Who Creates New Firms when Local Opportunities Arise?
Shai Bernstein, Emanuele Colonnelli, Davide Malacrino, and Tim McQuade⇤ August 29, 2018
ABSTRACT
New firm formation is a critical driver of job creation, and an important contributor to the responsiveness of the economy to aggregate shocks. In this paper we examine the characteristics of the individuals who become entrepreneurs when local opportunities arise due to an increase in local demand. We identify local demand shocks by linking fluctuations in global commodity prices to municipality level agricultural endowments in Brazil. We find that firm creation response is almost entirely driven by young individuals with generalist and managerial skills. In contrast, we find no such response within the same municipalities among skilled, yet older individuals. Those individuals who respond to local demand shocks are younger and more skilled than the average entrepreneur in the population. Entrepreneurial response of young individuals is larger in municipalities with better access to finance, more skilled human capital, and with overall younger demographics. These results highlight how the characteristics of the local population can have a significant impact on the entrepreneurial responsiveness of the economy.
⇤This is the substantially revised version of a paper previously titled “Marginal Entrepreneurs”. Shai Bernstein is with Stanford University, Graduate School of Business, and NBER; Emanuele Colonnelli is with Stanford University, Department of Economics; Davide Malacrino is with the IMF, and Tim McQuade is with Stanford Graduate School of Business. The views expressed in this article/presentation are those of the authors and do not necessarily represent the views of the IMF, its Executive Board or IMF management. We are grateful to Nick Bloom, Rebecca Diamond, Callum Jones, Arvind Krishnamurthy, Luigi Pistaferri, Amit Seru, as well as seminar participants at Alabama, Brigham Young, Duke, Statistics Norway, and Stanford University. We are grateful to the Stanford Institute for Innovation in Developing Economies (SEED), the Private Enterprise Development in Low-Income Countries (PEDL) Initiative by the Centre for Economic Policy Research (CEPR), the Stanford Center for International Development (SCID), and the Abdul Latif Jameel Poverty Action Lab (J-PAL) Governance Initiative for financial support.
I. Introduction
Entrepreneurship plays a critical role in aggregate job creation, with new businesses responsible for the majority of new employment in the economy (Decker et al.(2014);Haltiwanger et al.(2013b)).
Of course, entrepreneurship and the creation of new firms is a multi-faceted phenomenon. Some types of entrepreneurship can be described as Schumpeterian, in which talented individuals person- ally create new technologies or products that facilitate a creative destruction process in the econ- omy disrupting existing organizations. Other types of entrepreneurship would better be described as Kirznerian, in which alert individuals identify the existence of new and exogenous investment opportunities created by changing market conditions, and take advantage of them by forming new businesses (Kirzner(1973,1985)).
This latter form of entrepreneurship is increasingly recognized as an important driver of economic dynamics. A growing theoretical and empirical literature shows that new business formation is key to understanding how economies respond to aggregate shocks. For example, a variety of macroeconomic studies have emphasized the role new firm creation plays in the amplification and propagation of exogenous economic shocks (e.g., Bilbiie et al. (2012); Clementi and Palazzo(2016); Sedláček and Sterk(2017)). At the micro level,Adelino et al.(2017) show that new firms are responsible for the majority of jobs created in response to changes in investment opportunities driven by local demand, andDecker et al. (2017) find that new firms account for most of the employment growth in regions that experienced a significant economic expansion due to the discovery of shale oil and gas.
While recognizing the importance of firm entry response to local economic shocks, little is known about the characteristics of the individual entrepreneurs who identify and act upon such opportu- nities when they arise. Do such entrepreneurs share similar defining traits with other entrepreneurs in the economy, or do they differ in substantive and meaningful ways? This issue is particularly salient. If these entrepreneurs are concentrated in a particular segment of the population, then characteristics of the local population and long-term demographic trends may have significant im- plications for the entrepreneurial responsiveness of the economy. Moreover, understanding the key
traits of such individuals could be useful to policymakers in thinking how to foster vibrant, dynamic local economies. In this paper, we fill this gap and explore the personal and career characteristics of the individuals who create new firms in response to changes in local economic opportunities.
Tackling this question empirically poses two challenges. First, exploring this question requires employer-employee matched data that follows individuals over time and allows identifying the timing in which individuals choose to become entrepreneurs.1 Second, in order to study the entrepreneurial response of individuals to opportunities, we need a source of plausibly exogenous variation in local economic opportunities. For both these reasons, we choose to study the Brazilian economy.
First, our study relies on access to the administrative employer-employee matched data from the Brazilian Ministry of Labor that captures all the employees in the formal sector, and includes information on their work history, wages, education, gender and occupation. This data allows us to not only identify the founders of new firms, but also provides a rich set of information regarding their personal characteristics before the creation of the new firm. Second, the large agribusiness sector in the Brazilian economy allows us to identify exogenous local income and demand shocks arising from global commodity price fluctuations, and to study the firm creation response.2
Specifically, we interact municipality level historical production endowments of agricultural crops with contemporaneous changes in global commodity prices, a strategy similar in nature to Allcott and Keniston(2017) in the context of US oil and gas booms andBenguria et al.(2018) in the context of Brazil’s commodity cycles. These historical concentrations of agricultural crops are persistent due to the accumulation of expertise and economic activity over long time periods, as well as physical
1Studying the question in the U.S. is challenging due to limited information on individual level behavior. For example, the US Census Bureau’s Longitudinal Employer-Household Dynamics (LEHD) is an employer-employee matched data but its coverage is limited to few states only and excludes key states such as California, New York and Massachusetts (Babina(2015)). Alternatively, survey data in the U.S. such as the PSID (used by Hurst and Lusardi(2004) for example), is constrained to a sample of less than 5000 observations. The Survey of Small Business Finances compiles data on small businesses and offers information on their characteristics but does not follow firms over time. Moreover, while the survey provides a good description of the existing small businesses, it is not designed to specifically study the characteristics of new businesses - an important distinction noted by Haltiwanger et al.
(2013a).
2Brazil is among the largest producers in the world of coffee, sugarcane, orange juice, soybean, corn and ethanol, among others. These crops provide the basis for the large agribusiness industry in Brazil, which represents 22%
of Brazil’s GDP, a third of its employment, and almost 40% of its export (PwC,2013). The agribusiness industry includes not only farming production, but also the supply of farming inputs such as machinery, the selling, exporting and marketing the products, warehousing facilities, wholesalers, processors, and retailers, among others.
characteristics of the regions such as climate and soil. Similar to methods employed byBartik(1991) and Blanchard and Katz (1992), this strategy overcomes the reserve causality problems inherent in a simple regression of firm creation on changes in local income. In our case, the concern is that unobserved shocks to the investment opportunities of a particular segment in the population could mechanically impact local income.
We find that, in affected municipalities, increases in commodity prices lead to a significant in- crease in local income and local employment. Our estimated effects are economically meaningful.
At the top 10% of commodity price increases, municipalities experience a 2.9% increase in local aggregate income and a 4.1% increase in local employment. This increase in local aggregate income arguably creates new investment opportunities in the non-tradable sector, which is heavily depen- dent on local demand (Mian and Sufi (2012b); Basker and Miranda(2016);Mian and Sufi(2012b);
Stroebel and Vavra(2014)). Consistent with this idea, and similarly to Benguria et al.(2018) and Allcott and Keniston (2017), we find that the local demand shock triggers significant firm entry driven entirely by increases in the non-tradable sector, with the number of local firms increasing by 3.7%.
We then turn to our main question, and explore the characteristics of those entrepreneurs who respond to local demand shocks by forming new firms in the non-tradable sector. We find that such entrepreneurs are almost exclusively young individuals. Specifically we find that within munici- palities that experience a commodity price shock in endowed crops, entrepreneurship increases by almost 10 percent among individuals below the age of 30, while there is essentially a zero response for older individuals. These results are robust to the inclusion of industry fixed effects and a variety of covariates controlling for other demographic characteristics which may be correlated with age.
These results are consistent with the idea that lifecycle considerations strongly influence indi- vidual entrepreneurial responses to local economic opportunities. In particular, younger individuals have been shown to have higher degrees of risk tolerance than older individuals, and thus may be better able to tolerate the risk associated with fast transition to entrepreneurial activity (Kihlstrom and Laffont (1979b); Miller (1984b); Levesque and Minniti (2006)). Likewise, young individuals
may have less constraints in the form of family or looming retirement needs, and may therefore have sufficient flexibility to quickly respond to changes in economic opportunities. Finally, younger individuals, being at the early stage of their career, may have less attractive outside options, which may enhance their flexibility to respond to local economic opportunities when they arise.
Interestingly, we find that entrepreneurs who respond to local economic shock are significantly younger than the average entrepreneurs in the economy. Specifically, while roughly 40% of the new entrepreneurs in Brazil are less than 30 years old, we find this to be the case for more than 60% of the entrepreneurs responding to the demand shock. This is again consistent with the notion that the ability to rapidly respond to new local opportunities requires a degree of flexibility and risk tolerance that is uniquely possessed by the young.
While the results so far suggest that lifecycle considerations are important, it turns out that being young in itself is insufficient to explain entrepreneurial response. We find that among the young, those who have acquired certain skills through previous employment and education are more responsive to local opportunities. For example, it is those individuals who have the most industry experience, those that are more educated, and those that have worked in managerial occupations who are most responsive to these new economic opportunities.
Specifically, following Muendler et al. (2004), we classify individuals into generalists and spe- cialists according to their previous occupation, where the formers are those working in occupations that require multiple abilities and involve leadership, monitoring, and supervisory tasks. We find that within the young population, individuals in the generalist category are highly responsive to aggregate shocks, relative to specialists. In contrast, older individuals with generalist skills do not respond to these local opportunities. In addition, we followAutor et al.(2003) and use the occupa- tional data to identify occupations that are non-routine and require cognitive skills, which involve tasks that demand creativity, generalized problem-solving, and complex communications, such as selling, managing, and legal writing, among others. We find that young individuals who worked on cognitive non-routine tasks are significantly more responsive. Again, we find that older individuals who work in occupations that require high cognitive skills do not respond to the local shock. Finally,
we also find that among the young, more educated individuals are more responsive to local shocks by forming new firms.
Next, we compare the skills of the entrepreneurs who respond to local demand shocks with the skills of the average entrepreneur in the population. Focusing on young individuals, we find that the average young entrepreneur has relatively similar levels of skills when compared to the average in the population, measured by the generalist measure, cognitive non-routine occupations and past experience. However, these skill traits are much more pronounced among the responsive entrepreneurs who form businesses when local opportunities arise.
In sum, we find that entrepreneurial responsiveness to local economic opportunities is concen- trated among the young and the skilled. These findings are consistent with various theories that argue that entrepreneurship requires a variety of general business and managerial skills (Evans and Leighton (1989); Lazear(2005)).
Our finding that both age and skill matter for the individual-level decision whether to become an entrepreneur when local opportunities arise, suggest that several characteristics of the local economy may affect its entrepreneurial responsiveness. First, since the ability to create new firms hinges on access to finance (Evans and Jovanovic(1989);Hurst and Lusardi(2004)), and since young individuals have less time to accumulate wealth, we posit that in municipalities with better access to finance, we are likely to find an even stronger entrepreneurial response of the young. Moreover, Lucas(1988) andGennaioli et al.(2012) argue that the presence of other skilled individuals generates human capital externalities making it easier for potential entrepreneurs to learn how to start a business, suggesting that given the importance of skill, the overall stock of entrepreneurial knowledge might impact the firm creation response. Finally, given the importance of life-cycle considerations, younger demographics may lead to a stronger entrepreneurial responsiveness. In fact, younger age demographics may also lead to a potential indirect effects, as suggested by Liang et al. (2014).
Specifically, older populations may make it more difficult for younger individuals to move up the job ladder and thereby acquire the requisite skills for entrepreneurship. This suggests that younger individuals would generally be less responsive in economies with older populations.
We find suggestive support for all of these three hypotheses using cross-municipality regressions.
Younger individuals are indeed more responsive in municipalities with better access to finance, as proxied for by the number of banks or total value of credit to small businesses measured at the beginning of the sample. Moreover, younger individuals are also more responsive in municipalities where the population is endowed with more generalist and managerial skills and more entrepreneurs, consistent with human capital externalities. Finally, we find that individuals do take longer to acquire generalist and managerial skills in municipalities with older demographics. In line with this finding, we document that young individuals are more responsive to local opportunities in municipalities with younger demographics.
Our work relates to several strands of literature. First, as mentioned above, a variety of macroe- conomic studies have emphasized the crucial role that new firm creation plays in the amplification and propagation of aggregate economic shocks.3 We contribute to these studies by investigating the characteristics of the entrepreneurs who respond to local demand shocks by creating new firms. Our micro-level evidence highlights the importance of individual level heterogeneity, suggesting that the demographic characteristics of the local population, may affect the entrepreneurial responsiveness of economy to economic fluctuations.
Second, our paper contributes to a long-lasting literature on the nature and characteristics of entrepreneurs.4 In particular, little is known about the specific individuals who select into entrepreneurship in response to changes in local opportunities. Using rich individual-level data, we illustrate that these entrepreneurs are substantively different when compared to the average
3General equilibrium models of monopolistic competition linking firm entry and exit to aggregate fluctuations indicate the presence of various channels. Devereux et al.(1996),Chatterjee and Cooper(2014), andBilbiie et al.
(2012) are examples of models where entry of new firms generates greater product variety, while inJaimovich and Floetotto (2008) entry works through increased competition and lower markups. In related recent work, Clementi and Palazzo(2016) argue that increases in firm entry in response to aggregate shocks lead to large and persistent expansions because of lifecycle considerations. Sedlácek(2014) suggest the lack of startups during a downturn can lead to persistent employment declines in the economy, and Sedlacek et al. (2017) show that firm heterogeneity, and particularly the presence of high-growth startups, are key for aggregate gains. Several empirical studies further highlight how new and young firms act as important sources of job creation and employment (Haltiwanger et al.
(2013b); Pugsley and Sahin (2015)). In particular, Adelino et al. (2017) use US Census micro-data and regional variation in investment opportunities through Bartik shocks to show that it is the young and new firms that create the most jobs in response to positive local demand shocks.
4See, for example,Kihlstrom and Laffont 1979b;Blanchflower and Oswald 1998;Hamilton 2000;Moskowitz and Vissing-Jorgensen 2002;Hurst and Lusardi 2004;Hombert, Schoar, Sraer and Thesmar 2014, andHumphries,2016.
new entrepreneur in the population. Our evidence also highlights the importance of both life-cycle considerations and skill as important drivers of entrepreneurial dynamics.
Finally, our paper relates to a growing strand of literature that seeks to understand how de- mographic changes affect macroeconomic patterns and labor market dynamics.5 One of the most profound demographic transitions of the past 50 years has been towards aging populations. This trend is widespread, stemming from both declines in fertility rates and increased longevity. Our findings that young individuals are more responsive to local opportunities when they arise suggest that this trend may impact the entrepreneurial responsiveness of the economy. In that regard, our paper is also related to Kopecky (2017), who explore the relationship between aging populations and entrepreneurship across and within countries.
The remainder of the paper proceeds as follows. Section 2 provides our theoretical framework.
Section 3 describes the various data sources used in the analysis, while Section 4 describes the empirical strategy that combines local historical endowment of agricultural production with cur- rent movements of global crops prices. Section 5 presents municipality-level aggregate results, and Section 6 describes the individual-level analysis and reports the key results of the paper. Finally, sec- tion 7builds on our main findings and tests several theories that suggest that characteristics of local economies may affect entrepreneurial responsiveness. We test these theories using cross-municipality differences. Section 8 concludes.
II. Theoretical Framework
To motivate our empirical analysis, we construct a model of a local economy featuring exogenous profitability shocks to the local resource sector and an entrepreneurship decision in the spirit of Lucas (1978). The exact mathematical details of the model can be found in the appendix. Here, we describe the basic structure of the model and its key predictions. The local economy comprises three sectors, producing commodity goods, tradable goods, and local nontradable goods. The
5For example, see Jaimovich and Siu(2009);Jones(2010);Backus et al.(2014);Gagnon et al.(2016); Engbom (2017).
commodity and tradable sectors provide a single homogenous good, while the local nontradable sector is comprised of a continuum of differentiated goods, indexed by varieties !.
We assume that all individuals in the local economy have Cobb-Douglas preferences over the tradable and nontradable goods, with nontradable consumption being given by a standard CES aggregator. This implies that all individuals spend a constant fraction of their total income on tradable goods and a constant fraction on the nontradable composite. Each individual inelastically supplies one unit of labor. There are both skilled and unskilled individuals. Unskilled individuals can only supply a single unit of unskilled labor. Skilled individuals, however, have a choice over their occupation along the lines of Lucas (1978). They can either provide a single unit of skilled labor or choose to become an entrepreneur in the nontradable sector, producing a single differentiated vari- ety. Entrepreneurship involves non-pecuniary fixed costs which are heterogeneous throughout the population. We finally assume that unskilled labor is perfectly mobile within the overall economy, while skilled labor is immobile, so that the total stock of skilled individuals in the local economy is fixed.
The commodity sector and tradable sector are perfectly competitive, each with a composite firm producing a homogeneous good. The prices of the commodity good and tradable good are set by global demand and thus taken to be exogenous. The commodity sector hires unskilled labor, while the tradable sector hires both skilled and unskilled labor. Entrepreneurs operating in the nontradable sector compete via monopolistic competition. Each entrepreneur hires unskilled labor and operates a CRS production technology. In equilibrium, firms and individuals optimize, supply equals demand for skilled and unskilled labor, and the marginal entrepreneur must be indifferent between skilled labor and entrepreneurship.
Now suppose that there is an exogenous increase in the price of the commodity good. We have the following result:
Proposition 1. An increase in the price of the local commodity good leads to increased employment and new firm creation in the local nontradable sector.
To understand this result, first suppose that the number of entrepreneurs is fixed at its initial level. The higher price raises the marginal revenue productivity of the commodity sector . In the absence of unskilled mobility, this would raise unskilled wages. However, since unskilled workers are perfectly mobile, the increased revenue productivity leads to in-migration of unskilled workers until the marginal revenue productivity of the commodity sector is equal to the exogenous wage.
This also implies that skilled wages will remain constant. This increase in the total number of local unskilled workers raises aggregate income, which increases the demand for nontradable goods.
Since demand is homothetic and marginal costs are unchanged, the price of the non-tradable goods does not change. Therefore, there is increased output and higher employment in the nontradable sector. Under the assumption that the number of entrepreneurs does not change, this leads to higher entrepreneurial profits.
However, this then implies that entrepreneurial profits are now higher than skilled wages. If we now allow for the number of entrepreneurs to adjust, there will be firm entry. Skilled workers with sufficiently low non-pecuniary costs will become entrepreneurs, increasing the number of dif- ferentiated varieties, reducing entrepreneurial profits through greater competition, and increasing skilled wages. This will continue until the the marginal entrepreneur is again indifferent between entrepreneurship and skilled labor.
Proposition 2. If the skilled population is small or nonpecuniary costs are high among the popu- lation, there is less firm entry and the employment increase occurs more on the intensive margin than the extensive margin.
This result is intuitive. The smaller the skilled population, the less individuals there will be able to take advantage of the new economic opportunities, so the less firm entry there will be. Likewise, if nonpecuniary costs of entrepreneurship among the population are high, less individuals will find it worthwhile to switch from skilled labor to starting a business, again leading to a smaller firm entry response. If there is little firm entry, the the increase in output and employment will be largely accommodated by the existing nontradable firms.
In our empirical work, we will see that local agricultural commodity shocks do indeed lead to increased employment and firm entry in the nontradable sector. We then study the demo- graphic determinants of the individual firm entry response. That is, we seek to understand how the non-pecuniary costs of entrepreneurship are distributed among the population. We also seek to understand exactly which skills matter, and to what extent, in the ability of individuals to respond to new entrepreneurial opportunities.
III. Data
In this section we discuss the main datasets used in our analysis. We start by describing the RAIS dataset, which provides matched employer-employee information on all employees in the formal sector in Brazil. We supplement this data with aggregate municipality-level data on loans and firm credit. We further employ data on municipal agricultural crop endowments in Brazil, as well as data on global commodity prices.
A. Employer-Employee Data
The RAIS database (Relacao Anual de Informacoes Sociais) is an administrative database from the Brazilian Ministry of Labor (MTE) which provides individual level data on the universe of formal sector employees in Brazil. RAIS is widely considered a high quality Census of the Brazilian formal labor market (Dix-Carneiro (2014)). The database, created in 1976, is used by several Brazilian government agencies (such as the Brazilian Central Bank) to generate statistics for the Brazilian economy. The RAIS database also forms the basis for national unemployment insurance payments and other worker benefits programs. As a result, ensuring the accuracy of the information is in the interest of both firms (who would otherwise be subject to monetary fines) and individuals (who want to be eligible to receive government benefits), as well as the central government.
RAIS contains information on the firm and the establishment of each employee, including tax identifiers, locations, industry, and legal status. At the individual level, RAIS includes employee-
specific identifiers, called PIS, which allow individuals to be tracked over time and across firms (as well as across establishments of the same firm).6 Similar to other employer-employee matched data, for each employee we observe payroll, tenure in the firm, and hiring and firing dates. RAIS additionally has rich personal data on gender, nationality, age and education, as well as a few less commonly available variables such as hours worked, reasons for hiring and firing, and contract details. Finally, each employee is assigned to an occupational category specific to her current job.
There are more than 2,000 such categories, which follow the detailed Brazilian’s classification of jobs (Classificacao Brasileira de Ocupacoes - CBO) that is similar to the International Standard Classification of Occupations (ISCO-88).
Using data on occupations, we are able to identify individuals that are managers or CEOs of a firm, as well as lower ranked workers, both blue collar and white collar. Following standard practice in the entrepreneurship literature (e.g. Kerr et al.(2015);Babina(2015)), we define an entrepreneur as the CEO or the top paid manager of a new firm in the year of birth. Specifically, if no worker is classified as CEO or manager, we use the highest paid worker in the firm.7
Furthermore, the detailed data on occupations allow us to classify workers based on skills.
FollowingMuendler et al. (2004), we classify individuals into generalists, i.e. those working in oc- cupations that require multiple abilities and that involve leadership, monitoring, and supervisory tasks, and specialists, i.e. those working on specialized tasks and occupations. Additionally, follow- ingAutor et al.(2003) andGathmann and Schönberg (2010), we distinguish between workers who perform different types of tasks. Non-routine cognitive tasks require creativity and problem-solving ability, as well as negotiation and coordination skills. Non-routine manual tasks require physical work together with the ability to adapt to different situations. Finally, routine tasks are all other tasks based on well-specified processes and activities.
We focus on individuals that are within the ages of 18 and 65 and who have wage data for
6Individuals with multiple jobs in a given year therefore appear multiple times. Following standard practice in the literature (Menezes-Filho, Muendler and Ramey,2008), we keep only the highest paying job of the individual in a given year. If there are two or more such “highest paying” jobs, we break ties by keeping the earlier job.
7Our results are robust to various definitions of “entrepreneur.”
at least 3 years, during the 1998-2014 period. We restrict our focus to this time period due to the availability of information on agricultural resources at the municipality level. Under these restrictions, the sample includes roughly 69 million individuals.8
In Panel A of Table 1, we provide summary information on the relative importance of Brazilian industries. The two largest industries in the economy are the non-tradable and services sectors, which capture 54% and 20% of the annual number of firms, and 27% and 20% of annual employment respectively. Panel A also documents the annual creation of new firms across industries, with most new firms being created in the non-tradable and services sectors.
In the empirical analysis, we focus on municipalities as the local economic unit and explore how municipalities respond to plausibly exogenous economic shocks triggered by fluctuations in global commodity prices. Panel B of Table 1 provides municipality level summary statistics. The average municipality in the sample has a population of 24,122 and the GDP per capita is 3,093 (USD 2000).9 In our sample period, there is an average (median) of 237 (47) firms and an average (median) total number of formal private sector employees of 3,143 (342) per municipality, with significant dispersion in size across regions. The average (median) number of new businesses created in a given municipality in a given year is 32 (7). Once again, there is significant heterogeneity across municipalities. The RAIS data allow us to explore the characteristics of the newly created firms in the economy. In Panel C of Table1we find that on average, a newly created firm survives 6.8 years (median is 5) and, conditional on survivorship, employs 14 workers two years after its birth and 18 workers after five years.
Panel D of Table 1 provides summary statistics for all individual formal sector employees in our sample, excluding the set of entrepreneurs. On average, 61% percent of the workers are male.
Workers have an average of 11.45 years of education and the median level of education is high school (12 years). Based on the occupational status of the workers, we find that most workers can
8In the municipality-level analysis, we aggregate data using all individuals in this sample. When we move to the individual-level analysis, for computational reasons, we extract a random sample of 5% of all individuals on which we perform the analysis.
9All summary statistics on the municipalities are computed using the full aggregate municipality-level panel data from 1998 to 2014.
be characterized as either Blue Collar (48%) or White Collar (41.6%), while only a small fraction consists of Managers and CEOs (3.6%).1011 Finally, when we classify individuals according to their previous occupation into generalists and specialists, we find that 19.3% of the workers are in a job which entails a generalist skill set.
In Panel E of Table 1, we provide summary statistics on the the population of entrepreneurs in our sample. We find that, compared to the overall population, entrepreneurs are relatively more likely to be female and to have a higher educational level, with 12.11 years of education on average.The distribution is also skewed towards higher hierarchical jobs in their place of employment immediately preceding the founding a new firm, with 40.2% in Blue Collar positions, 50% in White Collar position , and 4.5% serving as Managers or CEOs.
B. Loans and Banking Sector
We supplement the RAIS data with municipal level data on the number and dollar amount of all loans to local businesses, as well as information on the location of bank branches in Brazil. These data are obtained from the Brazilian Central Bank datasets (Banco Central do Brasil, BCB). We also obtain confidential loan-level data from the BNDES, which represents the second largest national development bank in the world (after the Chinese Development Bank), and is a major lender of Brazilian companies. BNDES provides a significant share of long-term bank lending in Brazil and is amongst the largest sources of investment in industry and infrastructure (Colby, 2012). For each loan, we have information on the loan amount, the interest rate (and type), and tax identifier of the firm receiving the loan.
Panel B of Table1 reports summary statistics on the financial characteristics of municipalities.
We report information on both total volume of credit from private banks and the BNDES, expressed in millions of USD 2000, and total number of BNDES government loans. The data shows significant
10We do not observe occupational status for the remaining workers in the data.
11We match the CBO classification to the International Standard Classification of Occupations (ISCO-88) using the procedure outlined in Muendler et al. (2004). This correspondence allows us to categorize workers into four organizational layers, following a framework close toCaliendo and Rossi-Hansberg(2012). From bottom to top layers they are: Blue Collar, White Collar, Managers, CEOs. Please see Colonnelli and Prem(2017) for more details on the data construction.
variation across municipalities, with an average of 20.794 millions USD going to local business from private financial institutions (median is 4.129), and an average of additional 2.75 millions USD coming from the BNDES (median is 0.228).
C. Agricultural Crops in Brazil
The Brazilian economy relies heavily on agriculture. For example, Brazil is among the largest pro- ducers in the world of coffee, sugarcanes, orange juice, soybean, corn and ethanol. These crops, and others, provide the basis for the large agribusiness industry in Brazil, which represents 22% of Brazil’s GDP, a third of its employment, and almost 40% of its export (PwC(2013)). The agribusi- ness industry captures not only farming production, but also the supply of farming inputs such as machinery and seeds, as well as the selling and marketing of farm products, such as warehouses, wholesalers, processors, and retailers.
The empirical strategy in this paper relies on local demand shocks caused by fluctuations in the profitability of the local agricultural sector driven by global commodity prices. We obtain information on agricultural crops from the Brazilian Institute of Geography and Statistics (IBGE), which is responsible for the Brazilian census as well as most of the statistical analyses of the Brazilian economy. The data provides the annual production of all 66 different agricultural crops, at the municipality level, for the period 1993-2014. We standardize the different crops to the same unit measure (i.e., tons) to construct a panel dataset of the universe of agricultural crops production by Brazilian municipalities.
Panel B of Table 1 illustrates that the average aggregate dollar value of local crops in a mu- nicipality is equal to approximately 89% of local GDP, with the median equal to 12.6% of local GDP. Similarly, the value of local crops per capita is on average $2,926. Figure 1 illustrates the wide spatial distribution of agricultural resources across municipalities. Municipalities are divided into quintiles based on the production value of natural resources relative to GDP in 2000. The bottom quintile have production values of roughly 1% to 5% of municipality GDP. In contrast, in the top quintile, municipalities have production values worth more than 45% of local GDP. The
figure illustrates significant heterogeneity across municipalities. In fact, the heterogeneity across municipalities is even wider, given that different municipalities specialize in different portfolios of agricultural products, but this is not reflected in the figure.
International commodity prices are obtained from the Global Economic Monitor (GEM) Com- modities database of the World Bank. We focus on the period 1998-2014. For each crop, we create a yearly measure of commodity prices starting in 1998 by taking the average price within the year.
In some cases, there may be a single price that matches to multiple crops. For example, the price of tea is assigned to both “indian tea” and “yerba mate.” Hence, we consolidate several agricultural crops to match prices. We standardize all units of measure to US dollars per ton. In the final dataset, we have 17 different commodities present in Brazil which are traded on the international commodity markets. The detailed distribution of these agricultural crops across municipalities is provided in Table A.1of the Appendix.
IV. Empirical Strategy
Given our focus on the entrepreneurial response to new local opportunities generated by fluctuations in local income, we will focus specifically on new firm creation in the non-tradable sector, which is especially dependent on local demand. This approach is similar to Adelino et al.(2017), who show that job creation by new firms accounts for the bulk of new non-tradable employment in response to local demand shocks.12 In this paper, we seek to understand the defining characteristics of the actual individuals who are responsible for the creation of these new firms when such local opportunities arise. Simply running regressions of new firm creation on local income, however, is confounded by reverse causality concerns. In particular, unobserved shocks to the investment opportunities of particular sets of individuals could mechanically impact local income. For example, the introduction of local government programs providing start-up incentives to the young would likely increase both the firm creation rate of the young as well as local income. To the extent that such programs are
12Adelino et al. (2017) do not measure local demand shocks in the same way as we do, but instead identifying fluctuations in local income using the shift-share approach ofBartik(1991) and Blanchard and Katz (1992), which interacts local manufacturing shares with national trends in manufacturing.
unobserved by the econometrician, regressions of firm creation on local income would reflect this reverse causality.
To address this issue, we create a measure that isolates plausibly exogenous changes in munici- pality level local income over time. To do so, we identify fluctuations in the value of locally produced agricultural commodity crops, and thus also in the profitability of the local agricultural sector, by interacting the local agricultural endowment with movements in global commodity prices. Such com- modity price fluctuations are an important source of economic variability for emerging economies, as well as for developed economies rich with natural resources (Fernandez et al. (2017),Allcott and Keniston (2017)). Moreover, as shown by Allcott and Keniston (2017) in the context of US oil and gas booms and byBenguria et al.(2018) in the context of Brazil, such shocks do appear to increase local demand due to higher levels of local income, leading to an increased employment in the local non-tradable sector.
The agribusiness sector in Brazil is large, highly developed, and highly diversified. Different municipalities are endowed with different types of agricultural crops that they can grow locally.
We calculate the local value of a crop in a given year as the product of the local crop quantity (Q) with its unit price (P) in international commodity markets. While international prices are exogenous to current municipality-specific economic conditions, quantities are less likely to be so.
We therefore hold endowments fixed at their 1998 level, prior to the start of our sample period, so as to remove the endogenous component in the fluctuations of commodity values. We construct a proxy for the 1998 local endowment by averaging production quantities in the five years preceding the beginning of our analysis sample, i.e. between 1993-1997. Using this approach circumvents potential endogeneity concerns because historical production is likely to capture the (exogenous) spatial endowments of agricultural crops, rather than potentially endogenous production activity, which may correlate with unobserved local shocks. This approach is standard in the literature (see, for example,Dube and Vargas (2013)). These historical endowments of agricultural crops are persistent due to the accumulation of local expertise and economic activity over long periods of time, and because of the physical characteristics of the regions such as climate and soil.
Specifically, let Qkj,98 be our proxy for the regional endowment of crop k in municipality j in 1998, measured by the average production in the past 5 years. Let Pkt be the international price of crop k in year t. Thus, the annual Crops Index (CI) for municipality j in year t is the sum over all crops of local agricultural endowments, multiplied by the respective time-varying international prices:
CIjt =X
k
Qkj,98⇤ Pkt
Variations in this municipality-level measure allow us to approximate the ideal natural experiment where we randomly shock local income over time. The endowment part of the formula generates cross-sectional variation in the pre-existing exposure of different municipalities to different agri- cultural resources. International commodity price fluctuations generate time-series variation that is plausibly independent of shocks to local investment opportunities. Together, they provide a municipality-year varying series of exogenous demand shocks generated by the differential exposure of different Brazilian municipalities to the changing global value of agricultural commodities.
Our empirical strategy is inspired by the shift-share approach of Bartik (1991) and Blanchard and Katz (1992), which interacts local manufacturing shares with national trends in manufacturing employment to identify local income and demand shocks.13 The primary identification concern with this approach is that unobserved municipality level shocks in Brazil could impact global commod- ity prices, biasing the results. For example, one such concern is that local government programs designed to incentivize new firm creation might also lead to increased local agricultural output, driving down international prices. We address this and other concerns later in the text.
Figure 2 illustrates the variation we observe in the value of municipal endowments of crops, as captured by the annual Crops Index. Specifically, in Figure2we plot the residual value of crops in a municipality after removing municipality and year fixed effects. The thin grey lines provide the time series of these residuals for a 10% random sample of municipalities in our sample. The lines are median (solid line), 10th and 90th percentiles (dashed lines) of the distribution of residuals in each
13This strategy has been widely adopted by the economics literature. See, for instance, Gallin (2004), Saks and Wozniak (2011), Charles et al. (2013), Diamond (2016), and Adelino et al. (2017).
year. As the figure illustrates, there is both significant cross-sectional variation within a given year and considerable time variation within a given municipality in the value of agriculture commodities.
In what follows, we examine the impact of local endowment largest shocks in the 10th percentile relative to the municipality mean. Specifically, let Zjt be equal to one if the local agricultural endowment in municipality j is in the top 10th percentile change in the price distribution in year t relative to the municipality mean, and equal to zero otherwise. We consider municipality j to be “treated” in year t if Zjt = 1. As we discuss below, the choice of a binary shock allows us to transparently estimate the characteristics of the individuals who create new (non-tradable) businesses in response to local demand shocks. This choice does not affect the interpretation of our findings, as the results hold when using the binary or the continuous measure of the shock. We further discuss this and other robustness checks below.
V. Municipality-Level Analysis
We start by estimating the impact of global commodity price fluctuations on municipality level economic activity. As discussed above, our strategy interacts this time-series variation with cross- sectional differences in municipal endowments of agricultural crops; that is, we use municipality-year level variation in the Crops Index as a source of plausibly exogenous fluctuations in local income and demand. Our main specifications use the binary version of the shock, described earlier. Specifically we estimate the following model:
Yjt = ↵j+ Zjt+ Xjt+ ujt
where Yjt is the outcome of interest, ↵j are municipality fixed effects, Zjt is the binary shock of largest 10th percentile change in Crops Index, and Xjtis a set of controls that includes year dummies and the log-population. In the Appendix we illustrate that our results hold for continuous and other definitions of the shock.
A. Local Employment and Firm Creation
The main results are presented in Table 3. We find that positive shocks to the value of local crops generate both higher income and higher employment in the treated municipalities. Treated municipalities experience a highly significant increase of 4.1% in the level of formal employment (column 1) and a highly significant increase of 2.9% in total local income, as measured across all local firms. This increase in total local income is driven primarily by higher employment levels rather than higher wages.
Higher levels of local income suggest new profit opportunities available to be exploited by po- tential entrepreneurs, particularly in those sectors which are highly dependent on local demand conditions. We see that the commodity price shock does indeed lead to an increase in the total number of local firms. As reported in column 3 of Table3, there is a statistically significant increase of 3.7% in the number of local firms following the shock. Importantly for our purposes, this increase is primarily driven by the creation of new firms (column 4), which increases by 2.6%, rather than a higher likelihood of survival of existing firms, which instead seems unaffected given the small and statistically insignificant -0.004 effect on firm closures (column 5).
Table 4 illustrates the impact of the shock by economic sector, which we categorize using the Brazilian CNAE industry codes into Agriculture and Mining (columns 1), Manufacturing (columns 2), Non-tradable (columns 3), and Services (columns 4). Panel A focuses on employment, and shows a statistically significant increase in employment levels in all but the Services sector. As illustrated in column 1, this finding is consistent with rising commodity prices having a positive direct effect (8.4%) on the sectors responsible for the production of these commodities (Agriculture and Mining). Moreover, there seem to be a positive spillover effects on other industries that are potentially connected through various linkages or tied to local demand, as illustrated in columns 2 and 3. When studying the aggregate sectoral impact on number of firms, in Panel B, we find that the vast majority of new firm creation is accounted for by firms in the non-tradable sector (column 3), where we observe a highly statistically significant increase of 5.5%, compared to small
and statistically insignificant effects on other sectors.
Our aggregate results emphasize the importance of entrepreneurship for the dynamics of local economic activity. Our findings that local income shock lead to a significant entrepreneurial response in the non-tradable sector is consistent with Adelino et al.(2017) who study the US. These results provide a preliminary step towards our main analysis in which we study the entrepreneurs who account for the firm creation response to the local demand shock.
We next ask whether the creation of economic activity we observe in the non-tradable is econom- ically relevant. A potential concern is that new firms created in the non-tradable sector in response to the local demand shock may be short-lived or may not contribute to significant employment creation in the long-run, relative to the average firms created. We explore these concerns in Table 5. Using firm-level data, and focusing only on firm entry, we estimate specifications where we use as dependent variables indicators for whether a new firm survives for at least 3 and 5 years (columns 1 and 2, respectively), and indicators for whether a new firm has at least 3 or 5 employees five years after creation (columns 3 and 4, respectively).
We find that, if anything, firms created in response to the positive local demand shocks are significantly more likely to survive both after 3 and 5 years and that, conditional on survival after 5 years, these firms are likely to be larger when compared to the lifecycle of the average new firm.
While survivorship and size are just proxies for firm success, these results suggest that new firms who respond to local demand shocks are an important and persistent propagation mechanism. We further corroborate these results in column 5 of Table 5, where we find that newly created firms are also more likely to be incorporated, which can be viewed as an ex-ante proxy for future growth plans of the businesses created (Levine and Rubinstein(2017)). These findings are consistent with Sedláček and Sterk (2017) who find that the success of firms is strongly influenced by aggregate conditions at the time of their entry.
B. Robustness and Additional Results
In this subsection, we further describe additional tests to probe the robustness of our results and provide further characterization of the main aggregate effects. All the results are reported in the Appendix.
Alternative Treatment Definitions
First, TableA.2reports the main estimation results when we vary our designation of what constitutes a treated municipality, as one may be worried that the choice of a top 10% binary shock is arbitrary.
We find that all of the main findings continue to hold, when we focus on more moderate local endowment shocks defined as being in the top 25th percentile relative to the municipality mean (second row).
As further robustness, we also find statistically significant negative effects on local economic outcomes when we instead look at negative endowment shocks. In particular, as reported in the third row of TableA.2, when defining a negative endowment shock to be in the bottom 10% relative to the municipality mean, we find a 5.9% decline in local employment and 4.9% decline in local total income. The number of firms falls by 4.0%, driven again by a decline in firm entry instead of increased closures of existing firms. We find very similar findings if we define a negative endowment shock to be in the bottom 25% relative to the municipality mean. These results are reported in row 4 of Table A.2.
Finally, all of our results are robust to using a continuous log version of the shock (fifth row), which estimates the the elasticity of new firm creation and other outcomes to the value of the local agriculture endowment. For example, 10% increase in the value of the local endowment is associated with a 1.54% increase in employment and a 0.68% increase in the inflow of new firms. All estimated effects are highly significant, except for the small and insignificant effects on firm closures.
Influencing Global Commodity Prices
The key endogeneity concern with our approach is that the local agricultural sector is sufficiently large relative to global production so as to potentially influence international prices. If this were true, then unobserved municipality level shocks impacting local firm creation, such as government incentive schemes, might also impact global commodity prices and thus bias the results. Indeed, Brazil is a leading global player in the production of crops, accounting for more than 10% of world’s exports for some of them (e.g., sugar cane, coffee, soybeans, yerba mate, tobacco).
Of course, it is important to note that our empirical strategy relies on within-country variation in the value of agricultural endowments, and therefore the concern is only be valid to the extent that single municipalities can influence global prices. Regardless, to test the robustness of our results, we re-estimate the main specifications dropping those municipalities with high levels of production of specific crops. In particular, we complement our municipality-level data with data from the United Nations Food and Agriculture Organization (FAO) to compute the share of world production of each specific crop accounted for by Brazilian municipalities. Panel A of Table A.3 reports the results after dropping 42 municipalities that have ever produced, in any given year, 1% or more of the world production of any commodity in the period 1996-2015, while Panel B reports even more conservative results, obtained after dropping the 145 municipalities with at least a 0.5% share. The results remain unaffected in both cases.
Persistence of Treatment Effects
Finally, we explore the persistence of the effects generated by the local endowment shocks. As we may expect, new firm creation, and economic activity more generally, respond to changes in local opportunities, and the effect is persistent but decrease gradually. TableA.4reports our main results for different lagged definitions of the binary treatment variable. While the response is strongest in the year of the shock (especially for new firms), we find that local economic activity continues to positively respond one to four years after the commodity endowment shocks, and gradually
decreasing.
VI. Individual-Level Analysis
In this section we move to our primary analysis, which aims to determine the key characteristics of those individuals who respond to local demand shocks by creating a new firm. We model the decision to start a (non-tradable) business using a binary choice linear probability model. Let the binary indicator variable Tijt denote the decision in year t of an individual i in municipality j to become an entrepreneur. We again let Zjt = 1 denote an exogenous increase in local demand in municipality j, as proxied for by the local agricultural endowment shocks. We estimate the following linear probability model:
Tijt = ↵j+ t+ · Zjt+ "injt, (1)
where ↵j denotes municipality fixed effects, and t denotes year fixed effects. Here, captures the direct effect of the local endowment shock on individual firm creation. In this analysis, we focus on individuals that are already in the region rather than individuals that migrate from a different region, to ensure that we can identify those individuals who experience the change in local demand and investment opportunities.
In this section we explore the role of age and individual life cycle on the one hand, and skill and experience on the other hand, as potential determinants that can explain the entrepreneurial responsiveness of individuals to local opportunities.
A. The Importance of Lifecycle Considerations
What determines the entrepreneurial response to aggregate shocks? According to standard models such asLucas(1978), ability is the relevant dimension along which individuals sort into entrepreneur- ship. In this type of model, to the extent that ability is an innate characteristics, the age profile of the population does not matter per se. Other theories, however, would predict that age does play
a prominent role in the decision to start a business in response to changes in local opportunities.14 On the one hand, if ability reflects skills accumulated over time as individuals move up in their career, and such skills are necessary to take advantage of changes in local opportunities, it is reasonable to expect that individual responsiveness to aggregate shocks would increase with age (Lazear,2005;Evans and Leighton,1989). Similarly, to the extent that downpayment and financing constraints affect the ability to start a new business, older individuals again may be more responsive to new opportunities, having had more time to develop the necessary personal wealth (Quadrini, 1999). On the other hand, young individuals may be the most responsive if, for example, they have a greater tolerance for risk, limited outside options, or an overall higher degree of flexibility in their personal circumstances.15 Ultimately, the extent to which any of these forces matter, and their relatives magnitudes, is an empirical question.
Relying on the econometric framework outlined in the previous section, we take this question to the data by investigating the heterogeneous response of different types of individuals to our specific local demand shocks. We center our analysis in this sub-section squarely around the role of age.
Figure3reports the increase in entrepreneurial rates in response to the shock, estimated according to model (1) for different age quartiles. The results clearly illustrate that it is young individuals, those in the bottom quartile of the age distribution, who significantly respond to the shock. The likelihood of becoming an entrepreneur increases by 0.45 (out of 1000) under a positive shock. The increase becomes statistically insignificant and close to zero for individuals in the remaining quartiles of the age distribution. These findings are consistent with lifecycle considerations playing a key role in the individual entrepreneurial response.
We refine this analysis in Table6, in which we show that the impact of age on the firm creation response is robust to the inclusion of a variety of other demographic characteristics. Column (1) shows the simple entrepreneurial responsiveness to the shock, controlling only for year and munici-
14SeeParker(2018) for a comprehensive review.
15See, for example, Miller 1984a;Reynolds and White 1997;Levesque and Minniti 2006;Delmar and Davidsson 2000;Arenius and Minniti 2005;Rotefoss and Kolvereid 2005;Wagner 2006;Bergmann and Sternberg 2007;Uusitalo 2001.
pality fixed effects. As the table reports, the shock leads to an increase of 0.179 (out of a 1000) in the probability of becoming an entrepreneur. This reflects a 6.2% increase in entrepreneurial activity in response to the shock when compared to the average flow of entrepreneurs in the economy (2.89 out of a 1000). In column (2), we interact municipality fixed effects with dummies for whether indi- viduals are at the bottom quartile of the age distribution. This specification therefore accounts for municipality level heterogeneity in the age composition, and yet the main effect remains unchanged.
In column (3), we add an interaction term between the treatment variable and the young dummy.
This column relays the same message as Figure 3. This group exhibit a striking 5.7 times larger responsiveness when compared to the rest of the population.16 Relative to the average flow to en- trepreneurship in the economy, the shock leads to an increase of 12.5% in the probability to become an entrepreneur for the young, when compared to the rest of the of the population (0.3632.89 ⇡ 12.5%).
We next show that this result is extremely robust in sign and magnitude to the inclusion of several other controls, such as sector fixed effects in column 4 (defined as the sector of the newly created firm), education dummies in column 5, and indicators for whether the individual had a generalist occupation (column 6) or a white collar one (column 7), in the year before founding the new firm.
We next compare the distributional characteristics of the responsive entrepreneurs to the average new entrepreneurs in the population. The procedure for doing this is described in the Appendix.
We find that the average individual who starts a business tends to be younger relative to the overall population, but that this feature is significantly more pronounced among responsive entrepreneurs.
As Figure 4 illustrates, roughly 40 percent of individuals who start a new business are at the bottom quartile of the age distribution, younger than 29 years old. However, more than 60 percent of entrepreneurs who respond to the demand shocks are at the bottom quartile of age. As Figure 4 illustrates, the entire age distribution of those responsive entrepreneurs is tilted towards younger demographics, when compared to the average new entrepreneurs in the economy. These results are consistent with the notion that the ability to respond quickly to new economic opportunities depends crucially on flexibility and the willingness to take risks, traits that younger individuals are
16Specifically, the magnitude is obtained as: 0.0773+0.363 0.0773 ⇡ 5.69
significantly more likely to possess.
B. Do Skill and Experience Matter?
So far we have illustrated that young individuals are disproportionally more likely to start a busi- ness in response to local economic shocks. Lifecycle considerations therefore seem important in understanding entrepreneurial dynamics and shock propagation mechanisms. Do skills also affect individuals’ entrepreneurial responsiveness to local economic opportunities, as suggested by Evans and Leighton(1989),Lazear (2004) and many others? We evaluate this question by exploring het- erogeneity in firm creation within the population of young individuals, focusing on a battery of proxies for an individual’s skill set.
Table 7, Panel A reports the estimates of the treatment effect for individuals in the bottom quartile of the age distribution, split according to these observable characteristics. The analysis shows that skills and wealth are significant determinants of individual responsiveness within the young (and responsive) entrepreneurs. First, we find that within the young group, individuals who were previously employed in non-routine and cognitive occupations are significantly more responsive than others to the rise of local opportunities. The magnitude of the coefficients suggest that the former group is a staggering 52 times more responsive than the latter (Panel A, columns (1) and (2)). Similarly, individuals employed in generalist occupations are almost three times as responsive as those who do not have this type of experience (Panel A, columns (3) and (4)). Furthermore, columns (5) and (6) of Panel A show that within the young population, responsive entrepreneurs are also more likely to have at least a high school diploma. All these results are consistent with Lazear (2004) and other empirical studies emphasizing the importance of ability and skills for entrepreneurial responsiveness.17
Table 8 Panel A reports a similar analysis for other proxies of ability, such as high pay growth relative to the other workers in the same municipality (Panel A, columns (1) and (2)). Again, among individuals with above median pay growth, the shock increases the probability of starting a
17Examples includeWagner(2006);Djankov et al.(2005);Lazear(2005).
business by 0.731 in 1000, while it is only 0.202 in 1000 for those with lower relative pay growth.
Similarly, columns (3) to (6) of panel A show that, within the group of young individuals, both more experienced and wealthier individuals are more likely to respond to shock and create a new firm. Young individuals with higher than median experience are 2.8 times more responsive than individuals with lower experience. In terms of accumulated earnings, those in the upper half of the distribution are 66% more responsive than those in the bottom half.
Strikingly, but conceptually in line with our previous results, the two panels B in Table 7 and Table 8 show that there is no statistically significant heterogeneous response when we perform the analogous analysis for the old individuals, i.e. exploring the entrepreneurial responsiveness of skilled individuals in the top three quartiles of the age distribution.
Finally, as we did with age, we compare the skill characteristics of those (young) individuals who start a firm in response to the local demand shock with the skill characteristics of the average young entrepreneur. The results are shown in Figure5. We first note that the average young entrepreneur is, in fact, quite similar to the average young individual in the population in terms of skill. While the average young entrepreneur is slightly better educated, she has similar levels of generalist and non-routine cognitive skills to the average population, as well as similar levels of recent experience and pay growth.
In contract, all of these traits are significantly more pronounced among those entrepreneurs who specifically create new firms in response to the local demand shocks. For example, while only 18% of the entrepreneurs in the population worked previously in occupations that we classify as non-routine cognitive, it is the case for almost 50% of the responsive entrepreneurs. Similarly, while roughly 20% of entrepreneurs are defined as generalist, responsive entrepreneurs are defined as generalists in 40% of the cases. Similar findings apply to individuals experience and pay growth. We find that 85% of the responsive entrepreneurs experience a pay growth above the median, and 70% of them have more experience than the median. In contrast, among the overall entrepreneurs in the population these categories apply in only 45% and 50% of the cases, respectively.
In summary, within the young population, responsive entrepreneurs are more likely to be expe-
rienced, wealthy, educated, on a positive pay growth path, and to have generalist and non-routine cognitive skills. That is, after conditioning on age, skill, and experience all affect entrepreneurial responsiveness to local opportunities within the young.
C. Robustness and Additional Results
Negative Treatment Effects
As previously discussed for the municipality-level analysis in Table A.2, our results hold also when we explore alternative treatment definitions, and we also find a decrease in economic activity in response to declines in global commodity prices. In Table A.5, we further explore the individual responsiveness to shocks when economic opportunities exogenously decrease. In columns 1 to 3, we show that the likelihood of becoming an entrepreneur is lower (though not statistically significant) when the treatment is defined as a shock in the bottom 10th percentile change in the price distri- bution in year t relative to the municipality mean. Furthermore, in columns 4 to 7 and consistent with the presence of various constraints limiting the entrepreneurial activity of the young popula- tion, we find that the young are significantly less responsive to negative economic shocks, as seen in the coefficients on the interaction term T reatment X Y oung and independently of the addition of multiple sets of controls. These results illustrate that young individuals are particularly responsive to changes in local opportunities, both when such opportunities arise and when they disappear.
Attrition due to Early Formation of Businesses
An important concern may arise that the results are mechanically driven by compositional changes in the set of entrepreneurs over time. In particular, one may be worried of attrition. Young individuals start businesses at a young age, and therefore may not respond to local opportunities by forming a new businesses as they are already business owner. In the Appendix we provide a robustness test to address this concern.
We re-define the dependent variable F ounder as P reF ounder, to be equal to 1 in a given
