Entrepreneurial Reluctance: Talent and Firm Creation in China

Abstract

This paper examines the correlation between cognitive ability and firm creation. Drawing on administrative college admission data and firm registration records in China, we investigate who had created firms by their mid-thirties. We find a clear pattern of entrepreneurial reluctance: given the same backgrounds, individuals with higher college entrance exam scores are less likely to create firms. Through an exploration of firm performance, alternative career trajectories and variations across regions, we propose an explanation: the ability represented by exam scores is useful across occupations, yet higher-scoring individuals are attracted to waged jobs, particularly those of the state sector.

A growing literature in economics has shown that the ability of entrepreneurs is important for both firm-level productivity and aggregate-level economic growth (Bertrand and Schoar, 2003; Bloom and Van Reenen, 2007; Gennaioli et al., 2013; Queiró, 2022; see also the survey by Syverson, 2011), suggesting that the allocation of talent to the entrepreneurial sector may improve economic efficiency. The extent to which talented individuals are likely to become entrepreneurs, however, is not clear. On the one hand, canonical models of entrepreneurial choices predict a positive relationship between ability and entrepreneurial entry (e.g., Evans and Jovanovic, 1989). On the other hand, a theoretical literature on talent allocation has long noted that talented individuals are often attracted to non-entrepreneurial sectors, which can lead to misallocation (Baumol, 1990; Murphy et al., 1991; Acemoglu, 1995). Despite its importance to productivity and economic growth, the association between individual-level ability and entrepreneurship has not been extensively examined.¹

There are several empirical challenges to such an examination. It is difficult to measure entrepreneurship and ability using typical survey data. Most survey data do not differentiate firm creation from self-employment, even though the former is more important for the study of opportunistic entrepreneurship (e.g., Schoar, 2010; Levine and Rubinstein, 2017). Surveys of individuals or households also do not provide much information on the firms created by these individuals, making it difficult to study entrepreneurial success. In addition, because firm creation is of low probability, we need a large sample size of household surveys to have a sufficient number of entrepreneurs. Moreover, it is also difficult to separate the impact of individual ability from that of education and family background on firm creation.

In this paper, we study the relationship between individual ability and entrepreneurship in China. With over 20 million private firms, some of which are globally leading firms, China is active in entrepreneurial activities. Moreover, other than becoming entrepreneurs or working for a private firm, as in most developed economies, Chinese face an additional important choice: joining the state sector. With its various government positions and the world’s largest state-owned enterprises, having a job in the state sector in China is usually viewed as ideal by its citizens (Li et al., 2024). Thus, a priori, it is unclear whether the relationship between ability and entrepreneurship is positive or negative.

We address the empirical challenges by linking two administrative datasets—the universe of college admission records during 1999–2003 and the universe of Chinese firms and their owners. The large sample size allows us to study entrepreneurship, a small probability event. Specifically, our analyses use a random sample of 20% of the linked data,² including over 1.57 million college graduates who created approximately 105,000 firms by 2015.³ The linked administrative data have good measures of cognitive abilities, educational background (including both high schools and colleges), firm creation and firm success, as well as personal attributes. We supplement these linked administrative data with the Chinese College Student Survey, a large survey of Chinese college graduates that we conducted during 2010–15 to study waged jobs.

Our setting provides a clear measure of cognitive abilities: the National College Entrance Exam (known as Gaokao) score. The score is the criterion for college admission in China, which is popularly believed to determine the course of the life of the exam taker. As an influential cross-country literature has documented, exam scores outperform years of schooling as a proxy for assessing human capital in predicting economic growth (e.g., Hanushek and Kimko, 2000; Hanushek and Woessmann, 2015). While it is evident that ability is multi-dimensional, this body of research indicates that exam scores are a more accurate measure of cognitive abilities, which is the central focus of our paper, than merely counting years of formal education. Recognising that the exam score is affected by family investment, we consider a set of proxies for family background in our analyses.

Another advantage of our datasets is that we can separate the effect of individual-level ability from that of colleges (e.g., college selection, college education and peer influence). Conceptually, we can compare individuals who graduated from the same college by including college fixed effects. Such a within-college design, however, is not ideal if there exist systematic selection biases. For instance, higher-scoring individuals may perform worse in extracurricular activities (that can affect future firm creation) than do their lower-scoring college peers in the college admission process. We examine college selection and show that performance reflected in non-exam activities is not a critical concern in our setting, where the exam score is the primary criterion in college admission.

We document a phenomenon that we refer to as ‘entrepreneurial reluctance’: higher-scoring individuals are less likely to create firms. The raw correlation between firm creation and the Gaokao score is negative. More importantly, after controlling for college fixed effects, we still find a strong and negative association between the score and firm creation, suggesting that the college selection, education and network effect cannot explain the negative association between firm creation and the score. Our estimate removing college fixed effects suggests that a one-SD higher score is associated with a 14% lower probability of creating a firm. This negative correlation seems to be general as it holds when we examine the relationship by major and by college quality.

There are two potential interpretations of the negative relationship: the personal trait hypothesis and the opportunity cost hypothesis.⁴ The personal trait hypothesis posits that individuals with higher exam scores may exhibit behavioural traits—such as higher risk aversion or lower social engagement—that are not conducive to entrepreneurial success, thereby making them less inclined towards entrepreneurship. Conversely, the opportunity cost hypothesis suggests that, while the abilities indicated by higher exam scores are valuable in both entrepreneurial and waged employment, the waged sector may offer more attractive opportunities, thereby diverting talent away from entrepreneurship.

We discern which hypothesis aligns more closely with our data. First, we find a positive link between higher exam scores and entrepreneurial success in terms of firm size, expansion, entry into non-local markets and public listing, suggesting that it is unlikely that higher-score individuals have unfavourable personal attributes for becoming entrepreneurs. Second, exam scores also correlate positively with success in waged jobs, suggesting that opportunity cost is indeed a possible reason for why higher-score individuals are less likely to create firms. Moreover, we find that the state sector serves as a more important alternative than the private sector.⁵ As further evidence, we examine differences in state employment across regions and find that the reduced inclination among high-score individuals to establish businesses is more pronounced in areas with higher state employment. Finally, we do not find evidence supporting the personal trait hypothesis, as there are no strong correlations between exam scores and risk attitudes or engagement in social activities. While our findings do not establish causality, they lend stronger support to the opportunity cost hypothesis, suggesting that the allure of alternative career paths, particularly in the state sector, may explain the reduced entrepreneurial inclination among high-achieving individuals.

Our findings contribute to a broad literature on entrepreneurship, where evidence on ability and firm creation is still elusive. Using data from the National Longitudinal Survey of Youth, Hartog et al. (2010), Levine and Rubinstein (2017) and Hegde and Tumlinson (2021) investigated the relationship between ability and entrepreneurship.⁶ Our approach is closest to that of Levine and Rubinstein (2017), although we are able to mitigate the confounding roles of colleges (an issue recognised by the literature) by employing a within-college design. Most importantly, different from our results, Levine and Rubinstein (2017) found that individuals with higher scores in the Armed Forces Qualification Test are more likely to own incorporated businesses.⁷ This contrast with our findings might be the result of institutional differences between the two countries, i.e., the state sector is more prevalent in China. We indeed find that higher-scoring individuals are less likely to create firms in regions in which the state plays a more dominant role.

Our findings have nuanced implications for economic development. On the one hand, talent in the state sector is an important source of state capacity. A large political economy literature has documented the importance of Chinese policymakers for economic growth (e.g., Li and Zhou, 2005). On the other hand, there is also a risk that this talent may be diverted towards rent-seeking behaviours within the public sector, which generates limited social value and can lead to misallocation. Existing research has shown that the dominance of the state sector potentially hinders economic development in the Chinese context (Hsieh and Song, 2015). Although we cannot fully quantify these two aspects for economic development,⁸ our results suggest that certain regions could have more successful firms with reduced state sector involvement.

In addition, our study joins the recent empirical literature that tries to understand talent allocation in a variety of settings (e.g., Hsieh et al., 2019 on racial discrimination in the United States and Ashraf et al., 2021 on gender norms across countries). Our findings underscore the significant role of the state in shaping talent allocation, particularly in economies in which the public sector serves as a prominent career path for highly skilled individuals. This perspective can enrich our understanding of talent allocation across countries and economies.

1. Background and Data

China employs a centralised college admission system, wherein approximately 2,300 colleges admit students primarily based on the scores from a centralised college entrance examination, known as Gaokao. In early June of their senior year (Grade 12), students are required to undertake the Gaokao in either the sciences track (comprising Chinese, English, advanced mathematics and science subjects) or the social sciences track (including Chinese, English, basic mathematics and social sciences). This choice of track is made in Grade 10. Given that the exam and admission processes are provincially administered each year, a student’s exam scores are only directly comparable with those of peers from the same province, year and track (either social or natural sciences). While the Gaokao score is the predominant factor in college admissions, a minority of students can gain additional points for extracurricular achievements, such as medals in various Olympiads, which we consider in our empirical analysis.

1.1. Gaokao and Firm Owner Data

We employ comprehensive administrative data encompassing the entire cohort of Gaokao participants from 1999 to 2003, alongside their subsequent college admission outcomes across more than 2,300 institutions (CDC, 2015). The dataset comprises over 20.3 million entries,⁹ of which 12.4 million successfully secured college placements. This rich dataset offers extensive details on each student’s examination performance, including overall scores, scores by subject and, for those admitted, the names of the colleges and chosen majors. Additionally, it provides biographical data such as gender, Hukou status (indicating urban or rural origin), birth year, birth county, high school attended and any political affiliations.¹⁰

To analyse firm creation and performance, we use comprehensive administrative records of all firms registered in China up to February 2015 (SAMR, 2015).¹¹ These records encompass approximately 27 million firms, including 11 million that had been deregistered by 2015, which we interpret as an indicator of firm closure. The data detail the firm owners, or shareholders, who may be either individuals or other firms. Individual owners are referred to as firm creators or entrepreneurs within our study. Additionally, the dataset provides fundamental information about each firm, such as its industry sector, geographical location, registered capital and instances of cross-firm ownership.

The linkage between college entrance examination and admission data and the firm owner database is facilitated through an encrypted national identity number present in both datasets. Out of the 12.4 million available records, we successfully establish connections for 8.2 million entries,¹² encompassing 7.9 million unique individuals. The discrepancy is attributable to individuals who have undertaken the examination multiple times; our analysis, however, prioritises their initial attempt, considering it a more accurate reflection of inherent ability. In adherence to privacy and de-identification standards, our study utilises a random sample comprising 20% of the successfully linked data, equating to a subset of 1.57 million individuals.

1.2. Chinese College Student Survey 2010–15

We supplement the administrative data with our own Chinese College Student Survey data, which include information on wages (Tsinghua University, 2015). We conducted large-scale surveys of college graduates during the graduation months (May and June) of 2010–15, which cover approximately 30,000 students from ninety colleges, approximately 14,800 of whom reported detailed information on their first jobs. We designed the surveys to evaluate the elite college premium and intentionally asked about Gaokao scores (see Li et al., 2012 and Jia and Li, 2021 for a detailed description of these surveys and how the first job is important for future jobs). Moreover, the surveys include information on student performance and behaviours in college.¹³

1.3. Key Variables and Summary Statistics

1.3.1. Firm variables

We present firm-level variables in panels A and B of Online Appendix Table A.1. By 2015, the 1.57 million college graduates in our data had established 150,473 firms, and the probability of creating any firm is 7.36%. The median firm was established in 2010, or around six years after college graduation of the founder. The top five industries are wholesale and retail (30.8%), leasing and business services (20.1%), scientific research and technology services (14.1%), manufacturing (8.8%) and information technology and services (6.9%). The remaining fifteen industries accounted for 19% of the total.

We use registered capital in 2015 as a proxy for firm size, which is the maximum liability that a firm has and, hence, matters for doing business.¹⁴ As a verification, we employ another dataset, the Chinese Annual Survey of Industrial Firms data, to examine the correlations between registered capital and other firm success measures, including employment, sales and total factor productivity (TFP). As shown in Online Appendix Table B.1, there are strong correlations between registered capital and these measures in the survey.

We adopt several alternative measures for firm success. The primary indicator is a firm’s registration location relative to the founder’s home province. Given the well-documented tendency of provincial governments to shield local businesses from inter-provincial competition (Young, 2000), the ability to establish a firm outside the entrepreneur’s home province is a notable marker of success. In our dataset, a minority of firms, approximately 40%, achieved this feat, underscoring the challenges posed by regional protectionism in expanding beyond local boundaries. The second metric of success we consider is expansion, specifically defined as a firm’s direct investment in another company, thereby becoming an owner. Within our sample, 5% of firms demonstrated growth through this avenue. The final measure is whether a firm has been publicly listed on the stock market. Being publicly listed is regarded as a proxy for exceptional success due to its rarity, occurring in only 3.5 instances per 10,000 firms in our sample.

1.3.2. Exam scores

The exam score is our ability measure. We should note that exam scores vary greatly within colleges, and there is a lot of overlap of scores across different tiers of colleges (Online Appendix Figure C.1(a)). College fixed effects can explain only half (49.6%) of the variation in exam scores. As reported in panel C of Online Appendix Table A.1, the mean and SD of the exam scores in the raw data are 444.7 and 95, respectively. Once we control for province-year-track fixed effects, the SD decreases to 82.9. The SD decreases to 60.6 after we also control for college fixed effects, and 58.7 if we further control for twelve major fixed effects.

The large within-college variation in scores could be due to certain institutional reasons. First, the college application and admission process are highly uncertain (Li and Qiu, 2023). In our study period, students in most provinces applied for colleges before they knew their exam scores. Each exam taker needed to indicate college preference, via a pencil-and-bubble sheet, for up to three colleges (and three majors in each college) within a few days. Each student could be accepted by only one college, and priority was given to the first choice in the bubble sheet (second and third choices were nearly useless). As a result, the match between score and college was far from ideal.¹⁵ Second, scores for each college vary greatly across provinces due to the uneven distribution of admission quotas. Each province is assigned a quota for each college by the central government. Because of political and historical considerations, major metropolitan areas, such as Beijing and Shanghai, and minority provinces, such as Tibet, Xinjiang and Yunnan, typically get a larger quota, especially for elite colleges. Finally, there is a college-major trade-off. Some students may choose a lower-ranked college for a popular major. For example, within any college, admission scores for popular majors, such as economics, finance, law and STEM, are normally higher than those for humanities.

1.3.3. Personal background

We consider five sets of personal background variables: gender, Hukou (urban or rural status), age, political capital and high school dummies.¹⁶ As reported in panel C of Online Appendix Table A.1, 55% of college students are male, slightly higher than the male share in the population (51.3% in 2001), and 52% of students have urban Hukou, much higher than the urban share in the population (37% in 2001), which is consistent with the fact that fewer rural students are able to attend college. The median age of firm creators in our sample was thirty-three in 2015, very close to the median age of college-educated firm owners in the entire firm registration data (33.9).¹⁷

We know whether one was a member of the Chinese Communist Party or other political parties at the time of college application, which measures one’s political capital at an early stage in life. Additionally, we observe which high school each student went to. Different from a centralised college system, high school education in China is locally financed, and wealthier families are more likely to be able to afford better high schools (Ye, 2015). By including over 32,000 high school dummies in our analysis, we are able to compare individuals with relatively similar socioeconomic status.

2. Research Design and Evidence

2.1. Research Design

Leveraging detailed information on individuals and their education background, we employ the specification

$$\begin{eqnarray} \it{Firm}_{i,pyt,c} = \beta \it{Score}_{i,pyt,c} + \alpha X_{i} + \lambda _{pyt} + \theta _{c} + \epsilon _{i,pyt,c}, \end{eqnarray}$$

(1)

where |$\it{Firm}_{i,pyt,c}$| is a dummy variable that indicates whether individual i of province p, year y and track t in college c created a firm. The key independent variable of interest is the exam score (⁠|$\it{Score}_{i,pyt,c}$|⁠). Because |$\it{Score}_{i,pyt,c}$| is comparable only within the same province, year, and track, we always control for province-year-track fixed effects (⁠|$\lambda _{pyt}$|⁠) in our analysis. Here |$X_{i}$| indicates one’s personal characteristics, which include gender, Hukou (rural versus urban), age, political capital and the high school fixed effects. Although we do not have measures such as parental income, it is conceivable that high school fixed effects account for considerable variation in family status. We denote by |$\theta _{c}$| college fixed effects, which control for the potential influence of the college network and reputation on firm creation. We report SEs that are clustered at the college level.

2.2. Score and Firm Creation

In our preliminary assessment, which does not adjust for college fixed effects, we observe a negative association between firm creation and college entrance exam scores. Figure 1(a) graphically illustrates this relationship by depicting firm creation rates against exam scores, excluding province-year-track fixed effects and considering all individuals, irrespective of college admission status. As demonstrated in the figure, a pronounced negative correlation emerges. The corresponding quantitative analysis is detailed in column (1) of Table 1, where we present the impact of a one-SD increase in exam scores. Specifically, the coefficient of −0.519 implies that a one-SD surge in the exam score is associated with a decrease in the likelihood of firm creation by approximately 7% from the mean probability of 7.36%.

Fig. 1.

Firm Creation versus College Entrance Exam Score.

Notes: These plots report the fitted relationship between firm creation and college entrance exam scores, controlling for provincial-year-track fixed effects. Panel (a) includes both admitted and non-admitted students to colleges. Panel (b) focuses on college students, where we control for college fixed effects.

Table 1.

Exam Score and Firm Creation.s

Dependent variable: firm creation
					Economics,		Top ten	11–100	100|$+$|
	Full sample	College sample		STEM	finance, law	Humanity	colleges	colleges	colleges
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Exam score (SD)	−0.519***	−1.002***	−1.069***	−0.866***	−0.915***	−1.075***	−0.951*	−1.525***	−0.945***
	(0.013)	(0.040)	(0.039)	(0.053)	(0.071)	(0.067)	(0.467)	(0.110)	(0.039)
Male			3.452***	2.964***	4.114***	4.376***	3.834***	3.340***	3.465***
			(0.066)	(0.077)	(0.104)	(0.146)	(0.640)	(0.141)	(0.075)
Urban			0.193***	0.304***	−0.025	0.063	1.034**	0.568***	0.110
			(0.062)	(0.082)	(0.130)	(0.141)	(0.450)	(0.145)	(0.069)
Age			−0.029	−0.049	0.084*	−0.125**	−0.003	0.092	−0.052**
			(0.024)	(0.030)	(0.050)	(0.049)	(0.218)	(0.070)	(0.025)
Political capital			1.351***	1.595***	0.518	1.839*	−0.317	1.037	1.550***
			(0.387)	(0.483)	(0.725)	(1.027)	(1.606)	(0.846)	(0.470)
Province-year-track FEs	Y	Y	Y	Y	Y	Y	Y	Y	Y
High school FEs			Y	Y	Y	Y	Y	Y	Y
College FEs		Y	Y	Y	Y	Y	Y	Y	Y
Observations	2,187,037	1,571,990	1,560,038	804,644	426,615	306,269	29,515	239,213	1,283,493
Dependent variable (mean)	7.36	7.36	7.36	6.66	8.41	7.54	8.36	7.79	7.25

Notes: This table reports the correlation between firm creation and college entrance exam scores at the individual level, using administrative data. Column (1) includes both admitted and non-admitted students, while columns (2) through (9) focus on college students and include college fixed effects. In the regressions presented in columns (3) through (9), we control for personal characteristics and high school fixed effects. Columns (4) through (6) divide the samples by major, and columns (7) through (9) divide the samples by college rank. SEs reported in parentheses are clustered at the college level. ***|$p\lt .01$|⁠, **|$p\lt .05$|⁠, *|$p\lt .1$|⁠.

2.3. The College Effect

A pivotal challenge within existing research is the potential confounding influence of college effects on the observed correlation between exam scores and entrepreneurship. This ambiguity arises from two possibilities: firstly, that potential entrepreneurs might gravitate towards colleges with lower (or higher) academic standings; and secondly, that the reputation and networks associated with certain colleges could facilitate entrepreneurial ventures. By having access to data on each student’s college destination, we are uniquely positioned to empirically assess and account for college effects. We achieve this by incorporating college fixed effects into our analysis, thereby isolating the impact of college characteristics from the relationship between exam scores and firm creation.

Our findings indicate that college selectivity does not account for the observed negative correlation between exam scores and firm creation; this relationship persists, and even becomes larger, upon adjusting for college fixed effects. Illustrated in Figure 1(b), the analysis incorporating college fixed effects unveils a consistent negative correlation between firm creation and exam scores. Remarkably, the negative correlation strengthens, as evidenced in column (2) of Table 1, where a one-SD increase in the exam score is associated with a 13.6% reduction in the likelihood of firm creation. This suggests a positive association at the college level between average exam scores and entrepreneurial propensity. Indeed, our further analysis reveals that students from colleges with higher average scores exhibit a greater inclination towards entrepreneurship, as detailed in Online Appendix Figure D.1, which explains why the within-college estimate is even more negative. This negative correlation between firm creation and the score remains robust even after accounting for individual characteristics and high school fixed effects, as shown in column (3) of Table 1.

2.4. College Selection Based on Other Characteristics

College admissions might indeed involve a delicate trade-off between academic excellence and entrepreneurial flair. In the context of the United States, the admission framework extends well beyond mere scholarly achievements, embracing extracurricular participation, leadership prowess and empathetic qualities, all of which are reputed to be indispensable for entrepreneurial success. One concern is that Chinese institutions of higher learning adopt a similar approach, welcoming students who, despite lower academic scores, shine in these entrepreneurial competencies, as well as those who achieve high scores, but may not exhibit as strong entrepreneurial tendencies. This balanced admission strategy could potentially drive the negative correlation we observe between exam scores and the propensity for firm creation.

However, the probability of such a trade-off occurring within our context is relatively low, given the predominant emphasis on exam scores in Chinese college admissions, as previously discussed. The system does allow for the addition of extra points for exceptional achievements, such as performance in international-level Olympiads, but these opportunities affect only a minor segment of applicants. Our data reveal that merely 8.7% of college-admitted individuals received such extra points. Moreover, the average extra points awarded stand at 12.8, a figure that pales in comparison to the mean college entrance exam score of 444 points. We further show in Online Appendix Table E.1 that, despite a discernible negative relationship between extra points and exam scores, this correlation is negligible due to the limited number of students who benefit from extra points, consistent with the marginal impact of such achievements on the overall admission process.

Our regression analysis, detailed in Online Appendix Table E.1, confirms that the inclusion of extra points awarded for extracurricular achievements does not significantly affect the observed relationship between exam scores and firm creation. Initially, the analysis reveals only a marginal positive correlation between these extra points and the likelihood of firm creation, indicating that the presence of extra points does not substantially influence entrepreneurial outcomes. More crucially, even after accounting for these extra points in our model, the negative correlation between exam scores and firm creation remains unaltered. This finding confirms the robustness of the initial relationship, suggesting that the influence of exam scores on firm creation is not significantly mediated by the additional points from extracurricular achievements.

It is important to recognise that exam scores can be correlated with personal characteristics such as inclinations towards risk-taking and social skills. We examine these relationships in our later analysis.

2.5. Majors and College Tiers

We further investigate whether the correlation between exam scores and the propensity for firm creation exhibits variation across different academic disciplines and levels of college quality. For analytical simplicity, we classify the twelve majors into three broad categories: STEM, which encompasses science, technology, engineering and mathematics; economics, finance and law; and the humanities. STEM majors represent 52% of students—aligning with China’s strategic focus on STEM education. Students pursuing economics, finance and law constitute 28%, while those in humanities fields make up the remaining 20%. In parallel, we segment colleges into three quality-based tiers: the top ten institutions, those ranked between 11 and 100, and the rest. This stratification allows us to assess how the nexus between academic performance and entrepreneurial activity might be influenced by the field of study and the calibre of the educational institution.

The analysis reveals that the connection between college entrance exam scores and the likelihood of creating a firm remains relatively stable across different majors and tiers of college quality. As detailed in columns (4)–(6) of Table 1, students majoring in economics, finance and law exhibit a marginally higher propensity for entrepreneurship, aligning with the hypothesis that choosing such fields may reflect an entrepreneurial inclination. Crucially, within each academic discipline, the inverse relationship between exam scores and firm creation persists. Specifically, a one-SD increase in exam scores (within college) correlates with a 13% reduction in the probability of firm creation for STEM majors, 11% for those in economics, finance and law, and 14% for humanities students. Moreover, this negative trend is consistent across different levels of college quality, as indicated in columns (7)–(9) of the table. Despite a general tendency for students from more prestigious colleges to engage more in entrepreneurship, the adverse effect of higher exam scores on the likelihood of starting a firm is evident across all college quality categories.

2.6. Additional Checks

To ensure the reliability of our findings, we conduct a series of sensitivity checks addressing potential data-related issues. Initially, we confine our analysis to province-year combinations where the likelihood of data omission was below 5%; the outcomes of this refined analysis are presented in columns (1)–(2) of Online Appendix Table F.1. Subsequently, we exclude individuals who had taken the exam more than once in columns (3)–(4). Lastly, we remove observations with the top and bottom 10% of scores, considered potential outliers, from the dataset (columns (5)–(6)). These results all align with our original findings.

Our dataset includes family-owned enterprises, raising the possibility that some college graduates might join these firms through succession rather than founding new ventures. Preliminary scrutiny reveals that a vast majority, 98.2%, of the firms in our sample were inaugurated post the exam years of the students, implying a minimal role of family succession. To further assess the impact of familial transitions, we analyse the age disparities among shareholders. Specifically, we exclude potential family businesses, defined as those established prior to the proprietor’s exam year or those exhibiting a significant age gap—over twenty years—between the youngest college-educated owner and the oldest shareholder. After this filtering, our data consistently reveal a comparable negative relationship between exam scores and firm creation, as documented in columns (7)–(8) of Online Appendix Table F.1.

Furthermore, we provide our findings without college fixed effects in Online Appendix G. These results indicate that all our findings are of the same signs whether we include college fixed effects or not. However, due to the additional influences of colleges, the magnitudes are different.

3. Possible Interpretations

In this section, we examine two overarching hypotheses that might explain the observed negative correlation between firm creation and the score. The first, termed the personal trait hypothesis, posits that there may be an inverse relationship between exam scores and certain personality characteristics conducive to entrepreneurial success. For example, individuals with higher scores might lack a versatile skill set often associated with entrepreneurship, or they could exhibit greater risk aversion. The second hypothesis, known as the opportunity cost hypothesis, suggests that, while exam scores may be indicative of inherent abilities relevant both to entrepreneurship and salaried employment, the returns to these abilities in the salaried sector increase more steeply with exam scores than in entrepreneurship. Consequently, this disparity in returns may steer talented individuals away from entrepreneurship due to the elevated opportunity costs. To shed light on potential reasons underpinning this aversion to entrepreneurship, we next study personal characteristics, firm success and the allure of alternative job opportunities.

3.1. Exam Score and Personal Traits

To examine the importance of personal traits in explaining entrepreneurial reluctance, we examine participation in social activities and the levels of risk aversion in our Chinese College Student Survey. Our analysis does not yield definitive evidence linking academic scores to either social engagement or risk-taking propensities. Specifically, we observed no substantial correlation between students’ scores and their involvement in social activities, proxied by holding positions within the college student union and other social organisations, as reported in column (1) of Table 2.¹⁸

Table 2.

Exam Score and Personal Traits.

Dependent variable:	Social organisations in college	Lottery (2011)	Risky investment (2011)
	(1)	(2)	(3)
Exam score (SD)	−0.009	−0.029	−0.042
	(0.014)	(0.030)	(0.028)
Male	−0.090***	0.063**	0.036
	(0.013)	(0.030)	(0.023)
Urban	0.053***	−0.014	−0.003
	(0.011)	(0.035)	(0.038)
Age	−0.009	−0.011	−0.027**
	(0.005)	(0.010)	(0.013)
Political capital	0.026	−0.053**	0.009
	(0.017)	(0.022)	(0.024)
Province-year-track FEs	Y	Y	Y
High school types	Y	Y	Y
College FEs	Y	Y	Y
Observations	14,094	3,024	3,024
Dependent variable (mean)	0.50	0.36	0.67

Notes: This table reports the correlation between personal traits related to entrepreneurial activities and college entrance exam scores at the individual level, using data from the Chinese College Student Survey. Column (1) examines holding positions in college student unions and other social organisations, column (2) focuses on choosing a risky lottery and column (3) looks at taking a risky investment. SEs reported in parentheses are clustered at the college level. ***|$p\lt .01$|⁠, **|$p\lt .05$|⁠.

Additionally, our 2011 Chinese College Student Survey included questions designed to gauge risk attitudes: one concerning the preference between a risky lottery and a certain monetary outcome, and another related to the prioritisation of guaranteed investment returns over higher-risk, higher-reward opportunities. We define a ‘risk-loving’ individual as one who opts for the lottery or disagrees with the cautious investment approach. Our findings do not reveal strong associations between students’ scores and their risk attitudes, as indicated in columns (2)–(3). As a caveat, we acknowledge that we cannot definitively reject this hypothesis since there may be additional unobserved personal attributes that could play a role.

3.2. Exam Score and Firm Success

We could also test whether a higher score is associated with unobservables that hampers entrepreneurial ability by directly examining whether a higher exam score means that entrepreneurs are less successful. Specifically, we estimate the equation

$$\begin{eqnarray} y_{f,i,pyt,c} = \gamma \it{Score}_{f,i,pyt,c} + \alpha X_{i} + \lambda _{pyt} + \theta _{c} + \epsilon _{i,pyt,c}, \end{eqnarray}$$

where |$y_{f,i,pyt,c}$| refers to different success measures for firm f created by individual i. To mitigate the influence of colleges, we also incorporate college fixed effects. The measures of firm success include firm size, registration of the firm outside the individual’s home province, firm expansion and public listing status. This component of our study focuses on a subset of individuals who have embarked on entrepreneurial ventures. To make the firm outcomes more comparable, we report the results after including province-by-industry fixed effects.¹⁹

Our regression analyses reveal a positive correlation between academic scores and all four examined measures of firm performance. We first observe that a one-SD increase in within-college exam scores corresponds to a 1.2% increase in the firm’s registered capital (column (1) of Table 3). This positive association between scores and firm success becomes even more pronounced with alternative success indicators, such as the firm’s location outside the entrepreneur’s home province, firm expansion and achieving publicly listed status. Specifically, a one-SD increase in within-college exam scores boosts the likelihood of establishing a firm outside the home province and of firm expansion by approximately 5.8% and 12%, respectively (derived from the coefficients 0.023/0.4 and 0.006/0.05 in columns (2)–(3)). While less precise, the estimate for the probability of a firm becoming publicly listed suggests an even more substantial increase of about 39% relative to the mean, underscoring the significant potential impact of academic achievement on entrepreneurial success.

Table 3.

Exam Score and Firm Success.

Dependent variable:	ln registered capital	Out of home province	Firm expansion	Listed (× 1,000)
	(1)	(2)	(3)	(4)
Exam score (SD)	0.012*	0.023***	0.006***	0.137**
	(0.006)	(0.001)	(0.001)	(0.064)
Male	0.005	−0.008***	0.000	0.257*
	(0.012)	(0.002)	(0.002)	(0.138)
Urban	0.057***	−0.027***	0.006***	0.101
	(0.014)	(0.003)	(0.002)	(0.150)
Age	−0.023***	0.005***	−0.002***	−0.109*
	(0.006)	(0.001)	(0.001)	(0.063)
Political capital	0.061	−0.003	0.015	0.742
	(0.079)	(0.015)	(0.011)	(1.657)
Province-year-track FEs	Y	Y	Y	Y
High school FEs	Y	Y	Y	Y
College FEs	Y	Y	Y	Y
Province-industry FEs	Y	Y	Y	Y
Observations	144,277	144,278	144,278	144,278
Dependent variable (mean)	4.29	0.40	0.05	0.35

Notes: This table reports the correlation between firm success and college entrance exam scores, all at the firm level, drawing on administrative data. Firm success is measured by firm size or registered capital (column (1)), registration outside the individual’s home province (column (2)), firm expansion (column (3)) and public listing status (column (4)). In addition to the fixed effects controlled in the regressions in Tables 1 and 2, we also include province and one-digit industry fixed effects. SEs reported in parentheses are clustered at the college level. ***|$p\lt .01$|⁠, **|$p\lt .05$|⁠, *|$p\lt .1$|⁠.

We recognise the possibility of sample selection bias among the individuals who have established firms, and due to limitations in data availability, we admit this as a constraint in our study. To explore how selection bias could influence our results, we refer the reader to Online Appendix H, where we categorise our sample into two segments—the top 25% and the remaining—and calculate the Lee bounds (Lee, 2009). The results generally indicate that even the lower-bound estimates are positive, affirming a positive correlation between exam scores and firm success.

3.3. Exam Score and Waged Jobs

Leveraging data from our Chinese College Student Survey, we delve into the correlation between academic performance and first-job wages, employing a within-college framework akin to (1). Despite the general tendency for first-job wages to be relatively uniform, our findings indicate a positive association between exam scores and earnings. As detailed in column (1) of Table 4, a one-SD increment in academic scores correlates with a 2.6% increase in first-job wages.

Table 4.

Exam Score and Waged Jobs.

Dependent variable:	ln wage	Job w. Hukou	State sector = 1	Private sector = 1
	(1)	(2)	(3)	(4)
Exam score (SD)	0.026***	0.035***	0.022	−0.016
	(0.009)	(0.013)	(0.015)	(0.016)
Male	0.052***	0.107***	0.100***	−0.095***
	(0.018)	(0.016)	(0.019)	(0.018)
Urban	0.006	−0.038***	0.017	−0.010
	(0.010)	(0.012)	(0.012)	(0.013)
Age	−0.005	0.003	−0.002	−0.003
	(0.008)	(0.006)	(0.006)	(0.006)
Political capital	0.023**	−0.004	0.053***	−0.051***
	(0.011)	(0.013)	(0.012)	(0.010)
Province-year-track FEs	Y	Y	Y	Y
College FEs	Y	Y	Y	Y
High school types	Y	Y	Y	Y
Observations	13,326	13,326	14,094	14,094
Dependent variable (mean)	7.81	0.34	0.43	0.50

Notes: Column (1) reports a regression examining the correlation between log wage and college entrance exam scores. Column (2) reports a regression examining the correlation between whether the waged job offers local Hukou and college entrance exam scores. Columns (3) and (4) report regressions examining the correlation of the exam score with career choice. All regressions are at the individual level, based on survey data. SEs reported in parentheses are clustered at the college level. ***|$p\lt .01$|⁠, **|$p\lt .05$|⁠.

In addition to wage benefits, academic scores appear to confer advantages in securing non-wage job benefits, notably including eligibility for local Hukou. This status is crucial, as it affords workers and their families access to essential local public services, like education and healthcare. Employing the same within-college approach, our findings, presented in column (2), indicate that individuals with higher exam scores are more likely to secure employment positions that offer local Hukou, implying the broader benefits of higher academic achievement in the job market beyond just salaries.

3.4. The Lure of the State Sector

A defining characteristic of China’s job market is the prevalent role of the state sector. According to our survey data, 43.4% of graduates found employment in state enterprises, 50.4% in private entities and 6.1% embarked on entrepreneurial ventures.²⁰ Additionally, the survey explored graduates’ employment preferences, revealing a strong inclination towards state-owned enterprises, with 64% of respondents identifying them as their top employment choice (Li et al., 2024). This preference is often driven by the perceived benefits of state sector jobs, such as enhanced job security, superior non-wage perks and elevated social status.

Our analysis offers tentative evidence suggesting that the state sector may serve as a more appealing alternative to entrepreneurship for high-scoring individuals, compared to the private sector.²¹ In column (3) of Table 4, we analyse the likelihood of entering the state sector compared to the private sector or entrepreneurship, while in column (4), we examine the likelihood of entering the private sector compared to the state sector or entrepreneurship. The positive (and negative) coefficient of exam scores for joining the state (and private) sector indicates that the private sector is more effective in attracting individuals with higher scores. The lack of significance reflects that both state sector and private sector jobs serve as alternatives in attracting higher-score individuals. Because we include a large amount of fixed effects in our analysis, we do not further employ multinomial logit regressions.

As an alternative to further test whether the influence of academic scores on entrepreneurial activity varies with the state’s presence across different regions, we employ variation across college prefectures.²² We use the proportion of employment in the state sector as an indicator of the state’s dominance in each prefecture, categorising these data into four quartiles for comparative purposes. Our analysis indicates that an increased presence of state employment in one’s college location correlates with a more pronounced disinclination towards entrepreneurship. The results, presented in Online Appendix I, demonstrate that, compared to individuals originating from areas within the lowest quartile of state sector employment, those from the highest quartile regions show a significantly greater reduction in the propensity to engage in entrepreneurship. Specifically, individuals from third-quartile and fourth-quartile prefectures exhibit a sharper decline in the likelihood of starting a business, relative to their score, than their counterparts from the first quartile.

A back-of-the-envelope calculation based on our findings highlights the significant impact of state employment levels on entrepreneurial activity. For an individual at the 75th percentile, transitioning their college prefecture’s state employment from the fourth quartile to the first quartile would, according to the increase in entrepreneurial propensity indicated in Online Appendix Figure I, elevate their likelihood of starting a firm by approximately 1.26 percentage points. This represents a 16.9% rise from the mean firm creation rate of 7.46%. These estimations reveal the substantial influence that regional state employment levels can exert on individual decisions to engage in entrepreneurship.

In summary, we note a positive connection between exam scores and success in both entrepreneurship and salaried employment, although this relationship is not causally identified. Moreover, individuals with higher scores are more likely to work in the state sector rather than starting their own businesses, and the reluctance towards entrepreneurship seems more pronounced in regions with a more dominant state sector. These results collectively lend support to the opportunity cost interpretation.

4. Conclusion

Leveraging a comprehensive dataset of both administrative records and our own surveys, our study examines the intricate link between cognitive ability and the propensity for firm creation. A main finding from our investigation is that individuals with higher college entrance exam scores are, maybe unexpectedly, less likely to create firms. This pattern, however, does not imply a deficiency in entrepreneurial capabilities among high achievers. To the contrary, firms founded by high-scoring individuals are more likely to thrive than those initiated by their lower-scoring peers. This contrast between entrepreneurial entry and firm success highlights a critical insight: while college entrance exam scores may be indicative of general ability, those with higher scores often gravitate towards careers outside entrepreneurship, suggesting an allocation of talent influenced by opportunity cost.

In our context, the state sector emerges as a notable force in diverting talent away from entrepreneurial pursuits. This interpretation is supported by data showing that higher-scoring individuals are more inclined to join the state sector. In addition, the pattern of entrepreneurial reluctance is more pronounced in regions and industries with stronger state influence. Although we do not assert that this mechanism is universally applicable, our research offers a broader perspective: the way a society configures its reward system has an impact on how talent is distributed. This point is critical to the discussion of talent allocation, a fundamental topic in political economy and development studies. Hence, we speculate that the distribution of talent across sectors varies greatly from country to country, depending on each nation’s economic reward system.

Additional Supporting Information may be found in the online version of this article:

Online Appendix

Replication Package

Notes

The data and codes for this paper are available on the Journal repository. They were checked for their ability to reproduce the results presented in the paper. The authors were granted an exemption to publish parts of their data because access to these data is restricted. However, the authors provided the Journal with temporary access to the data, which enabled the Journal to run their codes. The codes for the parts subject to exemption are also available on the Journal repository. The restricted access data and these codes were also checked for their ability to reproduce the results presented in the paper. The replication package for this paper is available at the following address: https://doi.org/10.5281/zenodo.13958685.

We thank the editor, Marco Manacorda, four anonymous referees, Ying Bai, Sam Bazzi, Ting Chen, Julie Cullen, Gordon Dahl, Alexia Delfino, Hanming Fang, Roger Gordon, Josh Graff Zivin, Elizabeth Lyons, David Mckenzie, Rohini Pande, Yona Rubinstein, Imran Rasul, Andrei Shleifer, John Van Reenen, Michael Song, Yang Song, Noam Yuchtman, Xiaobo Zhang, Yu Zheng, Xiaodong Zhu and especially Xiao Ma for their comments. We also benefited from the seminar and conference presentations at the Bank of Canada-Tsinghua-Toronto conference, CEFPA, CEPR/LEAP, Fudan, Georgetown-World Bank, HKUST, LISER, Liverpool, LSE, Luohan Academy, NBER, Nottingham, Peking University, Queen Mary, SITE Stockholm, Stanford China Center, UBC, UCL, UCSD and USC.

Financial support from the National Natural Science Foundation of China (NSFC)/Research Grants Council of Hong Kong (RGC) Collaborative Research Scheme (NSFC, 72261160577; RGC, CRS_CUHK405/22) is gratefully acknowledged.

Footnotes

1

There are a few studies of this relationship that use survey data from the United States, including Hartog et al. (2010), Levine and Rubinstein (2017) and Hegde and Tumlinson (2021). We compare our research design and findings with theirs when discussing the literature.

2

Our firm data include firms that have exited.

3

The median age of firm creators in our sample was thirty-three in 2015, very close to the median age of college-educated firm owners in the entire firm registration data (33.9). According to the Global Entrepreneurship Monitor, China has a high proportion of young entrepreneurs, with 57% aged 18–34 and less than 25% falling in the older 45–64 category (Xavier et al., 2013).

4

A third interpretation might consider a negative association between exam scores and entrepreneurial ability, suggesting that individuals with higher scores lack the diverse skill set emblematic of a jack-of-all-trades, as discussed by Lazear (2004). However, this theory is not very likely in our setting, as our data on firm performance provide suggestive evidence to the contrary.

5

The state sector pays more than the private sector on average, with a 5% gap in our sample in terms of first-job wages.

6

For the measurement of entrepreneurship, Hartog et al. (2010) and Hegde and Tumlinson (2021) included self-employment as entrepreneurship, whereas Levine and Rubinstein (2017) highlighted that the relationship between ability and entrepreneurship differs, depending on whether one includes self-employment as entrepreneurship or not. For the determinants of entrepreneurship, Hartog et al. (2010) studied different ability measures and highlighted the importance of general ability and balanced ability. Levine and Rubinstein (2017) emphasised the interaction between cognitive ability and risk-taking attributes, whereas Hegde and Tumlinson (2021) noted the importance of information friction in regard to ability that affects selection into entrepreneurship.

7

In a different setting, Shu (2016) found that higher-GPA graduates from MIT are more attracted to science and engineering than to the finance sector, which also suggests that talented individuals are not deterred from entering the sector that is more closely related to productivity.

8

Knowing how the negative correlation between score and firm creation affects efficiency and social welfare is extremely difficult because one would need to know the full ‘social return function’ of ability in every occupation.

9

Excluded from this count are approximately 2 million individuals who pursued vocational education; our analysis focuses on high school graduates. Previous iterations of this study have demonstrated that the inclusion of vocational school attendees does not significantly alter our findings.

10

It is pertinent to note the remarkably low college dropout rates in China, which renders attrition a negligible factor in our analysis. The Chinese higher education ethos, characterised by the maxim ‘strict entrance, easy exit’, ensures a graduation rate exceeding 95%. According to the Mycos Institute based in Beijing, the college dropout rate in 2011 was estimated at 3%, while the Ministry of Education reported a lower rate of 0.75% for the same year.

11

For entrepreneurs, particularly those running small businesses, there exists a trade-off between remaining self-employed—a scenario not covered in this paper—and opting to register a firm. While certain sectors mandate firm registration for operational legality, a key advantage of registration is the attainment of limited liability status. It is noteworthy that choosing to stay unregistered, and thus self-employed, may facilitate tax evasion practices. Our analysis encompasses both small and large enterprises, with the latter invariably requiring formal registration.

12

The matching process between the two datasets is not without its challenges, primarily due to missing or incorrect identification numbers. Such discrepancies often manifest at the province-year level. To ensure the robustness of our findings, our analysis is confined to province-year combinations characterised by a minimal incidence of missing identifiers.

13

Other information, e.g., city-level employment data, is from China’s National Bureau of Statistics (NBS, 2007).

14

After registration, firms can adjust their registered capital by submitting requests to the firm registration office. Once approved, the registered capital is updated. These decisions usually stem from the company’s strategic considerations such as increasing access to credit, which are difficult to follow dynamically. If a firm exited before 2015, its registered capital size refers to the most recent information before the exit.

15

Such uncertainty was mitigated only in recent years, when the admission system was reformed to become a parallel system that, thanks to computer technology, allows students to apply to a few more colleges.

16

We also know one’s county of origin. Because each county has one or more high schools, including high school fixed effects in our analysis is stricter than including county fixed effects.

17

Azoulay et al. (2020) documented that, in the United States, the mean age at founding for the 1-in-1,000 fastest growing new ventures is forty-five. Our aim, however, concerns more than star firms, and thus it is reassuring to observe the comparability between our sample and administrative data and other sources, such as the Global Entrepreneurship Monitor, noted above.

18

All analyses utilising survey data take into account sampling weights; specifically, regressions are conducted using weights that are the inverse of the sampling probability in our surveys.

19

Our findings are similar if we do not include these fixed effects.

20

Unlike the administrative data that distinguish between firm owners and the self-employed, our survey did not make this differentiation.

21

It is noteworthy that in our context, state-owned enterprise (SOE) managers share similarities with bureaucrats, holding official ranks and capable of transitioning between SOE management and governmental roles.

22

We use the college prefecture as it is the labour market of a college graduate. The results are similar if we use one’s birth prefecture.