Policy Research Working Paper 10929 The Financial Premium and Real Cost of Bureaucrats in Businesses Ana P. Cusolito Roberto N. Fattal-Jaef Fausto Patiño Peña Akshat V. Singh Europe and Central Asia Region Development Research Group & Prosperity Practice Group September 2024 Policy Research Working Paper 10929 Abstract : This paper characterizes finance allocation distortions toward the private sector. The findings show that although in capital markets across state-owned and private-owned state-owned enterprises are on average subsidized relative enterprises. It does so by implementing Whited and Zhao’s to private firms, removal of state-owned enterprises from (2021) methodology to infer idiosyncratic financial distor- the market may lead to aggregate productivity losses of up tions on a novel firm-level database containing information to 40 percent due to their superior technical efficiency in on the ownership structure of firms operating in 24 Euro- some sectors. Targeted reforms that only shut down poorly pean countries during 2010–16. The analysis finds that performing state-owned enterprises lead to aggregate total firms with public authorities as direct shareholders (state- factor productivity gains in every country, reaching up to 15 owned enterprises) have subsidized access to debt and equity, percent. Reforms that in addition remove distortions before compared to their private counterparts. The paper then reallocating the released resources toward more productive quantifies the macroeconomic effects of removing state- firms increase productivity up to 83.7 percent. owned firms and reallocating their financial resources This paper is a product of the Chief Economist, Europe and Central Asia Region, Development Research Group, and Prosperity Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at acusolito@worldbank.org, rfattaljaef@ worldbank.org, fpatinopena@worldbank.org, and akshat.singh@economics.ox.ac.uk. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team The Financial Premium and Real Cost of Bureaucrats in Businesses⋆ Ana P. Cusolitoa , Roberto N. Fattal-Jaefa , Fausto Pati˜ naa , Akshat V. Singhb no Pe˜ a World Bank, Washington, D.C., US b Oxford University, Oxford, UK JEL classification : O1, O43, G3. Keywords: SOEs, state-ownership distortions, productivity, finance misallocation. ⋆ The authors would like to thank Brian Castro for excellent research assistance. We welcome comments by Aaditya Mattoo, Ayhan Kose, Daniel Lederman, Davide DeMare, Dorothe Singer, Ivailo V. Izvorski, Jean Pesme, Jesica Torres, Martha Martinez Licetti, Mary Hallward-Driemeier, Mona Haddad, Sebastian Eckardt, Sofia Bauducco, Ufuk Ackcigit, William F. Maloney, seminar participants at the Central Bank of Chile, the World Bank authors’ workshop, the MENA Chief Economist Office BBL, the EAP Chief Economist Office internal seminar and NASMES, WEAI, and EPSA conferences. The authors also thank the Bureau van Dijk data team for clarifying questions about Orbis raw financial and ownership vintage files. The opinions expressed in this paper do not represent the views of the World Bank Group, its Board of Directors, or the Governments they represent. All errors and omissions are the authors’ responsibility. All mistakes are our own. ⋆⋆ acusolito@worldbank.org Corresponding author. rfattaljaef@worldbank.org fpatinopena@worldbank.org akshat.singh@economics.ox.ac.uk 1. Introduction Despite decades of retrenchment, the entrepreneurial state’s footprint is notable across many countries’ economies. Such prevalence is not just concentrated in sec- tors where government intervention is rationalized on the grounds of market failures, coordination problems, and increasing returns to scale. By contrast, it is widespread across all sectors of the economy, with 70% of government-owned enterprises oper- ating in competitive industries, such as food, construction, and hospitality, typically dominated by private sector firms. Even when operating in sectors subject to market failures, state-owned enterprises (SOEs) may crowd out resources from more efficient private-owned firms (POEs). Thereby reducing aggregate productivity and output. In this paper, we study one particular source of interference from government- owned enterprises: the competition with private firms for external financing in capital markets. There are several well-understood mechanisms through which less efficient firms with a large government stake could access external financing more favorably than private counterparts, thereby distorting finance, innovation, investment, and resource allocation. These range from lower debt issuance costs due to the federal government’s backing of corporate debt to direct budget support from the government in the form of internal equity. While the mechanisms are well acknowledged, there is less clear understanding of their quantitative significance. Do capital markets favor SOEs with lower debt and equity financing costs? If so, how large is their financial premium? And what are the aggregate implications on productivity? To study these questions, we appeal to Whited and Zhao’s (2021) methodology for inferring financial distortions from firms’ balance sheet data and apply it to 24 European countries from 2010-2016. Mirroring the real misallocation literature (Hsieh and Klenow, 2009), the methodology delivers a distribution of firm-specific financial distortions resulting from deviations in the rates of marginal revenue products of debt and equity across firms, the properties of which we characterize across state- and private-owned firms. We find that firms with public authorities as direct shareholders (SOEs) get, on average, subsidized access to debt and equity compared to private- 2 owned enterprises (POEs). A 1 p.p. increase in government stake reduces the implicit average finance cost by 0.01 percent. We then leverage the theoretical structure to quantify the macroeconomic effects of removing state-owned firms and reallocating their financial resources toward the private sector. We find that despite being, on average, subsidized, shutting down SOEs may lead to aggregate productivity losses due to their superior technical efficiency. Targeted reforms that only close down poorly performing SOEs lead to aggregate TFP gains in every country, reaching up to 15%. A reform that adds to the previous scenario, removing distortions before the reallocation of released resources towards more productive firms, increases aggregate productivity up to 83.7%. Our approach for measuring financial frictions follows closely the work of Whited and Zhao (2021). The methodology postulates a production function, mapping fi- nancial resources (debt and equity) into value-added. The underlying assumption is that firms primarily raise costly external financing to acquire physical capital, hire workers, and purchase intermediate inputs. In the spirit of the real resource misalloca- tion literature (Hsieh and Klenow, 2009), the efficient allocation of financial resources prescribes that finance should be allocated to equalize the distribution of marginal revenue products of debt and equity across firms. Firm-specific financial frictions can be read off the data as wedges rationalizing the observed dispersion in the marginal returns. Unlike Hsieh and Klenow (2009) and following Whited and Zhao (2021), we consider degrees of substitutability between debt and equity that differ from Cobb- Douglas complementarity. The purpose of this is to flexibly accommodate sector-wide financial frictions that rationalize deviations from the Modigliani-Miller benchmark of irrelevance of the financial composition of the firm. A CES structure generates well- defined debt-to-equity ratios. Once the sector-wide financial frictions are captured through the elasticity of substitution, we are interested in gauging the firm-specific wedges that rationalize the dispersion in these ratios across firms within narrow in- dustries. 3 We implement the methodology developed by Cusolito and Vranic (2020), which recreates the ownership trees of all firms in the ORBIS database, endowing us with the firm-level financial and ownership information required for our investigation. We focus on direct ownership links, as this is the most prevalent public-related ownership link for the European countries under study.1 Our primary explanatory variable measures the percentage of direct shares owned by all the public authorities (PAs) that govern the country where a firm operates. Thus, it takes into consideration span-of-control issues. Equipped with a distribution of firm-specific financing wedges, we use fixed-effect regressions to assess the role of state ownership in the cost of finance. Controlling for lagged values of firm size, age, physical productivity, firm, sector-time, and country fixed effects, we regress the idiosyncratic cost of finance (i.e., the wedge) against the lagged fraction of state ownership in a firm. Next, we leverage the theoretical foundations of our methodology for identifying financial wedges to assess the economic significance of SOEs’ subsidized access to capital markets. To this end, we conduct a series of counterfactual reforms, where SOEs are removed from their sectors of operation, and their financial resources are reallocated to surviving enterprises. How much would the aggregate productivity of the economy increase or decrease from these interventions?2 The first reform imposed the shutdown of all SOEs in the economy and the real- location of their financial resources towards their sectoral counterparts. Since these firms also face capital-market distortions, we reallocated the financing while preserv- ing these distortions.3 Next, we explored a reform that dismantled poorly performing 1 We do not consider indirect linkages where the state is a shareholder of another company that holds shares in a particular firm. From this point of view, our results should be considered conser- vative estimates of the actual degree of financial subsidies confronted by state-related enterprises. 2 Notice that our reforms could be considered a privatization exercise where the physical produc- tivity and the idiosyncratic financial distortion of the firm post-privatization are drawn from the joint distribution of productivity and financial wedges across surviving firms in the sector. 3 Under a CES aggregator of individual varieties, there is a mechanical variety channel through which productivity would fall if the number of varieties is reduced, which would occur under a shutdown of SOEs. To abstract for this channel, we normalize TFP by the number of firms. 4 SOEs, those whose distortion-adjusted productivity lies at the bottom of their market distribution. We then reallocated the released resources while preserving the financial distortions affecting the rest of the firms. The final reform added to the previous scenario the removal of distortions before reallocating resources towards the most productive firms. We find that a one-percentage-point increase in government shareholding reduces the cost of finance by 0.01% for non-publicly listed firms. However, publicly listed SOEs face a higher cost of accessing finance. The largest subsidies occur in industries that facilitate the functioning of an economy, such as financial services, electricity, water, and information and communications, as they exhibit a higher elasticity of the financial subsidy to state ownership than the rest. Back-of-the-envelope calculations show that the fiscal burden of the SOE financial subsidy ranges from 0.001% to 0.955% of GDP for the year 2016. The first reform scenario (i.e., dismantling of all SOEs) leads to productivity losses in some countries and gains in others, ranging between -40% and 35%. There are two forces behind the mixed results. Firstly, even if more resources become available for the private sector, the severity of persisting financial distortions in the private sector could be so extreme that reallocating finance towards POEs is productivity-reducing. Secondly, it may be that SOEs are more productive than their private counterparts, so despite their implicit financial subsidy, it is productivity-reducing to shut them down. The latter channel is the most important in explaining the tails of the distribution of aggregate productivity gains and losses. In Bosnia and Herzegovina, for example, where TFP would decline by more than 40% if all SOEs were shut down, state-owned firms are three times as productive as their private sector counterparts. Therefore, the proposed reform backfires, as it reallocates financing toward private enterprises with substantially inferior technology. The converse is true in Ukraine, where private firms exhibit a significant technical edge over government-run ones. A natural implication of the previous result is that SOE reform should be imple- mented by targeting only poorly performing SOEs, an implication that we evaluate 5 quantitatively. Within each country and sector, we identify the poorly performing SOEs as those whose distortion-adjusted productivity lies at the bottom of the dis- tribution across all firms in their market. This targeted reform leads to TFP gains of up to 15% for all countries. These are most modest where the state sector showed superior technologies and are most pronounced where the private sector’s productiv- ity dominated government-run firms. The highest TFP gains, roughly 20%-80%, are obtained when the previous scenario is complemented with a reform that eliminates distortions before reallocating resources toward the most productive firms. While our analysis pursues a well-established theory-based approach for measur- ing financial distortions and deploys it globally, such an approach is subject to some caveats that should be considered when interpreting our results. Firstly, it is based on a theoretical structure where, absent any financial distortion, the equilibrium al- location is Pareto-efficient. As such, it abstracts from any rationale for state-owned enterprises’ financial subsidy to be interpreted as correcting some market imperfec- tion. Despite such a limitation, our analysis uncovers preferential access to credit by SOEs across many sectors of the economy where such market imperfections are arguably less pronounced. Moreover, even when market failures were to be argued in some sectors, our estimates provide a benchmark of observed financial subsidies against which the ”optimal” ones can be compared. Is a discount of 0.01 percent in the cost of credit for each percentage point of government ownership more or less than necessary to correct the market failure? A second caveat concerning the methodology pertains to the theory of allocation of financial resources. We assume finance should costlessly be allocated across firms to equalize marginal returns according to a CES production function. However, akin to adjustment costs in the physical capital misallocation literature, financial adjustment costs could also cause dispersion in the marginal returns to equity and debt across firms, which would be misattributed as distortions in our model. This is a latent limitation of the analysis that we expect to address in future research. The remainder of the paper is organized as follows. Section 2 reviews the relevant 6 literature. Section 3 presents the model. Section 4 describes the data. Section 5 outlines the identification strategy. Section 6 discusses the empirical results. Section 7 examines the productivity gains from counterfactual SOE structural reforms. The final section concludes. 2. Literature Review Our paper relates to two strands of research. The first one explores the effect of state ownership on finance misallocation before and after the global financial crisis of 2008/09 (GFC). The second strand of research studies the effect of finance misalloca- tion on real misallocation. We bridge both strands of research to analyze the effect of state ownership on real misallocation through the financial channel. In the first strand of research, early work by Borisova and Megginson (2011) shows that government ownership affects the cost of debt in Europe, immediately before and during the GFC. On average across firms, a one-percentage-point decrease in government ownership is associated with an increase in the credit spread. However, the spreads of fully divested firms are significantly lower than those of partially privatized companies. The authors attribute the findings to decreasing government guarantees, firm performance improvements, and ownership uncertainty, among others. Evidence after the GFC focuses primarily on China due to the large state footprint in the economy. Research by Bai et al. (2016), Cong et al. (2019), and Huang et al. (2020) examines allocative efficiency in the credit market, after the introduction of a large fiscal stimulus to cope with the GFC.4 Bai et al. (2016) find credit reallocation away from private-run firms and towards government-run firms, as commercial banks perceive that the former have higher levels of default risk than the latter due to their lack of government guarantees. Huang et al. (2020), correspondingly, present evidence of crowding-out effects, with private investments shrinking relatively more in locations with higher public-debt growth rates. Song et al. (2011) develop and 4 The 2008–09 Chinese economic stimulus plan is estimated to be worth RMB¥ 4 trillion (US$586 billion). 7 calibrate a growth model consistent with China’s economic transition. The authors show that private-run firms, which are more productive than government-run firms, must finance investments through internal savings due to financial distortions that favor state-owned firms. These distortions in turn cause allocative inefficiencies in the credit market. More recently, Jurzyk and Ruane (2021) analyze the role of state ownership on real misallocation among listed Chinese firms, during the period 2002-2019. The authors find that SOEs are, on average, 30 percent less productive than private firms, with large differences in capital productivity explaining the performance gap between both types of firms. The paper also shows that government-run enterprises have higher leverage levels and lower effective interest rates than their private counterparts. A counterfactual reform that both closes the productivity gap and equalizes the average capital intensity of both groups of firms can generate aggregate real gains of 7.5 percent in China. In related work, Nigmatulina (2021) studies the contribution of state ownership and political connections to misallocation in the Russian Federation, employing a unique natural experiment, whereby U.S. and EU sanctions targeted Russian firms connected to government elites. The paper shows that sanctions and the consequent government’s response worsened allocative efficiency, as private-sanctioned firms maintained their relative size, while state-sanctioned ones gained additional government support. De Haas et al. (2022) employ cross-country firm-level data from 89 countries from 2000 to 2019 to study the relationship between state ownership and corporate leverage. Their results show that state ownership, both at the extensive and intensive margins, is negatively related to firm leverage. Second, the negative relationship between state ownership and corporate leverage holds across most of the firm-size distribution, with the important exception of the largest firms. This relationship holds across most of the firm-size distribution, with the important exception of the largest companies, and is stronger in countries with weak political and legal institutions. Geng and Pan (2023) document that Chinese state-owned firms face lower credit costs than 8 their private counterparts, consistent with previous evidence. The authors show that financial regulatory changes introduced in 2018, which tightened credit conditions, further increased credit misallocation in favor of government-run firms because of their government guarantees. In the second strand of research, early work by Buera et al. (2011) explores the role of financial distortions in generating capital and talent misallocation, showing that distortions account for a significant part of the cross-country differences in real outcomes such as output, output per worker, and total factor productivity (TFP). Similarly, Midrigan and Xu (2014) exploit plant-level data for the Republic of Korea to evaluate the role of financial frictions in determining TFP. The parametrization of the developed model shows that financial frictions reduce aggregate real outcomes primarily through two different channels, inefficiently low levels of firm entry and technology adoption. Gopinath et al. (2017) also confirm the role of financial frictions in determining productivity losses from capital misallocation. Using data from manu- facturing firms in Spain, the paper shows that the decline in real interest rates, often attributed to the euro convergence process, led to important real losses, as capital inflows were allocated to unproductive firms. More recently, Whited and Zhao (2021) extend the seminal framework of Hsieh and Klenow (2009) to analyze the effect of finance misallocation on the real economy.5 The paper also quantifies the reallocation gains for China relative to the U.S., showing that state-owned firms stand to gain less in percentage terms than non–state-owned firms. Our paper differs from both strands of research in several regards. First, we un- pack the black box of financial distortions and identify one, state ownership, which generates non-trivial real losses by distorting financial markets. Second, we advance the literature and estimate the effect of such a distortion on the cost of production. Third, we use a mirror approach–finance instead of factor markets–to deal with mea- 5 The paper shows that financial distortions reduce aggregate output through two different channels–a scale and a composition effect–as distortions create a gap between the observed and efficient total liabilities and debt-to-equity mix firms get access to. 9 surement concerns and quantify the real reallocation gains countries can obtain from SOE-related structural reforms, unveiling the drivers behind these gains. This in- cludes distortions and productivity differences between private- and government-run firms. Finally, our paper presents evidence for a large set of European countries, right after the GFC–a period characterized by significant variation in state intervention– instead of focusing as often on extreme cases of (non)interventionism such as China, Russia, and the U.S. The latter allows us to present a broader characterization of the effect of state ownership on real misallocation through finance misallocation. Thus, enlightening the policy debate. The next section introduces the model we use to guide the empirical analysis. 3. Model This section presents the theoretical framework that we use to guide the empirical analysis. In doing so, we follow the model developed by Whited and Zhao (2021), which builds on the framework developed by Hsieh and Klenow (2009). The economy consists of S sectors. Total value added, Y , is the aggregate output value of all the sectors in the economy. Parameter θs measures the relative importance of Ys in Y . Thus, the economy’s production function is as follows: S S Y = Ysθs , where θs = 1. (1) s=1 s=1 Value added in sector s, Ys , is produced with Ms differentiated varieties. The market for each variety has a monopolistic competition market structure with σ as the elasticity of substitution between varieties. Sector s’s production function is as follows: σ Ms σ −1 σ −1 Ys = Yis σ . (2) i=1 Firm i, which operates in sector s, uses two different types of liabilities, debt (Dis ) 10 and equity (Eis ), to produce value added, Yis . The production function has a CES functional form, with elasticity of substitution between liabilities γs . Ais stands for total factor productivity and parameter αs represents the relative importance of debt to equity in producing one unit of the good. Firm i’s production function is as follows: γs γs −1 γs −1 γs −1 γs γs Yis = Ais αs Dis + (1 − αs ) Eis . (3) Firm i maximizes profits by choosing the optimal price (Pis ) at which to sell output Yis , as well as the debt (Dis ) and equity (Eis ) levels, taking as given the demand for its good and firm-specific distortions (τDis , τEis ), which alter the cost of finance, relative to the sectorial values, Rs and λs , respectively. A negative (positive) value of τJis means that firm i has preferential (disadvantageous) treatment when accessing to financial resource J , with J = D, E . Thus, firm i’s profit maximization problem is as follows: Psσ Ys {Pis , Dis , Eis } argmaxΠis = Pis σ − (1 + τDis ) Rs Dis − (1 + τEis ) λs Eis (4) Pis The profit maximization problem yields first-order conditions (FOCs) (5) and (6), from which the firm derives the optimal debt and equity demand. At the optimum, the level of debt and equity is such that the marginal revenue of one extra unit of the financial liability equals its marginal cost. Three factors determine the relative demand of debt to equity as shown in equation (7): the relative liability prices, the distortions ratio, and the relative importance of debt to equity in the production process. Given the assumption of price-taking behavior by firms within a sector, marginal costs are the same across firms within a sector, thus implying that marginal revenue products of debt and equity would be equalized across firms, absent any idiosyncratic financial friction. By substituting equation (7) in the FOCs (5) and (6), one can derive the final expression for firm i’s optimal demand of debt and equity as shown in equations (8) 11 and (9). σ−1 Pis Yis {Dis } : αs · γs −1 γs −1 = (1 + τDis ) Rs , (5) σ γs γs 1 γs αs Dis + (1 − αs ) Eis Dis σ−1 Pis Yis {Eis } : (1 − αs ) · γs −1 γs −1 = (1 + τEis ) λs , (6) σ γs γs 1 γs αs Dis + (1 − αs ) Eis Eis γs Dis αs (1 + τEis ) λs Zis ≡ = , (7) Eis (1 − αs ) (1 + τDis ) Rs 1−σ Pis γ −1 − s γs αs +(1−αs )Zis (1+τDis ) Dis = 1− σ × Ds , (8) Ms Pjs j =1 γ −1 − s γs αs +(1−αs )Zjs (1+τDjs ) 1−σ Pis γs −1 αs Zsi γs +(1−αs ) (1+τEis ) Eis = 1−σ × Es . (9) Ms Pjs j =1 γs −1 αs Zjsγs +(1−αs ) (1+τEjs ) σ Firm i’s optimal price, Pis , is a markup, σ −1 , over the marginal cost of producing one unit of value added. The marginal cost is, in turn, a function of two components. The first one is the inverse of firm i’s total factor productivity, Ais . The second ˜ gCostis , is the marginal cost of finance associated with producing one component, M unit of value added. The latter is a weighted average of the unit cost of debt and ˜ equity. Thus, Pis and M gCost is can be written as follows: 12 σ 1 ˜ gCostis (τDis , τEis , Rs , λs , γs , αs ), Pis = M (10) σ−1 Ais where ˜ M gCost ˜ ˜ is = M gCostDis + M gCostE is (11) −1 − γ γ− s − γs s1 ˜ Dis = (1 + τDis ) Rs αs + (1 + M gCost αs ) Zis γs (12) γs −1 − γ γ− s 1 s ˜ E is = (1 + τEis ) λs αs Z M gCost γs + (1 + αs ) (13) is 3.1. Efficient Allocation In preparation for the counterfactual SOE-related structural reforms assessed in section 7, we present here the theoretical properties of an efficient allocation of re- sources, that is, an allocation where there are no financial frictions. This represents a useful benchmark against which to compare the aggregate implications of the proposed reforms. Setting debt and equity distortions to zero in equation (7), it follows that, in an efficient economy, the debt-to-equity ratio is equalized across all firms operating in the same sector. However, equalization of debt and equity ratios does not imply equalization of levels of finance across firms, as shown in the optimal allocations of debt and equity, represented in equations (14) and (15). There, it follows that the planner allocates resources across firms within a sector based on their contribution to the sector’s productivity. Thus, the most productive firms receive a larger proportion of the total optimal level of debt and equity available for each sector. The aggregate output gains from eliminating distortions and reallocating financial 13 resources towards the most productive firms can be written as in equation (16). These gains are equivalent to aggregate productivity (TFP) gains as sector-level debt and equity remain invariant.6 −1 ˆ is = Aσ is D Ms −1 × Ds . (14) j =1 Aσ js −1 ˆis = Aσ is E Ms −1 × Es . (15) j =1 Aσ js ˆ Y Aggregate Output (TFP) gains= −1 × 100. (16) Y While equation (16) allows us to calculate the total gains an economy can obtain from eliminating the distortions and reallocating resources towards the most produc- tive firms, a similar expression, together with equations (8) and (9), can be used to calculate the counterfactual real gains from pursuing different types of SOEs-related structural reforms. 4. Data This section describes the firm-level and dynamic database we constructed to empirically identify the effect of state ownership on the marginal cost of production through the financial channel. We implement the theory on the ORBIS database, working with the data’s raw files to assess the role of firm ownership on the cost of finance. These files are collected by the Bureau van Dijk (BvD) and compiled into two separate modules, a financial 6 Aggregate output, Y , can be expressed as a function of sector-level debt, Ds , sector-level equity, γs θs γs −1 γs −1 γs − 1 S γs γs Es , and sector-level productivity, T F Ps : Y = s=1 T F Ps αs Ds + (1 − αs ) Es . This is derived by combining the expressions for the sector-level demand of debt, sector-level demand of equity, sector-level price, Ps , and aggregate output price, P . As sector-level debt and equity remain constant between the efficient economy and the economy with distortions, aggregate output gains in equation (16) are equivalent to aggregate productivity gains. 14 and an ownership module, which can be matched through a BvD firm identifier (ID).7 The raw database comprises firm-level information from 24 European countries during the period 2010-2016.8 We work with unconsolidated financial balance sheets, which better capture the activity of a firm in the country where it operates, to create the financial module. Building on the work of Kalemli-Ozcan et al. (2015) and Cusolito and Didier (2020), we clean the financial raw files. Appendix A describes the proce- dure implemented to do so. Appendix B presents the representativeness analysis for the final sample. We build on the work by Cusolito and Vranic (2020) to create the ownership module.9 Our main explanatory variable, SOE, measures the percentage of direct shares owned by all the public authorities (PAs) that govern the country where a firm operates. To create this variable, we work with ORBIS Links historical files (often known as Vintage disks) and the Entities file. A Links file is a matrix. Each row of the matrix contains information about the ownership structure of a firm at the first layer of the firm’s ownership tree. This includes the firm identifier, shareholders’ names, each shareholder’s ownership stake, source of information, and date the ownership information was validated. The Entities file contains additional information for each firm, such as the firm identifier, entity type, entity name, and whether the firm is a subsidiary or a parent company. To build the SOE variable, we first clean the raw Links files and eliminate branches (See Appendix C for details). Then, we merge the Links and Entities files to identify the shareholders that are public authorities. We do so using the entity category S, which corresponds to governmental agencies, public departments, and local authori- ties. We exclude sovereign wealth funds and institutional investors from this set, as 7 Each BvD firm ID has two components: a country code that refers to the country where the firm operates and a number. 8 Austria, Belgium, Bosnia and Herzegovina, Bulgaria, Croatia, Czechia, Estonia, Finland, France, Germany, Hungary, Italy, Luxembourg, Montenegro, North Macedonia, Norway, Poland, Portugal, Romania, Slovak Republic, Slovenia, Spain, Serbia and Ukraine. 9 The authors develop a methodology and SQL code to recreate the ownership trees for all the firms in the ORBIS database. 15 these types of investors often capture indirect ownership links. We also differentiate domestic from foreign public authorities, comparing the country code of a firm’s ID with that of the shareholder. Due to span-of-control issues, governments often spread ownership stakes in a firm across several agencies. Therefore, to finally construct our SOE variable, we collapse all the direct shares belonging to all the public authorities in which a firm operates. Figure 1 presents two examples of firms’ ownership trees. The red dots stand for public shareholders, while the blue ones are for private investors. As Figure 1 shows, the state’s ownership stake is widely spread across several firms. Indeed, we found plenty of cases like this one in our database. Moreover, our analysis focuses on direct shares because on average 75.4 percent of the public ownership stake in Europe is concentrated at the first layer of a firm’s ownership tree. Figure 1: Example of a Firm’s Ownership Tree Note: red circles represent public shareholders while blue circles represent private investors. Table 1 presents summary statistics for our main SOE variable. Three conclusions can be drawn from it. First, the state footprint in the economy–the proportion of SOEs in the economy and the average government shareholding–displays a U-shaped pattern with increasing state participation since 2012. Second, assuming a standard 16 class of shares (e.g., 1 share grants 1 voting right), on average, the state controls the firms where it invests, as direct government shareholding accounts for two-thirds, approximately, of the total ownership of a firm. Third, although the proportion of SOEs in a market is small relative to that of POEs, the total number of SOEs in the economy is not trivial. To understand the drivers behind the main state-ownership trends, Table 2 un- packs the sources of SOE variation. As Table 2 shows, ownership changes (type and stake) occur both at the extensive and inventive margins, as well as in both directions. That is, our database reflects situations where the state bailed out or bought POEs and vice versa. In addition, data show that state ownership stakes have increased and decreased, depending on the country and the year. Finally, Table 3 shows that, on average, SOEs have better access to finance, as they have higher debt and equity levels than POEs. Table 1: Summary Statistics - Ownership Variables Variables 2010 2011 2012 2013 2014 2015 2016 Average Proportion of SOE Firms 0.8% 0.7% 0.4% 0.5% 0.5% 0.7% 0.7% 0.6% Average Govt. Shareholdings 84.0% 83.7% 76.2% 77.8% 78.9% 83.1% 83.4% 81.0% Total Number of Firms 611,293 799,037 806,677 897,895 920,578 973,592 887,427 842,357 Total Number of SOEs 4,940 5,615 3,520 4,726 4,889 6,855 6,242 5,255 Note: The number of observations between 2010-2016 is 5,896,499. To obtain the values in this table, we calculate the statistic at the country level and then take the mean of the country statistic for each year. The statistic of the first column corresponds to the proportion of SOE firms in total firms. The statistic of the second column corresponds to the average level of government shareholdings in SOEs. 17 Table 2: Summary Statistics - Variation in State Ownership Extensive Margin Intensive Margin Period POEs POEs SOEs SOEs Share of % of SOEs Average % of SOEs Average →SOEs →SOEs →POEs →POEs SOEs with with Positive with Negative (as a % of (as a % of Change in Positive Change in Negative Change in total total State Change State Change State POEs) SOEs) Ownership relative to Ownership relative to Ownership % all SOEs % all SOEs % with with change change 2010 - 2011 684 0.1% 910 18.4% 6.5% 59.6% 6.2% 40.4% -12.3% 2011 - 2012 733 0.1% 836 14.9% 4.3% 62.7% 9.9% 37.3% -14.2% 2012 - 2013 1,331 0.2% 473 13.4% 6.9% 55.3% 13.8% 44.7% -10.5% 2013 - 2014 493 0.1% 426 9.0% 6.2% 41.8% 12.6% 58.2% -11.3% 2014 - 2015 1,960 0.2% 428 8.8% 6.6% 33.8% 13.4% 66.2% -7.8% 2015 - 2016 447 0.1% 778 11.3% 5.4% 65.5% 6.7% 34.5% -18.5% Note: The number of observations between 2010-2016 is 5,896,499. Column 2 reports the number of POEs that became SOEs in the two-year period. Column 3 is the ratio (in percentages) of the number of POEs that became SOEs in the two-year period relative to the total number of SOEs in the first year of the two-year period. Column 4 reports the number of SOEs that became POEs in the two-year period. Column 5 is the ratio (in percentages) of the number of SOEs that became POEs in the two-year period relative to the total number of SOEs in the first year of the two-year period. Column 6 is the share of SOEs that changed state ownership shareholding percentage, but remained as SOEs, in the two-year period relative to the total number of SOEs in the first year of the two-year period. Column 7 is the share of SOEs that reported a positive change in state ownership shareholding percentage relative to all SOEs that reported a change in state ownership shareholding percentage. Column 8 reports the average positive change in state- related shareholding percentage for SOEs that reported a change in shareholding percentage over the two-year period. Column 9 is the share of SOEs that reported a positive change in state ownership shareholding percentage relative to all SOEs that reported a change in state ownership shareholding percentage. Column 10 reports the average negative change in state ownership shareholding percentage for SOEs that reported a change in shareholding percentage over the two-year period. Table 3: Summary Statistics - Financial Variables: SOEs vs. POEs Variable State-Owned Enterprise Private-Owned Enterprise p-value (t-test) Debt 14.163 12.887 0.00000 Cost of Debt 0.653 0.645 0.00002 Equity 14.235 12.243 0.00000 Cost of Equity 0.512 0.585 0.00000 Number of Observations 36787 5859712 5896499 Note: The number of observations between 2010-2016 is 5,896,499. The descriptive statistics for debt and equity are in natural logarithm. Our measure of debt for firm i that operates in sector s, Dis , is equal to the sum of short-term and long-term debt. The latter includes creditors, loans, other current liabilities, long-term debt, and other non-current liabilities. Our measure of equity, Eis , is equal to total shareholders’ funds. This includes shareholders’ capital and other shareholders’ funds. The descriptive statistics for the cost of debt and cost of equity are expressed as the natural logarithms of the firm-level cost normalized by the weighted average industry cost. Monetary values are in USD 2005. To obtain the values in this table, we calculate the average of each variable across countries and years. The third column shows the p-values for the t-test comparing means across state-owned enterprises and private-owned enterprises. 5. Empirical Strategy To study the effect of state ownership on the cost of production through the ˜ financial channel, M gCost , we estimate the following equation: ˜ ln M gCostisct = α + β SOEisct−1 + γ Publicly Listedisct−1 × SOEicst−1 18 + κXisct−1 + λi + λst + λc + uisct , (17) ˜ where M gCost isct measures the cost of finance for firm i, which operates in sector s, and it is located in country c, at time t. Variable SOE captures total direct shares owned by all the public authorities of the country in which the firm operates. Publicly Listed is a dichotomous variable that takes value 1 if the firm is publicly listed and 0 otherwise.10 Vector X includes control variables that may affect the cost of capital Cusolito and Didier (2020). This includes firm size (ln assets), age, and productivity (ln TFPQ). ˜ We calibrate the model to calculate M gCost isct . We measure nominal value added, Pis Yis , as the difference between sales and intermediate inputs. We calibrate debt and equity prices, as well as the elasticity of substitution between varieties following Hsieh and Klenow (2009) and Whited and Zhao (2021) (Rs = 0.1, λs = 0.1, and σ = 1.77). 11 We estimate γs at the country-sector (2-digit NACE) level following the methodology developed by Kmenta (1967) and applied by Whited and Zhao (2021).12 . Further, using equations (5) and (6), we back-out the firm-specific wedges (1 + τDis ) and (1 + τEis ) with our measures of Pis Yis , Dis , Eis , prices of financial resources, and calibrated parameters. Last, we use equation (18) to calculate firm-level total factor productivity (TFPQ) as follows: σ (Pis Yis ) σ−1 Ais = γs −1 γs −1 . (18) γs γs αs Dis + (1 − αs ) Eis There are important identification concerns associated with eliciting the effect of state ownership on the cost of capital. One of the virtues of the pioneering work by Hsieh and Klenow (2009) and Whited and Zhao (2021) is the tractability of the frameworks they pro- 10 To identify publicly listed firms, we use the variable “Listed” from Orbis, which classifies firms into three categories: Listed, Delisted, Non-listed. We consider a firm listed if it is labeled as “Listed”. 11 The calibration of the sectoral debt and equity prices to arbitrary value is innocuous given our goal of assessing idiosyncratic taxes and subsidies to the purchases of debt and equity across firms within a sector since, in this case, it is the dispersion of marginal returns to debt and equity relative to the average marginal return in the sector that matters for financial misallocation. 12 Thus, γsc is estimated using a non-linear specification of value added on equity and debt con- trolling for firm fixed effects. 19 posed. However, the latter comes at the expense of relying on very restrictive assumptions,13 which have left several economists a bit uncertain as to what the distortion measures re- ally capture when the data do not validate them. This includes variations across firms in risk (Doraszelski and Jaumandreu, 2013), factor prices (De Loecker et al., 2016), markups (Haltiwanger et al., 2018), quality (Krishna et al., 2020), technology (Kasahara et al., 2017), adjustment costs (Asker et al., 2014), and informational asymmetries (David et al., 2021). One characteristic of those potentially confounded factors is that they are often struc- tural in that they are related to firms’ fundamentals. Thus, it takes time for the firm to change them. Therefore, given that our database covers a short period of time, 2010-2016, our identification strategy is to estimate a saturated econometric specification that controls for firm fixed effects. We also include sector-time fixed effects to control for industry trends that may affect the cost of finance (e.g., trade, technological change), as well as governments’ ownership stakes. Since the period of analysis began in 2010, a year in which substantial government intervention was still in place in Europe to recover from the financial crisis of 2008/9, these fixed effects allow us to control for sector-specific bailouts. Our econometric specification also includes country-fixed effects to control for the quality of financial institu- tions and sovereign risk. Last but not least, we lagged all the explanatory variables (except age) to control for endogeneity issues, as variations in the marginal cost of capital may affect governments’ decisions about their ownership stake in a firm, as well as firm size and firm productivity (TFPQ). 6. Empirical Results This section presents the main empirical findings. We start by describing the results from estimating our core specification for the entire sample. Then, we explore heterogeneous effects across sectors. We conclude by presenting back-of-the-envelope calculations to pin down the total government cost of granting preferential financial treatment to SOEs. Table F.9 presents the results from estimating equation (17). Columns (1) to (5) display the results from running OLS regressions with contemporaneous regressors, while column (6) controls for endogeneity. On average, SOEs get subsidized access to financial resources. 13 This includes monopolistic competition and a producer’s price elasticity of -1 with respect to its TFPQ level. 20 Under our most conservative estimate, where the marginal cost of finance is regressed on lagged values of the explanatory variables, a 1 p.p increase in government shareholding reduces SOEs’ cost of finance by 0.01 percent. Additionally, in line with previous empirical evidence for Europe (Cusolito and Didier, 2020), we find that large and mature firms face a lower cost of finance than small and young ones. An important feature of our empirical results is the positive elasticity between a firm’s physical productivity and idiosyncratic financial distortion. Such a property dictates that the nature of the underlying financial distortion is not just randomly creating dispersion in marginal returns to debt and equity but misallocating financial resources away from pro- ductive firms and into less productive ones. This is also a pervasive feature of idiosyncratic distortions in the real-misallocation literature and is particularly detrimental to Total Factor Productivity.14 Table 4: Financial Premium (Tax) of Bureaucrats in Business (1) (2) (3) (4) (5) (6) (OLS) (OLS) (OLS) (OLS) (OLS) (IV) State Ownership -0.0004*** -0.0004*** -0.0004*** -0.0006*** -0.0008*** -0.0001** (0.0001) (0.0001) (0.0001) (0.0001) (0.0000) (0.0001) (Publicly Listed=1) x State Ownership -0.0004 -0.0004 -0.0002 -0.0008** 0.0014** (0.0006) (0.0006) (0.0007) (0.0004) (0.0007) Age -0.0378*** -0.0594*** 0.0128*** -0.0210** (0.0090) (0.0099) (0.0035) (0.0098) Log(Total Assets) -0.3284*** -0.5358*** -0.1577*** (0.0009) (0.0003) (0.0011) Log(TFPQ) 0.4019*** 0.0638*** (0.0001) (0.0005) Observations 5896499 5896499 5896499 5896499 5896499 3917199 Country fixed effects Y Y Y Y Y Y Firm fixed effects Y Y Y Y Y Y Industry-time fixed effects Y Y Y Y Y Y Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. While we highlight the estimates from the specification with lagged explanatory vari- ables, it is worth noting that the estimates for contemporaneous regressors are statistically significant and notably higher in magnitude than the preferred specification. According to the proposed finance allocation theory, firms’ current characteristics determine their 14 As shown in Hsieh and Klenow (2014), the productivity-dependence of idiosyncratic distortions discourage firms’ investments in innovation, magnifying the detrimental effect of these distortions on aggregate productivity. 21 marginal returns to debt and equity. Thus, there is theoretical support for gauging the relationship between the marginal cost of finance and the contemporaneous share of state ownership in the firm, the firm’s idiosyncratic productivity, and other firm-specific charac- teristics. In such a situation, a one p.p. increase in the share of state ownership is associated with a 0.04 and 0.08 percent reduction in the marginal cost of finance. Furthermore, the elasticity of the marginal cost of finance with respect to firms’ idiosyncratic physical pro- ductivity is orders of magnitude higher in absolute value than in the baseline specification and in the ballpark of the estimated values in the real misallocation literature from similar specifications.15,16 To explore heterogeneous effects across sectors, we split the sample and estimate equation (17) for each 1-digit NACE sector, separately. Figure 2 displays blue bars (grey) with the value of the estimated coefficients for the sectors where the SOE variable is (not) statistically significant at the 5 percent or lower percentage level. As Figure 2 shows, the largest state- ownership subsidies appear in sectors that are the greatest facilitators of the economy. This includes finance, electricity, water, transport, as well as agriculture. The average financial premium varies between 0.05 p.p for a sector like transport and up to 3.73 p.p for a sector like finance. 15 See, for instance, Fattal-Jaef (2022) and Bento and Restuccia (2017) for cross-country estimates of the elasticity between idiosyncratic distortions to capital and labor allocations and idiosyncratic productivity. 16 We also conduct a robustness analysis to control for the possibility that our variable of interest may capture the effect of real shocks, as lenders and shareholders may grant access to finance at a lower cost when firms face a positive or a negative shock. To test our hypothesis, we control for a real shock variable. The latter is measured as a dummy that indicates whether a firm’s value-added level is an outlier for that year relative to the period 2010-2016. A value-added level is considered an outlier if it exceeds 2.5 times the inter-quartile range of that firm’s value-added observations within the panel. Table F.9 in the Appendix displays the results. As observed, even though the real shock variable is negative and statistically significant as expected, the estimated coefficients for the variables of interest do not change in value or statistical significance. 22 Figure 2: Heterogeneous SOE Financial Premium (across sectors) Note: This figure plots the financial premiums (taxes) of SOEs differentiated by aggregate industry. Premiums are estimated using equation (17) separately for each 1-digit NACE sector. SS stands for statistically significant at the 5% level or lower. Then, with the main estimated coefficient handily and the information on state owner- ship, the marginal cost of production, and value added for each firm in a country during the year 2016, we conduct back-of-the-envelope (BoE) calculations to determine the cost for the government (in GDP% terms) from subsidizing SOEs. Equation (19) specifies the BoE calculation we performed, and Figure 3 portrays the costs for each country.17 As Figure 3, the total SOE subsidy ranges from a low of 0.001% of GDP for Spain up to 0.955 percent of GDP for Slovenia. ˆ × SOEisct × M gCost ˜ Total SOE Subsidyct i∈Ωct β isct × Yisct = (19) GDPct Yct 17 Ωct in equation (19) represents the set of SOEs in country c at time t. 23 Figure 3: Total SOE Subsidy (% of GDP) Note: The total cost of subsidizing SOEs relative to GDP is calculated using equation (19) and 2016 data. In order to shed light on the policy debate, the next section calculates the economic gains that countries can obtain from implementing other types of SOE reforms than eliminating the financial premium of having bureaucrats in businesses. 7. Productivity Gains from SOE Structural Reforms Our empirical analysis allows us to conduct back-of-the-envelope calculations to de- termine the fiscal savings a government can obtain by implementing policy reforms that eliminate state-ownership financial subsidies. We complement the regression-based analy- sis by proposing a series of policy counterfactuals to quantify the aggregate productivity gains or losses countries can get from pursuing these reforms. More concretely, we con- sider alternative equilibrium stationary allocations where all or subsets of SOEs are shut down and their financial resources are reallocated back into the private sector or remaining firms, respectively. This type of reform captures elements of a privatization exercise where, rather than converting the SOE’s productivity and distortion to those of any particular 24 private firm, resources are reallocated to the private sector in accordance with their rela- tive distortion-adjusted productivity. As reflected in Megginson and Netter (2001), various privatization schemes assigning public ownership of firms to private stakeholders have been pursued throughout history. Our approach is to make financial resources in hands of the state available to the private sector and be allocated across market participants following market forces. Lastly, since the gains from reallocating the SOEs’ financing to the private sector depend on persisting financial frictions in capital markets, we also consider comple- mentary reforms where the shutdown of state enterprises is accompanied by capital market reforms that eliminate financial frictions. 7.1. Productivity Gains: Dismantling all SOEs and Distorted Capital Markets Our first counterfactual characterizes an extreme scenario, where all SOEs are shut down, while the private sector remains active but subject to its intrinsic financial distortions. By closing firms with the state as a shareholder, additional financing can be channeled toward private-run firms. However, because of persisting financial frictions, such reallocation will still be inefficient. While unrealistic, this counterfactual is instructive to emphasize the importance of complementarities in the implementation of structural reforms. The cross-country distribution of TFP changes arising from the first counterfactual is illustrated in Figure 4. A revealing pattern in the Figure is the observation of both gains and losses in aggregate productivity across countries. In some cases, such as Bosnia and Herzegovina, there would be a significant contraction in aggregate productivity from re- movng SOEs, reaching 40%, whereas there are substantial gains in countries like Ukraine, where TFP would increase by close to 35%. There are two confounding forces driving the mixed results. Firstly, even if private-run firms were more productive than government-run ones, the severity of financial distortions in the private sector could be so extreme that further reallocating finance towards POEs is productivity-reducing. Secondly, SOEs may be on average more productive than their private counterparts. Through these mechanisms, aggregate productivity could fall in response to an elimination of SOEs and a reallocation of finance to the private sector. 25 Figure 4: TFP Gains or Losses: Dismantling SOEs with Distorted Capital Markets Note: This figure portrays the TFP gains or losses countries experience from shutting down all SOEs and reallocating the resources towards POEs when financial markets are distorted. To better understand the relative contribution of the two forces governing the reform- driven output change, we decompose the average liability ratio between SOEs and POEs into two components, as shown in equation 20, for both the liabilities in our model: Lsi = {Dsi , Esi }.18 The first one measures the average productivity gap between SOEs and POEs. This component is common to the debt and equity ratios. The second component measures the liability-specific weighted average distortion gap between both groups of firms.19 ¯ SOE L s ¯SOE A s τ SOE ¯Ls ln ¯ P OE = ln ¯P OE − ln P OE . (20) L s A s ¯Ls τ Average Liability Ratio Average Productivity Gap Weighted Average Liability Distortion Gap The tails analysis reveals three interesting patterns. Overall, regardless of whether 18 Bars above variables in equation 20 denote averages. Please refer to Appendix E for the explicit derivation of the equation. 19 This ratio depends on both types of distortions, the price of both liabilities and the parameters of the production function. The weights are the relative contribution of the productivity of each firm to the aggregate productivity of the group they belong to. 26 the reform is productivity-enhancing or -reducing, on average SOEs face higher distortions than POEs. Therefore, aggregate output changes are mainly explained by productivity differences between SOEs and POEs in relevant sectors. Indeed, it is the productivity gap in mainly enabling sectors–with SOEs being more productive than POEs–the driving force behind aggregate real losses in Bosnia and Herzegovina, Finland, and Hungary.20 In contrast, it is the productivity gap in the manufacturing and retail sectors–with POEs being more productive than SOEs–the driving force behind aggregate real gains in the Ukraine.21 Given that productivity differences between SOEs and POEs are important to determine the aggregate effect of an SOE reform, the next subsection discusses the results of a targeted approach, where only unproductive SOEs are shut down. 7.2. Productivity Gains: Targeted SOE Reform and Distorted Capital Markets Given that an indiscriminate SOE reform could backfire in terms of raising aggregate productivity, in this section we consider a targeted intervention where only the poorly per- forming SOEs are withdrawn from the economy and their resources diverted to the private sector. In the distorted economy, the one we observe in the data, a firm’s relative per- formance is given by its debt and equity demands, which are a function of the physical productivity of the firm and the idiosyncratic distortions. Therefore, we design the targeted intervention to close down only the SOEs, if any, whose debt and equity demands are be- low the median demand among their POEs market counterparts in their given country and 4-digit industry. The results illustrated in Figure 5 dictate that once the interventions are targeted at under-performing SOEs, all countries experience a moderate increase in aggregate produc- tivity in response to the policy reform. The gains are moderate because despite reallocating resources away from poorly performing firms, the reallocation takes place in distorted finan- 20 The sales-weighted average productivity gap is 2.7, 6.9, and 1.3 for these countries, respectively. In the case of Bosnia and Herzegovina, SOEs are 93.4 and 9.3 times more productive than POEs in the ICT and energy sectors, respectively. In Finland, the productivity differential is large in the transport, 41.7, and construction, 29.9, sectors. Lastly, in Hungary, most of the real losses are ex- plained by the productivity gap in the ICT, 7.1, energy, 1.5, and transport, 1.4, sectors. Additionally, productivity gaps in manufacturing sectors appear to be relevant for Bosnia and Herzegovina, 3.1, and Finland, 8.2. 21 Ukraine’s POEs are 9.5 and 4.3 times more productive than SOEs in the retail and manufacturing sectors, respectively. 27 cial markets. Indeed, the highest TFP gains are 14.1% for Germany, 11.1% for Ukraine, and 8.4% for Montenegro. If financial distortions among the private sector and surviving SOEs are relatively more severe, aggregate productivity declines in response to the intervention. Overall, however, the targeted intervention proves to be more effective at raising aggregate productivity than an indiscriminate elimination of government-run firms. The next section explores the effect of fixing markets before reallocating resources to make the SOE reform more effective. Figure 5: TFP Gains: Targeted SOE Reform with Distorted Capital Markets Note: The figure illustrates the T F P aggregate changes from a targeted removal of the SOEs with debt and equity levels below the median level among private sector enterprises in the same industry and country. The T F P gains in the counterfactual economy are measured relative to the distorted allocation we observe in the data, with both SOEs and financial distortions. The numbers reported in the histogram correspond to the average for each country across all years in our sample. 7.3. Productivity Gains: Targeted SOE Reform and Fixed Capital Markets In our final counterfactual, we consider a scenario where the targeted SOE intervention is implemented alongside a reform that withdraws the dispersion in distortions from financial markets. In this case, the reallocation of the financing absorbed by the underperforming SOEs is conducted efficiently. Figure 6 shows that targeted SOE interventions combined with financial reforms aimed at making markets more efficient lead to gains in every country. The latter ranges from 26.6% for a country like Austria up to 85.5% for Ukraine. 28 Figure 6: TFP Gains: Targeted SOE Reform with Undistorted Capital Markets 1.0 0.8 0.6 0.4 0.2 0.0 Norway Austria Montenegro France Poland Luxembourg Finland Slovenia Portugal Bosnia and Herzegovina North Macedonia Hungary Germany Czech Republic Italy Belgium Estonia Slovak Republic Croatia Serbia Spain Ukraine Romania Bulgaria Note: This figure illustrates T F P gains from a targeted removal of SOEs with debt and equity levels below the median among private sector enterprises in the same industry and country. The T F P gains in the counterfactual economy are measured relative to the economy as implied by the data. The numbers reported in the histogram correspond to the average for each country across all years in our sample. 8. Conclusion For more than a century, economists and policy makers have debated about the rationale for and potential effects of state participation in the economy. Despite the latest privatiza- tion waves and further structural reforms that broadly mirrored the global consensus about the need to shrink a bossy business government, the footprint of the state in the local and global economy remains indelible. The debate has recently regained momentum amid the notorious return of industrial policy as the cornerstone of the policy toolbox to promote private-sector-led growth despite the past pitfalls of its own. Evidence about the distortionary effect of state ownership in the economy is surprisingly thin and mainly China-centric. However, while China offers the typical case for studying the economic implications of state participation in the market-based economy, the speci- ficities of the Chinese case impede establishing a broader cross-country and cross-sectoral characterization that helps understand and, more importantly, quantify the fiscal cost and productivity effect of having the state heavily involved in the economy as a market player. This paper comes to fill this gap. Our paper shows that firms with public authorities as direct shareholders (SOEs) get, on average, subsidized access to finance compared to private-owned enterprises (POEs). A 1 p.p increase in government direct shareholding reduces the average cost of finance (e.g., 29 debt and equity) by 0.01 percent. The total SOE subsidy has an annual fiscal cost of up to 1% of GDP for a country like Slovenia during the year 2016. The largest subsidies appear in sectors that are the greatest facilitators of the economy. This includes finance, electricity, water, transport, as well as agriculture.22 Moreover, the paper shows that one size does not fit all. Our counterfactual analysis re- veals that indiscriminate interventions aimed at dismantling SOEs may backfire in economies where government-run enterprises outperform private-sector peers and where severe finan- cial distortions affecting POEs remain in place. Leveraging the theoretical underpinnings of our analysis, we constructed counterfactual allocations where SOEs were shut down un- der alternative assumptions about distortions in capital markets. We found that in many economies, SOEs perform relatively well compared to POEs. Hence, their dismantlement would not translate into aggregate productivity gains. Targeting SOE reforms to dismantle those government-run firms with relatively poor performance increases the number of countries benefiting from the reforms. However, while necessary, the latter is not enough to maximize the economic gains from SOE reforms. Nonetheless, as expected, all interventions will translate into larger productivity gains if governments make financial markets less distorted. That is, if the SOE policies are accom- panied by financial market reforms that improve the allocation of capital across all types of firms. We conclude our paper with some reflections about the relative importance of intensive (reducing the subsidy) versus extensive (dismantling) SOE reforms to improve economic outcomes. It may be fair to argue that the small value of the estimated coefficient that measures the average effect of state ownership on the cost of finance suggests that reforms at the extensive margin seem to be the most effective mechanism to maximize the economic gains from SOE reforms. However, changes in the intensive margin, often implemented through corporate governance reforms, may be prominent if SOEs are, as these days, highly indebted. In the end, answering this type of policy question highlights the relevance of firm-level evidence to inform the current policy debate. 22 We replicated the analysis for factor markets and found that distortions in factor markets are larger than in financial ones, probably as they accumulate the upstream distortions related to getting access to finance. The results are available upon request (Cusolito et al., 2023). 30 Appendix A. Historical Financial Data Cleaning Procedure Following Bureau van Dijk (2011), Kalemli-Ozcan et al. (2015), Cusolito and Didier (2020), and Kalemli-Ozcan et al. (2023), we document the steps we apply to clean the financial information. 1. Fill time-invariant data gaps : for a given BvD.ID-year combination, with BvD.ID standing for firm unique identifier, replace missing highly likely time-invariant in- formation with information available for previous years (e.g., US SIC code, NAICS, NACE, NACE main sector, company name, city, region, postal code, legal form, in- corporation date). To perform this step, the team first worked with auxiliary raw tables, which collect legal and sectoral information of the firm, and collapsed the time-invariant variables at the BvD.ID level. 2. Harmonize timeframe : convert variable closedate from string to numeric format. Then create a new variable, name it year, and assign a year to the observation according to the following rule. If closing month corresponding to the observation is June or any other month after June, then make Year take the year reported in closedate. Otherwise, make Year the year reported in closedate minus 1. 3. Drop duplicates : the raw database presents a large number of duplicates at the BvD.ID-year level. The team noticed that the information was the same, except in the SIC primary code variable. Thus, we collapsed all the SIC primary codes re- ported by the same BvD.ID-year in one variable, using semicolons to list all the SIC primary codes, and eliminated duplicates. 4. Drop firms with missing relevant information : drop all the firms with no information for the following set of variables: US SIC code, NAICS, NACE core code, NACE main sector. 5. Drop observations with missing information for the currency code : eliminate obser- vations with missing information for the currency code. 6. Drop observations with missing information for variable closedate : eliminate observa- tions with missing information for the close date of the financial statement. 31 7. Drop observations with relevant missing information eliminate observations that at the BvD.ID-year level have missing information in all the following variables: operat- ing revenue (turnover), sales, employment, total assets. 8. Drop duplicates and keep most updated information : keep observations with the most recent closing date if there are duplicates at the BvD.ID-year-first letter of consolida- tion code (e.g., C, U) level. 9. Drop duplicates and keep information from annual reports : keep observations with annual report in Use FillingType variable if there are still duplicates and keep the standardized information. Using annual reports (IFRS preferred, instead of local reports) guarantees standardization of reporting protocol at international level. 10. Eliminate firms with noisy data : drop all the observations corresponding to a specific BvD.ID if any of the following variables has a negative value in a specific year – total fixed assets, tangible fixed assets, intangible fixed assets, other fixed assets, current assets, sales, and employment. 11. Deflate values : use country GDP deflators from the World Bank database to deflate nominal variables and set year 2005 as the base year.23 12. Harmonize currencies : convert values in local currency to USD dollars, using the average of the monthly exchange rate for year 2005. 23 https://data.worldbank.org/indicator/NY.GDP.DEFL.ZS. 32 Appendix B. Validation of Final Database We validate the representativeness of the final database by calculating the ratio of the sum of employment and gross output in the database to their corresponding aggregates, in the same manner as Gopinath et al. (2017). Aggregates for employment and gross output are obtained from Eurostat’s Structural Business Statistics Database (SBS). Table B.5 com- pares the coverage of our final database to that of Gopinath et al. (2017) for the Spanish manufacturing sector. Our coverage is smaller than theirs because not all firms have the appropriate information in the historical ownership module to determine their linkage to the state or not, resulting in additional attrition in our sample. Tables B.6 and B.7 show the coverage of our sample by country, separately for manufacturing and non-manufacturing sectors. Table B.5: Coverage of Final Database Relative to Gopinath et. al. (2017) - Spain Manufacturing Employment Gross Output Year Final Database Gopinath et. al. Final Database Gopinath et. al. 2010 41.9% 68.0% 37.3% 74.0% 2011 43.5% 69.0% 35.6% 75.0% 2012 39.8% 65.0% 31.5% 72.0% Note: This table only compares the years for which our database overlaps that of Gopinath et al. (2017), 2010 - 2012. 33 Table B.6: Coverage of Final Database Relative to Eurostat (SBS) - Manufacturing 2010 2011 2012 2013 2014 2015 2016 Country Employment Gross Output Employment Gross Output Employment Gross Output Employment Gross Output Employment Gross Output Employment Gross Output Employment Gross Output Austria 7.1% 5.6% 13.4% 10.3% 27.2% 28.7% 39.3% 47.7% 39.5% 48.9% 42.6% 54.1% 43.9% 55.1% Belgium 54.0% 69.5% 56.2% 66.4% 57.8% 71.4% 58.7% 71.3% 61.6% 69.9% 61.9% 73.8% 61.2% 71.2% Bosnia and Herzegovina 45.6% 41.9% 34.7% 37.2% 52.2% 49.6% 50.7% 49.6% 52.0% 53.9% 47.7% 48.1% Bulgaria 35.0% 26.4% 42.5% 27.6% 48.4% 27.3% 66.3% 38.2% 67.4% 40.5% 71.1% 43.6% 68.0% 43.8% Croatia 34.2% 41.2% 34.8% 44.4% 46.6% 55.5% 46.1% 57.1% 49.7% 58.2% 52.0% 55.6% 51.3% 53.1% Czechia 61.7% 42.2% 60.1% 43.6% 61.6% 43.1% 65.8% 46.8% 69.1% 48.7% 70.2% 53.7% 69.0% 56.9% Estonia 35.5% 30.5% 38.5% 30.2% 37.0% 29.4% 40.0% 32.3% 38.8% 32.5% 39.4% 33.2% 40.7% 35.1% Finland 27.2% 19.0% 29.9% 20.1% 31.2% 21.3% 29.6% 18.1% 34.6% 23.1% 35.4% 25.7% 33.7% 26.4% France 17.9% 15.5% 17.4% 14.9% 15.2% 12.7% 18.9% 16.9% 23.3% 22.3% 25.3% 25.0% 25.2% 24.5% Germany 24.7% 33.9% 26.8% 36.2% 27.9% 37.1% 26.5% 33.2% 27.0% 34.3% 27.8% 36.8% 25.0% 31.5% Hungary 48.1% 67.5% 52.1% 69.4% 53.1% 70.5% 58.1% 78.5% 54.5% 82.1% 58.8% 85.6% 55.9% 85.8% Italy 39.6% 42.8% 54.9% 53.4% 55.8% 51.5% 57.6% 54.5% 58.4% 54.7% 61.3% 55.8% 60.4% 57.1% Luxembourg 49.3% 71.6% 64.5% 98.6% 49.6% 84.5% 59.3% 75.6% 56.9% 72.2% 52.5% 63.9% 44.1% 57.0% North Macedonia 57.3% 39.0% 53.4% 42.2% 39.3% 29.3% 33.4% 25.7% 53.2% Norway 3.0% 1.8% 4.0% 2.5% 3.0% 2.6% 1.6% 0.9% 2.6% 1.9% 52.9% 44.1% 55.4% 48.3% Poland 20.5% 25.8% 17.5% 21.9% 13.3% 17.8% 8.4% 10.4% 6.3% 9.0% 5.5% 8.1% 13.5% 19.7% Portugal 41.6% 33.6% 58.7% 46.2% 41.7% 30.7% 62.7% 47.9% 63.7% 48.6% 64.4% 51.2% 64.3% 51.8% Romania 44.7% 40.5% 49.8% 45.5% 51.3% 49.3% 56.2% 52.7% 56.5% 52.3% 58.9% 54.1% 57.6% 53.9% Serbia 59.5% 78.5% Slovak Republic 46.0% 33.1% 44.4% 37.8% 49.1% 48.0% 58.3% 55.8% 57.1% 46.6% 52.9% 40.3% 49.8% 40.3% Slovenia 46.2% 46.1% 48.6% 48.3% 51.8% 52.3% 50.2% 49.3% 52.5% 50.6% 50.5% 50.4% 53.4% 52.7% Spain 41.9% 37.3% 43.5% 35.6% 39.8% 31.5% 47.6% 40.2% 48.7% 38.4% 49.0% 40.4% 34.4% 28.7% Note: Blanks correspond to country-year pairs for which the Eurostat’s SBS Database does not report information. Montenegro and Ukraine are excluded from this table as the SBS Database does not include information for any of the years in our sample. Table B.7: Coverage of Final Database Relative to Eurostat (SBS) - Non-Manufacturing 34 2010 2011 2012 2013 2014 2015 2016 Country Employment Gross Output Employment Gross Output Employment Gross Output Employment Gross Output Employment Gross Output Employment Gross Output Employment Gross Output Austria 9.8% 9.2% 10.1% 9.4% 17.5% 21.4% 21.6% 27.4% 19.4% 26.9% 24.6% 30.9% 20.3% 27.9% Belgium 33.9% 43.1% 32.9% 43.2% 35.5% 43.4% 36.3% 44.5% 36.0% 45.4% 36.3% 44.7% 35.3% 44.2% Bosnia and Herzegovina 49.7% 54.6% 27.9% 33.7% 49.1% 54.3% 50.7% 53.2% 43.4% 50.8% 49.0% 51.5% Bulgaria 30.5% 31.2% 36.7% 35.5% 42.6% 36.3% 61.2% 51.5% 64.3% 52.9% 68.4% 54.2% 64.0% 53.4% Croatia 24.3% 33.9% 26.9% 38.0% 42.0% 52.4% 42.9% 53.3% 45.6% 56.7% 44.6% 56.2% 45.1% 56.4% Czechia 51.5% 36.4% 51.5% 32.4% 52.0% 33.7% 56.6% 37.3% 58.0% 41.1% 59.2% 43.6% 56.9% 43.7% Estonia 33.7% 30.1% 36.4% 31.4% 35.3% 33.7% 37.9% 33.7% 37.2% 34.1% 37.5% 34.5% 40.0% 37.0% Finland 30.1% 39.3% 30.1% 44.0% 31.5% 41.6% 28.9% 44.3% 32.3% 45.2% 32.8% 48.7% 30.9% 45.1% France 14.1% 18.2% 13.7% 17.5% 12.3% 15.4% 15.6% 20.5% 17.0% 23.1% 17.7% 23.0% 15.2% 22.3% Germany 23.0% 39.6% 24.3% 41.8% 25.7% 41.9% 25.1% 41.7% 25.9% 41.8% 25.3% 38.1% 23.6% 37.5% Hungary 30.1% 49.6% 31.9% 53.5% 31.6% 56.8% 32.7% 60.4% 33.6% 62.3% 32.3% 60.6% 29.8% 59.0% Italy 31.5% 34.1% 45.8% 45.1% 46.7% 42.6% 45.1% 41.4% 47.3% 42.7% 49.8% 45.2% 48.1% 46.5% Luxembourg 19.8% 46.4% 24.1% 57.8% 20.7% 48.6% 32.0% 63.0% 36.9% 65.2% 55.4% 80.1% 41.4% 76.0% North Macedonia 46.5% 61.5% 53.7% 69.7% 37.9% 50.3% 36.1% 49.3% 59.0% 68.5% Norway 1.0% 1.4% 4.3% 4.1% 1.0% 1.1% 3.4% 3.0% 1.7% 1.9% 65.1% 52.7% 65.4% 57.7% Poland 13.1% 16.2% 12.7% 14.9% 8.5% 9.9% 4.9% 5.5% 4.6% 5.8% 3.1% 4.4% 8.9% 11.0% Portugal 27.2% 31.9% 36.7% 41.7% 27.6% 31.6% 39.1% 43.8% 40.4% 46.4% 40.6% 47.4% 39.7% 46.6% Romania 35.2% 42.6% 42.3% 51.4% 43.4% 52.8% 46.8% 57.8% 48.8% 59.6% 53.7% 67.6% 53.2% 68.2% Serbia 44.8% 78.3% Slovak Republic 35.7% 37.5% 36.6% 39.3% 38.6% 39.2% 51.8% 51.2% 55.3% 53.5% 51.2% 53.4% 48.7% 48.0% Slovenia 35.3% 45.0% 40.1% 48.2% 40.0% 47.7% 41.7% 46.0% 40.9% 48.0% 39.0% 46.3% 42.5% 46.8% Spain 31.5% 38.0% 32.6% 37.9% 30.0% 35.0% 34.8% 41.2% 34.8% 40.2% 35.0% 41.5% 25.3% 30.2% Note: Blanks correspond to country-year pairs for which the Eurostat’s SBS Database does not report information. Montenegro and Ukraine are excluded from this table as the SBS Database does not include information for any of the years in our sample. Appendix C. Historical Ownership Data Cleaning Procedure To develop the historical ownership module, we apply the following sequence of cleaning steps, following the methodology and SQL code developed by Cusolito and Vranic (2020). 1. Merge all Links files : create a unique file that compiles the individual annual Links files. 2. Remove duplicates and keep the most updated information : eliminate duplicate obser- vations and keep the most updated information. 3. Harmonize the time frame of the information : use the information-date variable to identify the latest month where the ownership data was collected. If the latest month is June or after, then assign to the firm the Links file of the same year. Otherwise, assign the Links file of the previous year. We apply this rule to keep consistency with the timing rule applied when cleaning the financial information. 4. Replace BvD ownership codes with numeric values : in some specific cases, BvD has missing information about the ownership stakes of a particular shareholder. However, using secondary sources of information, BvD collects imprecise, though valuable, in- formation that can be used to fill the data gaps. The following table presents the codes that BvD uses, their meaning, and the numeric value the company assigns to each link (Bureau van Dijk, 2018). Table C.8: BvD Ownership Codes BvD code Meaning Definition Numeric Value Assigned WO Wholly Owned The shareholder has at least 98% of the company 98% MO Majority Owned The shareholder has at least 50.01% of the company 50.01% JO Jointly Owned The shareholder has 50% of the company 50% CQP1 General Partner The shareholder has 50% of the company plus 1 share 50.01% NG Negligible The shareholder has 0.01% of the shares or less than that . 35 Appendix D. Misallocation and Model Mis-specification The framework proposed by Whited and Zhao (2021), which we adopt to conduct our empirical and quantitative analysis, inherits all of the limitations of any model-driven ap- proach to elicit distortions from the data. Part of it is an actual distortion, and another part is the forces unaccounted for in the model. In the context of real-resource misallocation, un- accounted forces, such as adjustment costs to labor and capital accumulation, heterogeneity of markups across firms, technological differences across firms, and informational frictions, were all shown to account for some fraction of the overall misallocation inferred without consideration for these channels. However, substantial residual dispersion of marginal re- turns across firms remained, suggesting that misallocation is effectively a rooted feature of the data in less developed countries. In the context of our assessment of the role of SOEs in misallocating financial resources, further limitations should be acknowledged. For instance, the theoretical framework does not endow state enterprises with any role in fixing market failures or contributing to a country’s welfare function other than through allocative efficiency. All we are characterizing in our model is the amount and type of financing received compared with their private marginal returns. If, in reality, SOEs brought non-marketable returns that merited a larger size, the methodology would incorrectly treat it as a distortion. We acknowledge this as a possibility and offer a few caveats that attenuate the concern. Firstly, we estimate the relationship between financial distortions and state ownership sep- arately across broad economic sectors. So, if we found robust patterns across sectors, and if the non-private role of SOEs was more notable in some sectors and not in others, then we would be reassured that, indeed, there is distorted access to finance from SOEs. As we will see, this is what we report later. Secondly, even in the economy wide econometric specifications, we control for sector, time, and firm fixed effects, providing another layer of attenuation to the concerns. And lastly, we do not endow private firms with any social role either. It is conceivable that some private firms bring externalities to the rest of the econ- omy through knowledge spillovers or other channels. In this way, we are also misattributing excessive finance to a private firm when indeed it is warranted by their externalities. 36 Appendix E. Decomposition The expressions for the optimal debt and equity are: 1−σ Psi γ −1 − s γs αs +(1−αs )Zsi (1+τDsi ) Dsi = 1−σ × Ds Ms Psj j =1 γ −1 − s γs αs +(1−αs )Zsj (1+τDsj ) 1−σ Psi γs −1 αs Zsi γs +(1−αs ) (1+τEsi ) Esi = 1−σ × Es . Ms Psj j =1 γs −1 αs Zsjγs +(1−αs ) (1+τEsj ) The optimal price of the differentiated firm is given by: − γ γ− s σ 1 − γs −1 s 1 Psi = (1 + τDsi ) R αs + (1 − αs ) Zsi γs σ − 1 Asi γs −1 − γ γ− s 1 s γs + (1 + τEsi ) λ αs Zsi + (1 − αs ) . Replacing Psi in the expression for optimal debt, Dsi : 1−σ Psi γ −1 − s γs αs +(1−αs )Zsi (1+τDsi ) Dsi = 1−σ × Ds Ms Psj j =1 γ −1 − s γs αs +(1−αs )Zsj (1+τDsj )   γs γs 1−σ γ −1 −γ − γs −1 −γ − − s s 1 s 1  σ 1  σ −1 Asi (1+τDsi )R αs +(1−αs )Zsi γs +(1+τEsi )λ αs Zsi γs +(1−αs )  γ −1 − s γs αs +(1−αs )Zsi (1+τDsi ) ⇒ Dsi = 1−σ × Ds Ms Psj j =1 γ −1 − s γs αs +(1−αs )Zsj (1+τDsj ) 37   γs γs 1−σ γ −1 −γ − γs −1 −γ − − s s 1 s 1 σ σ Zsi 1  σ −1 Asi (1+τDsi )RZsi αs +(1−αs )Zsi γs +(1+τEsi )λZsi αs Zsi γs +(1−αs )  γ −1 − s γs Zsi αs +(1−αs )Zsi (1+τDsi ) ⇒ Dsi = 1−σ × Ds Ms Psj j =1 γ −1 − s γs αs +(1−αs )Zsj (1+τDsj )   γs γs 1−σ γ −1 −γ − γ −1 −γ − − s s 1 − s s 1 σ −1  σ Zsi 1  σ −1 Asi (1+τDsi )RZsi αs +(1−αs )Zsi γs +(1+τEsi )λ αs +(1−αs )Zsi γs  γ −1 − s γs αs +(1−αs )Zsi (1+τDsi ) ⇒ Dsi = 1−σ × Ds Ms Psj j =1 γ −1 − s γs αs +(1−αs )Zsj (1+τDsj ) γs (σ −1)−1 −1 σ −1 σ −1 σ −1 [(1+τDsi )RZsi +(1+τEsi )λ] 1−σ − γs −1 γs −1 Aσ si Zsi σ (1+τDsi ) αs + (1 − αs ) Zsi γs ⇒ Dsi = 1−σ × Ds Ms Psj j =1 γ −1 − s γs αs +(1−αs )Zsj (1+τDsj ) SOE Define ΩSOE = 1, ..., Ms P OE as the set of SOEs and ΩP OE = 1, ..., Ms as the set of −1 σ −1 Aσ Asi POEs. Define χSOE si = si Aσ −1 ∀i ∈ ΩSOE and χP si OE = Aσ −1 ∀i ∈ ΩP OE . i∈ΩSOE si i∈ΩP OE si SOE : The sum of total debt for SOEs in sector s, Ds SOE Ds = Dsi i∈ΩSOE   γs (σ−1)−1 γ −1 γs − 1 σ −1 [(1+τDsi )RZsi +(1+τEsi )λ]1−σ  − s −1 σ −1 σ −1 γs  Aσ si Zsi σ αs + (1 − αs ) Zsi (1+τDsi ) SOE ⇒ Ds = 1−σ × Ds i∈ΩSOE Ms  P sj γ −1  j =1 − s γs  αs +(1−αs )Zsj  1+τDsj    γs (σ−1)−1 γ −1 γs −1 σ −1 [(1+τDsi )RZsi +(1+τEsi )λ]1−σ  − s σ −1 σ −1 γs  Zsi σ αs + (1 − αs ) Zsi −1 (1+τDsi ) SOE σ −1 Aσ si ⇒ Ds = Asi × −1 1−σ × Ds i∈ΩSOE i∈ΩSOE i∈ΩSOE Aσ si Ms  P sj γ −1  j =1 − s γs  αs +(1−αs )Zsj  1+τDsj    γs (σ−1)−1 γ −1 γs −1 σ −1 [(1+τDsi )RZsi +(1+τEsi )λ]1−σ  − s σ −1 σ −1 γs  Zsi σ αs + (1 − αs ) Zsi (1+τDsi ) SOE σ −1 SOE ⇒ Ds = Asi × χsi 1−σ × Ds i∈ΩSOE i∈ΩSOE Ms  P sj γ −1  j =1 − s γs  αs +(1−αs )Zsj  1+τDsj  σ−1 σ −1 Ds SOE σ −1 ⇒ Ds = Asi × 1−σ i∈ΩSOE σ Ms  P sj γ −1  j =1 − s γs  αs +(1−αs )Zsj  1+τDsj    γs (σ −1)−1 γs − 1 SOE σ −1 [(1 + τDsi ) RZsi + (1 + τEsi ) λ]1−σ − γs γs −1 × χsi Zsi αs + (1 − αs ) Z si  . i∈ΩSOE (1 + τDsi ) 38 γs −1 (σ −1)−1 σ −1 [(1+τDsi )RZsi +(1+τEsi )λ]1−σ − γs γs − 1 Define τ SOE ¯Ds 1 = i∈ΩSOE χSOE si Zsi (1+τDsi ) αs + (1 − αs ) Zsi γ s , which represents the level of distortions that SOEs in sector s face, weighted by firms’ productivity, χSOE si . This term is a function of firm-level wedges. Also, define the average productivity σ −1 ¯SOE = i∈XSOE Asi of SOEs as A s MsSOE . Hence, the average level of debt of SOEs in sector s, SOE Ds ¯ SOE = D , can be defined by: s SOE Ms σ −1 ¯ SOE = A ¯SOE × σ−1 Ds 1 D s s 1−σ × SOE . σ Ms Psj ¯Ds τ j =1 γ −1 − s γs αs +(1−αs )Zsj (1+τDsj ) A similar expression can be written for the average level of debt of POEs in sector s, P OE Ds ¯s D P OE = P OE : Ms σ −1 ¯sP OE ¯P OE σ−1 Ds 1 D =A s × 1−σ × P OE , σ Ms Psj ¯Ds τ j =1 γ −1 − s αs +(1−αs )Zsj γs (1+τDsj ) γs (σ −1)−1 1−σ − γs −1 γs −1 where 1 = χP OE Z σ −1 [(1+τDsi )RZsi +(1+τEsi )λ] αs + (1 − αs ) Zsi γs . τ P OE ¯Ds i∈ΩP OE si si (1+τDsi ) Hence, the ratio of average debt level of SOEs relative to average debt level of POEs can be expressed as: ¯ SOE D ¯SOE A τ P OE ¯Ds s s ¯sP OE = ¯P OE × SOE D A s ¯Ds τ Average Productivity Gap Weighted Average Debt Distortion Gap ¯s D SOE ¯SOE A s τ SOE ¯Ds ln ¯ P OE = ln ¯P OE − ln P OE . Ds A s ¯Ds τ Average Productivity Gap Weighted Average Debt Distortion Gap A similar expression can be obtained for the ratio of average equity level of SOEs relative to POEs: ¯ SOE E ¯SOE A τ P OE ¯Es s s ¯sP OE = ¯P OE × SOE E As ¯Es τ Average Productivity Gap Weighted Average Equity Distortion Gap 39 ¯ SOE E ¯SOE A τ SOE ¯Es s s ln ¯sP OE = ln ¯P OE − ln P OE , E A s ¯Es τ Average Productivity Gap Weighted Average Equity Distortion Gap γs 1−σ γs −1 γs −1 (σ −1)−1 1 σ −1 [(1+τDsi )R+(1+τEsi )λZsi ] where τ SOE ¯Es = i∈ΩSOE χSOE si Zsi (1+τEsi ) αs Zsi γs + (1 − αs ) γs 1−σ γs −1 γs −1 (σ −1)−1 and where 1 = χP OE Z σ −1 [(1+τDsi )R+(1+τEsi )λZsi ] αs Zsi γs + (1 − αs ) . τ P OE ¯Es i∈ΩP OE si si (1+τEsi ) Hence, for either liablity of our model, Lsi = {Dsi , Esi }, we can decompose the average liablity ratio between SOEs and POEs into: ¯ SOE L s ¯SOE A s τ SOE ¯Ls ln ¯ P OE = ln ¯P OE − ln P OE . L s A s ¯Ls τ Average Liability Ratio Average Productivity Gap Weighted Average Liability Distortion Gap 40 Appendix F. Controlling for Real Shocks Table F.9: Financial Premium (Tax) of Bureaucrats in Business - Robustness OLS OLS OLS OLS OLS OLS IV SOE -0.0004*** -0.0004*** -0.0004*** -0.0006*** -0.0008*** -0.0008*** -0.0001** (0.0001) (0.0001) (0.0001) (0.0001) (0.0000) (0.0000) (0.0001) (Publicly Listed=1) x SOE -0.0004 -0.0004 -0.0002 -0.0008** -0.0008** 0.0014** (0.0006) (0.0006) (0.0007) (0.0004) (0.0004) (0.0007) Age -0.0378*** -0.0594*** 0.0128*** 0.0129*** -0.0207** (0.0090) (0.0099) (0.0035) (0.0035) (0.0098) Log(Total Assets) -0.3284*** -0.5358*** -0.5359*** -0.1575*** (0.0009) (0.0003) (0.0003) (0.0011) Log(TFPQ) 0.4019*** 0.4018*** 0.0640*** (0.0001) (0.0001) (0.0005) Real Shock 0.0107*** -0.0432*** (0.0006) (0.0019) Observations 5896499 5896499 5896499 5896499 5896499 5896499 3917199 Country fixed effects Y Y Y Y Y Y Y Firm fixed effects Y Y Y Y Y Y Y Industry-time fixed effects Y Y Y Y Y Y Y Standard errors in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01 Note: The real shock variable is measured as a dummy that indicates whether a firm’s value-added level is an outlier for that year. A value-added level is considered an outlier if it exceeds 2.5 times the interquartile range of that firm’s value-added observations within the panel. 41 References John Asker, Allan Collard-Wexler, and Jan De Loecker. Dynamic inputs and resource (mis) allocation. Journal of Political Economy, 122(5):1013–1063, 2014. URL https://www.journals.uchicago.edu/doi/full/10.1086/677072. Chong-En Bai, Chang-Tai Hsieh, and Zheng (Michael) Song. The long shadow of china’s fiscal expansion. Brookings Papers on Economic Activity, pages 129– 165, 2016. ISSN 00072303, 15334465. URL http://www.jstor.org/stable/ 90000434. Abhijit Banerjee. Land Reforms: Prospects and Strategies. Working papers 99-24, Massachusetts Institute of Technology (MIT), Department of Economics, October 1999. URL https://ideas.repec.org/p/mit/worpap/99-24.html. Pedro Bento and Diego Restuccia. Misallocation, establishment size, and productiv- ity. American Economic Journal: Macroeconomics, 9(3):267–303, July 2017. doi: 10.1257/mac.20150281. URL https://www.aeaweb.org/articles?id=10.1257/ mac.20150281. Timothy Besley and Maitreesh Ghatak. Property rights and economic development. volume 5, chapter Chapter 68, pages 4525–4595. Elsevier, 2010. URL https: //EconPapers.repec.org/RePEc:eee:devchp:v:5:y:2010:i:c:p:4525-4595. Philipp Boeing, Elisabeth Mueller, and Philipp Sandner. China’s R&D explo- sion—Analyzing productivity effects across ownership types and over time. Re- search Policy, 45(1):159–176, 2016. doi: 10.1016/j.respol.2015.07. URL https: //ideas.repec.org/a/eee/respol/v45y2016i1p159-176.html. Ginka Borisova and William L Megginson. Does government ownership affect the cost of debt? evidence from privatization. The Review of Financial Studies, 24 (8):2693–2737, 2011. Loren Brandt, Trevor Tombe, and Xiadong Zhu. Factor Market Distortions Across Time, Space, and Sectors in China. Review of Economic Dynamics, 16(1):39–58, January 2013. doi: 10.1010/j.red.2012.10.002. URL https://ideas.repec.org/ a/red/issued/11-95.html. Francisco J. Buera, Joseph P. Kaboski, and Yongseok Shin. Finance and development: A tale of two sectors. The American Economic Review, 101(5):1964–2002, 2011. ISSN 00028282. URL http://www.jstor.org/stable/23045628. Bureau van Dijk. Orbis - User Guide. Technical report, Bureau van Dijk, 2011. Bureau van Dijk. Bureau van Dijk historical ownership databases - User Guide. Technical report, Bureau van Dijk, 2018. 42 Yiming Cao, Raymond Fisman, Hui Lin, and Yongxiang Wang. Soes and soft incentive constraints in state bank lending. American Economic Journal: Economic Policy, forthcoming. Hans Christiansen. Balancing commercial and non-commercial priorities of state- owned enterprises. (6), 2013. doi: https://doi.org/https://doi.org/10.1787/ 5k4dkhztkp9r-en. URL https://www.oecd-ilibrary.org/content/paper/ 5k4dkhztkp9r-en. Lin William Cong, Haoyu Gao, Jacopo Ponticelli, and Xiaoguang Yang. Credit allo- cation under economic stimulus: Evidence from china. The Review of Financial Studies, 32(9):3412–3460, 2019. Ana Cusolito and William Maloney. Productivity Revisited: Shifting Paradigms in Analysis and Policy. 12 2018. ISBN 978-1-4648-1334-4. doi: 10.1596/ 978-1-4648-1334-4. Ana Paula Cusolito and Tatiana Didier. Methodology, SQL, and Stata Codes to Create a Clean Financial Firm-level Data using Orbis Raw Files. Working Paper, 2020. Ana Paula Cusolito and Goran Vranic. Methodology and SQL Code to Create Own- ership Trees and a Firm-level Ownership Module based on Orbis Data. Working Paper, 2020. no Pe˜ Ana Paula Cusolito, Roberto N. Fattal-Jaef, Fausto Pati˜ na, and Akshat V. Singh. (Anti)preneurial State and (Mis)Allocation. Working Paper, 2023. Joel M. David and Venky Venkateswaran. The Sources of Capital Misallocation. American Economic Review, 109(7):2531–2567, July 2019. URL https://ideas. repec.org/a/aea/aecrev/v109y2019i7p2531-67.html. Joel M. David, Lukas Schmid, and David Zeke. Risk-Adjusted Capital Allocation and Misallocation. Working Paper Series WP-2020-34, Federal Reserve Bank of Chicago, December 2020. URL https://ideas.repec.org/p/fip/fedhwp/ 92319.html. Joel M David, Venky Venkateswaran, Ana Paula Cusolito, and Tatiana Didier. Capital allocation in developing countries. The World Bank Economic Review, 35(4): 1102–1121, 2021. URL https://academic.oup.com/wber/article/35/4/1102/ 5902750. Ralph De Haas, Sergei Guriev, and Alexander Stepanov. State ownership and corpo- rate leverage around the world. 2022. Jan De Loecker, Pinelopi K Goldberg, Amit K Khandelwal, and Nina Pavcnik. Prices, markups, and trade reform. Econometrica, 84(2):445–510, 2016. URL https: //drive.google.com/file/d/1ZDP0o-ZasoKO7j0Ev0p7ioRctYAr9e24/view. 43 Klaus Deininger and Gershon Feder. Land institutions and land markets. In B. L. Gardner and G. C. Rausser, editors, Handbook of Agricultural Economics, volume 1 of Handbook of Agricultural Economics, chapter 6, pages 288–331. Elsevier, 00 2001. URL https://ideas.repec.org/h/eee/hagchp/1-06.html. Walter Erwin Diewert. Productivity measurement in the public sector. The Oxford Handbook of Productivity Analysis, 2018. David Dollar and Shang-Jin Wei. Das (wasted) kapital: Firm ownership and invest- ment efficiency in china. Working Paper 13103, National Bureau of Economic Research, May 2007. URL http://www.nber.org/papers/w13103. Ulrich Doraszelski and Jordi Jaumandreu. R&d and productivity: Estimating en- dogenous productivity. Review of economic studies, 80(4):1338–1383, 2013. URL https://academic.oup.com/restud/article/80/4/1338/1580686. Isaac Ehrlich, Georges Gallais-Hamonno, Zhiqiang Liu, and Randall Lutter. Productivity Growth and Firm Ownership: An Analytical and Empiri- cal Investigation. Journal of Political Economy, 102(5):1006–1038, October 1994a. doi: 10.1086/261962. URL https://ideas.repec.org/a/ucp/jpolec/ v102y1994i5p1006-38.html. Sanford B. Ehrlich, Alex F. De Noble, Tracy Moore, and Richard R. Weaver. After the cash arrives: A comparative study of venture capital and private investor involvement in entrepreneurial firms. Journal of Business Venturing, 9(1):67–82, 1994b. URL https://EconPapers.repec.org/RePEc:eee:jbvent:v:9:y:1994: i:1:p:67-82. European Comission. European chips act. 2023. Pablo Fajgelbaum, Gene M. Grossman, and Elhanan Helpman. A linder hypothesis for foreign direct investment. The Review of Economic Studies, 82(1 (290)):83– 121, 2015. ISSN 00346527, 1467937X. URL http://www.jstor.org/stable/ 43551465. Roberto N. Fattal-Jaef. Entry barriers, idiosyncratic distortions, and the firm size distribution. American Economic Journal: Macroeconomics, 14(2):416–68, April 2022. doi: 10.1257/mac.20200234. URL https://www.aeaweb.org/articles? id=10.1257/mac.20200234. Zhe Geng and Jun Pan. The soe premium and government support in china’s credit market. Technical report, National Bureau of Economic Research, 2023. Gita Gopinath, S ¨ ¸ ebnem Kalemli-Ozcan, Loukas Karabarbounis, and Carolina Villegas-Sanchez. Capital Allocation and Productivity in South Europe*. The Quarterly Journal of Economics, 132(4):1915–1967, 06 2017. ISSN 0033-5533. doi: 10.1093/qje/qjx024. URL https://doi.org/10.1093/qje/qjx024. 44 Nezih Guner, Gustavo Ventura, and Xu Yi. Macroeconomic Implications of Size- Dependent Policies. Review of Economic Dynamics, 11(4):721–744, October 2008. doi: 10.1016/j.red.2008.01.005. URL https://ideas.repec.org/a/red/issued/ 07-73.html. John Haltiwanger, Robert Kulick, and Chad Syverson. Misallocation measures: The distortion that ate the residual. Technical report, National Bureau of Economic Research, 2018. URL https://www.nber.org/system/files/working_papers/ w24199/w24199.pdf. Ann Harrison, Marshall Meyer, Peichun Wang, Linda Zhao, and Minyuan Zhao. Can a tiger change its stripes? reform of chinese state-owned enterprises in the penumbra of the state. Technical report, National Bureau of Economic Research, 2019. Hugo Hopenhayn and Richard Rogerson. Job Turnover and Policy Evaluation: A Gen- eral Equilibrium Analysis. Journal of Political Economy, 101(5):915–938, October 1993. doi: 10.1086/261909. URL https://ideas.repec.org/a/ucp/jpolec/ v101y1993i5p915-38.html. Chang-Tai Hsieh and Peter J. Klenow. Misallocation and Manufactur- ing TFP in China and India. The Quarterly Journal of Economics, 124(4):1403–1448, 2009. URL https://ideas.repec.org/a/oup/qjecon/ v124y2009i4p1403-1448..html. Chang-Tai Hsieh and Peter J. Klenow. The Life Cycle of Plants in India and Mexico. The Quarterly Journal of Economics, 129(3):1035–1084, 2014. URL https:// ideas.repec.org/a/oup/qjecon/v129y2014i3p1035-1084.html. Chang-Tai Hsieh and Enrico Moretti. Housing constraints and spatial misallocation. Working Paper 21154, National Bureau of Economic Research, May 2015. URL http://www.nber.org/papers/w21154. Yi Huang, Marco Pagano, and Ugo Panizza. Local crowding-out in china. The Journal of Finance, 75(6):2855–2898, 2020. doi: https://doi.org/10.1111/jofi.12966. URL https://onlinelibrary.wiley.com/doi/abs/10.1111/jofi.12966. International Monetary Fund. 2022 global debt monitor. Technical report, 2022. eka Juh´ R´ asz, Nathan J Lane, and Dani Rodrik. The new economics of industrial policy. Technical report, National Bureau of Economic Research, 2023. Emilia Jurzyk and Mr Cian Ruane. Resource misallocation among listed firms in China: The evolving role of state-owned enterprises. International Monetary Fund, 2021. Sebnem Kalemli-Ozcan, Bent Sorensen, Carolina Villegas-Sanchez, Vadym Volosovych, and Sevcan Yesiltas. How to Construct Nationally Representative 45 Firm Level Data from the Orbis Global Database: New Facts and Aggregate Im- plications. NBER Working Papers 21558, National Bureau of Economic Research, Inc, September 2015. URL https://ideas.repec.org/p/nbr/nberwo/21558. html. Sebnem Kalemli-Ozcan, Bent Sorensen, Carolina Villegas-Sanchez, Vadym Volosovych, and Sevcan Yesiltas. How to construct nationally representative firm level data from the orbis global database: New facts on SMEs and aggregate implications. American Economic Journal: Macroeconomics, Forthcoming, 2023. Jonathan Karpoff. The impact of shareholder activism on target companies: A survey of empirical findings. SSRN Electronic Journal, 08 2001. doi: 10.2139/ssrn.885365. Hiroyuki Kasahara, Mitsukuni Nishida, and Michio Suzuki. Decomposition of Ag- gregate Productivity Growth with Unobserved Heterogeneity. RIETI, 2017. URL https://ies.keio.ac.jp/upload/20170922appliedpaper-1.pdf. Jan Kmenta. On estimation of the ces production function. International Economic Review, 8(2):180–189, 1967. Pravin Krishna, Andrei A Levchenko, and William F Maloney. Growth and risk: A view from international trade. The World Bank, 2020. URL https://documents1.worldbank.org/curated/ar/699231592936066374/ pdf/Growth-and-Risk-A-View-from-International-Trade.pdf. Paul H. Malatesta and Kathryn L. DeWenter. State-Owned and Privately Owned Firms: An Empirical Analysis of Profitability, Leverage, and Labor Intensity. American Economic Review, 91(1):320–334, March 2001. URL https://ideas. repec.org/a/aea/aecrev/v91y2001i1p320-334.html. es Carmen. Productivity and resource mis- Busso Matias, Madrigal Lucia, and Pag´ allocation in Latin America. The B.E. Journal of Macroeconomics, 13(1):1–30, June 2013. doi: 10.1515/bejm-2012-0087. URL https://ideas.repec.org/a/ bpj/bejmac/v13y2013i1p30n3.html. Mariana Mazzucato. The entrepreneurial state must lead on climate change, 2022. URL https://www.project-syndicate.org/commentary/ entrepreneurial-state-only-solution-to-climate-change-by-mariana-mazzucato-2022-1 barrier=accesspaylog. David Mckenzie. Weekly links may 26: cash, national identity, learning, industrial policy, don’t follow your passions, and more. . . . 2023. William L. Megginson and Jeffry M. Netter. From state to market: A survey of empirical studies on privatization. Journal of Economic Literature, 39(2):321–389, June 2001. doi: 10.1257/jel.39.2.321. URL https://www.aeaweb.org/articles? id=10.1257/jel.39.2.321. 46 Virgiliu Midrigan and Daniel Yi Xu. Finance and misallocation: Evidence from plant-level data. American Economic Review, 104(2):422–58, February 2014. doi: 10.1257/aer.104.2.422. URL https://www.aeaweb.org/articles?id=10.1257/ aer.104.2.422. Ministry of Finance of Norway. Better Taxation – A TAx Reform for Transformation and Growth. Technical report, Government of Norway, 2015. URL https://www. regjeringen.no/en/dokumenter/meld.-st.-4-20152016/id2456324/. Mark Muro. Bidens big bet on place-based industrial policy. 2023. Dzhamilya Nigmatulina. Sanctions and misallocation. how sanctioned firms won and russia lost. How Sanctioned Firms Won and Russia Lost.(April 12, 2021), 2021. Dzhamilya Nigmatulina. Misallocation and state ownership: Evidence from the rus- sian sanctions. Technical report, mimeo, 2022. Oluremi Akin Olugbade, Gareth Anderson, Maria Atamanchuk, Iacovos Ioannou, Tannous Kass-Hanna, Wei Shi, and Joyce Wong. State-owned enterprises in middle east, north africa, and central asia: Size, costs, and challenges. Departmental Papers, 2021(019):A001, 2021. doi: 10.5089/9781513594088.087.A001. Nina Pavcnik. Trade Liberalization, Exit, and Productivity Improvements: Evidence from Chilean Plants. Review of Economic Studies, 69(1):245–276, 2002. URL https://ideas.repec.org/a/oup/restud/v69y2002i1p245-276.html. Diego Restuccia and Richard Rogerson. Policy Distortions and Aggregate Produc- tivity with Heterogeneous Plants. Review of Economic Dynamics, 11(4):707–720, October 2008. doi: 10.1016/j.red.2008.05.002. URL https://ideas.repec.org/ a/red/issued/07-48.html. Mr Ernesto Ramirez Rigo, Christine J Richmond, Oluremi Akin Olugbade, Gareth Anderson, Maria Atamanchuk, Mr Hatim Bukhari, Iacovos Ioannou, Deeksha Kale, Tannous Kass-Hanna, Mr Maximilien Queyranne, et al. State-Owned En- terprises in Middle East, North Africa, and Central Asia: Size, Costs, and Chal- lenges. International Monetary Fund, 2021. Dani Rodrik. Where are we in the economics of industrial policies? 2019. Dani Rodrik. An industrial policy for good jobs. Hamilton Project, Brookings Insti- tution, 2022. Andrei Shleifer. State versus private ownership. Journal of Economic Perspectives, 12(4):133–150, 1998. Reprinted in Liberal Thought, The Association for Liberal Thinking, Ankara, Turkey, 2001. Reprinted in Michael A. Crew and David Parker, eds., Developments in the Economics of Privatization and Regulation, Edward Elgar Publishing Inc., 2008. 47 Andrei Shleifer and Robert W. Vishny. Politicians and Firms. The Quarterly Journal of Economics, 109(4):995–1025, 1994. Zheng Song, Kjetil Storesletten, and Fabrizio Zilibotti. Growing Like China. American Economic Review, 101(1):196–233, February 2011. URL https://ideas.repec. org/a/aea/aecrev/v101y2011i1p196-233.html. The Economist. Welcome to the era of the bossy state, 2022. URL https://www.economist.com/leaders/2022/01/15/ welcome-to-the-era-of-the-bossy-state. Daniel Trefler. The long and short of the canada-u.s. free trade agreement. The American Economic Review, 94(4):870–895, 2004. ISSN 00028282. URL http: //www.jstor.org/stable/3592797. Shang-Jin Wei, Zhuan Xie, and Xiaobo Zhang. From ”made in china” to ”innovated in china”: Necessity, prospect, and challenges. Journal of Economic Perspectives, 31 (1):49–70, February 2017. doi: 10.1257/jep.31.1.49. URL https://www.aeaweb. org/articles?id=10.1257/jep.31.1.49. Toni M. Whited and Jake Zhao. The misallocation of finance. The Journal of Finance, 76(5):2359–2407, 2021. doi: https://doi.org/10.1111/jofi.13031. URL https:// onlinelibrary.wiley.com/doi/abs/10.1111/jofi.13031. World Bank. International debt report 2022: updated international debt statistics. The World Bank, 2022. World Bank. The business of the state. Technical report, World Bank, 2023. 48