Recession and the Resurgent Entrepreneur; National-Level Effects of the Business Cycle on European Entrepreneurship. Senior Thesis.

Transcription

1 Recession and the Resurgent Entrepreneur; National-Level Effects of the Business Cycle on European Entrepreneurship Senior Thesis Presented to The Faculty of the School of Arts and Sciences Brandeis University Undergraduate Program in Economics Adam Jaffe, Advisor In partial fulfillment of the requirements for the degree of Bachelor of Arts by Samuel Stemper May 2013 Copyright by Samuel Stemper

2 Stemper 2 Abstract: I examine entrepreneurship data from France, Germany, Spain and the United Kingdom to identify the relationship between entrepreneurship and the business cycle on a national level. I find strong support for divergence between country-level entrepreneurial responses to fluctuations in the business cycle. Country-level differences are most pronounced and consistent between France, whose population responds to high unemployment with a rise in entrepreneurial entry, and Germany, where is opposite is true. Additionally, I examine attitudinal and experiential drivers of entrepreneurship, such as perceived entrepreneurial skill and individual financing of early-stage businesses. In most countries, fluctuations in attitudes and experiences appear to reinforce existing direct effects of business cycles on entrepreneurship, but do so with less economic magnitude.

3 Stemper 3 Times of innovation are times of effort and sacrifice, of work for the future, while the harvest comes after. [ ] There is, thus, a good deal of truth in the popular saying that "there is more brain in business" at large during recession than there is during prosperity. -Joseph Schumpeter, 1939 Introduction Questions related to the contemporaneous effects of economic downturns on entrepreneurial activity have a long history in economic discussions, with conclusions that both validate and contradict Schumpeter s above assertion. These conclusions produce subsequently divergent conclusions regarding the possibility for recession as a mechanism for innovation and thus long-term social value. One of the primary avenues through which research exhibits conflicting empirical results is geographic; researchers have consistently identified variation in regional responses to recession on both an individual and aggregate level. While these findings are well-established in economic literature, the mechanisms through which this geographic variation operates remain undocumented. This paper identifies national differences in entrepreneurial responses to fluctuations in national labor markets and the economic, experiential, and attitudinal forces behind them. Europe serves as an excellent source of data in this regard. For one, the national level of aggregation provides a relatively diverse set of economic and social conditions that may have far-reaching implications on individual entrepreneurial flows. Additionally, there is a compelling public policy interest in promoting European entrepreneurship. In the wake of recessions across

4 Stemper 4 the European continent between 2009 and 2012, The Economist asserts that Europe s culture is deeply inhospitable to entrepreneurs and that business creation is particularly important for the future growth prospects of Europe s largest economies ( Time bomb at the heart of Europe 2012). In 2010, early-stage entrepreneurs made up just 4.2 percent and 5.8 percent of Germany and France s adult populations, respectively, well below America s 7.6 percent, China s 14 percent, and Brazil s 17 percent ("Les misérables" 2012). However, beyond the baseline gap in startup rates, there is reason to believe that entrepreneurship may be of even greater importance both during and immediately following economic downturns. Simon Parker notes that, particularly in the presence of slack labor markets, entrepreneurship stimulates growth directly, by removing a newly self-employed worker from the threat of unemployment, as well as indirectly, through eventual job creation (Parker 2004). Further, more than half of Fortune 500 companies were founded during recessions or bear markets ( Downturn, Start Up 2012), and managers who start in recessions are more risk-averse, take on less debt, and deliver less stock return volatility (Schoar et al 2011). In response to the rise in American entrepreneurship following the late 2000s recession, Robert Fairlie asserts that one positive byproduct of the recent severe recession is that a wide range of eventually-successful firms might emerge and contribute to the long-run economy (Fairlie 2011). A thriving entrepreneurial base in Europe could be a force for recovery and eventual economic stability across the continent. Using individual-level survey data provided by the Global Entrepreneurship Monitor, I propose a means to assess both national entrepreneurial responses to changes in unemployment as well as the mechanisms through which differences across countries may develop. I first consider the direct effects of unemployment on entrepreneurial entry as a measure of economic

5 Stemper 5 factors that influence the entrepreneurial response to recession. In addition, I estimate the indirect effects of business cycles on entrepreneurship that resonate as a result of change in attitudes and experiences amidst recession. Literature Review In accordance with my methodology, my literature review first considers the economic drivers of entrepreneurship, followed by more far-ranging and less documented attitudinal and experiential effects. The economic relationship between entrepreneurship and the business cycle is often considered within the context of a model first proposed by Evans and Jovanovic, in which would-be entrepreneurs pursue business formation only when the expected net income associated with self-employment exceeds that of wage work (Evans and Jovanovic 1989). Evans and Jovanovic offer the set of equations below to represent the entrepreneurial decision. Net income associated with wage work is:, where is the wage earned in the market, is the market interest rate, and represents the consumer s assets. Net income stemming from entrepreneurship is: ( ) ( ), where is a measure of entrepreneurial ability, ( ) is a production function whose only input is capital, is the amount of capital invested in the business, and is a disturbance.

6 Stemper 6 Within this static framework, individual workers maximize their net income by evaluating the expected payoffs of each venture and choosing the higher option. Given its specifications, such a model produces a wide range of contemporaneous effects associated with recession and there are subsequently multiple reasons to suggest that economic downturns might either encourage or discourage entrepreneurship that a recession may push entrepreneurs to start businesses or that an economic boom may pull them to do so. I refer to these hypotheses as recession-push and boom-time pull theories, respectively. Recession-Push Theory During recession, wage earners face lower wages within traditional venues of work, a reduction in and thus an increase in the relative attractiveness of entrepreneurship, ceteris paribus. Workers who remain employed in wage during recession face potential wage reductions due to reactionary cuts in hours or salaries. On the other end of the business cycle, recession-push theory suggests that robust economic growth discourages business formation. Indeed, an increase in associated with a booming private sector would, in this model, discourage entrepreneurship within the population of would-be entrepreneurs by making wage work more attractive. Evans and Jovanovic s model does not consider the potential for unemployment, suggesting that all workers will either find employment that reflects their level of education and experience or pursue entrepreneurship instead. However, adjusting the model to consider unemployment as a third designation of work status provides support for the recession-push theory for two reasons. First, workers who lose their jobs face dramatically lower wages when searching for new employment opportunities. Henry Farber has documented these wage reductions at length. Using

7 Stemper 7 data of recently unemployed American workers from 1981 to 1995, he found reductions in weekly earnings of 11 percent among respondents who were able to find work following a job loss (Farber 1997). Recessions increase the stock of unemployed workers and thus reduce the wage expectations of a large segment of the working population. One potential explanation for this reduction is geographic workers who cannot find employment that reflects their level of education and experience in local labor markets may be pushed into lower-wage work due to the high cost of moving in pursuit of more lucrative work. Secondly, a recessionary increase in this stock of unemployed workers may push that population increasingly towards alternatives to wage work, such as entrepreneurship. Henry Farber documented such an effect among American job losers in 1994 and 1996, finding that this population was more likely to pursue alternative venues of employment such as on-call or contract work during the following year (Farber 1999). Recessions increase the size of a population with a documented propensity for alternative work. Boom-time Pull Alternatively, there are reasons to believe that a recessionary environment may discourage business formation. First, a reduction in aggregate demand reduces the available market for goods and services provided by new firms. Concerns over constrained initial sales may discourage business creation during recession. While Evans and Jovanovic treat aggregate demand as exogenous, it can be interpreted as a fall in within their proposed framework. Second, constrained lending, modeled as a more binding restriction on, may discourage entrepreneurship. This may be particularly true with regard to capital-intensive business ventures, such as those in manufacturing. Evans and Jovanovic formally argue that liquidity constraints discourage would-be entrepreneurs (Evans and Jovanovic 1989). Parker notes that

8 Stemper 8 more labor-intensive jobs, such as those in one-to-one personal services and seasonal jobs with erratic demand, display naturally higher rates of entrepreneurship (Parker 2004). Assuming the marginal product of the entrepreneurial production function, ( ), is higher than the market interest rate,, a restriction on lending will reduce the value of entrepreneurial net income. This effect may be aggravated by a reduction in the value of investment assets, such as house prices or savings in securities. A business owner that must borrow to invest in a new business idea needs collateral, and if the owner s assets are worth less, there is less collateral and hence less liquidity. Empirical Observations Just as there is a broad range of theoretical predictions associated with recession and entrepreneurial entry, there is a similarly diverse set of empirically reported results. Due to the breadth of existing research, it is most helpful to consult research reviews, such as those produced by Nigel Meager (1992) and Simon Parker (2004). In his analysis of the existing empirical literature related to self-employment and the business cycle, Meager emphasizes the necessity of using flow, rather than stock, selfemployment measures in assessing the validity of recession-push and boom-time pull theories; self-employment must be measured with respect to the movement into and out of business ownership. He concludes, there are good theoretical reasons to believe that there will be no consistent time-series relationship between the stock of self-employment and the economic cycle (emphasis mine), though with respect to flows into self-employment it would appear that both the unemployment and the GDP growth variables are seen as "push" factors, affecting entry to self-employment within published empirical research (Meager 1992).

9 Stemper 9 Parker echoes Meager s methodological assertion, stating that the search for cyclical patterns in self-employment stocks is inappropriate. From there, the author divides the existing literature between cross-sectional, time series and panel analyses. His conclusions are mixed. Specifically, the conclusions drawn from cross-sectional and time-series diverge, displaying negative and positive effects of unemployment on self-employment, respectively. Finally, Parker asserts that In view of the differences between empirical results obtained from the cross-section and time-series approaches, it is perhaps unsurprising that panel data studies that combine crosssection and time-series elements have generated mixed results. Parker rightly points out that Most of this work has been conducted using UK and US data and later acknowledges that empirical research consistently exhibits large and significant regional difference in rates of entrepreneurship (Parker 2004). David Blanchflower formally documents these effects in the analysis of self-employment panel data across 23 countries between 1975 and Through the use of an interaction term between national unemployment and country dummy variables, Blanchflower finds considerable variation across countries in the influence of unemployment, both in terms of the direction and magnitude of any effect. While the author notes a stronger base of support for negative effects of unemployment on self-employment (boom-time pull theory), there is evidence of positive effects in a number of countries. Beyond the recognition of this divergence, researchers have not yet provided an explanation for how or why these regional differences exist (Blanchflower 2000). Regional Variation As mentioned previously, independent of responses to unemployment, baseline entrepreneurial rates between countries are extremely varied. In addition to cultural traditions

10 Stemper 10 and role models that support or discourage entrepreneurship, Parker counts low levels of demand, [ ] poorly educated workers, [and] [ ] high concentrations of capital-intensive industries that sustain effective barriers to entry among the primary forces reducing relative rates of entrepreneurship within a particular region (Parker 2004). Parker presents these factors as explanations for static differences in levels of entrepreneurship. In an attempt to identify attitudinal and experiential drivers of recessionary entrepreneurship, let us investigate how these relationships might operate in response to business cycles in a dynamic framework. Traditions and Role Models Regarding culture, one could infer that recessionary environments could bolster positive conceptions of entrepreneurship, through the recognition of the entrepreneur as an agent for economic growth and recovery. Countries in which positive conceptions of entrepreneurship proliferate during recession are likely to see these indirect effects of unemployment increase entrepreneurial flows. Also plausible is the possibility that entrepreneurs become culturally associated with the business failures of the recession itself, leading to the opposite negative effect. Such countries will exhibit negative indirect effects of unemployment. Researchers have found that additional attitudinal factors, namely economic expectations, strongly impact personal economic decisions, such as those related to employment. Specifically, research in psychology confirms that higher risk tolerance in everyday money matters is associated with increased economic expectations (Grable 1999). Recessions may cause would-be entrepreneurs to undervalue the expected returns associated with risky business ventures, independent of whether such concerns are warranted. In this paper, a variable representing individuals fear of entrepreneurial failure serves as the primary mechanism through which the effects of traditions, role models, and economic

11 Stemper 11 expectations may indirectly affect entrepreneurship through unemployment. Specifically, countries that display lower levels of fear during recession will, ceteris paribus, display stronger support for recession-push theory, as fear of business failure is assumed to be inversely related to entrepreneurial entry. Levels of Demand Because the unemployment rate, my independent variable of interest, relates directly to fluctuations in levels of demand, evaluating its net effect would entail delineating between longterm differences in demand between countries and short-term deviations from these differences. As an attempt to do so, I run additional regressions as robustness checks, in which unemployment rates are measured in standard deviations from their country-level means. The methodological details of this process are described later. However, the nature of recession-push and boom-time pull theories would suggest that there is no reason to suspect that recessionary entrepreneurial responses might differ based on national baseline levels of demand, particularly because my analysis is concerned with flows into entrepreneurship, rather than existing stocks of business owners. Education Blanchflower reports that the self-employment is more prevalent at the two ends of the education spectrum (Blanchflower 2000). While there is evidence that exposure to recession may have long-term effects on individual educational trajectories (Shu 2011), there is little reason to believe that these decisions immediately and significantly adjust the stock of would-be entrepreneurs within any specific range of the education spectrum. However, one relevant measure that does change in response to business cycles is individual perception of skills and abilities. As is detailed later, I control my fully-specified

12 Stemper 12 regressions using a binary variable representing a self-reported positive perception of entrepreneurial skills. Within this framework, it is reasonable that recessions may produce educational cohort effects that produce differences in population-level perception of skills. For example, if highly-educated cohorts are more likely to perceive their skills positively in light of recession, it follows that countries with more highly-educated populations would display stronger evidence for recession-push theory. Capital Intensity Considering industry concentrations, capital intensity, and local economic conditions provides a more structured framework with regard to understanding national-level variation in entrepreneurial responses to recession. Independent of macroeconomic conditions, countries dominated by large firms with high capital requirements limit the fraction of the workforce whose experiences are conducive to entrepreneurship. Further, recessionary liquidity constraints may most strongly discourage entrepreneurial entry within countries with large bases of capitalintensive industry or capital-intensive entrepreneurship more specifically. Recessionary borrowing constraints may disproportionately hamper the formation of firms in capital-intensive industries due to the barriers to entry associated with recession, which may be felt more strongly among would-be entrepreneurs in manufacturing than those in service work. It would follow that countries with a relatively smaller base of manufacturing and industry would be expected to display evidence for the recession-push theory with greater magnitude and significance. This paper centers on the countries of France, Germany, Spain, and the United Kingdom from 2001 to Using the percentage of GDP composed of industry (measured in 2008) as a proxy for sectoral capital intensity, the sample countries rank as follows: Germany (33.4

13 Stemper 13 percent), Spain (30.1 percent), the United Kingdom (23.4 percent), and France (20.6 percent) ( World Factbook Download ). If, as proposed above, recessions impact entrepreneurial inflows primarily through increased liquidity constraints, the United Kingdom and France, given the prominence of their service-sector, would show evidence with regard to the recession-push theory that diverges strongly from that of Germany and Spain. Within this framework, the direction of these effects is dependent on the disproportionate effect of borrowing constraints during recession on capitalintensive industries. Though there is a wide array of economic, attitudinal, and experiential factors that drive geographic differences in entrepreneurial responses to recession, my data allows for a robust assessment of these potential effects. Data My empirical questions demand a dataset that contains individual-level data across years, countries, and ages with a sufficiently large number of observations, adequate demographic information, and clear measures of entrepreneurship. Survey data provided by the Global Entrepreneurship Monitor s (GEM) Adult Population Survey meets these requirements. The largest ongoing study of entrepreneurial dynamics in the world, the Global Entrepreneurship Monitor provides an annual assessment of entrepreneurial activity, aspirations and attitudes of individuals across a wide range of countries. My dataset includes individual-level responses to surveys from France, Germany, Spain, and the United Kingdom from 2001 to The responses by year and country are summarized below: France Germany Spain UK Total ,216 4, ,882

14 Stemper ,303 9, ,460 20, , , , , ,511 5,226 14,918 9,297 30, ,408 3,165 20,415 35,411 60, , ,201 32,649 55, ,471 3,634 25,184 4,909 35,198 Total 11,124 26,348 81,718 91, ,916 Missing country-year combinations are a result of missing income data, which is treated as a necessary control. Additionally, a number of observations have been dropped because of missing endogenous variables related to attitudes towards entrepreneurship. Methodological processes and issues associated with missing data are addressed within the methodology section below. Global Entrepreneurship Monitor surveys contain a multitude of questions related to individual entrepreneurial activity and attitudes towards business formation. There is a broad set of information with respect to both individual inflows to entrepreneurship as well as relevant attitudes and experiences related to business formation. In addition to exogenous control variables representing age, gender, education (dummies representing secondary, post-secondary, and graduate education), and income brackets (dummies representing top, middle, and bottom thirds of national income distributions), I generate a group of endogenous variables meant to measure entrepreneurship at large as well as experiences and attitudes with respect to entrepreneurship more generally. As a baseline measure of entrepreneurial inflows, I generate a dummy variable, labeled, representing a yes response to the question:

15 Stemper 15 "Are you, alone or with others, currently trying to start a new business, including any self-employment or selling any goods or services to others?" Without regard to explanatory factors, there is a high degree of variation in rates of entrepreneurship across countries and years. Entrepreneurial Inflows by Year Year France Spain Germany UK

16 Stemper 16 Drawing upon each observation s survey year and country, I generate a variable for each individual observation representing the unemployment rate within the country in the year the survey was conducted. Data is sourced from the OECD. Unemployment data displays a relatively high level of variation, particularly in 2007 and 2008, when falling unemployment rates in Germany and France stood in stark contrast to climbing unemployment rates in Spain. National Unemployment by Year Year France Spain Germany UK

17 Stemper 17 Absent of controls for individual demographic factors, entrepreneurship displays a mildly positive relationship with respect to national unemployment rates, reflecting mild support for the recession-push theory. However, there is a high level of entrepreneurial variation around the median range of unemployment values between seven percent and nine percent. More importantly, this level of analysis addmittedly omits all controls, particularly country-fixed effects and country-unemployment interactions that allow the direction of the unemployment effect to vary by country. Entrepreneurial Inflows and National Unemployment Rate of Unemployment France Spain Germany UK

18 Stemper 18 Finally, I generate four endogenous variables related to entrepreneurial attitudes and experiences. As measures of attitudes towards entrepreneurship, I generate the binary variables and to represent respective yes responses to the following prompts: Fear of failure would prevent you from starting a business. You have the knowledge, skill and experience required to start a new business. GEM data displays means of 42.9 and 49.1 percent of respondents for the above equations respectively. These means are noteworthy particularly because their magnitude eliminates the possibility that, when included in regressions, these variables might simply serve as proxies for entrepreneurship. Nearly half of respondents believe they are adequately prepared to start a business. However, in no country-year combination does the rate of entrepreneurship exceed nine percent. The fact that only a fraction of people who believe they are qualified to become an entrepreneur actually do so indicates that these questions asked are measuring different qualities of the respondent. Equivalently, over 40 percent of respondents would let fear of failure prevent them from becoming an entrepreneur yet a much larger percentage do not engage in business creation, again indicating that the questions measure two unique characteristics among respondents. Finally, I measure recent entrepreneurial experiences using two additional binary variables, and, representing yes responses to the following prompts:

19 Stemper 19 You have, in the past three years, personally provided funds for a new business started by someone else, excluding any purchases of stocks or mutual funds You have, in the past 12 months, sold, shut down, discontinued or quit a business you owned and managed, any form of self-employed, or selling goods or services to anyone These binary endogenous variables representing experiences display far lower means than attitudinal responses. 2.9 and 2.0 percent of respondents answer yes to the above questions, respectively. Methodology I use a probit model to estimate the country-specific effect of changes in national unemployment, measured by the interaction term of national unemployment and country dummy variables, on individual-level probability of pursuing entrepreneurship, holding demographic, country fixed-effects, and year fixed-effects constant. To account for missing data in the initial dataset Germany, Spain, and the United Kingdom between 2001 and 2004, primarily I include country and year-fixed effects as controls in all regressions. My structural-form, fully specified regression is represented below: ( is included to represent individual age, gender, education, and income characteristics as well as country and year fixed-effects.)

20 Stemper 20 In the structural-form regression above, represents the remaining direct effect of country-level unemployment when holding the remaining attitudinal ( and ) and experiential ( and ) variables constant. However, it can be inferred that variations in the business cycle impact all of the endogenous variables in the system. For example, rising unemployment may change entrepreneurial decisions due to direct push and pull effects as well as indirectly through an increase or decrease in the proportion of the population that fears business failure. These endogenous variables also have causal relationships with country-unemployment interaction variables; these relationships are specified in the system represented comprehensively below:

21 Stemper 21 [1] [2] [3] [4] [5] Combining intercepts and error terms produces the following reduced-form equation. [6] ( ) ( ) where:

22 Stemper 22 There are thus two categories of effects stemming from country-unemployment interactions on individual entrepreneurial probabilities: direct and indirect. The prior refers strictly to, the remaining coefficient on country-unemployment interactions after controlling for endogenous attitudinal and experiential differences. The latter refers to,,, and, the coefficients on endogenous variables that enter the reduced form equation only through their indirect effects on current attitudes and recent experiences. These indirect effects function as a result of business cycles and affect entrepreneurship only circuitously. My analysis estimates equations [1], [2], [3], [4], [5], and [6] using probit. While probit estimated coefficients do not represent marginal effects, their values can be expressed as marginal coefficients evaluated at the mean of the data. Doing so allows for a more comprehensive evaluation of economic significance in addition to existing measures of statistical significance such as p-values and confidence intervals. I report all regression results as probit coefficients as well as probit estimated marginal effects at the mean. Country-unemployment interactions in equations [1] and [6] are evaluated for significance, both individually and with respect to their equality. The latter analysis serves as an evaluation of Blanchflower s observation that the direct effects of national unemployment on entrepreneurship vary geographically and is performed using the test command in STATA, producing a Chi-Square value and a corresponding probability for the equality of coefficients. If Blanchflower s observation holds true within the GEM Adult Population Survey data, Chi- Square tests should find that coefficients on country-unemployment variables that are unequal with statistical significance. Estimating equations [1], [2], [3], [4], [5], and [6] allows for an approximation of the components of the total effect of recession on entrepreneurship, both direct and indirect.

23 Stemper 23 Estimating these values presents two methodological difficulties. Firstly, Global Entrepreneurship Monitor data is missing for a select group of variables and observations within relevant countries and years. Secondly, the system cannot be modeled comprehensively without an exclusion restriction, allowing for an instrumental variable regression, which is unavailable within the data. I will examine each of these difficulties individually. Missing Data The primary missing variables are the two endogenous attitudinal controls, and. As is necessary to estimate the fully-specified relationships in equation [1] as well as the endogenous relationships in equations [2], [3], [4], and [5], I generate a reduced dataset by dropping all observations for which,,, and are missing. The demographic characteristics of this new dataset are very similar to its larger source, as evidenced by the values represented in the table below. This suggests that there doesn t appear to be a systematic demographic explanation associated with missing responses. However, there is a clear upward bias with respect to my primary independent variable,, as well as the two experiential entrepreneurial variables, and. Full Dataset Mean Partial Dataset Mean Age Gender Middle Income High Income Secondary Education Postsecondary Education Graduate School Startup Fear Failure Skills Required Funding Shutdown

24 Stemper 24 The source of this bias originates primarily from and, which are missing for a number of non-entrepreneurs. Given that the interview prompt for appears in Global Entrepreneurship Monitor questionnaires prior to that of the other endogenous variables such as and, it can be inferred that interviewers were unable to elicit responses to additional questions after non-entrepreneurs identified themselves as such. Instruments Given its partially recursive nature, the specified entrepreneurial system cannot be modeled comprehensively without an exclusion restriction, allowing for an instrumental variable regression. Such a framework necessitates a variable that explains the endogenous variables,, for example, but which does not affect the likelihood of except through its impact on. Unfortunately, the dataset does not offer any clear candidates for instruments, particularly because of the potential for correlation between error terms of endogenous equations in the entrepreneurial system. While this limits the statistical exactness of the system as a whole, this biasedness does not affect the estimation of the reduced form, equation [6]. To be clear, this allows for an unbiased estimation for the value of ( ), the combined (direct and indirect) effect of unemployment on entrepreneurship. However, potential correlation between error terms implies that equations [2], [3], [4], and [5] cannot be estimated without bias. Additionally, because estimating the reduced form doesn t require the inclusion of and, it can be completed using the complete dataset. Doing so allows for an additional check with respect to the magnitude and direction of the predicted value of the comprehensive effect.

25 Stemper 25 The estimation of these relationships, though imperfect, remains informative with respect to the mechanisms through which push and pull effects resonate in European entrepreneurship data. Specifically, while the estimated values of,,, and are econometrically flawed, regressions can provide insight into their direction and magnitude. Additionally,,,, and have relationships with entrepreneurship which can be statistically estimated, albeit with a biased dataset. Finally, my estimation of the reduced form with both restricted and unrestricted datasets allows for an evaluation of the extent of bias stemming from correlated error terms between endogenous variable regressions. Results Entrepreneurship and Direct Effects of Unemployment My fully-specified regression, detailed fully in Appendix A, provides strong support for the hypothesis that the direct effects of national unemployment on entrepreneurship vary geographically, in both direction and magnitude. Respondents in France and the United Kingdom exhibit strong, resurgent entrepreneurial responses to rising unemployment rates; their counterparts in Germany do the opposite. The Spanish entrepreneurial response to high unemployment is mildly positive. These effects are summarized in the table below.

26 Stemper 26 Direct Country-Unemployment Interactions within Probit Regressions on using Restricted Dataset (t-statistics in parentheses) France Germany Spain United Kingdom Probit Coefficient 0.403*** (6.89) ** (-2.81) *** (3.81) 0.225*** (3.92) Marginal Effects (evaluated at mean of data) 0.030*** (6.89) ** (-2.81) 0.004*** (3.81) 0.017*** (3.92) * p<0.05, ** p<0.01, *** p<0.001 Taken individually, all country-unemployment interactions are statistically significant with 99 percent confidence. Further, results for France and the United Kingdom are strongly economically significant, given their magnitude relative to the population mean of within the data, 5.5 percent. Based on the marginal probit results evaluated at the mean of the data, an increase of one percent in the French or UK unemployment rate increases the likelihood of an individual in their population creating a business by 3 percent and 1.7 percent, respectively. The corresponding effects are mildly positive in Spain (0.4 percent) and mildly negative in Germany (-0.3 percent). In addition to the significance of the variables themselves, I test the results with respect to their differences between countries. The results strongly support the view that the direct effect of unemployment varies geographically, as the equality of unemployment coefficients between all six national combinations is rejected with 95 percent confidence. With the exception of France

27 Stemper 27 and the United Kingdom, the equality of the remaining five combinations is rejected with 99 percent confidence. In the fully-specified regression, most demographic variables operate as expected. Specifically, the variables which predict entrepreneurship with the most economic significance are those related most closely to the entrepreneurial decision. Respondents who identified themselves as current students or retired or disabled workers were far less likely to be entrepreneurs. Age and income display limited economic or statistical significance in entrepreneurial regressions. With respect to education, there appear to be strong entrepreneurial returns to graduate school, but limited predictive power when comparing groups with lower education levels. It is quite likely that the inclusion of attitudinal variables, specifically, the measure of perceived entrepreneurial ability, diminished the significance of age, income, and education variables. Individuals who are older, belong to higher income brackets, or have higher-level degrees are much more likely to view themselves as highly skilled for business creation, which indirectly controls for demographic factors that would otherwise display greater significance. Fixed-effects related to respondent nationality and year of survey are strongly predictive of entrepreneurship. Regarding the prior, country fixed-effect coefficients are most significant in Germany and France. German respondents are much more likely to become entrepreneurs, when controlling for demographic and attitudinal factors. This effect, coupled with the mildly negative unemployment effect reported above, suggests that entrepreneurship in Germany appears to be largely attributed to innate differences in Germans, unrelated to demographic factors. Alternatively, respondents in France are mildly less likely to become entrepreneurs, controlling for demographic factors.

28 Stemper 28 Country fixed-effects are somewhat reflective of baseline gaps in entrepreneurship in the data, though it is worth noting that the negative coefficient associated with France exists in spite of the spike in French entrepreneurship between 2004 and 2008 in these five consecutive years, French entrepreneurship rates exceeded six percent and surpassed those of all other countries in the data. This fact, coupled with the significance of the French unemployment interaction coefficient, suggests that this spike can be attributed more heavily to mildly higher French unemployment rather than genuinely higher rates of baseline French entrepreneurship. With respect to the significance of year fixed-effects, it is plausible that country unemployment rates do not wholly reflect the economic conditions in a country or group of countries. Specifically, positive coefficients associated 2005 and 2007 may reflect increased concerns of the impending financial crisis that were not yet reflected in national unemployment data, pushing potential entrepreneurs away from traditional work venues. The similarly negative coefficient in 2003 may reflect the opposite confidence in the stability and growth prospects of European business that guided would-be entrepreneurs towards wage work. Entrepreneurship and Combined Effects of Unemployment Running regressions on the reduced form, when controls for attitudinal and experiential factors are removed from regression specifications, can be performed using both the restricted dataset, as was used in the fully-specified regression above, or the full dataset, as observations with missing endogenous attitudinal and experiential variables no longer present a methodological impediment. Reduced form results using the restricted dataset, detailed in Appendix B, are very similar to those reported previously. The directions country-level effects of unemployment are identical; again, my regression finds strongly positive entrepreneurial responses to recession in

29 Stemper 29 France and the United Kingdom, mildly positive responses in Spain, and mildly negative responses in Germany. Further, each coefficient increases in magnitude. This increased scale is a strong indication that the combined indirect effects of experiential and attitudinal factors reinforce, rather than weaken the existing direct effects of unemployment on entrepreneurial entry. These effects are reported in the table below. Combined Country-Unemployment Interactions within Probit Regressions on using Restricted Dataset (t-statistics in parentheses) France Germany Spain United Kingdom Probit Coefficient 0.405*** (7.46) ** (-3.12) *** (3.75) 0.294*** (5.46) Marginal Effects (evaluated at mean of data) 0.040*** (7.46) ** (-3.12) 0.005*** (3.75) 0.029*** 5.46 * p<0.05, ** p<0.01, *** p<0.001 The increased push effects exhibited with respect to French unemployment lessen the previously reported inequality with the United Kingdom. However, all five previous inequalities hold with 99 percent confidence, again a strong confirmation of geographic variance in direction and magnitude of recession-push and boom-time pull effects. Transitioning to the full, unrestricted dataset produces country-level unemployment coefficients that are similar in direction, but lessen in magnitude and statistical significance. Appendix C documents these regressions. Of the four countries represented, only unemployment in Spain is found to be economically significant with 99 percent confidence. Despite more

30 Stemper 30 pronounced boom-time pull effects represented in Germany, the unemployment variable is significant with only 80 percent confidence. France displays push effects with 85 percent confidence, while the United Kingdom shows no statistically significant effects of unemployment on entrepreneurial entry. Again, these effects are reported in the table below. Combined Country-Unemployment Interactions within Probit Regressions on using Unrestricted Dataset (t-statistics in parentheses) France Germany Spain United Kingdom Probit Coefficient (1.54) (-1.43) * (2.34) (0.56) Marginal Effects (evaluated at mean of data) (1.54) (-1.43) 0.002* (2.34) (0.56) * p<0.05, ** p<0.01, *** p<0.001 Collectively, these results suggest that the observations dropped due to missing attitudinal variables display unemployment responses that are either less explicable by changes in unemployment or divergent from country-level patterns. Nonetheless, given their direction, the results are not inconsistent with the previously reported recession-push effects in France, Spain, and the United Kingdom and the boom-time pull effects in Germany. To this end, the reported coefficient on German unemployment remains statistically unequal to both France and Spain with 95 percent confidence. Tests of equality provide evidence that combined boom-time pull effects dominate among German entrepreneurs and that this is not the case in their counterparts elsewhere in Europe.

31 Stemper 31 Indirect Experiential and Attitudinal Effects There are limited methodological means of analyzing the country-level components of the indirect effects of unemployment, primarily because the algebra necessary to evaluate indirect effects that are products of two coefficients is only possible when coefficients represent marginal effects and are estimated via OLS. Given the use of a nonlinear probit model in this paper, I am unable to approximate the indirect components of the combined effect numerically. For example,, the indirect effect of unemployment stemming from increased or decreased fear of failure, cannot be estimated arithmetically due to the use of probit in producing estimates of and. Nonetheless, regression estimates for system as a whole allow for qualitative observations with respect to the mechanism through which experiential and attitudinal factors reinforce direct recession-push effects in France, Spain, and the United Kingdom and direct boom-time pull effects in Germany. In other words, it is valuable not only to know how endogenous variables, fear of failure, for example, impact business startup, but also how fear of failure is impacted by changes in unemployment. Allowing for these changes to vary by country produces even more robust results. The table below reports the country-level marginal effects of unemployment on the four endogenous attitudinal and experiential variables.

32 Stemper 32 Marginal Effects in Probit Regressions using Restricted Dataset (t-statistics in parentheses) France Unemp 0.086*** (6.57) 0.047*** (3.42) 0.019*** (5.14) 0.016*** (4.67) Germany Unemp (-0.37) (1.28) *** (-5.14) *** (-4.83) Spain Unemp (0.003) (0.60) 0.003*** (3.12) (1.62) UK Unemp *** (-5.96) 0.072*** (5.50) 0.016*** (3.79) 0.016*** (4.94) Additionally, the table below reports the probit and marginal coefficients with respect to these endogenous variables effects on entrepreneurship, assumed to be constant across the dataset. Probit Regressions on using Restricted Dataset (t-statistics in parentheses) Probit Coefficient *** (-18.92) 0.822*** (59.74) 0.451*** (18.62) 0.413*** (14.39) Marginal Effects at Mean of Data *** (-18.92) 0.066*** (59.74) 0.049*** (18.62) 0.043*** (14.39) * p<0.05, ** p<0.01, *** p<0.001 Most striking within the endogenous variable regressions is the extent to which the direction of coefficients on experiential factors ( and ) are correlated with

33 Stemper 33 previously documented recession-push and boom-time pull effects. All of the coefficients operate in the same direction as those on and, with the exception of in Spain, all are significant with 99 percent confidence. Given that can also be considered an experiential variable, it is logical that there is a high level of correlation between entrepreneurial experiences. My results suggest that in recession-push countries such as France and the United Kingdom, economic downturns are met with an increase in entrepreneurial activity of all sorts: entry, funding, and entrepreneurial exit. Further, the latter two indirectly increase the rate of business formation in response to recession. The opposite is true in boomtime pull countries such as Germany, where recession decreases all three measures of entrepreneurial experiences. However, the indirect effect of experiential variables on entrepreneurship is likely a small one, given the small population of effected individuals; within the dataset as a whole, only 2.9 percent of survey respondents identified themselves as engaged in funding a new business, while the percent experiencing recent entrepreneurial exit stood at 2.0 percent. While it appears that experiential factors reinforce documented recession-push and boom-time pull direct effects, there is little evidence to suggest that these effects are equal in magnitude as the documented direct effects in the fully specified regression in Appendix A. There is less consistency with respect to attitudes and predominance of recession-push and boom-time pull effects. The United Kingdom displays the strongest consistency, where recession decreases fear of failure and increases perceived entrepreneurial skills, both of which indirectly increase rates of entrepreneurial entry. Both of these effects reinforce the positive direct effects of unemployment in the United Kingdom. Survey respondents in France display increased perceived skills and increased fear of failure during recession, which have offsetting

34 Stemper 34 effects with respect to entrepreneurial entry. Unemployment doesn t appear to alter German or Spanish attitudes in any statistically significant direction. High unemployment positively impacts perceived entrepreneurial skills most strongly in established recession-push countries France and the United Kingdom. Given the relatively high coefficient associated with perceived skills and probability of entrepreneurial entry (probitestimated marginal effects of 0.066) and the relatively high base rate of positive responses (49.1 percent overall in the data),the results suggest that increased perception of entrepreneurial skills has a strong indirect effect on entrepreneurial entry during recession within recession-push countries. This effect may in part be a reflection of education cohort effects; the percent of respondents with a postsecondary degree is highest in France (40.4 percent), Spain (38.4 percent), and the United Kingdom (37.3 percent), compared to Germany, which exhibits much lower levels of higher education (27.2 percent). It is reasonable to assert that higher-skilled populations may display resurgent perceptions of skills during economic downturns. This cohort effect would in part explain the relatively high coefficients associated with perception of skills in response to unemployment observed in France and the United Kingdom. Robustness Checks As specified, my regressions account for differences in baseline levels of unemployment. To ensure that differences in unemployment effects appropriately reflect business cycles at the national level, I run separate regressions using a standardized measure of unemployment, in which each unemployment rate is measured as its difference from the national average over the eight year period and divided by that period s standard deviation. In this specification,

35 Stemper 35 unemployment numbers are represented in standard deviations from their national average, rather than absolute percentages. Standardizing my measure of unemployment to represent standard deviations from national averages yields quite similar results, with identical directions and t-statistics, as detailed in Appendix D. One very minor exception, however, relates to the equality of unemployment variables between countries. In my initial specification, country-unemployment interactions in Spain and the United Kingdom were statistically different. However, when unemployment is measured in standard deviations, the coefficients do not exhibit statistically different effects on entrepreneurship. This difference operates as a result of the large difference in unemployment variation between the two countries. Between 2001 and 2008, the unemployment rate in the United Kingdom remained relatively stable, while unemployment in Spain was far more volatile. The standardized unemployment measures in the United Kingdom and Spain subsequently display relatively similar values, which generate numerically closer coefficient estimates. These estimates thus display less statistical inequality. All other national differences observed previously remain present. Conclusion The positive, push effects of recessionary environments operate most forcefully in countries in which individuals perceive higher levels of skills in the midst of economic slowdowns. In such countries, attitudinal, experiential, and direct effects appear to operate in conjunction with one another; recessions produce populations with higher perceived skills, less fear of failure, and more recent exposure to entrepreneurial experiences, all of which reinforce the existing positive direct effects of unemployment on entrepreneurship. Despite this reinforcement, the magnitude of the direct effect of unemployment appears to outweigh the