measure at the first audit) relative to the final target. Source: ISUD. 64 IAB‑Discussion Paper 01|2024 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 Two tables are used to measure firm‑level information: basis_kennziffern and basis_kennziffern_91. The table basis_kennziffern_91 comprises most of the information and, therefore, is the main table. In case of missing values, we search for information in basis_kennziffern to complement and to construct a comprehensive cross‑section of firm information for the year 1990.29 The information 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 percent are observed for the first time in 1990.30 We complement the data with additional industry information from the SOESTRA survey (see Mergele/Hennicke/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. 28 Section C describes the merge between contracts and external firm‑level data, the Mannheim Enterprise Panel (MUP), to study dynamics beyond the commitment period. 29 The information can be combined to construct a yearly panel with information at the firm level between 1989 and 1994. This dataset cannot be used to study the evolution of firms over time because the firm disappears from the dataset once the firm transitions out of the THA portfolio either because of a privatization or liquidation. 30 THA created legal entities over time, and, as a result, 5.1 percent of firms are observed for the first time in 1991, and 1.2 percent in 1992, and 0.48 percent in 1993. 65 IAB‑Discussion Paper 01|2024 Eröffnungsbilanz THA Period Privatizations Market Period (01.07.1990) Firms Firm ID Contract ID Liquidation t Balance Sheets THA Planning VM Control Management MUP Baseline numbers Guarantee Payments Growth (Basiskennziff ern) Employment (Bürgschaft en) (Raten) Financials Legacy Debt Outlays & Inflows Exit (Finanzierungsdaten) (Altkredit) (Budgetverfolgung) Liquidation Liabilities (Abwicklungsdaten) (Verbindlichkeiten) Commitments Audits (Zusicherungen) (Überprüfung) Figure B.1: Timeline from reunification to the market period Source: Own presentation. 66 IAB‑Discussion Paper 01|2024 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, privatizations and liquidation decisions. These data allows us to not only track changes in ownership, 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 ISUD system. The tables ASVA02T, VATVT, ASVA22T, ASVA50T, and FE3_VT are identified to be candidates 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 firms and consequently do not have a privatizer attached to it. Linking contracts to contain privatizer information, labor commitment contracts, and firm links generates a sample of 11,194 contracts as shown in Section 5. 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 67 IAB‑Discussion Paper 01|2024 Table B.1: Contract‑Level Data Tables Table names Description A: Baseline tables ASVA01T The tables contains master data and status information for contracts signed with the THA. It combines many variables from different tables. The table contains the contract ID (sysnr), the date of the contract signed with the notary, and the name of the privatizer. Total number of unique contracts: 256,842. ASVA02T The table provides information on partial contracts. It contains the link between the con‑ tracts and the firms, the fixed price payed by the contract partner, and the assignment to THA offices. Total number of unique contracts: 213,052. Unique contracts with a non‑ missing contract‑firm link: 22,837. VATVT The table provides information on partial contracts. It contains the link between the con‑ tracts and the firms. Total number of unique contracts: 37,967. Unique contracts with a non‑missing contract‑firm link: 30,745. ASVA22T This table provides information on mappings. It contains the link between the contracts and the firms. Total number of unique contracts: 40,036. Unique contracts with a non‑ missing contract‑firm link: 9,784. ASVA50T This table provides header data for concerted action. It contains the link between the contracts and the firms. Total number of unique contracts: 82. Unique contracts with a non‑missing contract‑firm link: 82. FE3_VT This table provides information on processes/operations of main tables related to finan‑ cials. It contains the link between the contracts and the firms. Total number of unique contracts: 1,723. Unique contracts with a non‑missing contract‑firm link: 1,710. B: Labor Commitments & Audits VAPST This table provides information on labor commitments of the contract partner. Total num‑ ber of unique contracts: 17,753. Total number of observations: 52,438. VAPIT This table provides information on labor audits. Total number of unique contracts: 16,583. Total number of observations: 116,619. VAPITH This table provides information on labor audits and is labeled as history in the documen‑ tation. Total number of unique contracts: 19,052. Total number of observations: 102,933. ASVA12T This table, among others, provides information on labor commitments. Total number of unique overall contracts: 275,054. Total number of unique contracts with positive number of committed labor: 22,535. Total number of observations: 322,829. ASVA13T This table, among others, provides information on labor audits. Total number of unique overall contracts: 47,111. Total number of unique contracts with positive number of au‑ dited labor: 15,702. Total number of observations: 153,155. C: Investment Commitments & Audits VAZST This table provides information on investment commitments of the contract partner. Total number of unique contracts: 18,120. Total number of observations: 20,366. VAZIT This table provides information on investment audits. Total number of unique contracts: 16,806. Total number of observations: 32,096. VAZITH This table provides information on investment audits and is labeled as history in the documentation. Total number of unique contracts: 26,195. Total number of observations: 60,159. ASVA15T This table, among others, provides information on investment commitments. Total number of unique overall contracts: 274,375. Total number of unique contracts with positive num‑ ber of committed investment: 24,220. Total number of observations: 280,370. ASVA16T This table, among others, provides information on investment audits. Total number of unique overall contracts: 47,111. Total number of unique contracts with positive number of audited investment: 15,619. Total number of observations: 64,725. Source: Own presentation. 68 IAB‑Discussion Paper 01|2024 documentation by the German Federal Archives. The 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 of observations in both tables is higher because there can be multiple commitments for different years of the commitment period as well as several audits per commitment. Figure B.2: Paper File: Labor Commitment Notes: The figures show the original template used by the THA to document labor commitments. Source: Handbuch Treuhandanstalt. 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 observations where the number of employees at the audit is zero, but the variable that states whether employee information is reported is set to zero. This reduces the sample further by 2,536 observations. 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 69 IAB‑Discussion Paper 01|2024 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 the date of the contract with the notary (7 observations), (ii) if the time between two consecutive commitments is negative, and (iii) if the final employment commitment is zero (224 observations). This 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 3,385 additional contracts. In terms of labor audits, however, these contracts are not observed in VAPIT. 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 version of VAPIT, called VAPITH, is suitable to fill parts of the missing audits from ASVA12T. Among the 3,385 additional contracts after basic data cleaning steps, we are able to merge the audit information for 2,702 contracts. Together, these data tables generate our final sample of 18,235 contracts with labor 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. After basic data cleaning steps and combining commitment information in VAZST with audit information 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 31 Out of the 5,125 additional contracts with labor commitments ASVA12T, 17 contracts are found in VAPIT and 22 contracts are found in ASVA13T. 70 IAB‑Discussion Paper 01|2024 Figure B.3: Paper File: Investment Commitment Notes: The figures show the original template used by the THA to document investment commitments. Source: Handbuch Treuhandanstalt. 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 percent among the labor commitment contracts, this share is 65.2 percent. Due to the flow nature of investment commitment, there are fewer audits during the commitment period. Combining labor with investment contracts results in a sample 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 any labor commitment condition we merge this combined dataset with the 58,544 contracts with cleaned privatizer names. 71 IAB‑Discussion Paper 01|2024 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). The Mannheim Enterprise Panel (MUP), is the most comprehensive micro database of companies 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 percent coverage (for further details, see Bersch et al. (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, we observe for each firm owner. In the contract‑level data, we have access to the contract partner, 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 observations correspond to 11,199 contract partners. These individuals usually have multiple links to 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 72 IAB‑Discussion Paper 01|2024 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 et al. (1996) and Mergele/Hennicke/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 percent 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 percent of the revenue questions ask for revenue numbers during the first half of the year, and the remaining belongs either to the first quarter of the year (9.4 percent) or to the third quarter of the year (6.6 percent). Likewise, the survey covers 17,896 investment information belonging to 6,743 firms. The majority of 95.3 percent of the investment numbers are related to the full calendar year, and the remaining 4.6 percent relate to the first six months of the year. Therefore, we harmonize the data to the yearly level by assuming linearity e.g., if we only observe revenue/investment information for the first six month of the year, 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 information and for 834 firms investment information to the end of the year. Regarding employment, we construct the average 79 IAB‑Discussion Paper 01|2024 employment level out of the monthly information. We complement the survey data on yearly employment and revenue with basis_kennziffern as described in Appendix B. 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 percent 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, which drops the initial sample of 11,105 firms to 10,390 firms. In the occurrence that employment and 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 measured in 1990, add investments, and assume a 10 percent 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 percent of all firms do have information on revenue, whereas this share decreases to 65 percent in 2000. Table D.1: Actual and Imputed Investment Information Year (1) N (2) N with investment (3) N with imputed investment (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: The table shows the number of firms in the final survey data as well as the number of firms with actual and imputed revenue and investment information. Source: SOESTRA. 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 80 IAB‑Discussion Paper 01|2024 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 second degree polynomial. Due to the fact that Treuhand firms got restructured (to different degrees) 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 percent 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 decreases, indicated by a higher optimal cross‑validated penalty parameter. Table D.2: Lasso Results: ln(investment) N (1) Optimal lambda (2) Number of non‑zero coefficients (3) Cross‑validated minimum prediction error (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: The table shows summary results from yearly lasso regressions with ln(investment) as the outcome variable. Source: SOESTRA. 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 81 IAB‑Discussion Paper 01|2024 information is missing and the selected covariates are not missing. Column (4) of Table D.1 shows the number of imputed observations over time. Figure D.1: Correlation Actual and Predicted Values 10 12 14 16 18 Investment 10 12 14 16 18 Predicted investment Notes: The figure plots actual vs. predicted investment numbers pooling all years between 1991 and 1999 with the cross‑validated lambda. Source: SOESTRA. 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 percent depreciation 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. 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. Specifically, we estimate yi = α + βlli + βkki + ϵi 82 Investment IAB‑Discussion Paper 01|2024 Figure D.2: Main variables used from Soeastra firm survey A: Capital stock 20000 25000 30000 35000 40000 Capital stock (in 1000 DM) 1990 1992 1994 1996 1998 2000 Year B: Investment 3000 4000 5000 6000 Investment (in 1000 DM) 1990 1992 1994 1996 1998 2000 Year C: Revenue D: Employment 20000 30000 40000 50000 60000 Revenue (in 1000 DM) 1990 1992 1994 1996 1998 2000 Year 100 200 300 400 Employment 1990 1992 1994 1996 1998 2000 Year Notes: The figures plot average firm‑level capital stock, investment, revenue, and employment numbers between 1991 and 1999. Source: SOESTRA. where yi is the logarithm of the firm’s output, in our case, revenue. li and ki are the logarithm of the firm inputs, in our case, the number of employees and the capital stock. We construct TFP as ωi = exp(yi − β ˆ lli − β ˆ kki). Table D.3 provides the regression results separately for each year between 1991 and 1999. In Panel A, we provide the results using the baseline imputation approach of firm‑level investment. Except for the first and the last year of the sample, we estimate βl to be around 0.65 and βk to be around 0.35. Towards the end of the sample, both coefficients increase with a significant 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 provides the results with the baseline imputation procedure using the proportionality approximation of the initial capital stock. While Panels B and C show rather similar results, Panel D shows that βk is higher by a magnitude of around 0.1, 83 IAB‑Discussion Paper 01|2024 Table D.3: Regression Results: ln(revenue) 1991 1992 1993 1994 1995 1996 1997 1998 1999 (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A: baseline, covariates: employment & revenue dummies Ln(Empl.) 0.4899*** 0.6679*** 0.6937*** 0.6482*** 0.6683*** 0.5895*** 0.6070*** 0.6887*** 0.7290*** (0.025)(0.020) (0.015) (0.013) (0.012) (0.014) (0.019) (0.019) (0.023) Ln(Capital) 0.4582*** 0.3497*** 0.3454*** 0.3859*** 0.3654*** 0.4019*** 0.4407*** 0.4093*** 0.4002*** (0.023)(0.018) (0.015) (0.014) (0.014) (0.014) (0.018) (0.019) (0.021) 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 Panel B: 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*** (0.026)(0.020) (0.016) (0.013) (0.012) (0.014) (0.019) (0.020) (0.023) Ln(Capital) 0.4927*** 0.3613*** 0.3502*** 0.3897*** 0.3634*** 0.3988*** 0.4347*** 0.4056*** 0.3927*** (0.022)(0.018) (0.015) (0.014) (0.013) (0.014) (0.018) (0.018) (0.021) 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 Panel C: covariates: ln(employment) & ln(revenue) with second polynomial order Ln(Empl.) 0.4293*** 0.6509*** 0.6843*** 0.6383*** 0.6628*** 0.5805*** 0.5996*** 0.6763*** 0.7214*** (0.026)(0.020) (0.016) (0.013) (0.012) (0.014) (0.019) (0.019) (0.023) Ln(Capital) 0.5007*** 0.3603*** 0.3499*** 0.3909*** 0.3633*** 0.3947*** 0.4300*** 0.4092*** 0.3984*** (0.023)(0.018) (0.015) (0.014) (0.013) (0.014) (0.018) (0.018) (0.021) 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 Panel D: baseline with proportional initial capital stock Ln(Empl.) 0.3012*** 0.5456*** 0.6002*** 0.5460*** 0.5764*** 0.4876*** 0.4872*** 0.5961*** 0.6467*** (0.033)(0.022) (0.020) (0.017) (0.015) (0.017) (0.024) (0.024) (0.029) Ln(Capital) 0.6537*** 0.4454*** 0.4109*** 0.4728*** 0.4569*** 0.4836*** 0.5351*** 0.4759*** 0.4508*** (0.029)(0.020) (0.021) (0.018) (0.018) (0.018) (0.023) (0.024) (0.027) 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 Mean revenue 15.33 15.474 15.434 15.55 15.66 15.744 15.684 15.602 15.556 Mean employment 4.636 3.966 3.676 3.484 3.394 3.338 3.436 3.526 3.714 Mean capital 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 size dummies. All lasso specification include 259 4‑digit sector dummies. Source: SOESTRA. whereas βl is lower by about the same magnitude. The reason 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 84 IAB‑Discussion Paper 01|2024 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 environment contains 47,322 contracts merged to 10,023 Treuhand firm IDs. In order to select the contracts 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: (emplsurvey − emplISUD) employmentdiff = , (emplsurvey + emplISUD) where emplsurvey and emplISUD refer to the respective employment figures in both datasets and keep 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. To judge the success of the linkage, we define a match to be close or acceptable if the employment difference is smaller or equal to the following threshold value: 1 abs(employmentdiff ) ≤ √ . (min[emplsurvey, emplISUD] + 1) Out of the 5,221 linked contracts, 73.07 percent 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 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. 85 IAB‑Discussion Paper 01|2024 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 privatizer‑level 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 percent points. This point estimate is rather stable across different empirical specifications and close to the ϵ transformed labor productivity results of 0.145 presented in Tables 4 and A.9. The point estimates decrease by about 6 percent points to 10 percent 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 percent level). Again, compared to Panel C of Table A.9 with documented point estimates of around 0.55, these results are very consistent. Table D.4: OLS Robustness Results: TFP Growth Covariates: ln(rev) & ln(empl) Covariates: ln(rev) & ln(empl), poly 2 Covariates: Baseline/proportional initial capital No imputation (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Binding contracts 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 TFP growth .474 .474 .474 .486 .486 .486 .556 .556 .556 .386 Baseline controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Individual controls Yes Yes Yes Yes Yes Yes Yes Yes Yes No Industry controls Yes Yes Yes Yes Yes Yes Yes Yes Yes No Purchasing price No Yes Yes No Yes Yes No Yes Yes No Investment target No No Yes No No Yes No No Yes No Notes: The table shows OLS regression results of TFP growth on binding contracts for different lasso specifications to impute 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) use as covariates the baseline revenue and employment introduced with size dummies. All lasso specification include 259 4‑digit sector dummies. Column (10) provides the results without investment imputation. All regression specifications control for fully interacted THA agency and year fixed effects. Binding contracts are defined as initial firm size below the committed target level. Baseline controls are as in the baseline specification. The purchasing price is flexibly introduced using decile dummies. Investment targets is a dummy is the contract contains investment commitments. Standard errors are two‑way clustered at privatizer and THA office level. *p<0.1, **p<0.05, ***p<0.01. Source: ISUD, SOESTRA. 86 IAB‑Discussion Paper 01|2024 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 positive point estimates between 0.07 and 0.12 percent 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. 87 IAB‑Discussion Paper 01|2024 List of Figures Figure 1: THA Headquarters and Subsidiaries.............................................. 12 Figure 2: Profits across Firm Productivity................................................... 16 : Contracts and Labor Audits ....................................................... 20Figure 3 Figure 4: Employment Distribution around the Commitment Level .................... 22 Figure 5: First‑stage analysis.................................................................. 26 Figure 6: Employment Dynamics by Initial Size to Target ................................ 32 Figure 7: Productivity Growth and the Degree of Binding Contracts ................... 35 Figure 8: Market Exit and the Degree of Binding Contracts .............................. 37 Figure 9: Employment Growth................................................................ 40 Figure 10: Total Employment in the Post‑Commitment Period ........................... 40 Figure 11: Total Employment under Counterfactual Economies .......................... 42 Figure A.1: Privatizations per Privatizer........................................................ 50 Figure A.2: Persistence of Privatizer Characteristics ......................................... 54 Figure A.3: Labor Productivity across Firm Size.............................................. 54 Figure A.4: Correlation Productivity Measures ................................................ 60 Figure A.5: Bunching with different polynomials............................................. 61 Figure A.6: Bunching with symmetric R....................................................... 62 Figure A.7: Bunching with percent deviation bin ............................................ 63 Figure B.1: Timeline from reunification to the market period ............................. 66 Figure B.2: Paper File: Labor Commitment ................................................... 69 Figure B.3: Paper File: Investment Commitment............................................. 71 Figure C.1: Comparison of Employment Figures between Contracts and MUP .......... 76 Figure C.2: Close Matches between Contracts and MUP .................................... 77 Figure C.3: Employment Distribution around the Commitment Level using Firm‑Level Data ................................................................................... 78 Figure D.1: Correlation Actual and Predicted Values ........................................ 82 Figure D.2: Main variables used from Soeastra firm survey ................................ 83 88 IAB‑Discussion Paper 01|2024