Managers, Entrepreneurs, and the Allocation of Talent: Evidence from Hungary's Transition

Managers, Entrepreneurs, and the Allocation of Talent: Evidence from Hungary’s Transition∗ Mikl´os Koren†and Krisztina Orb´an‡ September 30, 2025 Abstract Management quality drives firm performance and aggregate productivity, yet the supply of managerial talent remains poorly understood. A key friction is that hired managers cannot fully appropriate the surplus they generate, unlike entrepreneurs who own their firms, creating a wedge between private and social returns to management. Here we develop a general equilibrium model to quantify how this corporate governance friction distorts talent allocation between entrepreneurship, management, and employment. Using the universe of Hungarian firms and CEOs (1986–2022), we exploit the transition to capitalism—when the count of enterprises increased from 21,000 to 115,000 in three years—to identify the parameters of the model. We find that managers capture only 60% of the surplus they create, resulting in too few professional managers and too many less-productive entrepreneurs. Eliminating this friction would raise GDP per worker by 4% through improved occupational composition. Uniform subsidies fail to correct the misallocation, raising GDP by only 0.1%. Our results show that management interventions’ aggregate effects depend critically on targeting the specific friction between hired managers and entrepreneurs rather than expanding the overall pool of business leaders. ∗We thank Jan Eeckhout, Hugo Hopenhayn, Chris Edmond, Mih´aly Laki, Jaume Ventura, Diego Restuccia, and seminar audiences at CEU, the ¨ Osterreichische Nationalbank, Monash University, University of Melbourne, the University of New South Wales Firms in the Macroeconomy workshop, Deakin University Macro Development workshop for comments and B´alint Szil´agyi, Melinda Tir, J´ulia Varga and Andr´as Vereckei for help with the data. This research has been supported by the Forefront Research Excellence Program of the National Research, Development and Innovation Office of Hungary (grant no. 144193) and by a European Research Council Advanced Grant (grant no. 101097789). Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union, the European Research Council or National Research, Development and Innovation Office. Neither the European Union nor the granting authority can be held responsible for them. †Central European University, HUN-REN KRTK, CEPR, and CESifo ‡Monash University 1 1 Introduction A large literature documents the importance of management for firm performance and aggregate productivity (Bloom et al., 2014). While firm-level management interventions can be effective (Bloom et al., 2013; Bruhn et al., 2010, 2018; Bloom et al., 2020; Giorcelli, 2019, 2021), less is known about the supply of management skills and its macroeconomic consequences. Who becomes a manager? How easily can policies influence this supply? Do policies that raise returns to management increase aggregate output, or merely reshuffle talent across occupations? We build a general equilibrium model that integrates occupational choice between entrepreneurship, management, and paid employment. Individuals differ in managerial skill and in an occupation-specific endowment (“entrepreneurial spirit”). The key friction is a corporate governance wedge: professional managers cannot capture the full operating surplus they generate, whereas entrepreneurs, who own their firms, can. Consistent with Jensen and Meckling (1976), we interpret this wedge as the shadow value of monitoring and incentive provision. This friction leads to underprovision of management skills: too few people become managers. The friction also creates misallocation: because there are too few managers, worker wages are depressed, making entrepreneurship more attractive. The equilibrium features too few managers and too many entrepreneurs relative to the first-best allocation. We use data on the universe of Hungarian firms and their CEOs between 1986 and 2022 (Koren et al., 2025) to calibrate the model and evaluate its mechanisms. We exploit Hungary’s transition from communism to capitalism as a natural experiment to identify key parameters. The transition created a sudden shock increasing demand for managers and entrepreneurs. As markets became liberalized, business enterprises exploded from 21,000 in 1989 to 115,000 in 1992. This increase put more than 160,000 people, many without prior management experience, in top managerial positions. This demand shock helps us trace out the supply curve.1For a sample of CEOs, we observe their earnings from linked employer-employee data based on social security records. Three empirical patterns emerge from the data. First, we find a novel empirical fact: a strong inverse relationship between the size of the cohort of newly entering individuals managing firms and average firm size. This relationship is consistent with heterogeneous manager quality and a downward sloping supply curve of managerial talent. The relationship is robust to controlling for demographic factors, age, and experience.2Second, entrepreneurs—CEOs who own their firms—are three times as numerous as hired man1Our work is motivated by case studies and interviews with business leaders during this period (Laki and Szalai, 2004, 2013). 2This fact is different from the fact in Sedl´aˇcek and Sterk (2017) and Moreira (2016) who find that the size of firms that enter during downturns stay permanently low compared to firms that enter during different times. 2 agers and run firms half as large. This is consistent with earlier work on the private benefits and low productivity of entrepreneurship (Evans and Leighton, 1989; Hamilton, 2000; Hurst and Pugsley, 2011). Third, annual gross compensation of full-time CEOs amounts to 10–20 percent of their firms’ operational performance (EBITDA), consistent with managers not capturing the full surplus they generate, though some of this gap may reflect unobserved firm-level factors. There are three key mechanisms in our model, each of which we can discipline using micro moments from the data. First, the extent of selection of managers by skill depends on how heterogeneous managers are in terms of quality. To aid tractability, we model manager skill with a Pareto distribution, with its tail parameter θdirectly calibrated to the degree of selection measured in the data. Our estimates reveal moderate selection: increasing the number of managers by 10 percent (holding their supply constant) reduces their average firm size by 0.4–1.4 percent, implying a tail parameter between 7 and 25. This moderate selection (1/θ) is combined with a high supply elasticity (θ), meaning many managers are near the occupational margin. Second, the relative productivity of entrepreneurs versus managers, together with the corporate governance friction, determines the strength of talent misallocation across occupations. We calibrate this relative productivity by exploiting within-person variation in firm size of CEOs who become entrepreneurs at some point in their careers. The within-person nature of this measurement is important because it allows us to control for management quality. Using this variation we find that entrepreneurial firms are about 44 percent smaller than managerial firms (that is, about 56 percent of the size). Third, the earnings data allow us to discipline the corporate governance wedge, which determines how much of the firm’s surplus a manager captures. With different estimates ranging from 0.2 to 0.6 to calibrate the corporate governance friction, we pick a conservative value of 0.6, that is, managers capture 60 percent of the surplus they generate for the firm. We conduct a number of policy counterfactuals in our calibrated model. First, we compute what the first-best allocation would be without the corporate governance friction. In a simpler version of the model, with only managers and workers (the “M-economy”), GDP per worker would be about 1.5 percent higher. We then study the full model with both entrepreneurs and managers (the “E–M economy”). Here, the first-best would raise GDP per worker by about 4 percent. The effect is larger than in the M-economy, because the subsidy that induces choices as if the corporate governance friction did not exist would also correct the misallocation of talent across occupations by simultaneously increasing the number of (more productive) managers and reducing the number of (less productive) entrepreneurs. These effects require large interventions. Managers need subsidies to reach their socially optimal role because the corporate governance friction reduces private returns relative 3 to social returns.3The first-best requires a 67 percent subsidy to manager earnings, offsetting the corporate governance friction. This would increase the share of managers from 10 to 15 percent of the workforce in the M-economy and from 3 to 15 percent in the E–M economy. (In the baseline E–M economy, three quarters of “managers” are entrepreneurs, as in the data.) We consider more modest policies. A 10 percent subsidy to manager earnings raises GDP per worker by 0.5 percent in the M-economy and 1.6 percent in the E–M economy. This subsidy reduces the total number of managers in the E–M economy, as more entrepreneurs exit than managers enter. The GDP elasticity with respect to the subsidy differs markedly: 0.05 in the M-economy versus 0.16 in the E–M economy. The E–M economy responds more elastically because the subsidy removes unproductive entrepreneurs from the market, reducing product market competition and increasing profit per worker for the remaining firms. This reallocation from low-productivity entrepreneurial firms to high-productivity professional management amplifies the subsidy’s effects. In practice, setting different subsidies for managers and entrepreneurs is difficult because “entrepreneurial spirit” is unobservable and ownership patterns are easily manipulated. We therefore consider a uniform subsidy to all business leaders. The second-best uniform subsidy that maximizes GDP is 17 percent. Even this large subsidy raises GDP by only 0.1 percent in the E–M economy because it incentivizes both occupations to enter, failing to correct the composition distortion. Finally, we consider an education-like intervention that raises managerial skill by 10 percent. This increases output by 1.6 percent in both economies without requiring reallocation across occupations. While our model is silent about specific trainings and costs, this exercise suggests such interventions can raise aggregate productivity without disrupting labor and manager markets. Our results highlight that macroeconomic elasticities of management interventions depend crucially on the economic environment and can be very different from the microeconomic elasticities. Our maintained assumption is that managers differ in innate manager skills. Recent evidence confirms that CEOs differ systematically in their effects on firm performance. This has been established using CEO fixed effects as they move across firms (Bertrand and Schoar, 2003; Cust´odio et al., 2013; Quigley and Hambrick, 2015; Schoar and Zuo, 2016), 3In our framework, managers earn “too little,” not “too much,” as often suggested in the executive compensation debate (Frydman and Saks, 2010). The settings differ: we study small and medium-sized private businesses in an emerging economy, whereas the executive compensation literature focuses on large publicly listed firms in industrial economies. 4 CEO characteristics (Graham et al., 2012), and quasi-experimental variation from CEO deaths (Fee et al., 2013; Becker and Hvide, 2022; Sauvagnat and Schivardi, forthcoming) and sickness (Bennedsen et al., 2020). General management skills have become increasingly important relative to firm-specific knowledge in recent decades (Frydman and Saks, 2010; Cust´odio et al., 2013). Our contribution is to build an equilibrium framework that integrates occupational choice between entrepreneurship, management, and paid employment. Prior work has studied either professional CEOs of large corporations or entrepreneurs running small businesses, but not their interaction. For example, Bloom et al. (2014); Akcigit et al. (2021) study management quality and organization in large corporations, while Lucas (1978); Banerjee and Newman (1993) focus on entrepreneurial talent and selection into small business ownership. In our model, these two groups face different incentives: managers cannot capture all the surplus they generate for the firm, while entrepreneurs can. This leads to an underprovision of manager skills in the aggregate economy, but also a misallocation as too many people become entrepreneurs, drawn in by depressed wages. Following Engbom et al. (2024), we show that this distinction between managers and entrepreneurs is crucial for understanding the rise of managerial capitalism. Most existing models with managerial skills or organizational capital are set up at the aggregate level (Burstein and MongeNaranjo, 2009; Gennaioli et al., 2012; Akcigit et al., 2021; Hjort et al., n.d.), with no role for such heterogeneity and selection. Our approach with micro data is similar to Akcigit et al. (2020), who emphasize selection and equilibrium response of innovators to policy changes. The model has implications for large economic transitions. Fuchs-Sch¨undeln and Masella (2016); Fuchs-Sch¨undeln and Sch¨undeln (2020) explore occupation choice after German reunification. We study a very large shock with business activity increasing 20-fold. Recent evidence from repeated large-scale management development programs in the US during WWII (Giorcelli, 2023) and management training in Italy (Bianchi and Giorcelli, 2022) shows that such interventions can have lasting effects on manager careers and firm outcomes. Our results can be informative about the general equilibrium implications of such programs. 2 An equilibrium model of managers We build a general equilibrium model to study managers and entrepreneurs. The key friction is that managers, unlike entrepreneurs, cannot capture all rents from their skills. This results in underprovision of management skills and opens the door for policy interventions. We study the steady-state equilibrium and show how policy can achieve the first-best allocation. 5 Individuals choose whether to become a manager, an entrepreneur, or a worker. Workers are identical, but managers and entrepreneurs differ in “management skill,” an input to production. Higher returns to management skills induce more people to choose managerial careers. Policy influences this occupational choice through taxes and subsidies. 2.1 The M Economy We first study an economy with only managers and workers (the M-economy) as a benchmark. We then extend the model to include entrepreneurs (the E-M economy). There is a mass Lof individuals, who choose to be either a worker or a manager. Each individual is characterized by an innate manager skill z∈R+. The distribution of skills in the population is Pareto with lower bound Λ(1 −1/θ) and shape parameter θ > 1, so that Pr(z > x)=Λθ(1 −1/θ)θx−θ. Note that E(z) = Λ. A manager with skill zcan hire hworkers to produce output with the production function q= (Az)νh1−ν.(1) Here Ais a productivity shifter reflecting technologies, institutions, or market conditions affecting all firms. Manager skill zenters multiplicatively with A: better managers raise productivity, and this effect is amplified in high-Aenvironments. Better managers organize production, motivate workers, implement technologies, and have larger spans of control (Lucas, 1978). The parameter ν∈(0,1) captures returns to management. The ν fraction of revenue represents the joint contribution of Aand skill z; since Ais exogenous, managers compete to capture rents from their skill. Workers are hired in competitive labor markets. Their value marginal product is equal to their wage, (1 −ν)(Az/h)ν=w, and, without loss of generality, we have normalized the price of output to one. This pins down the employment of a firm with manager zas h(z) = Azw−1/ν(1 −ν)1/ν, so that firm revenue is r(z) = q(z) = Azw1−1/ν(1 −ν)1/ν−1. Employment, wage bill, and revenue are all linear in Aand z. Workers receive a 1 −νfraction of revenue. The remaining fraction goes to owners who bid for managers competitively. Manager wages cannot exceed a ϕ < 1 fraction of 6 operating surplus. Leftover rents go to firm owners. Firms are owned by a mutual fund that redistributes profits to all workers.4This ensures GDP equals production and ϕ represents an allocation constraint rather than wealth destruction. If ϕ= 1, managers capture full operating surplus without friction. The corporate governance friction ϕ < 1 captures agency problems, monitoring costs, and limited commitment in the manager-owner relationship. In equilibrium, shareholders cannot write enforceable contracts that commit all surplus to hired managers, reflecting hold-up problems and asymmetric information about managerial effort. Formally, the wage of a manager with skill level zis ω(z). Profit maximization of firms involves max zr(z)−wh(z)−ω(z) subject to ω(z)≤ϕ[r(z)−wh(z)]. Bertrand competition among owners for managers drives the wage to the contractual upper bound, so the constraint binds: ω(z) = ϕνr(z) = ϕνAzw1−1/ν(1 −ν)1/ν−1. Definition 1. The equilibrium of the M-economy is (i) a worker wage rate w, (ii) a manager wage rate ω(z), (iii) employment function h(z), and (iv) a measure of management skills µ(z) for those active in the economy such that 1. firms maximize profit subject to the corporate governance constraint 2. labor market clears Zz [1 + h(z)]µ(z)dz =L 3. market for managers clears at each skill level z, µ(z)≤θΛθz−θ−1L 4. individuals choose occupations to maximize their income. At each skill level z, all individuals become managers if the manager wage exceeds the worker wage (ω(z)> w); some or all individuals become workers if work pays weakly better (ω(z)≤w). Firms maximize profits given governance constraints (1); labor market clears (2); the measure of active managers cannot exceed the potential supply from the skill distribution (3); and individuals choose the higher-paying occupation (4). Together, these conditions ensure that the equilibrium respects both resource constraints and individual incentives. 4This technical assumption abstracts from capital market frictions. 7 As the proposition shows, optimal occupational choice combined with the governance friction generates a threshold property: there exists a cutoff zmin such that higher-skilled individuals become managers while lower-skilled individuals become workers. Define n:= Z∞ Λ(1−1/θ) µ(z)dz/L as the fraction of managers in the workforce and ˜z:= Z∞ Λ(1−1/θ) zµ(z)dz/ Z∞ Λ(1−1/θ) µ(z)dz their average skill. Proposition 1. An equilibrium of the M-economy exists, is unique, and exhibits a threshold property: there is a threshold zmin such that all individuals with z > zmin become managers, while all individuals with z < zmin become workers. The equilibrium variables are characterized by the following equations: n=ϕ ϕ+κ(2) zmin = Λ(1 −1/θ)n−1/θ (3) ˜z= Λn−1/θ (4) h(z) = z1 ϕΛ 1−ν νθ n1/θ (5) w= ΛνϕνAννν θ(1 −ν)1−νn−ν/θ (6) ω(z) = θ θ−1zΛν−1ϕνAννν θ(1 −ν)1−νn(1−ν)/θ (7) with νθ:= ν(1 −1/θ)∈(0,1) the “selection adjusted” returns to management and κ= (1 −ν)/νθ>0 the relative importance of labor in production. The proof of this and all subsequent propositions can be found in the Appendix. The threshold property follows from single crossing: since zis valuable only for managers, the relative wage ω(z)/w is strictly increasing in z. The threshold zmin is determined by free entry and labor market clearing. The equilibrium share of managers ndepends only on ϕ,νand θ—not on Aor Λ—due to free entry. Once we know n, all other variables follow. The threshold zmin decreases in nand increases in Λ. Average skill ˜zis a constant multiple of the entry threshold by Pareto properties. Employment h(z) increases in both zand n. While more managers (n↑) leading to larger firms (h(z)↑) seems counterintuitive, this reflects general equilibrium: as nrises, the entry threshold falls and marginal managers have lower skill. To maintain indifference, worker wages fall, making labor cheaper for all managers. Hence h(z) rises for any z > zmin. 8 Aggregate employment (1 −n)Lstill falls because the composition effect (fewer workers) dominates the scale effect (larger firms). Proposition 2. GDP per worker in the M-economy is y:= Y L=AνΛνnνθ(1 −n)1−ν=AνΛνϕνθκ1−ν(ϕ+κ)−(1−ν/θ).(8) Productivity Aand skill Λ directly affect GDP per worker, though the skill effect is mitigated by ν < 1. The manager share nenters GDP with exponent νθ=ν(1 −1/θ) rather than νdue to selection: as nincreases, marginal managers have lower skill. The Pareto parameter θgoverns how quickly average quality declines with quantity. With stronger selection (smaller θ), expanding managers yields smaller GDP gains. The rent share ϕ reduces managers and total skill, reducing GDP. This effect is larger when management is more important (larger ν) and managers are more similar (smaller θ). While more people work in production when ϕis low, this is dominated by reduced managerial talent. Definition 2. The first best allocation of the M-economy is (i) an employment function h(z), and (ii) a measure of management skills µ(z) for those active in the economy such that 1. GDP is maximal 2. labor market clears Zz [1 + h(z)]µ(z)dz =L 3. market for managers clears at each skill level z, µ(z)≤θΛθz−θ−1L. Proposition 3. Afirst best allocation of the M-economy exists, is unique, and exhibits a threshold property: there is a threshold zmin ∗such that all individuals with z > zmin ∗ become managers, while all individuals with z < zmin ∗become workers. The solution is characterized by the following equations: n∗=1 1 + κ(9) zmin ∗= Λ(1 −1/θ)n−1/θ ∗(10) ˜z∗= Λn−1/θ ∗(11) h∗(z) = z1 Λ 1−ν νθ n1/θ ∗(12) w∗= ΛνAννν θ(1 −ν)1−νn−ν/θ ∗(13) ω∗(z) = θ θ−1zΛν−1Aννν θ(1 −ν)1−νn(1−ν)/θ ∗(14) The first-best allocation maximizes GDP subject to resource constraints, ignoring the 9 3 Calibration 3.1 Data Manager data comes from the Hungarian Manager Database (HUN-REN KRTK, 2024a), compiled from the C´egjegyz´ek (corporate registry). This records, for all corporations, officers as specified in corporate law, including name, mother’s name, address, position, and exact dates (Koren et al., 2025). We use name-based matching to identify managers moving across firms and infer nationality. We create an annual panel of CEOs by taking a snapshot of main directors on June 21 each year. We keep only chief executive officers with Hungarian names (Koren and Telegdy, 2023) to study the domestic labor market. For CEOs at multiple firms, we keep only the largest. We merge balance sheets and financial statements (HUN-REN KRTK, 2024b) with nearly universal coverage. Table 1: Manager Entry by Cohort Cohort Non-Entrepreneur Entrepreneur Total 1990 58,725 155,988 214,713 1995 59,058 156,117 215,175 2000 64,990 148,526 213,516 2005 66,804 118,981 185,785 2010 74,485 106,447 180,932 2015 56,202 54,840 111,042 2020 36,573 44,647 81,220 Total 427,234 785,546 1,212,780 Notes: This table reports the number of managers entering each cohort by entrepreneurship status. Cohorts are defined by the first year a manager appears in the data, grouped in 5-year bins. Entrepreneurs are defined as CEOs who are also founders of the firm they manage. The sample is restricted to Hungarian managers with non-missing revenue data and at most 4 simultaneous positions. Table ?? shows the breakdown of manager entries by cohort and entrepreneur status. We track 147,000 CEOs and 206,000 entrepreneurs over time. Across all cohorts, but especially in earlier cohorts, entrepreneurs are overrepresented at about 60–70% of all CEOs. 3.2 Overview of Calibration Strategy Our calibration proceeds in three steps. First, we identify the rent-sharing parameter ϕusing firm size differences, profit rates, and CEO wage shares. Second, we estimate selection intensity θfrom cohort variation during Hungary’s transition. Third, we pin down relative productivity AM/AEusing within-person variation. 16 These three parameters—θ,ϕ, and AM/AE—determine allocation between entrepreneurs and managers and aggregate productivity. The selection parameter θgoverns entry response to policy, ϕmeasures the governance friction, and AM/AEcaptures the efficiency gap between organizational forms. We also calibrate ν(output elasticity with respect to management) and α(entrepreneurial spirit) directly from data moments. 3.3 Identifying the Rent-sharing Parameter ϕ We employ three complementary approaches to estimate ϕ, the fraction of rents that managers can capture. Each method provides an independent moment to discipline this key parameter. Method 1: Firm Size Differences The model delivers a sharp prediction for the relative firm sizes of entrepreneurs and managers. In equilibrium, the ratio of average firm sizes is: ˜ hE ˜ hM =ϕ. (39) Managers can only capture fraction ϕof rents, requiring them to run proportionally larger firms to achieve the same return as entrepreneurs. Since both occupations face the same outside wage w, the marginal manager must run a firm 1/ϕ times larger than the marginal entrepreneur’s to be indifferent. Method 2: Profit Rate Comparison An alternative calibration strategy measures the profit rate of entrepreneur-run firms relative to manager-run firms. Under the assumption that entrepreneurs pay themselves only the minimum wage required for tax compliance, the profit rate of entrepreneur-run firms reveals the full rent ν. In contrast, manager-run firms report a share ϕν of revenue as manager wages in the wagebill and only a share (1 −ϕ)νas profit. We find that the median profit rate of entrepreneur-run firms is 17.2% (once corrected for the minimum tax burden of the CEO), while that of manager-run firms is only 13.8%, implying 1 −ϕ= 0.138/0.172 = 0.8, so ϕ= 0.2. Method 3: CEO Wage Shares A third estimate of ϕcomes from direct measurement of CEO compensation. In our model, νshare of revenue is retained as operating surplus (EBITDA), with managers capturing ϕν as wage while the remainder goes to firm owners. We use linked employer-employee data based on social security records, which cover a 50% random sample of all Hungarian employees.6We identify CEOs using occupation codes 6The linked administrative data are the property of the National Health Insurance Fund, Hungarian State Treasury, National Tax and Customs Administration, Ministry of Innovation and Technology, and the Educational Authority (and their legal successors). The data were processed and harmonized by the HUN-REN KRTK Databank. Tax return data originate from the National Tax and Customs Administration and were harmonized by the HUN-REN KRTK Databank (HUN-REN KRTK Databank, 2024; Seb˝ok, 2019). 17 and match them to our CEO panel from the corporate registry, restricting to full-time CEOs with no other jobs and at least twice the minimum wage. The detailed results appear in Appendix Table 8. The median CEO wage share is approximately 10% each year, while the mean is around 22%, implying that ϕlies in the 10–22% range. Reconciling the Estimates The three methods yield a range of estimates: size differences suggest ϕ≈0.60, profit rates imply ϕ≈0.20, and CEO wage shares indicate ϕ∈[0.10,0.22]. We adopt ϕ= 0.60 as our baseline, representing a conservative choice that avoids overstating governance frictions. This higher value is consistent with the firm size evidence and may reflect additional compensation channels not captured in wage data alone. 3.4 Identifying the Selection Parameter θ The selection parameter θgoverns how strongly average quality responds to entry. To identify it, we exploit the relationship between cohort entry rates and average firm size. The model predicts that average firm size for entrepreneurs follows: ˜ he= Λ(Ne/αe)−1/θL1/θAew−1/ν(1 −ν)1/ν, Taking logs yields our estimating equation: ln ˜ he= ln Λ −1 θln Ne+1 θln αe+1 θln L+ ln Ae−1 νln w+1 νln(1 −ν).(40) The coefficient on ln Neidentifies −1/θ if we can induce variation in manager entry that is orthogonal to other determinants like supply αeor productivity Ae. Following Sedl´aˇcek and Sterk (2017) and Moreira (2016), we exploit variation in cohort sizes when managers first became CEO. The Hungarian context provides particularly compelling variation: the transition to capitalism in 1990 led to a sudden, demand-driven increase in the number of firms and managers. By comparing cohorts entering in different years while controlling for competition and productivity measures, we can identify the selection effect. We implement this strategy by regressing log firm revenue on the log number of entrants in each cohort, controlling for manager and firm characteristics. The coefficient on log entry rate equals −1/θ, providing our identification of the selection parameter. Table 2 presents our main results across five specifications. The dependent variable is log firm revenue, and we control for CEO age, entrepreneur status, and firm age, with industry-year fixed effects throughout. The industry-year fixed effects absorb variation in worker wages wand competitive conditions in product markets, which are common to all cohorts within an industry-year. Because individual cohorts are small relative to the total labor force, their entry has negligible impact on aggregate wages, ensuring that the 18 identifying variation in ln Neis orthogonal to ln w. The baseline specification (Column 1) includes industry-year and demographic group fixed effects, the latter being an interaction of birth cohorts and gender. The coefficient on log entry rate is −0.139, implying θ= 7.19. This suggests moderate selection: when cohort entry doubles, average firm size falls by 13 percent. Columns 2-4 test robustness across different samples. Excluding entrepreneurs (Column 2) yields a weaker selection effect (θ= 9.26). Restricting to post-transition entrants (Column 3) or post-EU accession data (Column 4) shows weaker selection effects (θ= 11 −18), suggesting selection was strongest during the rapid expansion of the 1990s. Column 5 adds firm fixed effects, identifying the selection parameter from manager turnover within the same firms. The estimated θ= 24.39 indicates weaker selection when comparing managers across time within firms, but still confirms the negative relationship between entry and average quality. For our baseline calibration, we use θ= 11.0 from Column 3, which focuses on the post-transition period most relevant for our policy analysis. 3.5 Identifying Relative Productivity AM/AE The entrepreneur productivity penalty observed across all specifications provides initial evidence that entrepreneurs run smaller firms than managers. However, this size difference conflates two distinct mechanisms: selection (entrepreneurs may have different skills) and productivity (entrepreneurs may be less productive per unit of skill). To separate these effects, we need within-person variation. Table 3 exploits variation in the entrepreneur status of CEOs to identify key model parameters. The identification strategy relies on comparing the same individuals when they run their own firms (entrepreneurs) versus when they manage others’ firms (professional managers). Column 1 includes no firm or person controls, measuring the average size difference between entrepreneur-managed and non-entrepreneur-managed firms. We interpret the former as entrepreneurial (E) firms and the latter as managerial (M) firms. The entrepreneur coefficient of −0.46 log points in this specification identifies ln ϕ, as it captures the equilibrium size ratio predicted by the model. Column 3 adds person fixed effects, fundamentally changing what we identify. By controlling for individual skill z, selection no longer affects the estimates. The entrepreneur dummy now measures the size difference between E and M firms for the same manager skill level, effectively identifying ln(AE/AM). The coefficient of −0.58 log points implies AE/AM= exp(−0.58) = 0.56, or equivalently AM/AE= 1.78. 19 Table 2: Manager Selection and Cohort Entry Effects Dependent Variable: Log Revenue (1) (2) (3) (4) (5) Baseline No Founders Post-1992 Post-2004 Firm FE (ln n) Number of CEO Entrants, log -0.137∗∗∗ -0.104∗∗∗ -0.091∗∗∗ -0.055∗∗∗ -0.039∗∗∗ (0.019) (0.019) (0.015) (0.017) (0.003) CEO Age 0.126∗∗∗ 0.162∗∗∗ 0.112∗∗∗ 0.124∗∗∗ 0.087∗∗∗ (0.012) (0.011) (0.010) (0.010) (0.003) CEO Age Squared -0.001∗∗∗ -0.002∗∗∗ -0.001∗∗∗ -0.001∗∗∗ -0.001∗∗∗ (0.000) (0.000) (0.000) (0.000) (0.000) Entrepreneur CEO -0.415∗∗∗ 0.000 -0.412∗∗∗ -0.362∗∗∗ -0.131∗∗∗ (0.037) (.) (0.036) (0.034) (0.008) Firm Age 0.081∗∗∗ 0.024∗∗∗ 0.076∗∗∗ 0.055∗∗∗ 0.076∗∗∗ (0.005) (0.005) (0.003) (0.006) (0.004) Firm Age Squared -0.003∗∗∗ -0.000∗∗ -0.003∗∗∗ -0.002∗∗∗ -0.003∗∗∗ (0.000) (0.000) (0.000) (0.000) (0.000) Constant 7.686∗∗∗ 6.994∗∗∗ 7.679∗∗∗ 7.320∗∗∗ 7.661∗∗∗ (0.455) (0.425) (0.481) (0.441) (0.101) Fixed Effects: Industry ×Year Yes Yes Yes Yes Yes Birth Cohort ×Gender Yes Yes Yes Yes Yes Firm No No No No Yes Selection Parameter: θ7.30*** 9.62*** 10.98*** 18.22*** 25.32*** (1.01) (1.72) (1.76) (5.60) (2.16) Observations 7,087,435 1,649,978 6,756,480 5,065,112 7,003,390 Adjusted R-squared 0.173 0.152 0.175 0.150 0.753 Notes: This table reports results from regressions of log firm revenue on cohort entry characteristics for Hungarian CEOs, 1992-2022. The key variable of interest is Log Entry Rate (ln n), which measures the log number of managers entering in each cohort. Demographic groups are defined by birth cohort (5-year bins) interacted with gender, capturing systematic differences in baseline skills and demographics across manager cohorts. Column (1) shows the baseline specification with industry-year and demographic group fixed effects. Column (2) excludes entrepreneurs from CEOs. Column (3) restricts to post-transition entrants (first year ≥1992). Column (4) uses only post-EU accession data (2004 onwards). Column (5) adds firm fixed effects. The selection parameter θis computed as θ=−1/βln nusing the delta method for standard errors. Sample restricted to Hungarian managers aged 18-75 with non-missing revenue data and at most 4 simultaneous positions. Standard errors clustered by first entry year in parentheses. Significance levels: * p <0.10, ** p <0.05, *** p <0.01. Data source: Hungarian Manager Database (CEU MicroData) merged with firm financial statements, 1992-2022. 20 Table 3: Entrepreneur Discount Dependent Variable: Log Revenue (1) (2) (3) No Controls With Controls CEO FE Entrepreneur CEO -0.460∗∗∗ -0.430∗∗∗ -0.577∗∗∗ (0.005) (0.005) (0.008) CEO Age 0.112∗∗∗ 0.176∗∗∗ (0.002) (0.003) CEO Age Squared -0.001∗∗∗ -0.002∗∗∗ (0.000) (0.000) Firm Age 0.082∗∗∗ 0.047∗∗∗ (0.001) (0.001) Firm Age Squared -0.003∗∗∗ -0.002∗∗∗ (0.000) (0.000) Constant 9.749∗∗∗ 6.861∗∗∗ 5.066∗∗∗ (0.005) (0.072) (0.110) Fixed Effects: Industry ×Year Yes Yes Yes Birth Cohort ×Gender No Yes No CEO No No Yes Observations 7,087,436 7,087,435 6,967,913 Adjusted R-squared 0.146 0.170 0.713 Notes: This table reports results from regressions of log firm revenue on an entrepreneur CEO indicator for Hungarian CEOs, 1992-2022. Column (1) shows the baseline specification with industry-year fixed effects. Column (2) adds controls for CEO age and firm age as well as skill group fixed effects defined by birth cohort (5-year bins) interacted with gender, capturing systematic differences in baseline skills and demographics across manager cohorts. Column (3) adds CEO fixed effects, so the entrepreneur coefficient is identified off CEOs who switch between entrepreneur and non-entrepreneur roles. The sample is restricted to Hungarian managers aged 18-75 with non-missing revenue data and at most 4 simultaneous positions. Standard errors clustered by manager in parentheses. Significance levels: * p <0.10, ** p <0.05, *** p <0.01. Data source: Hungarian Manager Database (CEU MicroData) merged with firm financial statements, 1992-2022. 21 The similarity between estimates (−0.46 vs −0.58 log points) has economic significance. It suggests individuals are roughly indifferent between careers when ϕAM≈AE: managers receive ϕAMzwhile entrepreneurs receive AEz. This near-indifference validates our occupational choice framework. Within-person comparison is crucial—without it, we cannot distinguish whether entrepreneurs run smaller firms due to lower skill (selection) or lower productivity per unit of skill. 3.6 Direct Calibration of νand α The remaining parameters can be directly calibrated from data moments. We calibrate α, the fraction of the population with entrepreneurial spirit, to target the share of entrepreneurs among CEOs. From Table 1, we observe that 60-70% of CEOs are entrepreneurs. Given our estimates of ϕ,θ, and AM/AE, we choose α= 0.86 to match a baseline entrepreneur share of 75% in the model.7 The elasticity of output with respect to management, ν, is calibrated to match the share of CEOs in the workforce. The average firm size in the economy is: ˜ h=NE NE+NM ˜ hE+NM NE+NM ˜ hM=θ θ−1 1−ν ν αAθ E+ (1 −α)Aθ Mϕθ−1 αAθ E+ (1 −α)Aθ Mϕθ, which pins down ν= 0.16 given our estimates of the other parameters. 3.7 Summary of Calibrated Parameters 4 Counterfactual Policies This section uses the calibrated model to quantify how alternative policies affect the allocation between entrepreneurs and hired managers, skill composition, and aggregate output. 4.1 Understanding the Policy Response Magnitudes The policy responses in our calibrated economy reflect two key features of our parameter estimates: moderate selection (1/θ= 1/11 ≈0.09) and substantial governance frictions (ϕ= 0.6). The moderate selection means that quality declines only gradually as more people become managers—when entry increases by 10 percent, average firm size falls by only 0.9 percent. At the same time, the high supply elasticity (θ= 11) means that many high-ability individuals are close to the occupational margin and will switch occupations 7The model-predicted 75% entrepreneur share is slightly above the 60-70% range observed in the data, reflecting the conservative choice of ϕ= 0.6. Alternative calibrations with lower ϕvalues (0.2-0.4) would match the observed range more closely but would imply stronger governance frictions and larger policy responses. Our baseline choice represents a conservative approach that avoids overstating the potential gains from policy intervention. 22 Table 4: Model Parameter Calibration Parameter Description Target Moment Preferred Value ϕManager rent share Entrepreneur firm size 0.60 Table 3, Col (1) Alternative: Profit rates (0.20) Alternative: CEO wages (0.10–0.22) θSelection parameter Cohort selection elasticity 11.0 Table 2, Col (3) AM/AERelative productivity Within-person entrepreneur discount 1.78 Table 3, Col (3) νManagement elasticity CEO share in workforce 0.16 Average firm size equation αEntrepreneurial spirit Share of entrepreneurs among CEOs 0.86 Table 1 Notes: This table summarizes the calibrated parameter values. Each parameter is identified from specific empirical moments. For ϕ, we present three independent estimation methods (firm size differences, profit rates, CEO wage shares) with our preferred estimate based on firm size evidence to avoid overstating governance frictions. The selection parameter θis identified from cohort variation during Hungary’s transition. The relative productivity AM/AEuses within-person variation to separate selection from productivity effects. The parameters νand αare directly calibrated to match data moments. in response to moderate policy changes. The governance friction (ϕ < 1) creates a wedge between social and private returns to management, leaving room for welfare-improving interventions. The interaction between entrepreneurs and managers amplifies policy effects through a competition mechanism. When we subsidize managers, not only do more people become managers (direct effect), but the increased competition in the goods market also reduces entrepreneur profitability, causing additional reallocation from less productive entrepreneurs to more productive managers (indirect effect). This explains why the E-M economy shows larger policy responses than the M economy. We report steady-state outcomes normalized to the baseline calibration (GDP per capita = 1) and track the number of managers n, average skill ˜z, and wage outcomes for workers and managers. We first build intuition in the manager-only (M) economy, then analyze the economy with both entrepreneurs and hired managers (E–M). Policy levers include a manager subsidy sthat shifts the occupational margin, a skill intervention that raises the scale parameter Λ by 10% (“training”), and—in the E–M economy—uniform subsidies to both occupations alongside the second-best uniform rate that maximizes GDP. We also report a first-best allocation as a benchmark and later decompose GDP gains into expansion of the overall manager pool versus reallocation between E and M. 23 Table 5: Counterfactual Policies in the M-Economy Baseline First best 10% subsidy Training Manager subsidy — 0.667 0.100 — s GDP per capita 1.000 1.015 1.005 1.016 (baseline = 1) Managers 0.097 0.151 0.105 0.097 n Avg. skill 1.000 0.960 0.992 1.100 ˜z Worker wage 1.000 1.080 1.014 1.016 w, (baseline = 1) Manager wage 1.000 1.125 1.022 1.016 ω(˜z)/z, (baseline = 1) Notes: This table presents counterfactual policy analysis using the calibrated model. Baseline: baseline calibration values. First best: first-best allocation. 10% subsidy: 10% subsidy to manager earnings. Training: business training that increases the skill parameter Λ by 10%. GDP per capita: Normalized to 1 in the baseline calibration. Skill measures: Average manager skill, normalized to 1 in baseline. Wage premiums: Ratio of occupation-specific average compensation to worker wages. 24 4.2 Manager-Only Economy Table 5 presents counterfactuals for the manager-only economy. An important caveat: because all parameters are calibrated in the E-M economy, and we adopt conservative values for both rent-sharing (ϕ≈0.6, keeping us near first best) and relative productivity (AM/AEchosen so entrepreneur-run firms are not much worse than manager-run firms), the M-economy baseline does not exactly match the data on the number of managers. In particular, there are fewer managers in the M economy (n= 0.097) than in the data (n= 0.13), because in reality, many CEOs are entrepreneurs. Implementing the first-best allocation (here, an s≈0.667 manager subsidy) expands the manager share from 0.097 to 0.151 while average manager skill falls by about 4 percent to 0.960 due to selection. General-equilibrium forces keep the entry elasticity below one: as nrises, 1 −nshrinks, worker wages increase (to 1.080), and the return to managerial skill compresses, dampening additional entry. The aggregate payoff is modest—GDP rises by 1.5 percent—even as both worker and manager wages increase, implying higher wage inequality at the first best. Policy magnitudes matter. A realistic 10 percent subsidy barely moves aggregates: n increases to 0.105, average skill slips to 0.992, and GDP rises by only 0.5 percent; worker and manager wages rise roughly in tandem (1.014 and 1.022). By contrast, a 10 percent rightward shift in the managerial skill distribution (training that raises Λ) delivers a larger 1.6 percent GDP gain without distorting occupational choice: headcounts are unchanged, and both wages move one-for-one (1.016 and 1.016), leaving the manager premium unchanged. While we do not model specific training programs, our result that a skill increase can raise productivity without large reallocations or changes in inequality is consistent with empirical evidence from Italy and the United States that management-oriented training can work at a large scale (Giorcelli, 2019; Bianchi and Giorcelli, 2022). 4.3 Economy with Entrepreneurs and Managers Table 6 shows that introducing entrepreneurs dramatically changes how the economy responds to policy. The baseline calibration is set to match the observed composition: nE≈0.097, nM≈0.032. The first-best manager subsidy (sM≈0.667) triggers a large reallocation: hired managers expand to 0.150 while entrepreneurs nearly disappear (0.002). This dramatic shift occurs because entrepreneurs and managers compete in the goods market. The almost fivefold increase in the number of managers reduces the profitability of entrepreneurs by 27 percent. Combined with worker wages rising by 6 percent, this competition effect drives most entrepreneurs to exit. Due to selection, the average skill of the remaining entrepreneurs 25 Table 8: CEO Wage Share of Adjusted EBITDA, 2013–2017 Year Number of CEOs Median wage share (%) Mean wage share (%) 2013 7,974 10.2% 23.2% 2014 7,751 10.4% 22.5% 2015 8,507 10.7% 22.3% 2016 9,558 11.1% 23.0% 2017 18,237 9.3% 20.3% Total 52,027 10.2% 22.1% Notes: Wage share equals CEO wage (inclusive of 27% payroll tax) divided by adjusted EBITDA. CEOs are selected from a 50% sample of all Hungarian employees (ADMIN3) and must: (i) be classified as CEO by occupation code; (ii) have a full-time employment contract with the firm for all 12 months of the year; (iii) have no other contractual relations reported in ADMIN3; and (iv) receive at least 2×the minimum wage. Adjusted EBITDA = Sales −Personnel −Materials + CEO wage. The series begins in 2013 because wages are top-coded in earlier years of the data. Optimal hiring and returns. Firms solve (1−ν)(Aez/he)ν=w, yielding he(z) = Aez w−1/ν(1− ν)1/ν and re(z) = Aez w1−1/ν(1 −ν)1/ν−1. The payment schedules before subsidies are ωF M(z) = ϕνrM(z) and ωE(z) = νrE(z). After subsidies, workers receive ωM(z) = (1 + sM)ωF M(z) and ωE(z) unchanged when sE= 0. Threshold and indifference. Single crossing implies thresholds zmin(e) satisfy ωe(zmin(e)) = wfor the received income. This gives zmin(M)=(xM)−1ν−1A−1 M(1−ν)1−1/νw1/ν, zmin(E)=(xE)−1ν−1A−1 E(1−ν)1−1/νw1/ν. Active shares and composition. With Pareto tails and threshold selection, ne= Λθ(1 − 1/θ)θzmin(e)−θand ˜ze=θ θ−1zmin(e). Taking the ratio eliminates w: nM nE =AMxM AExEθ =: R. Note that R= (xM/xE)R, where Ris defined in the main text as R:= (AM/AE)θ(xM/xE)θ−1. Labor market clearing. Aggregate labor demand equals the supply of production workers: αnEAEw−1/ν(1 −ν)1/ν ˜zE+ (1 −α)nMAMw−1/ν(1 −ν)1/ν ˜zM= 1 −n, where n=αnE+ (1 −α)nM. Substituting ˜ze=θ θ−1Λ(1 −1/θ)n−1/θ eand using the indifference conditions to eliminate w−1/ν term-by-term yields the linear identity α1 + κ xEnE+ (1 −α)1 + κ xMnM= 1, κ =1−ν ν θ θ−1. 32 Solution. Define D:= α1 + κ xE+ (1 −α)1 + κ xMR. The unique solution is nE=1 D, nM=R D, n =αnE+ (1 −α)nM. Prices and GDP. Using zmin(e) = Λ(1 −1/θ)n−1/θ eand the indifference conditions gives w= Λν(1 −1/θ)ν(AExE)ννν(1 −ν)1−νn−ν/θ E, ωE(z) = zΛν−1(1 −1/θ)ν(AExE)ννν(1 −ν)1−νn(1−ν)/θ E, ωM(z) = zΛν−1(1 −1/θ)ν(AMxM)ννν(1 −ν)1−νn(1−ν)/θ M. GDP per worker equals y= ΛναAEn1−1/θ E+ (1 −α)AMn1−1/θ Mν(1 −n)1−ν. First best. Setting xE=xM= 1 removes all distortions. One implementation is sE= 0 and sM=ϕ−1−1, which yields n=1 1 + κ,nM nE =AM AEθ. This is the unique optimal allocation that maximizes GDP. B.2 Proof of M-economy equilibrium Special case of §B.1 with α= 0, sE=sM= 0. B.3 Proof of M-economy first best Special case of §B.1 with α= 0, sM=ϕ−1−1. B.4 Proof of M-economy with manager wage subsidy Special case of §B.1 with α= 0, sE= 0. B.5 Proof of E-M equilibrium with Pareto tails Special case of §B.1 with sE=sM= 0. B.6 Proof of E–M first best Proof of Proposition 7. Special case of §B.1 with sE= 0, sM=ϕ−1−1. This sets xE=xM= 1, yielding nM/nE= (AM/AE)θand n∗= 1/(1 + κ). 33 Proof of Proposition 3. Compare equilibrium (xM=ϕ < 1) with first best (xM= 1). Since ϕ < 1 raises the coefficient on nMin the labor identity, monotone comparative statics imply nM< n∗ M,nE> n∗ E, and n<n∗. Proof of Proposition 8. The between-group term is maximized when nM/nE= (AM/AE)θ, which holds iff ϕ= 1. B.7 Proof of GDP per worker in the E-M economy See §B.1. The formula holds for any (nE, nM) regardless of subsidies. B.8 Derivation: Elasticities near zero subsidies Using the general equilibrium formulas from §B.1, differentiate GDP with respect to ln xMand ln xEand evaluate at sE=sM= 0 (i.e., xE= 1, xM=ϕ). Let u:= 1 −1/θ, ρ:= (ϕAM/AE)θ,D:= α(1+κ)+(1−α)(1+κ/ϕ)ρ, and wE:= αAEnu E/B,wM:= 1−wE where B:= αAEnu E+ (1 −α)AMnu M. Using ∂ln xMln R=θand ∂ln xEln R=−θ, the elasticities are ∂ln xMln nE=−δM, ∂ln xMln nM=θ−δM, ∂ln xEln nE=δE, ∂ln xEln nM=−θ+δE, where δM:= (1 −α)ρ[θ+κϕ−1(θ−1)] D, δE:= ακ + (1 −α)θ(1 + κ/ϕ)ρ D. Applying the chain rule to y= Λν[αAEnu E+ (1 −α)AMnu M]ν(1 −n)1−νyields dln y dln(1 + sM)0 =νuwE(−δM)+wM(θ−δM)−1−ν 1−nαnE(−δM)+(1−α)nM(θ−δM), dln y dln(1 + sE)0 =νuwE(δE) + wM(−θ+δE)−1−ν 1−nαnE(δE) + (1 −α)nM(−θ+δE). 34