Wealth, Industry and the Transition to Entrepreneurship Johanna Francis ∗ Department of Economics Fordham University

Berna Demiralp Department of Economics Old Dominion University

MAY 2008

Abstract Although the debate about the effect of wealth on entrepreneurship is now almost two decades old, there is little consensus among researchers about the significance of wealth as a determinant for self-employment. We re-visit the relationship between wealth and entrepreneurship using data from the National Longitudinal Survey of Youth. Like Hurst and Lusardi (2004), our results suggest the relationship between wealth and the probability of entering entrepreneurship is nonlinear. However, unlike Hurst and Lusardi, we find the probability of entrepreneurship increases at an increasing rate with wealth, starting at lower quantiles of the wealth distribution. We also observe that the aggregate relationship masks differences among entrepreneurs with respect to their industry. While high capital requirement industries and professional services display a convex relationship between wealth and the probability of self-employment, low capital requirement industries display a concave relationship. Since we find a positive relationship between wealth and the probability of entering entrepreneurship at lower quantiles of the wealth distribution, it is critical to check whether this relationship is caused by wealth endogeneity. In order to account for the possible endogeneity of wealth we instrument for wealth using changes in housing equity and the value of unexpected inheritances. The results of instrumental variable estimation reveal that there is no significant relationship between wealth and entering entrepreneurship for the full sample as well as for each of the three industries. JEL classification: E21, G11, J24 Keywords: Entrepreneur, wealth, industry, liquidity constraints ∗

Acknowledgements: We are grateful to Christopher Carroll, Thomas Lubik, Bart Moore and Johns Hopkins Macro lunch seminar participants for helpful comments. Berna Demiralp, College of Business and Public Administration, Department of Economics, Constant Hall 2039, Norfolk, VA 23529. email: [email protected]. Johanna Francis, Department of Economics, E-507 Dealy Hall, 441 East Fordham Road, Bronx, NY, 10458. email: [email protected].

1

1

Introduction

A key question that has emerged in academic literature and policy debates is whether potential entrepreneurs are constrained in obtaining capital to start businesses. The answer to this question has important implications: if the main reason people do not enter entrepreneurship is because they cannot obtain adequate capital, then there is a role for public intervention in capital markets, for example through programs to provide loans to small businesses. Otherwise, focusing on the provision or guarantee of loans may not be the best way to foster entrepreneurship and small business development. Entrepreneurship is widely recognized as an important contributor to economic growth. In the U.S., independently owned and operated small enterprizes account for up to half of non-farm private output, and employ half of all private sector employees (Kobe, 2007), so the health and stability of these businesses is a critical factor for overall economic prosperity. Several empirical papers find a positive relationship between assets and the likelihood of becoming an entrepreneur, and this result is interpreted as evidence that costly external financing for entrepreneurial startups prevents otherwise aspiring entrepreneurs from making that choice, thereby restricting the growth of small business (Holtz-Eakin et al. 1994; Gentry and Hubbard 2000; Quadrini 1999; Fairlie 2004; Evans and Jovanovic 1989). This view has recently been challenged by Hurst and Lusardi (2004) who argue it is the nonlinearity of the relationship between wealth and the probability of entrepreneurship that leads to the positive relationship that emerges when wealth and the probability of entrepreneurship are modeled linearly. Over a majority of the wealth distribution, they show that the effect of wealth on the probability of transitioning into entrepreneurship is insignificant. They conclude that the correlation between wealth and the subsequent decision to become an entrepreneur is driven by individuals at the top of the wealth distribution and so is not indicative of significant liquidity constraints. One reason for these conflicting results is that the relationship between wealth and entrepreneurship may be nonlinear, an idea which is receiving renewed support in the literature (Petrova, 2005; Buera, 2003). Evans and Jovanovic (1989) proposed the original static model for entrepreneurial en-

2

try which implied that a positive relationship between wealth and entrepreneurial entry is evidence of the existence of binding liquidity constraints. Their model assumes that individuals, endowed with an initial entrepreneurial ability and assets, are restricted in their ability to finance their business by the need to provide collateral for loans. The implication of their model is that in the presence of collateral constraints individuals with low wealth are likely to be prevented from becoming entrepreneurs so that the probability of becoming an entrepreneur increases with wealth. A positive relationship between wealth and entrepreneurial entry can be evidence that liquidity constraints are meaningful and unduly restrictive for many aspiring entrepreneurs if wealth is exogenous in the occupational decision. However, wealth may be correlated with unobservable characteristics that affect entrepreneurial choice. In order to avoid this potential endogeneity problem, Holtz-Eakin et al. (1994); Blanchflower and Oswald (1998) use receipt of an inheritance to proxy for wealth and find a positive relationship between recent receipt of an inheritance and transition into entrepreneurship. Hurst and Lusardi (2004)argue that the receipt of an inheritance is also likely correlated with unobservable traits affecting the entrepreneurial decision, and thus subject to the same endogeneity problem as wealth. Instead, they advocate for an instrumental variable approach and instrument for wealth using regional variation in housing prices. The focus of this paper is to re-visit the data set, the 1979 National Longitudinal Survey of Youth (NLSY), used for the original results (e.g., Evans and Jovanovic 1989; Evans and Leighton 1989; Holtz-Eakin et al. 1994) and carefully examine the relationship between wealth and entrepreneurship taking into consideration the likely endogeneity of wealth through instrumental variables regression. Similar to earlier work, we find a positive relationship between wealth and entrepreneurial choice based on the results of pooled probit regressions, although the effect is relatively small. We also find evidence that the probability of entering entrepreneurship is nonlinear in wealth; however, in contrast to the results of Hurst and Lusardi (2004), we find that the convex relationship between wealth and entrepreneurial entry is positive and significant over most of the wealth distribution, including the lower quantiles. We address the potential endogeneity of wealth in our analysis by proposing a new instru-

3

ment: amount of unexpected inheritances, which is based on a survey question from the 1989 wave of the NLSY, which asks whether the respondent is expecting to receive an inheritance in the future. In addition, we use regional housing price changes net of local economic factors as a second instrument. When wealth is instrumented, the relationship between wealth and entrepreneurial entry becomes insignificant. We also find substantial heterogeneity among entrepreneurs with respect to the industries that they enter. We divide aspiring entrepreneurs into three groups based on the industries that they enter (low-capital and high-capital requirement industries and professional services), and consider the relationship between wealth and entrepreneurial choice for those groups separately. Ordinary probit regressions, in which wealth enters linearly, yield a significantly positive relationship between wealth and entrepreneurial entry in high-capital industries and a statistically insignificant relationship in the low-capital industries and professional services. We also consider a nonlinear specification and find that the relationship between wealth and entrepreneurial entry is convex in high-capital industries. Professional services display a relatively flat relationship with only a small amount of convexity. Furthermore, the relationship between wealth and the probability of becoming an entrepreneur is concave for low-capital industries, suggesting that the insignificant relationship given by the ordinary probit estimation may be driven by this concavity. When we instrument for wealth using unexpected inheritances and housing price changes, we find that wealth does not affect entrepreneurial choice for the low-capital and high-capital industries, but it negatively affects transition into entrepreneurship for professional services. This paper consists of 5 sections. In section 2, we develop the theoretical framework for our analysis, and in section 3 we describe the data we used. Section 4 presents our multivariate probit regression results as well as our instrumental variables results, including tests of our instruments in each industry context. Section 5 concludes with a summary of our findings and suggestions for future research.

4

2

Theoretical Framework

Our empirical investigation is based on Evans and Jovanovic’s (1989) model, which describes the role of liquidity constraints in an individual’s decision to become self-employed. We assume that the total income of individual i in the wage-salary sector is yiW = wi + rAi , where wi is the individual’s wage earnings, Ai is her assets and r is the gross rate of return. The individual’s income from self-employment is yiE = θi f (ki )i + r(Ai − ki ), where θi is entrepreneurial talent, f (·) is the production function, ki is the amount of capital used in the entrepreneurial venture, and i is an idiosyncratic shock. Aspiring entrepreneurs can be restricted in the amount they can borrow due to capital market imperfections. Following previous research (Evans and Jovanovic, 1989), we assume that their borrowing ability depends on the value of the assets they have available for collateral. In particular, the maximum amount of capital available to an individual with asset level, Ai , is g(Ai ). We assume that this constraint loosens as assets grow, so g 0 (Ai ) > 0. The individual chooses the optimal level of capital investment to maximize self-employment income, yiE , subject to the liquidity constraint, ki ≤ g(Ai ). The optimal capital investment in the presence of liquidity constraints is k˜i∗ = min(ki∗ , g(Ai )), where ki∗ is the optimal choice of capital in the absence of constraints. It can be shown that

dki∗ dAi

= 0; therefore, the individ-

ual’s assets affect her choice of capital if and only if the individual faces a binding liquidity constraint. In other words, constraint; otherwise,

˜∗ dk it dAit

˜∗ dk it dAit

= g 0 (Ait ) > 0 if the individual faces a binding liquidity

= 0.

This observation has formed the basis of a simple test for liquidity constraints used by many researchers (Gentry and Hubbard, 2000; Evans and Leighton, 1989; Quadrini, 1999). If the individual is not liquidity constrained, the optimal level of capital investment is independent of the individual’s assets, and therefore should not affect the decision to become self-employed. Consequently, we should not observe an effect of assets on one’s likelihood to be self-employed in the absence of liquidity constraints. If we could directly observe start-up capital, then a good test of the existence of liquidity constraints could consider whether the entrepreneur’s lagged assets had any effect on start-up capital. Since we do not observe the

5

level of start-up capital for each firm, but just the worker’s choice to remain a worker or to start a business, we have to consider this test indirectly by looking at how the level of assets impacts the decision to become an entrepreneur. Selection into entrepreneurship depends on the utility that individuals gain from entrepreneurship versus wage work. Individuals will choose entrepreneurship over wage work if the expected utility from entrepreneurship is greater than the expected utility of wage work. We assume that utility depends on income, Yi , and nonpecuniary factors (δ) in each mode of employment. In particular, individual i chooses to become an entrepreneur if


EU yiE , δiE > EU (yiW , δiW )

(1)

In our empirical work, we use a probit specification to investigate the determinants of entrepreneurial entry and the existence of liquidity constraints faced by individuals. Based on the theoretical framework presented above, our reduced form econometric model of entrepreneurial entry can be expressed as

P (Di,t+1 = 1) = F (Zit , Ait )

(2)

where Di,t+1 is a dummy variable indicating entry into entrepreneurship at t + 1 conditional on not being self-employed at t, Zit is a vector of observables that affect the individual’s earnings as well as her nonpecuniary gains in each mode of employment, Ait is a vector of net assets (or wealth) and F is the cumulative normal distribution function. If people face a binding liquidity constraint, we would expect a positive coefficient on wealth in Equation (2). However, a positive relationship between wealth and entrepreneurial probability may also suggest that entrepreneurs accumulate more wealth. In order to mitigate such reverse causality, we investigate the effects of wealth and other exogenous variables dated before the entry into entrepreneurship on the entrepreneurial decision. We are aware that this specification may not guarantee the exogeneity of our variables, especially wealth, so we use more robust methods taking into account the endogeneity of wealth in the empirical results presented below. 6

3

Data Description

We use data from the 1979 National Longitudinal Survey of Youth (NLSY) to estimate the influence of wealth on the decision to become an entrepreneur. The NSLY provides a nationally representative sample of 12,696 men and women who were between ages 14 to 22 when they were first surveyed in 1979. We use the NLSY primarily because it is used in much of the early work detailing the positive relationship between wealth and entrepreneurial entry (for example, Evans and Jovanovic 1989; Evans and Leighton 1989). In addition, using the NLSY provides several advantages for studying the existence of liquidity constraints. First, its panel nature allows us to follow wage workers’ transition into self-employment and thus control for explanatory variables before entry into self-employment. Second, it collects detailed information on jobs, assets and demographic characteristics of respondents. Finally, the NLSY follows the labor market experiences of a young cohort from the beginning of its entry into the labor market. Given that age-earnings profiles tend to be upward sloping, liquidity constraints are more likely to be binding for younger individuals. Thus, the NLSY sample allows us to focus on the entrepreneurial decisions of a cohort that is likely to face binding liquidity constraints. We chose to use data from the years 1985 to 1990 and 1992 to 1994 because during these years the NLSY collected detailed information on assets. Prior to 1985 as well as for the 1991 survey wave, individuals were not asked specific details about the value of their assets. We restrict our sample to individuals who are in the labor force in the first year that they appear in the dataset in order to focus on transitions to entrepreneurship. We use data from the Bureau of Labor Statistics on the Consumer Price Index (CPI) and regional CPI for all urban consumers with 1982-1984 as the base year to adjust for inflation. We also use data from the Bureau of Economic Analysis for state GDP to create our variable for housing price appreciation (described below). Before we begin our analysis, we need to define what we mean by an entrepreneur. There are two main ways to define an entrepreneur: as a self-employed individual, or as someone who reports owning a business (or both). We use the more common approach in the literature

7

and consider a self-employed person to be an entrepreneur. We consider an individual to be self-employed in a particular survey year if the ‘class of worker’ category question for the current (CPS) job indicated self-employment. We restrict the sample to those who have never been an entrepreneur in order to ensure that wealth is not accumulated while being an entrepreneur which would bias our sample. We use net worth as the definition for wealth in our analysis because one’s ability to obtain credit is most likely determined by assets net of debt obligations.1 This is also consistent with other studies, such as Hurst and Lusardi (2004). Net worth is calculated as assets less debt obligations, where debt is comprised of mortgage debt owed, other property debt (including for business or farm property), debt owed on automobiles and other debt valued over $500. Assets are calculated as the sum of savings accounts, farm and business property assets, the market value of residential property owned, the market value of automobiles plus other assets which are valued above $500. The NLSY top-codes both income and assets at levels that change yearly. The earlier waves of the survey (1979-1983) had restrictive topcoding particularly compared to other surveys, such as the Survey of Consumer Finances (SCF). Cut-off values for responses to asset questions ranged from $30,000 for questions about vehicle values and debts to $500,000 for financial assets and farm or business property. Since the NLSY does not over-sample the wealthy, as the SCF does, the percentage of survey respondents for whom asset top-coding is relevant is less than 10 percent of the survey. Table 1 presents the mean asset levels held by entrepreneurs and wage workers. The statistics from our sample are consistent with the common finding that entrepreneurs hold more assets than wage workers on average. In our sample, the mean asset level for entrepreneurs is about 2.7 times that for wage workers. The difference in average assets between the two groups is driven by higher business property and housing valuations for entrepreneurs. Entrepreneurs also have higher debt obligations than wage workers as presented in Table 2. However, the wide gap between entrepreneurs and wage workers remains when we focus on net worth: an average entrepreneur has almost 2.5 times the net worth of an average wage worker. 1

Net worth will be used interchangeably with wealth in our discussion below.

8

Our analysis also includes the following covariates: age, gender, race (Black or nonBlack), marital status (married or not married), number of children in the household, years of education, unemployment experience (how many weeks of unemployment experienced in the previous year), and number of job separations (how many past jobs held, not concurrent jobs). Table 2 provides summary statistics for our sample and highlights the differences between wage-workers and entrepreneurs. Entrepreneurs are more likely to be older, male, married, and to have at least one child. They also have slightly less education, more frequent past job separations, and fewer weeks of unemployment. They are also more likely to have received an inheritance and the average inheritance received by an entrepreneur is almost twice the average amount of inheritance received by wage workers.

4

Results

In this section, we provide evidence from the NLSY on the relationship between net worth and entrepreneurial entry by estimating the model presented in Equation (2) with a probit technique. We start by presenting results of the multivariate analysis for the entire sample and by industry. Then we discuss the instrumental variable estimation results.

4.1

Multivariate Analysis

We estimate a pooled probit model of the transition to entrepreneurship as a function of household net worth and a set of control variables. We use a quadratic in age to capture a possible concavity in the relationship between the probability to become an entrepreneur and age. Our covariates also include years of education, race, family structure (marital status, number of children in the household, whether the wife is working), whether the individual has experienced unemployment, and how many job separations she has experienced. The results are presented in Table 3 with robust standard errors in brackets. We will start with discussing the common covariates across all 5 specifications. We find that being older, male, and living in an urban area (measured by being a resident of a Metropolitan Statistical Area) increases the probability of becoming an entrepreneur. Being married has no effect

9

on choosing entrepreneurship. The fact that the spouse is working and thus diversifying the household’s income source (provided the spouse works outside the business) has no significant impact on one’s likelihood of becoming an entrepreneur. Interestingly, having more children increases the probability of entering self-employment and given that marriage or the spouse working has no effect, this might be related to household stability. Blacks have a lower likelihood of becoming entrepreneurs. The reasons for this finding are complicated by many factors explored in Fairlie (1999). We also found that past unemployment spells have a positive effect on entrepreneurial entry, similar to Fairlie and Krashinsky (2006). The mean unemployment spell for a given year is 12 weeks and the median is 8 weeks. Having more job separations also has a positive effect on entrepreneurial entry. There are various explanations for this: individuals who are unhappy with work in the wage sector may change jobs more often seeking a better match or individuals who are not high quality workers may move from job to job. Either of these circumstances may make self-employment a more attractive occupational choice. In specification (1) of Table 3, we consider a linear relationship between household wealth and entrepreneurial entry. Similar to other studies, we find a positive and significant coefficient on net worth, implying that individuals with more wealth are more likely to become entrepreneurs. Our theoretical framework suggests that this relationship signifies the existence of liquidity constraints. The marginal effects coefficient on net worth shows that an extra $100,000 worth of assets would increase the probability of choosing entrepreneurship by approximately a tenth of a percent. In order to more accurately determine the relationship between net worth and entrepreneurship, we look carefully at how an individual’s place in the wealth distribution affects her entrepreneurial entry decision. We re-estimate our probit model including dummy variables for quantiles of the wealth distribution. We divide households into 5 groups: the first 3 groups represent the first 3 quantiles of the wealth distribution (quantile 1 = percentiles 0 through 25, quantile 2 = percentiles 26 through 50, quantile 3 = percentiles 51 through 75), the fourth quantile represents percentiles 76 through 94, and the fifth quantile represents the top 5 percent of the distribution, ranked by wealth holding. We can see from specification

10

(2) in Table 3 that belonging to the 5th quantile has the strongest effect on becoming an entrepreneur and that increases in net worth monotonically increase the probability of becoming an entrepreneur. If a wage worker, for example, is in the 2nd quantile of the wealth distribution, then her net worth has no effect on the decision to become an entrepreneur. For workers in the 3rd or higher quantiles, wealth is increasingly strongly related to the decision to become an entrepreneur. This result differs from Hurst and Lusardi (2004) in that we find a relationship between wealth and entrepreneurship at much lower levels of the wealth distribution. When wealth enters the entrepreneurial decision making process linearly, the marginal effect of an incremental change in wealth is small though significant. However, the results of specification (2) suggest that the effect of net worth on entrepreneurial entry may vary across the wealth distribution. Thus we consider a non-linear relationship between wealth and the likelihood of becoming an entrepreneur. Other studies, most notably Hurst and Lusardi (2004) and Buera (2003), provide evidence that there is a non-linear relationship between assets and entrepreneurial entry and that a 5th order polynomial fits the data well. We experimented with a variety of polynomial specifications and found that a 5th order polynomial fits our wealth data the most closely as well, since our data include a small fraction of households with very large levels of net worth and a fraction with low or negative net worth. Specification (3) of Table 3 uses a 5th order polynomial in wealth for the entrepreneurial entry decisions. We find that all the coefficients for the polynomial terms are significant. We also find, like Hurst and Lusardi (2004), a likelihood ratio test rejects the specification in which wealth is included linearly in favor of the 5th order polynomial version. Testing the joint significance of the coefficients on the polynomial terms shows that they are all significant at the 1 percent level. The marginal effect of net worth on the probability of becoming an entrepreneur, evaluated at mean values, is 0.012, which is almost ten times the marginal effect given by the linear specification. We also present the results of the polynomial model by graphing the relationship between wealth and the probability of entrepreneurship. Figure 1 is drawn using the estimates from specification 3 in Table 3. The graph is slightly convex indicating that wealth has an increas-

11

ing effect on the probability of entrepreneurship for individuals in the top half of the wealth distribution. Contrary to the results of Hurst and Lusardi, we find that wealth positively effects one’s probability of becoming an entrepreneur even at the lower levels of the wealth distribution. At the 25th percentile of the net worth distribution, increasing net worth by $100,000 leads to a 1.16 percent increase in the probability of becoming an entrepreneur. The marginal effect is 1.17 percent at the median net worth and is 1.26 percent at the 99th percentile. These marginal effects are all statistically significant. Our results, so far, suggest that there is a positive relationship between wealth and the probability of choosing entrepreneurship and that the positive relationship is robust to the nonlinear specification of our model. Figure 1 approximately here.

4.2

Stratification by Industry Type

The wealth of an aspiring entrepreneur may not matter for entering some industries while it might be critical for others. Because our sample mixes entrepreneurs from all types of industries, our results may obscure heterogeneity in the effect of wealth on entrepreneurship. There are two basic ways that industrial type and wealth could be related. Some industries have higher capital requirements for start-up, which implies that in the presence of capital market imperfections, potential entrepreneurs need to accumulate assets prior to entering these industries. Liquidity constraints would be more relevant for aspiring entrepreneurs in these industries. The second way in which industrial type and wealth could be related is through the skill required to start a business in some industries. In higher technology industries, for example, aspiring entrepreneurs must acquire significant knowledge or education before being able to start their businesses. In this section we focus on the distinction between low and high capitalization requirement industries and their relationship to net worth. We divide entrepreneurs into three groups: industries with low capitalization requirements, industries with high capitalization requirements and professional services. We follow Hurst and Lusardi (2004) and use the 1987 National Survey of Small Business Finances (NSSBF), which includes 1,099 small busi-

12

nesses, to determine cut-offs for low and high capital industries. We use data from newly created businesses only (which excludes inherited businesses or firms that were purchased). Agricultural firms (farms) are not included in the NSSBF data. We find that our results are not sensitive to the inclusion of agriculture in either the high or low capital categories, thus we place it in the high capital starting industries, where the median starting capital requirement is $45,000. Professional Services are also not surveyed in the NSSBF, so we do not have information on capital requirements for this category. This industry category also is unique in that in order to become self employed in professional services, which includes doctors, lawyers, portfolio managers, accountants, etc., one needs to meet certification requirements, unlike other small business ventures where having sufficient capital may be the only impediment. Thus we include professional services as a separate category. In Table 4 we show the breakdown of new businesses by industry types. Low starting capital industries account for the majority of small businesses in our sample. These industries include construction and most service industries, aside from professional services, such as hair and nail salons and business and repair services such as mechanic shops and copy services and account for 58.2 percent of the sample. High capital requirement industries are agriculture, mining, transportation/communication and public utilities industries, manufacturing, retail and wholesale sales and the finance, insurance and real estate industries which account for 33.8 percent of the sample. Professional services includes doctors, lawyers, accountants, etc. and account for 8.1 percent of the new businesses in our sample. Table 5 presents the probit regression results for these three groups. The top panel considers low-capitalization industries, the middle panel considers high-capitalization industries and the bottom panel considers professional services. The three specifications (in columns) follow the analysis for the full sample: specification 1 is the linear model, specification 2 includes wealth quantiles and specification 3 is the polynomial model. We use the same set of covariates as reported previously in Table 3 but only report the coefficients on net worth, net worth quantiles, or polynomial terms as appropriate. Looking at low-capitalization industries first, the coefficient on net worth in the linear model (specification 1), is positive but not significant. Comparatively, we find that it is

13

positive and significant in the high-capitalization industries (second panel) which is the same effect as for the full sample, while it is negative but not significant for the professional services industry (third panel). Then we look at the relationship between net worth quantiles and entrepreneurship, using the same quantile groups as with the full sample. For wage workers who start businesses in low capital industries we find the quantile coefficients are positive, significant and increasing, similar to the full sample, meaning that wage workers in higher wealth quantiles are more likely to start businesses, even though their chosen industry has a lower capitalization requirement. For high capital industries, the results are similar but somewhat stronger; the quantile coefficients are positive and strongly increasing between the 4th and 5th quantile. For the high capital industries, even more so than the low capital industries, workers at the top of the wealth distribution, in quantile 5, are driving most of the effect of wealth on entrepreneurial entry. Interestingly for wage workers who choose to enter self-employment in professional services, the coefficients are mostly negative and not significant. The only significant coefficient is for workers who start in quantile 3, and for them, wealth has a negative effect on self-employment entry relative to the omitted first quantile. Turning to the polynomial model for net worth, we find that the coefficients on the polynomial terms are extremely small, though significant for low capital industries. For the high capital industries, the coefficients on the first two polynomial terms are positive and significant and the rest of the coefficients are significant, similar to the full sample. For professional services, the coefficients are again extremely small on the polynomial terms and only significant for the square of net worth, the 4th and 5th power. From this analysis, the high capital industries appear to be responsible for the sign, size and significance of the coefficients in the full sample. In figure 2 we present the results of specification 3 (in Table 5) graphically for the lowcapitalization, high-capitalization, and professional services industries. For the high capital industries, the graph is slightly convex, suggesting that wealth has a positive effect on the probability of entrepreneurship across the wealth distribution and that effect increases with wealth. Based on our theoretical framework, this result suggests that individuals wishing to

14

enter entrepreneurship in a high-capitalization industry face binding liquidity constraints at all wealth levels. At the 25th percentile of the net worth distribution, increasing net worth by $100,000 leads to a 1.08 percent increase in the probability of becoming an entrepreneur. The marginal effect is 1.09 percent at the median net worth and 1.16 percent at the 99th percentile of net worth. These marginal effects are all statistically significant, suggesting that the probability of becoming an entrepreneur in a high capital requirement industry is increasing in wealth. Figure 2 approximately here. For the professional services industry, the graph is similar to the high capital industries, though flatter through the lower quantiles, suggesting that wealth is not very important for the decision to become self-employed in the lower quantiles. This result indicates that people at the lower quantiles of the wealth distribution are less constrained by liquidity in their decision to enter a professional business. The intuition provided by the graph is confirmed by looking at the marginal effects at various percentiles of the wealth distribution. At the 25th percentile of the net worth distribution, increasing net worth by $100,000 leads to a decrease in the probability of becoming an entrepreneur of 0.4 percent although the effect is not significant at the 10 percent level. At the median and the mean, increasing net worth also has a negative but statistically insignificant effect on the probability of becoming an entrepreneur. At the 99th percentile, we find that increasing net worth by $100,000 increases the probability of becoming an entrepreneur by 1.33 percent, but this effect continues to be statistically insignificant. For the professional services industry, we cannot find a percentile of the net worth distribution from which increasing net worth would have a positive effect on entrepreneurship. The graph for the low capital industries reveals that the probability of entering entrepreneurship initially increases and then decreases as wealth increases. This concave relationship suggests that while wealth positively affects the probability of entrepreneurship at lower wealth levels, it negatively affects that probability at higher wealth levels. The negative relationship between wealth and the probability of becoming self-employed at high wealth levels could be explained by a selection bias that may be occurring when we compare low and

15

high wealth individuals. Wage workers with high wealth might be very successful as wage workers, so their probability of choosing entrepreneurship would be low. To the extent that wealth among wage workers proxies for wage work ability relative to entrepreneurial ability, we would observe a negative relationship between wealth and entrepreneurial entry. However, it is interesting that we are observing this relationship only in low capital industries. The concave relationship between wealth and the probability of entrepreneurship may be the reason why the linear specification (specification 1) yielded no significant relationship between wealth and entrepreneurship for low capital industries. Another way of looking at these results for low-capitalization industries is to consider how the probability of becoming an entrepreneur changes with percentiles of the wealth distribution, as we discussed above for the other two industry types. At the 25th percentile of the net worth distribution, increasing net worth by $100,000 leads to a 1.13 percent increase in the probability of becoming an entrepreneur, which is stronger than in the high capital industry and is statistically significant. However, at the median, the probability decreases while it remains statistically significant: the marginal effect is 1.09, while at the mean (which is above the median in our sample), the marginal effect is much smaller at 0.10 percent. At the 99th percentile, increasing net worth by $100,000 actually reduces the probability of becoming an entrepreneur by 3.8 percent, although this result is not statistically significant at the 10 percent level. Within the top 1 percent of the distribution, increasing net worth by $100,000 further reduces the probability of becoming an entrepreneur by 6.4 percent, which is statistically significant at the 5 percent level, thus accounting for the concave shape of the graph for the low-capitalization industry.

4.3

Inheritance

Previous research has noted that a positive relationship between wealth and entrepreneurial entry can be driven by an omitted variable that is correlated with both wealth and entrepreneurial choice. Traditionally studies have looked at the receipt of an inheritance as a better indicator of the relationship between the liquidity aspect of wealth and entrepreneurial entry as inheritances are more likely to be uncorrelated with omitted variables affecting both

16

wealth and entry. The windfall nature of an inheritance may also give individuals an additional incentive to choose entrepreneurship. Alternatively, most significant inheritances are received by people from wealthy families who are likely already relatively wealthy themselves and this variable may still suffer from the same endogeneity problems as net worth. We think that people for whom inheritance proxies for more than wealth enter their labor market careers with the expectation of an inheritance in the future. The NLSY asked respondents in 1987 about whether they expect to receive an inheritance in the future. We use this information along with information about who received an inheritance to test whether the effect of inheritance on entrepreneurial entry varies with whether the inheritance was expected or not. We create 3 dummy variables: one for those who expected and received an inheritance, one for those who expected but did not receive an inheritance and one for those who did not expect but did receive an inheritance. Specification 1 in Table 6 presents results of a pooled probit regression for entrepreneurial entry including the dummy variables on inheritance receipt. We find that people who expect to receive an inheritance are more likely to become entrepreneurs regardless of whether they actually received the inheritance or not. On the other hand, people who did not expect to receive an inheritance but received it do not have a statistically different probability of choosing entrepreneurship compared to people who did not expect to receive an inheritance and did not receive it (which is our omitted category for specification 1 of table 6). We also include net worth and interaction terms between the dummy variables and net worth in the probit regression as shown in specification (2) of Table 6. The coefficients on ‘Expected and Received’ and ‘Expected and Did not Receive’ remain significantly positive when net worth is included, signaling that the inheritance expectation captures more than just an increase in wealth in our analysis. The effect of net worth does not seem to vary across people with different inheritance expectations and realizations since the coefficients on the interaction terms are statistically insignificant. These results support our hypothesis that attributes related to expectations of receiving an inheritance seem to affect entrepreneurial decisions rather than the actual receipt (or amount) of an inheritance. Therefore, an inheri-

17

tance that was not expected by the recipient is not related to the unobservable characteristics affecting both entrepreneurial entry and the amount of inheritance received.

4.4

Endogeneity of Wealth: Instrumenting for Wealth

We are concerned with the potential endogeneity of wealth in our probit regressions of entrepreneurial entry since being wealthy correlates with a variety of independent factors that may also be important for entrepreneurship. In this section, we identify two instruments for net worth: non-expected inheritances and housing price residuals. The analysis above suggests that the amount of inheritance that is not expected by the individual is likely to be uncorrelated with unobservables affecting entrepreneurial choice. In this section we present the results of instrumental variables probit regressions and in the next section we provide the results of tests for the validity and relevance of our instruments as well as for the endogeneity of net worth. We instrument for net worth using non-expected inheritance, discussed above, and also lagged implicit capital gains on housing. Lagged capital gains on housing, originally proposed by Hurst and Lusardi (2004) as an instrument for wealth, show considerable regional variation over our sample period across the 4 major census regions in the US (based on housing price indices from the Office of Federal Housing Enterprise Oversight). Furthermore, these changes have affected the wealth of most households since housing comprises the majority of household wealth. Housing price changes are likely to be endogenous since they depend on variables, such as state and regional economic conditions, which also affect the transition to entrepreneurship. Therefore, we calculate a measure of regional appreciation in house prices, excluding the effects of household demographics and the health of the local economy. In particular, we regress changes in the self-reported real value of houses for our sample between 1985-1988 on household demographic variables (age, gender, race, marital status, family size, education, and real household income) and two regional economic indicators (state real GDP and the regional unemployment rate). The regional unemployment rate is reported in the NLSY for each individual based on the unemployment rate in their MSA. We calculated state real GDP

18

from the Bureau of Economic Analysis data on state GDP and regional CPI measures from the Bureau of Labor Statistics. We run this regression separately for each region and collect the regional residuals which are then linked to each respondent. These residuals should capture the lagged regional variation in housing prices that is not driven by local economic conditions or household demographics. We use regional housing price residuals as an instrument for household wealth assuming that households living in areas with larger increases in housing prices should be more likely to become self-employed, everything else held constant. We then regress entrepreneurial entry after 1991 (between 1992 and 1994) on net worth instrumented with the residuals from past changes in housing prices and unexpected inheritances. We report the results of an instrumental variables probit analysis using both unexpected inheritances and housing price residuals as instruments for net worth, including the same covariates used in previous probit regressions, in Table 7. We find that for the full sample, the coefficient on net worth is positive but not significant, in contrast to the results we had with the ordinary probit, which showed net worth had a positive and significant effect on entrepreneurial entry in both the linear and polynomial specification. For entrepreneurs entering industries with low capital requirements, we find a negative but not significant coefficient on net worth, similar to what we found for the ordinary probit, where the coefficient was also not significant, though it was positive. For high capital requirement industries, we have a positive but not significant coefficient on net worth in contrast to the positive significant coefficient we found with an ordinary probit regression. The professional services industry has a puzzling result: the coefficient on net worth is large, negative and significant, suggesting that for individuals entering professional services, higher net worth makes them less likely to become self-employed. This result is consistent with the results we found for lowcapitalization industries and may be due to the fact that high wealth wage workers, who have experienced considerable success in wage-work, are less likely to select into self-employment. Thus the instrumental variables estimation results suggest that there is no effect of net worth on the probability of entering entrepreneurship for the full sample as well as for low and high capital industries and that there is a negative effect of net worth on the probability of entry for professional services. Thus, the positive relationship between wealth and

19

entrepreneurship for the full sample and the high capital industries given by the ordinary probit regressions disappears when the endogeneity of wealth is taken into account by the use of instrumental variables. The statistically insignificant relationship between wealth and entrepreneurship for the low capital industries remains when wealth is instrumented, and the statistically insignificant coefficient on wealth for the professional industries become significantly negative. The high capital industry results most closely match the full sample results, suggesting that the results we get for the full sample are driven by entry into high capital requirement industries and mask the negative effect of net worth on entry for workers becoming self-employed in professional services.

4.5

Testing Instruments

The results we presented above are interesting and somewhat surprising. In order to confirm the conclusions that we draw from these results, we need to consider the quality of our instruments and whether instrumental variables is the appropriate specification for our sample. We start with determining whether assets and inheritances are endogenous in our sample as we expect them to be. We perform a Smith Blundell test (Smith and Blundell, 1986) for each of our specifications. The Smith Blundell test is a test of exogeneity for a probit model, similar to the Hausman test, which is particularly useful when there is more than one instrument. The test statistic, under the null that net worth is exogenous, is distributed as Chi-squared(m) where m is the number of potentially endogenous variables (in our case m = 1). We report the results from the Smith Blundell test in the first line of Table 8 with the p-values for the test statistic in brackets. We find that we can reject the assumption that net worth is exogenous. Therefore the instrumental variables specification is the correct one for our full sample and each of our industry groups. A Wald test and Hausman tests (not reported here) give us the same results. We turn to considering the relevance and validity of our instruments. There are two main conditions that our instruments have to satisfy: they have to be correlated with the endogenous variable that is being instrumented and uncorrelated with the error term. We

20

consider these two conditions separately in Table 8. The first condition relates to the problem of weak instruments. Often the candidates for instruments are only weakly correlated with the endogenous variable they are instrumenting for. It is well known that weak instruments are likely to produce estimates with large standard errors. Since we are using a large microlevel data set, the problems of weak instruments may be obscured by the large sample size which gives reasonable standard errors. The Partial F-statistic provides us with a test of the relevance of our instruments. In Table 8 we report a joint Partial F-statistic for our instruments with p-values in brackets. We find that our instruments satisfy the relevance requirement for the full sample. According to the cut-off of 10 provided in Staiger and Stock (1997), housing price residuals and non-expected inheritances satisfy the relevance requirement for the full sample. We consider whether our instruments are valid for the full sample by doing an overidentification test. In the final line of Table 8, we present Sargan’s over-identification statistic, which is distributed as χ2 (1) for the probit model with one endogenous variable. We cannot reject the null of instrument validity, so both of our instruments are uncorrelated with the error term and satisfy the validity condition for the full sample. For the industry groups, we find that all of the instruments pass the weak instruments tests: all of the Partial F-tests have significant coefficients much greater than the Staiger and Stock minimum. As reported in Table 8, we also find that our instruments are relevant in each of the industry groups since we cannot reject the null of instrument validity in the over-identification tests for any of our groups. We conclude that instrumental variables is the correct specification for our full sample and individual industry groups and that the instruments we use are valid and relevant. Looking back at our instrumental variables regression results in Table 7 with new information about the validity of the technique and strength of our instruments, we conclude that wealth, as measured by net worth, has no bearing on the entrepreneurial entry decision for our full sample drawn from the NLSY or for any of the industry groups except for professional services. For the professional services group, higher wealth proves to be an impediment to self-employment. This result is likely due to the problems of sample selection we discussed

21

earlier, that individuals who find success (as measured by high wages) in wage-work due to high wage-work ability are less likely to choose entrepreneurship, in part because of the higher opportunity cost. Individuals who work in professional services industries are also more likely to have significant work autonomy already as employees and therefore may not be as interested in being their own boss. We have no data to support these speculations, but they present an intriguing line of inquiry for future research.

5

Conclusion

In this paper we investigate the extent to which individuals face binding liquidity constraints in their entrepreneurial entry decision by revisiting the relationship between wealth and entrepreneurial entry. Pooled probit results suggest that the relationship between wealth and entrepreneurial entry is positive. Although the relationship appears to be convex, it is statistically significant even at the lower tail of the wealth distribution. However, when we instrument for wealth using unexpected inheritance amounts and regional housing price changes, wealth has no effect on the average aspiring entrepreneur’s decision, supporting the hypothesis that constraints in obtaining capital do not affect entrepreneurial decisions. We also find substantial heterogeneity among entrepreneurs with respect to their industries. When we stratify entrepreneurs by their aspirant industry, we find that wealth positively affects entrepreneurial entry for high-capital industries although the effect disappears when wealth is instrumented. For low-capital industries, the relationship between wealth and the probability of choosing entrepreneurship is concave, which may explain the probit results that wealth has no affect on entrepreneurial entry. Instrumental variables probit results show that net worth negatively affects the probability of entrepreneurial entry for professional services, which may be a result of sample selection. This paper demonstrates that the relationship between wealth and entrepreneurial choice varies across the wealth distribution and across industry categories. Further work should address why we observe the patterns described in this paper. A dynamic model of occupational choice, which incorporates implications of self-selection, can help us to better understand the

22

transitions between wage-work and entrepreneurship (and vice versa) and how this relates to saving behavior. This work is left for a future project.

References Blanchflower, D., Oswald, A., 1998. What makes an entrepreneur? Journal of Labor Economics 16, 26–60. Buera, F., 2003. A dynamic model of entrepreneurship with borrowing constraintsManuscript. University of Chicago. Evans, D., Jovanovic, B., 1989. An estimated model of entrepreneurial choice under liquidity constraints. Journal of Political Economy 97(4), 808–826. Evans, D., Leighton, L., 1989. Some empirical aspects of entrepreneurship. The American Economic Review 79(3), 519–535. Fairlie, R., 1999. The absence of the African-American owned business: An analysis of the dynamics of self-employment. Journal of Labor Economics 17(1), 80–108. Fairlie, R., 2004. Self-employed business ownership rates in the united states: 1979-2003. Small Business Research Summary, SBA No. 243(December). Fairlie, R., Krashinsky, H., 2006. Liquidity constraints, household wealth and entrepreneurship revisited. IZA discussion papers 2201Institute for the Study of Labor (IZA). Gentry, W., Hubbard, R., September 2000. Entrepreneurship and household saving. NBER Working Paper 7894. Holtz-Eakin, D., Joulaian, D., Rosen, H., 1994. Sticking it out: Entrepreneurial survival and liquidity constraints. Journal of Political Economy 102(1), 53–75. Hurst, E., Lusardi, A., 2004. Liquidity constraints, household wealth and entrepreneurship. The Journal of Political Economy 112(2), 319–347.

23

Kobe, K., 2007. The small business share of GDP: 1998-2004. Small Business Research Summary, Small Business Administration. Petrova, K., 2005. Part-time entrepreneurship and wealth effects: New evidence from the panel study of entrepreneurial dynamics. The World Conference Proceedings of the International Council of Small Business June. Quadrini, V., 1999. The importance of entrepreneurship for wealth concentration and mobility. Review of Income and Wealth 45. Smith, R., Blundell, R., 1986. An exogeneity test for a simultaneous equation tobit model with an application to labor supply. Econometrica 54(3), 679–85. Staiger, D., Stock, J., 1997. Instrumental variables regression with weak instruments. Econometrica 65(3), 557–586.

6

Tables

24

Variable Entrepreneurs

Table 1: Asset Holdings of Entrepreneurs and Wage Workers Mean Standard Dev

Total Assets Saving Account Housing Wealth Farm/Business Property Auto Assets Other Assets

102,676 5,539 36,200 48,236 6,385 6,601

217,952 26,189 64,651 179,917 8,002 34,560

37,757 5,121 20,314 4,162 4,939 3,278

137,710 115,621 45,642 47,103 5,820 15,952

Wage Workers Total Assets Saving Account Housing Wealth Farm/Business Property Auto Assets Other Assets

Note: Data are from 1985-1994 waves of the NLSY excluding 1991. Income and Assets are measured in 1982-1984 dollars using regional CPI data from the Bureau of Labor Statistics.

Table 2: Descriptive Statistics of Entrepreneurs and Wage Workers: Pooled sample Variable Wage Workers Self-Employed p-value of Difference Age 27.11 28.4 < 0.01 Gender: Head = Male 0.51 0.60 < 0.01 Dummy: Head = Black 0.25 0.14 < 0.01 Dummy: Head = MSA resident 0.81 0.79 < 0.01 Dummy: Head = Married 0.45 0.58 < 0.01 Number of Children in HH 0.78 1.0 < 0.01 Years of Education 12.9 12.7 < 0.01 Unemployment Experience 37.6 29.1 < 0.01 Job Separations 7.1 7.8 < 0.01 Assets $37,757 $102,676 < 0.01 Debts $22,745 $64,601 < 0.01 Net Worth $19,993 $49,366 < 0.01 Ever Received Inheritance 0.08 0.11 < 0.01 Amount of Inheritance $5,483 $11,296 < 0.05 Note: Financial data are from the 1985-1994 waves of the NLSY excluding 1991. Income, Assets, Debt, Inheritances and Net Worth are deflated using 1982-1984 as a base year based on CPI data from the Bureau of Labor Statistics. Responses are mean values where appropriate and frequencies elsewhere.

25

Table 3: Pooled Probit Regressions of Entrepreneurial Entry for Full Sample (marginal effects reported) Specification (1)

(2)

(3)

Age

0.006 (.003)*

0.006 (.003)*

0.007 (.003)**

Age2

-.0001(6e-4 )**

-.0001 (.0001)**

-.0001 (6e-5 )**

***

Gender

0.006 (.002)

***

***

0.005 (.002)***

***

-.011 (.002)***

0.005 (.002)

Black

-.014 (.002)

-.011 (.002)

Married

0.003 (.003)

-.001 (.003)

0.001(.003)

Children in HH

0.004 (.001)***

0.005 (.001)***

0.004 (.001)*** -.002 (.002)

Spouse Working

-.002 (.003)

-.002 (.003)

MSA resident

0.003 (.002)*

0.003 (.002)

0.003 (.002)

Yrs of Education

-.0007 (.001)*

-.001 (.0004)***

-.001 (.0004)***

Unemployment

0.006 (.004)*

0.008 (.004)**

0.006 (.004)*

*

Job Separations

0.002 (.0002)

Net Worth

0.001 (.0003)***

***

0.002 (.008)

0.010 (.001)***

Quantile 2

0.002 (.002)

Quantile 3

0.006 (.003)**

Quantile 4

0.021 (.004)***

Quantile 5

0.053 (.009)***

Networth

0.002 (.0002)***

2

0.001 (6e-5 )***

Networth3

-.00003 (6.17e-5 )***

4

-3.15e-7 (5.4e-8 )***

Networth5

8.17e-9 (1.67e-9 )***

Networth

2

Pseudo R

0.020

0.028

0.027

Sample Size

41,491

41,491

41,491

Note: Sample includes individuals who are part of the labor force. Analysis is restricted to those who are not self-employed in period t; the dependent variable is a dummy variable, indicating transition into entrepreneurship in period t + 1. The covariates are described in Table 2. Huber-White standard errors reported in brackets. Net worth is divided by $100,000 to make the coefficients conform in size and is corrected for inflation using 1982-1984 as a base year for the CPI from the Bureau of Labor Statistics. *** Significant at 1 %, ** significant at 5 %, * significant at 10 %.

26

Table 4: Industry Type for New Business Owners Industry Type Percent of Firms Low starting-capital industries: Construction 17.6 Services: Business and Repair 15.06 Personal 22.75 Entertainment and Recreation 2.74 Professional Services industries: Lawyers, Doctors, Accountants, etc. 8.12 High starting-capital industries: Agriculture Mining Transportation, communication and public utilities Manufacturing Durables Non-durables Wholesale trade Retail trade Finance, insurance and real estate Public administration

9.62 0.28 3.3 2.57 2.19 2.14 11.39 2.23 0.1

Note: The division among professional services, high and low starting capital industries is described in the text, section 4.2. The data describe the industrial composition of firms in our sample between 1985 and 1994. The percentages of firms in each industry only add up to 97.1 because some of our firms could not be classified and so were excluded.

27

Table 5: Pooled Probit Regressions of Entrepreneurial Entry by Industry Low Capital Industries Marginal Effects (1) (2) (3) Net worth Quantile 2 Quantile 3 Quantile 4 Quantile 5 Networth2 Networth3 Networth4 Networth5 Pseudo R2 Sample Size

1.22e-60 (1.2e-58 )***

0.002 (.001) *

0.015(.009)

0.021(.010)** 0.059(.014)** 0.065(.027)** -4.85e-61 (4.9e-59 )*** 4.67e-62 (4.8e-60 )*** -1.53e-63 (1.5e-61 )** 1.30e-65 (1.3e-63 )** 0.0242

0.0313

0.0369

6,925

6,925

6,925

(1)

(2)

High Capital Industries Marginal Effects Net worth Quantile 2 Quantile 3 Quantile 4 Quantile 5 Networth2 Networth3 Networth4 Networth5 Pseudo R2 Sample Size

0.001 (.0003)***

(3) 0.009 (.001)***

0.002(.003) 0.011(.004)*** 0.024(.005)*** 0.074(.014)*** 0.0002 (6.0e-5 )*** -2.95e-5 (6.6e-6 )*** -3.02e-7 (5.6e-8 )*** 8.35e-9 (1.8e-9 )*** 0.0192

0.0342

0.0310

23,174

23,174

23,174

(1)

(2)

(3)

Professional Services Marginal Effects Net worth Quantile 2 Quantile 3 Quantile 4 Quantile 5 Networth2 Networth3 Networth4 Networth5 Pseudo R2 Sample Size

-7.86e-17 (2.0e-15 )

-.0005 (.001) -.003(.003) -.006(.003)* -.002(.004) 0.053(.008)

1.62e-16 (4.2e-15 )*** -1.99e-17 (5.1e-16 ) -1.51e-63 (3.9e-17 )* 2.33e-20 (1.3e-19 )* 0.0443

0.0476

0.0553

8,056

8,056

8,056

Note: Results are based on NLSY data from 1985-1994 excluding 1991. The regressions presented are probit regressions with Huber-White standard errors reported in brackets. Net worth is divided by $100,000 and corrected for inflation using 1982-84 as a base year for the CPI. *** Significant at 1 %, ** significant at 5 %, * significant at 10 %.

28

Table 6: Effect of Inheritance on Entrepreneurship Specification (1) Inheritance Amount Net Worth Expected and Received Expected and Did not Receive Did not Expect and Received NW e r NW ne r NW e nr Pseudo R2 Sample Size

(2)

0.006 (.004) 0.0008 (.0003)*** 0.018 (.006)***

0.0140 (.006)**

***

0.016 (.005)

0.0150 (.006)***

0.002 (.003)

0.0010 (.003) 0.0002 (.004) 0.0003 (.001) 0.0010 (.005)

0.0213

0.0220

29,231

25,345

Note: NW e r is net worth interacted with a dummy reflecting that the individual expected to receive an inheritance and did receive one. NW ne r is net worth interacted with a dummy reflecting that an inheritance was not expected but was received. NW e nr is net worth interacted with a dummy reflecting an inheritance was expected but not received. Huber-White standard errors reported in brackets. Net worth is divided by $100,000 and corrected for inflation using 1982-1984 as a base year for the CPI reported by the Bureau of Labor Statistics. *** Significant at 1 %, ** significant at 5 %, * significant at 10 %.

Table 7: Instrumental Variables Estimates of the Effect of Wealth on Entrepreneurial Entry Instrumental Variable Estimation of Specification (1) IV: Full Sample

IV: Low Capital

IV: High Capital

IV: Professional Services

Coefficient on Net Worth

0.010 (.179)

-.276 (.367)

0.285 (.209)

-.656 (.375)*

Sample Size

6,700

1,007

3,609

1,423

Note: The two instruments are non-expected inheritance received and housing price residuals. We get similar results for using each instrument separately so we only report the regressions with both instruments. The sample is comprised of individuals who are part of the labor force. Analysis is restricted to those who are not self-employed in period t (the previous period). Data about asset values was collected by the NLSY in 1985 to 1990 and 1992 through 1994. Robust standard errors reported in brackets. Net worth is divided by $100,000 and corrected for inflation using 1982-84 as the base year for the CPI reported by the Bureau of Labor Statistics. *** Significant at 1 %,** significant at 5 %,* significant at 10 %.

29

Table 8: Testing the Relevance and Validity of Instruments Full Sample

Low Capital

High Capital

Professional Services

Smith-Blundell test of exogeneity

0.395 (.530)

1.245 (.265)

0.708 (.400)

1.650 (.199)

Partial F statistic

1052.45 (.000)

203.85 (.000)

494.73 (.000)

346.22 (.000)

Sargan’s Over-ID statistic

0.550 (.458)

0.175 (.676)

0.406 (.524)

0.621 (.431)

Notes: The Smith Blundell test is a test of the instrumented variable’s exogeneity. The test statistic is distributed under the null as χ2 (m) where m is the number of explanatory variables specified as exogenous in the model. For the Partial F-statistic, the p-values are in brackets. Sargan’s OverID statistic is distributed χ2 (1) for the probit model, where 1 signifies the number of instrumented variables. P-values are reported in brackets for each statistic.

30

0.22

0.18

0.14

0.1

0.06

0.02 0

1

2

3

4

5

Net worth: $100,000

Figure 1: The probability of becoming an entrepreneur for different levels of net worth for the full sample. Graph is based on a polynomial specification, with all covariates held at their mean values. 95 percent confidence intervals shown.

Low Capitalization

0.22 High Capitalization

0.19

Professional Services

0.16 0.13 0.1 0.07 0.04 0.01 1

1.5

2

2.5

3

3.5

Net w orth $100,000

Figure 2: The probability of becoming an entrepreneur for different levels of net worth by industry. Graph is based on a polynomial specification, with all covariates held at their mean values.