Finding Success in Tragedy: Forced Entrepreneurs after Corporate Bankruptcy Isaac Hacamo and Kristoph Kleiner †

May 2016

Abstract Using linked employer-employee data that covers most of U.S. firms and contains full resum´es of workers, we document a channel through which corporate bankruptcies leads to a productive reallocation of human capital: self-employment and new firm creation. To causally identify the effect of forced unemployment on entrepreneurship, we compare workers in firms that experienced different bankruptcy outcomes as measured by the severity of layoffs around the filing date. The identification assumption is that two employees in the same occupation, with the same years of experience, education, and firm tenure, do not anticipate the bankruptcy outcome when joining the firm at least 4 years prior to the first bankruptcy filing. Comparing these employees, we find that forced unemployment increases self-employment rates by 2 percentage points, and starting an employer firm by 0.3 percentage points within two years after the bankruptcy. A displaced worker is likely to create a new firm (relative to self-employment) when she is college-educated and/or younger, but only when she had a high-skilled occupation in the bankrupt firm. Given the success of these firms in both survival and employment, our results highlight the potential productivity gains from reallocating skilled labor through bankruptcies.

†

Finance Department, Kelley School of Business, Indiana University. Address: 1309 E 10th St, 47405, Bloomington, IN. Email: [email protected] and [email protected]. We thank Noah Stoffman, Andrew Ellul, Matt Billett, Antoinette Schoar, Eitan Goldman, Scott Smart, and Nandini Gupta for very helpful suggestions. We also thank participants of the Kelley Finance Brownbag at Indiana University for their comments.

“Nobody offered me a job, I was probably too proud to go look for one, and I said well why not start your own company.” —Michael Bloomberg

1.

Introduction

The misallocation of resources hinders economic productivity [Davis and Haltiwanger, 1992, Hsieh and Klenow, 2009], suggesting a potential role for reallocating labor and assets away from distressed firms and towards more productive sectors of the economy [Aghion, Moore, and Hart, 1992]. Currently, empirical support for the success of the reallocation process has been mixed. On one hand, due to frictions including illiquid asset market with high specificity [Shleifer and Vishny, 1992], and a lack of unwilling buyers [Williamson, 1988], liquidated firms experience low occupancy and utilization of assets [Bernstein, Colonnelli, and Iverson, 2015]. Yet, in the event of a buyer, assets are more likely to be sold by low productive firms to high productive firms, and experience a gain in productivity upon acquisition [Maksimovic and Phillips, 2001, Schoar, 2002]. In the case of labor reallocation, workers caught in corporate bankruptcies experience long-term wage loss, especially after leaving the firm or even industry, and researchers suggest a loss of human capital [Graham, Kim, Li, and Qiu, 2013a]. Evidence of labor productivity gains through firm distress remain scarce, handicapped by absent data on labor productivity at the worker level [Schoar, 2002, Maksimovic and Phillips, 2001]. This paper documents a mechanism through which human capital is productively reallocated after bankruptcy: the promotion and success of new firms. After worker displacement, employees commonly face search frictions [Sahin, Song, Topa, and Violante, 2012], potentially resulting in wage and unemployment risk. Since prospective entrepreneurs equate these risks against entrepreneurship risk [Hamilton, 2000], worker displacement after bankruptcy might lead to self-employment and firm creation. While some fraction of this entrepreneurship activity may only be a self-employement stop-gap until the entrepreneur finds a new employer, another fraction might lead to successful firm creation. Firm creation is a reasonable proxy of productivity given young firms are adept at seizing economic opportunities [Adelino, Ma, and Robinson, 2014], potentially explaining their disproportionate share of net employment growth [Haltiwanger, Jarmin, and Miranda, 2013]. Using a hand-collected database that links employers to employees over the worker history, this paper doc-

1

uments that forced unemployment through bankruptcy leads to a significant fraction of successful firm creation; and that conditional on creating a firm, the survival rate and employment growth is similar to firms that are created in general circumstances, presumably by voluntary entrepreneurs. A priori, it is not clear such a mechanism would be economically significant, or even exist at all. Given the high levels of risk associated with entrepreneurship [Scott, Shu, Lubynsky, et al., 2015, Hombert, Schoar, Sraer, and Thesmar, 2014], displacement may affect a worker’s opportunity set: displacement is followed by a long-term decrease in wages as well as an increase in wage volatility [Jacobson, LaLonde, and Sullivan, 1993], and job search imposes heavy constraints on time [Aguiar, Hurst, and Karabarbounis, 2013]. This is particularly relevant to more senior workers as tenure leads to firm-specific human capital and decreases the likelihood of unforced exit [Topel, 1991]. However, job losses may impose additional financial constraints, making it harder to start a small business; this will be particularly common in a period of low economic demand [Parker, 2009]. Compounding these concerns, any newly-created firms may be only a stop-gap measure until the founder rejoins the workforce [Delmar and Davidsson, 2000]. Therefore, displacement may only lead to temporary self-employment, not transformational entrepreneurship [Schoar, 2009]. Documenting that such a mechanism for small firm creation exists—and results in successful entrepreneurs—requires detailed work histories on both salaried and self-employed individuals, and measures of firm growth over the lifecycle of each new establishment, which have not previously been available. We overcome this obstacle by developing a new dataset of individual employment histories obtained from public profiles available on one of the largest online business networking service. This online business platform includes employment histories for over 365 million users in over 200 countries, with approximately 30% of users (110 million people) living in the United States. The employment histories come from self-reported r´esum´es that detail the career-path of the individual. As we explain in detail below, we construct a large panel dataset by sampling from this data. Similar to panels of employee-employer linked data used in prior research, our data includes employment histories over long periods of time [Jacobson et al., 1993]. However, in contrast to existing studies, we also have information on occupations, including firm founder, detailed education achievement, including degrees and majors, and firm characteristics, for even the smallest start-ups. To causally identify the effect of job displacement during bankruptcy on the propensity of becoming an entrepreneur, we use a double matching procedure in which we first match firms and 2

then match employees. We match firms that file for Chapter 11 Bankruptcy within two years apart. We then match employees between this firms that have equal job occupations, years of experience, educational degrees, and firm tenure. When comparing employees between different firms, the source of variation is the size of layoffs during the year when the firm files for bankruptcy for the first time. However, employees are able to anticipate financial distress in the year leading up to the filing for bankruptcy [Brown and Matsa, 2012a]. This is because founder characteristics such as management ability [Nanda, 2008], risk tolerance and ambition [Hurst and Pugsley, 2011], or age [Ouimet and Zarutskie, 2014] may affect the likelihood of entrepreneurship or success, and might be correlated with the self-selection into firms in the two years leading up to the bankruptcy. To address this potential violation of the exclusion restriction, we impose that employees joined the firm at least four years prior to the bankruptcy. The exclusion restriction is guaranteed if employees with higher propensity to become entrepreneurs do not self select themselves to work in firms that are likely to suffer large job losses during the bankruptcy that happens at least four years after the employee joins the firm. Using this sample of worker matches, we study how corporate bankruptcy results in new firm creation and facilitates economic growth in both employment and productivity. We find that displaced workers are 3% likely to engage in self employment or creating a new firm outright within two years, with 25% of these cases actually resulting in employer firms. The result is driven predominantely by the significant displacement during this period: upon filing bankruptcy, firms lose 38% of their workforce that year with 27% of the remaining workers leaving the following year. Yet, even conditional on leaving the firm, we find a significant increase in the number of employer firms. The results hold under a range of specifications and are not driven by a particular time period, geographic location, or industry. Instead, we determine that human capital plays a predominant role in new firm creation after displacement. While age and number of years of work experience is a strong predictor of self-employment, employer firms are largely started by workers with managerial experience and graduate educations. Specifically, displaced workers with managerial experience are four time more likely to start an employer firm than other workers; even more striking, 4.4% of workers with executive experience start an employer firm.

3

We observe three outcomes for self-employed entrepreneurs, and two outcomes for founders of new firms. Over half of self-employed remains self-employed within ten years, other returns to the workforce, and 2% of self-employed succeeds into turning their entrepreneurship endeavor into a new firm that subsequently hires new employees. For founders of new firms immediately after corporate bankruptcy, we observe that a higher level of firm survival, while a small fraction fails and returns to the workforce. Mean employment for these firms is 16 employees with 10% have at least forty employees. We observe that the firms created by forced entrepreneurs are more likely to be employer firms (and have more employees) than firms voluntarily created by plausible counterfactual entrepreneurs. While displacement results in longer self-employment, we find equal survival rates among all employer firms. In Section 2, we review the related literature. In section 3, we develop the motivation for the research with aggregate level data. In section 4, we discuss the empirical methodology. In section 5, we discuss the data, section 6 describes the results, and section 7 concludes.

2.

Literature Review

Closest to our work is a recent paper by Babina [2015]. Using linked employee-employer data from the Census, she determines that financially-distressed firms are especially at risk of losing productive workers to found or join start-ups. We view this paper as complementary to our own as we instead focus on a separate group, individuals that lose their employment during a corporate bankruptcy and corresponding mass layoff. The distinction is important: while Babina [2015] focuses on the cost of financial distress to the firm we focus on displacement costs to the worker. Specifically, Babina [2015] argues that the costs of financial distress to the firm are potentially greater than previously known, using new firm creation as a measure of an individual’s talent. Alternatively, our work instead focuses on displacement costs at the worker-level. Under our alternative measure, new firm creation, our research suggests that the cost of displacement might be overstated to both the individual and the economy. Only by adding both forces can we estimate the full impact of firm distress on new firm creation. The separate focus between the two projects leads to distinct data sources. Babina [2015] depends on Census data which (unlike our data) is representative of all industries and worker types.

4

However, her analysis is significantly limited to employer firms and does not include any Schedule C self-employment. Given that the median firm in the US is defined as nonemployer, she excludes over half of all firms and is unable to determine the unconditional success of these new enterprises. In comparison, our data is biased towards particular industries and younger workers with college education. Yet, we are not limited by firm size and so can make statements on the success of all workers, not just those that reach employer status. The richness of the dataset allows us to study the job creation of small and young firms, focusing on a specific barrier to entrepreneurship: the risk of wage loss.Early papers by Birch [1979, 1987] illustrate the high employment growth of small firms, and these insights have held in light of finer data [Neumark, Wall, and Zhang, 2011]. Perhaps closest to our own efforts is Haltiwanger et al. [2013], who detail that it is specifically young firms-not small firms- that facilitate job creation. We expand this literature: by following the flow of workers between firms we are able to study the role of new firms in offsetting forced unemployment of older companies. In this way we can more broadly examine the economic implications of alleviating barriers to entrepreneurship, similar to studies on financial constraints [Hurst and Lusardi, 2004, Schmalz, Sraer, and Thesmar, 2013b], and government regulation/reforms [Cetorelli and Strahan, 2006, Kerr and Nanda, 2009]. We add to this work by (i) focusing on displacement, (ii) developing a unique control worker for each treatment, (iii) including substantial information on each worker, (iv) extending the estimation to a range of time periods, and (v) focusing on the US economy. Particularly close to our work are studies on individual or aggregate interventions that defray the risk of entrepreneurship [Scott et al., 2015, Hombert et al., 2014], yet the participants of any intervention are rarely random. In our paper, we instead argue that potential business owners must also equate entrepreneurial risk with alternative career paths, most notably, the stable income provided by employment in existing firms. By focusing on this displacement we can abstract from the issue of worker participation. We are certainly not the first to document this relationship between labor income shocks and entrepreneurship. Unemployed workers are twice as likely to transition to self-employment [Evans and Leighton, 1989, 1990]. Other researchers have instead suggested it is the history of job changes [Alba-Ramirez, 1994] or unemployment duration that actually predicts entry to entrepreneurship [Praag and Ophem, 1995]. In contrast, Farber [1999] finds that individuals are less likely to enter into self-employment following a layoffs compared with those not laid-off; however, the results switch 5

conditional on employment. Regional or national data similarly develops a positive relationship between unemployment rates and future entrepreneurship [Thurik, Carree, Van Stel, and Audretsch, 2008, Fairlie et al., 2010, Koellinger and Roy Thurik, 2012]. Yet, largely missing from this literature is a strategy to identify displaced workers that lossed jobs through no fault of their own. Our identification depends on evidence of significant long-term wage loss after worker displacement. Starting with Jacobson et al. [1993] labor economists have estimated that displaced workers suffer long-term losses of about 25 percent per year. These losses mount even prior to separation from the firm and are significant even when the employee finds a similar job in another firm. Related work has found these effects are amplified when workers who switch industries after displacement [Neal, 1995], face a tougher job market [Davis and Von Wachter, 2011], or are younger individuals [Couch, 1998, Kletzer and Fairlie, 2003], or are nearing retirement age [Ruhm, 1991, Farber, Hall, and Pencavel, 1993]. Similar to this literature we also focons on both displaced and non-displaced workers over a long-time horizon; however, we focus not on wages, but rather firm creation and measures of firm health. Finally, our work add to the large literature on the reallocation of labor and assets during bankruptcy. For instance, Maksimovic and Phillips [1998], Bernstein et al. [2015] study the reorganization, redeployment, and reallocation of firm assets after bankruptcies. Similarly, we detail the reallocation of human capital due to the bankrupcty process. Particularly close to our work, Graham, Kim, Li, and Qiu [2013b] estimate that the loss of human capital composes 29-49% of bankrupcty costs. The cost may be unestimated given prospective workers are less likely to apply for positions in distressed firms, exacerbating future distress [Brown and Matsa, 2012b]. These costs are so significant that they may impact the firm prior to distress through optimal capital structure decisions [Verwijmeren and Derwall, 2010, Bae, Kang, and Wang, 2011].While worker displacement is typically associated with bad economic outcomes such as the deadweight loss of search costs for a new job and loss of productivity, in this paper we highlight one potential benefit of such displacement: the increase in entrepreneurship by displaced workers. for whom the opportunity cost of starting a firm has declined.

6

3.

Motivation

We motivate our micro-level estimation by illustrating that entreprenuership is correlated with unemployment at the aggregate level. Prior to this paper, the implications of unemployment on firm success is unclear [Delmar and Davidsson, 2000]. While Audretsch and Fritsch [1996], Fritsch [1997] estimate only a minor effect of new firm creation on local employment, more recent findings suggest a substantial and statistically significant impact[Koellinger and Roy Thurik, 2012, Thurik et al., 2008]. According to our preliminary results, establishment births for the smallest firms move countercyclically with employment when compared to all larger firms. Our data is derived from the Business Register, the Census Bureau’s source of information on establishments and is included in the program on Statistics of U.S. Businesses (SUSB). In accordance with the U.S. Census Bureau, we define births as establishments that have zero employment in the first quarter of the initial year and positive employment in the first quarter of the subsequent year. We estimate the regression %∆Firm Birthsi,s,n,t = β1 × Unemps,t + β2 × Smalli,s,n,t, + β3 × Unemps,t × Smalli,s,n,t + Θ × Controlsi,s,n,t + εi,s,n,t

where i denotes the NAICS Industry Classification, s is the state, n is the firm size group, and t is the year. The data source groups firms into five size bins: 1–4 employees, 5–9 employees, 10–19 employees, 20–99 employees, 100–499 employees, 500+ employees. The data includes the number of firm births within each size group, state, and industry. We include all years with NAICS Industry classifications, specifically 1999-2011. Smallest is an identifier for the 1–4 employee group classification and U nemp is the state level unemployment rate. As detailed in Table 1, a 1% increase in the unemployment rate decreases all establishment births by 1.6%; however, this result is highly dependent on firm size. Relative to other firms, small firms establishment is greater when unemployment is greater. The effect cannot be explained by

7

fixed differences between firm groups, or by including year fixed effects. Overall we find that a 1% increase in unemployment actually increases the small firm establishment birth rate by 2% relative to all other firms. To further identify this relationship, we next sort industries by small firm accessibility based on the proportion of small firms in each industry in the initial year of our sample. The smallest firms compose only 8% and 22% of all firms in NAICS Classification 55 (Management of Companies and Enterprises), and 22 (Utilities). On the other side of the spectrum small firms are particularly common in for Code 11 (Agriculture, Forestry, Fishing, and Hunting) and Code 54 (Professional, Scientific, and Technical Services). According to Table 2 we find no significant unemployment effect in industries where small establishments compose under 25% of the population. However, the effect is significant in the other quartiles, reaching 6.4% for small-firm heavy industries: therefore a 1% unemployment results in 6.4% more small firm establishments relative to the rest of the industry. Finally, in unreported regressions we estimate the results separately for each the industry level and find the greatest effects in Wholesale Trade, Information Services, and Retail Trade. While the aggregate results are in line with our predictions, we are not able to fully isolate the direct impact of worker displacement on firm creation. In addition, we have no evidence on the survival or growth of these firms, and have little to say on the overal macroeconomic impacts. Therefore, we next discuss our methodology using micro-level data.

4.

Methodology

Firm creation or survival is known to be based on founder characteristics including management ability [Nanda, 2008], risk tolerance and ambition [Hurst and Pugsley, 2011], or age [Ouimet and Zarutskie, 2014]. To causally identify the effect of job displacement during bankruptcy on the propensity of starting a new firm or self-employment, one needs to address several endogeneity concerns. Would the employee have started a new firm voluntarily regardless of the corporate bankruptcy? Do employees in bankrupt firms have different propensities to start a new firm than employees in other firms? To address these issues, one needs to find a plausible counterfactual employee in a different firm that either did not go through a bankruptcy or if it went through one, the bankruptcy outcome was different. Below we present our empirical design that tackles these

8

issues, and allows us to credibly identify the effect of job displacement during bankruptcy on the propensity of starting a new firm or self-employment.

4.1.

Identification

A common identification strategy to study treatment effects on entrepreneurs is to employ a parametric regression and include a fixed effect for each individual in the sample. This way, the effect of the treatment is the mean change in the outcome variable before and after treatment for an individual. Time fixed effects are also usually added to control for macroeconomic differences affecting all individuals, while other individual characteristics can also be included in the regression. However, this strategy is not reasonable in our framework. As the treatment is worker displacement, by definition the worker has not founded the firm prior to treatment. Therefore, the panel structure is not possible when evaluating the survival and success of the new firm. We cannot compare individuals before and after treatment and instead need an alternative method to develop a counterfactual. By relying on the sheer number of potential individuals in the data as well as the level of detail observed for each individual, we instead incorporate a matching approach on observable characteristics, such as, the detailed occupation description, education degrees attained by the worker, age, and firm-specific and overall years of experience. To introduce our matching process, we first only compare firms that went through a bankruptcy process within the same three-year window. Employees might still anticipate the distress of the bankruptcy as the firm the filing date (Brown and Matsa [2012a]), and as a result, those with high propensity to form new firms and enter self-employment might self-select to work in firms that experience the most severe outcomes after filing for bankruptcy.1 The exclusion restriction would not be guaranteed. To overcome this potential issue we only study the behavior of employees who have worked at the firm for four or more years prior to firm’s first filing for bankruptcy. This implies that on average an employee has worked for the bankruptcy firm for six years. Next, we match workers on their resume. We require an exact match on education (No Bachelors, Bachelors, Graduate Degree), work experience (three year bins), firm tenure (three-year bins) and occupation. Definitions of occupation come from the US Department of Labor Dictionary of 1

Prospective employees are aware of the bankruptcy risk as Brown and Matsa [2012a] find that an increase in an employer’s distress results in fewer and lower quality applicants.

9

Occupation Titles Fourth Edition. The Dictionary classifies all common job titles to a particular occupation. Using these occupational definitions, we sort employees into a particular occupation based on the textual similarity between their self-reported job title and the Dictionary Job Title.

4.2.

Estimating Entry to Entrepreneurship

Using the matching process above we are able to easily determine the cumulative impact of worker displacement on likelihood of entrepreneurship. It is cumulative in the sense that displacement may have an affect on (i) the reservation wage for the individual, (ii) the source and timing of future cash flows, (iii) changing risk aversion, and (iv) unexpected cash sums due to severance. The estimation does not distinguish between these potential channels. As such, we estimate the linear probability model:

Entrepreneuri∈I,T +s = β%∆EmpLossI,T,T +1 + MatchFE + Controlsi,T +s + εi,s

We use the subscript i to refer to each worker in our sample and I to refer to bankrupt firm employing worker i. Subscript T denotes the year of the firm bankruptcy, while subscript s denotes number of years relative to the bankruptcy event. Finally, Entrepreneur is a binary variable that denotes entering into entrepreneurship. Therefore, we are estimating the probability a worker employed at time T + s starts a new firm. The variable %∆EmpLossI,T −1,T +1 is the percent employment decline of firm I during the two year surrounding the bankruptcy filing. We focus on the coefficient β, the effect of likely employment loss on an individual’s likelihood to enter into entrepreneurship. We estimate β over the full sample, and later consider if the estimate differs over subsamples of the data. We exclusively compare workers with their match: (i) firm files for bankruptcy in same year and is in same industry, and (ii) worker has same age, same education level, and same occupation. To make this comparison we include a fixed effect for each matched group, titles M atchF E. We also consider worker location controls in the analysis. We retain the matching sample when considering

10

the survival and success of the newly created firms. However, there is a low probability that both a displaced worker and match both created firms. Therefore, rather than rely on the same empirical strategy, we instead test for firm differences between the two samples.

5.

Data

This section introduces the data from the business networking service that provides us the full employment history of employees, the bankruptcy firm-level data, and other sources of data. We summarize: (i) the complete database, (ii) the subset of the data we include in our estimation, and (iii) differences across the sample four years prior to bankruptcy filing.

5.1.

Data Source

Our primary analysis requires (i) employment histories for a sample of displaced and non-displaced workers and (ii) firm-level information for all companies founded by each individual. The bulk of this data comes from a large scale business networking service that includes self-reported career history of each user. To identify displacement we then use the UCLA-LoPucki Bankruptcy Research Database to isolate firms that suffer bankruptcy. We match our displaced workers to firms in the UCLA-LoPucki dataset, and to Compustat. We also use the headquarter locations and the locations reported in the employees’ resumes, to merge our data to geographical based data, such bank concentration, or house price level. Business Networking Service Data

Our data is constructed from the largest online business

networking service, which covers employment histories for over 365 million users in over 200 countries. The site include 110 million U.S. individuals (seventy percent of the U.S. workforce). Each individual self-reports a resume including education backgrounds, and employment histories complete with firm name and occupation. In addition, the website contains firm profiles, while the firm-level employment history for each firm can be estimated through aggregating employee histories. All data is publicly-available and is obtained through a web searches and then parsed into a panel dataset. We compare our data to three related sources. First, our data resembles the employer-employer linked data from the U.S. Census Longitudinal Employer-Household Dynamics or administrative

11

datasets [Jacobson et al., 1993, Graham et al., 2013b]. These sources suffer from three primary disadvantages: (i) limited information on young and small firms, (ii) missing occupation information that denotes a firm founder, and (iii) missing information on unemployment2 . Secondly, employment histories are frequently studied using both general datasets such as the Panel Study of Income Dynamics or more specialized sources including the Survey of Displaced Workers [Ruhm, 1991, Farber et al., 1993]. However, there is no survey dataset that includes corresponding information on the employer. Third, our data is most similar to online job search websites as discussed in Brown and Matsa [2012b] and Agrawal and Tambe [2014]. Yet, while [Agrawal and Tambe, 2014] covers approximately 13% of the U.S. workforce, we cover approximately seventy percent. With more complete data we can develop reasonable measures of employment growth overtime for all firms in the sample. As age is not explicitly included, we determine age under the following rule: (i) age at high school graduation is 18 plus any years of prior full-time employment, (ii) otherwise, age at starting a college degree is 18 plus any years of prior full-time employment, (iii) otherwise, age at completing an Associates Degree is 20 plus any years of prior full-time employment, (iv) otherwise, age at completing a Bachelor’s Degree is 22 plus any years of prior full-time employment. We split education into: Associates Degree, Three-Year College, Bachelors, and graduate degrees. We distinguish between the following graduate degrees: Masters, Juris Doctor (JD), Masters of Business Administration (MBA), and PhD. We assume missing education signals high school graduate or less. Missing education also implies missing age since we infer age based from education dates. Therefore, instead of age we match years since first work experience. We define firm creation using the following criteria. We include individuals that classify as the owner or founder of a firm. Secondly, we also consider all Chief Executive Officers of firms that were founded when the person joined the firm. Finally, we allow individuals classified as self-employed or entrepreneurs, or job titles that contain the phrase ”independent”. We next impose additional restrictions on these criteria in the results section to determine a misleading report job title is not driving the estimates. In particular, if the firm has employees prior to the entrepreneur joining, 2 Workers that leave the sample may be unemployed, have retired, or moved to alternate states. Therefore, most analysis with US data has instead had to focus solely on wages.

12

we require the entrepreneur to be explicitly classified as an owner or founder. We note that our definition of new firm results in an underestimation of the true level of firm formation, primarily due to our exclusion of franchises. While it is possible to identify a franchise in our data, it is not possible to determine the total employment of a specific establishment. First, self-employment is defined as any firm that has only one employee (the owner) during our sample. Next, nonemployer-to-employer firms are all firms in the sample that start as nonemployer but transition to employer firms at anytime. Finally, employer firms are all firms with employees within the same year as formation. Second, we calculate firm survival as the tenure of the entrepreneur at the firm. This again will underestimate our survival if the founder may transfer firm ownership to another party. However, using the 2007 Survey of Business Owners (SBO) Public Use Microdata Sample (PUMS), we estimate that even five years after formation, ninety percent of all firms are owned by the initial owner. Third, employment is defined as the total number of employees working at the establishment in a given year. Additional Data Sources We match the business networking service data with the UCLA-LoPucki Bankruptcy Research Database by developing a list of firm names most commonly self-reported by workers in the service data. The database includes all public companies with over 100 million in assets (in 1980 dollars) that filed between 1979 and 2015. We use the information on the judges associated with the bankruptcy process for the instrumental variables technique. A firm is considered to emerge if at least one operating company emerged from the bankruptcy, has the intent to continue indefinitvely, does not continue to operate only for the purpose of liquidation. Importantly, a firm emerges if it is acquired by another firm at confirmation, even if the acquiror contributes capital or credit enahancements to the company. Not surprisingly, the firms in our sample are financially distressed. According to Table ?? they are relatively illiquid, have negative profictability and productivity, and are more levered than comparable firms. In addition, we follow the UCLA Research Database and identify assets sales as any firm that has (or intends to) sell a ”substantial” portion of their assets at bankruptcy according to the court document. We then match the Bankruptcy Research Database to CRSP and Compustat using GVKEY for further information on the original firm. In addition, we require firm financial information when 13

developing our nearest neighbor matches. Besides the NAICS Industry classification codes we also estimate: size, liquidity, profitability, productivity, and leverage. We define size as log of assets. Liquidity is measured as the ratio of working capital to total assets. Profitability is defined as retained earnings over total assets, while asset productivity is income before interest and taxes to total assets. In addition, we use two measures of leverage: book leverage is estimated to be current liabilities+long term debt+deferred taxes+investment tax credit all over total assets. Market leverage is the same numerator over the closing price times common shares outstanding. Similar to earlier, we match financial variables two years prior to firm bankruptcy. We compare our results to the 2007 Survey of Business Owners (SBO) Public Use Microdata Sample (PUMS). The SBO PUMS is a cross-sectional dataset on entrepreneurs and surveys a random sample of business from a complete list of all firms operating during 2007 with receipts of at least 1,000 USD. The list of all forms is compiled from business tax returns, specifically: Form 1040 Schedule C (Profit or Loss from Business), Form 1065 (US Return of Partnership Income), 1120 Corporation Tax Forms, Form 941 âĂIJEmployer’s Quarterly Federal Tax ReturnâĂİ, and Form 944 âĂIJEmployer’s Annual Federal Tax Return. For firms with paid employees, the Census Bureau also collected employment, payroll, receipts, and kind of business for each plant, store, of location during the 2007 Economic Census. To control for response bias, three report forms are re-mailed to employer firms and two report forms are re-mailed to nonemployer firms at one month intervals to all delinquent respondents. We use the data between 2002 and 2007 for our results. Finally, we match local data to each worker in the business networking service data using their self-reported current location. House price data is obtained from the Federal Housing Finance Agency. The location and size of bank deposits is available from the Federal Deposit Insurance Corporation from 1994 to 2014. Local population and per capita income data is made available by the Bureau of Economic Analysis. Finally, local unemployment data is obtained from the Bureau of Labor Statistics. We follow Cortes [2013], Adelino et al. [2014] and define a local bank as one with at least seventy-five percent of its deposits in a single MSA. The local bank share for each MSA is then the share of all deposits in the MSA that are held by banks local to that same MSA. Due to limited

14

information on prior locations of the workers, we instead the current location of the workers in 2016 as a proxy for location at the time of the banktruptcy filing.3

5.2.

Summary of Data

Summary of Complete Dataset To document the coverage of the full US Online Business Network Service (OBNS) dataset we match each OBNS industry to the industry definitions from the Bureau of Economic Analysis. We then determine which industries are overstated and understated in the complete OBNS database. Significantly, we find that all Two-Digit NAICS Industry Codes are represented in the database. According to Figure ??, OBNS overstates the relative size of the Information Sector (Code 51), Professional Services (Code 54), Administrative Services (56), Arts and Entertainment (72), Finance (52), and Real Estate (53). In contrast, OBNS understates the relative size of the Public Administration Sector (Code 81), Retail and Wholesale Trade (42,44-45), Accomodation and Food Services (72). Summary of Our Subsample Summary of Workers According to Table 3 we estimate that 1% of employees start a new firm within one year, 2% within three years, and 3% within five years. We also distinguish between nonemployer firms, nonemployer-to-employer, and employer firms. First, nonemployer firms are defined as any firm without any additional employees within three years of formation. Next, nonemployer-to-employer firms are all firms in the sample that start as nonemployer but transition to employer firms within three years of formation. Finally, employer firms are all firms with employees within the same year as formation. We estimate that 3% of workers enter self-employment with five years, compared to 0.01% for nonemployer-to-employer, and 0.02% for employer firms. We next summarize the worker characteristics in our sample. In Table 3 we first note near the majority have completed a Bachelor’s Degree (48 %), with fewer completing Graduate Degrees: 7% have a Masters Degree, 2% have a law degree, 8% have an MBA and 1% have a PhD. Lastly, 7% report an Associates and 9% a Three-year degree. Not surprisingly, the sample has a higher education than the actual population of the US;according to the National Center for Education 3 Alternatively, we could use firm headquarter location as a proxy for worker location at the time of job loss. However, Garcia and Norli [2012] estimate that the average public firm has operations in 7.9 states with a median of 5.5.

15

Statistics as of 2014, 34% of the US population 25-29 has completed a Bachelor’s Degree, while 7.6% have a Master’s or Higher. In addition, our workers are on average 37 years old and have 12 years of work experience. Finally, the MSA per capita income is about $36,000 and the population of those MSAs is slightly over five million residents. The average MSA sees a 32% bank concentration and finds an average of 8% house price growth over the prior two years. Summary of Firms Filing Chapter 11 Bankruptcy

We next focus our attention on our partic-

ular data sample of interest. We define full employment as employment four years prior to Chapter 11 Bankruptcy. According to Table 4 employment drops sixty-three percent within the first two years of bankruptcy filing. Nearly 70% of the sample emerges from the bankruptcy, and half of the sample experience a substantial sale of the firm’s assets. Due to data limitations, we are not able to match up the financial history of the public firm at the time of bankruptcy given a large portion do not emerge and there have no financial history to report. Instead, we match the data to the prior year. Not surprisingly, the firms in our sample experience a decline in liquidity (0.7%), productivity (20%), and asset size (2%); both book and market leverage is actually increasing for these firms. In addition, employment according to Compustat also decreases on average by 5 percent the year before filing for bankruptcy. Given the novely of our employment measure, we estimate the correlation between the financial condition of the firms and our employment measure in Table 4. The firms with the greatest employment loss according to the online business networking service are also less likely to emerge and instead more likely to sale a substantial percentage of assets. These firms are less productive and profitable, more highly levered, and see significant employment loss according to Compustat. Summary of New Firms

As prior studies have suggested government reform that promoting

entreprenuership leads to initially smaller or lower quality Jensen, Leth-Petersen, and Nanda [2014] firms, we also summarize the new firm establishments in our data. We include a total of 2,871 individuals new firms; 90% are self-employment, 2% are nonemployerto-employer firms, and 8% start out as employer firms. Among the firms with employees, we estimate mean employment at 15 workers including founders with 2 at the 10th percentile and a full 41 at

16

the 90th percentile. In addition, 2% of these firms are acquired by larger corporations. The survival rates of both employer and non-employer firms are similar: about 84% of firms still exist after three years, after five and ten years, these number decrease to 72% and 56-58%, respectively. Using the 2007 Survey of Business Owners (SBO) Public Use Microdata Sample (PUMS) Data, we estimate that an average survival rate of: 58 percent within about six months, 41 percent after 1 year, 32 percent after 2 years, 26 within three years, 21 within four years, and 18 percent within five years. Again, using the SBO Data, we estimate that at formation, only four percent of firms have employees. These numbers gradually increase (twelve percent after the first year, fifteen by year two, seventeen by year three, nineteen by four and twenty by year five). Total employment and survival are two common measures of economic impact; however, displacement also has the potential to move human capital from older, less productive firm to more productive industries even if the new firm has only a handful of employees. Therefore, we end this section by documenting the change in industry concentration through displacement. We begin by summarizing the industry concentration of the initial firms. As before, we compare this breakdown to the 2007 Survey of Owners (SBO) Public Use Microdata Sample. Surprisingly, a high 19% of the new firms in our sample are in the manufacutirng sector (NAICS Codes 31-33). This is much higher than the 5.3% according to the SBO, but may partially explain the larger employment in the sample. Finance and Insurance makes up 13% (compared to 4.4% in the SBO), Information composes 12% (compared to 2%), and Retail Trade makes up 7% (compared to 12%).

6.

Results

The results discussion is split into two sections. First, we estimate the impact of displacement on new firm creation. Second, we evaluate the survival, employment and productivity of these new firms.

6.1.

Entry to Entrepreneurship Results

The first step is estimating the impact of displacement on new firm creation. After discussing the baseline regression, we (i) include firm and location controls, (ii) break down the sample for a number of robustness checks, and (iii) determine the value of human capital and financial access.

17

Entry Baseline

In Table 6, we estimate that firm employment around the time of bankruptcy

increases the rate of new firm creation. We first estimate the results without any controls and estimate that displacement around bankruptcy increases new creation by 1% within one year, 3.1% within three years, and 3.7% after five years. According to the constant, firms that experience no employment loss see entrepreneurship rates of 0.7% within three years as a result of workers starting firms while still employed. Next, we include our match fixed effects to confirm the result is not driven by differences across workers. The results decline slightly- we estimate new firm creation at 2.9% after three years- but are still highly statistically significant. There are limitations of focusing on long-horizon effects if the employee match is imperfect as we may overstate the impact of displacement on entrepreneurship. The value, however, is that we are able separate between two explanations of the short-term results. One possibility is that displacement pushes individuals to early entrepreneurship; alternatively, those workers may have never started a new firm without first losing a job. If the impact of displacement only accelerates the timing of firm creation (as opposed to the overall likelihood) then we find no difference between displacement effects four and five years out in the future. Instead, the year five coefficient is still increasing from year four; in other words, workers from the matched firms are not catching up the firm creation rates of the displaced workers. Displacement has an effect on the total number of new firms, not just the timing of creation. We compare the estimates to the prior literature. For instance Fairlie et al. [2010] finds a five percentage point increase in unemployment leads to a 0.04 percent increase in the entreprenuership rate; in other words, assuming a baseline 95% employment rate, his results imply that losing a job increases entrepreneurship by 0.76%. The difference in estimates is likely due to contrasting empirical framework. Notice that Fairlie et al. [2010] only has information on regional unemployment. In comparison we are able to document worker displacement at the individual level. It should not be surprising we find substantially larger estimates as a result. Prior to Bankruptcy Next in Table 7 we estimate the probability of new firm creation during each year around the Bankruptcy Filing. As before, we estimate that displacement leads to immediate new firm creation: 1.3% at the year of filing and 1.4% the following year. However, we find no difference in firm creation four years prior to the bankruptcy. This supports our identification

18

strategy: there is no evidence that entrepreneurial workers sort into the more financially-distressed firm prior to bankrupty. Distinguishing Between Employer and Nonemployer Firms

Similar to the data summary, we

split firms by type: permanent nonemployer, nonemployer-to-employer, and initially employer in Table 8. We find that displacement increases self-employment by 2.3%, representing roughly 80% of all new firm creation. The likelihood of nonemployer-to-employer firms is unaffected by displacement, while the rate of employer firms increases 0.3%. Firm Creation Conditional on Exit The results illustrate that displacement increases the rate of new firm creation; we next test if this is due to only an increase in firm exits or alternatively if firm creation actually increases during bankruptcy conditional on worker exit from the firm. First, in Table 9 we determine that worker exit is especially large during the years of bankruptcy: 38% of workers leave in the year of filing with 27% of the remaining leaving the following year. We document a clear run-up prior to the bankruptcy with only 11% of workers leaving four year prior to filing. Second, we include only workers that exit the firm in the same year and then test the likelihood of new firm creation conditional exit from the firm. We estimate a slight increase in the selfemployment rate, reaching 3% the year following the bankruptcy filing. The result is a 14% increase from four years prior to banktupcy. In comparison, we find a significant increase in the rate of starting an employer firm. Four years prior to bankruptcy, we estimate that 0.22% of workers start a firm conditional on exit; by the year following the bankruptcy, the estimate increases to 0.36%, representing a a 63% increase in the rate of employer firms when compared to four years prior to the filing. The results suggests not only an increase in new firm creation, but an increase in the rate of new firm creation upon displacement. Comparing the results to Babina [2015], she finds that during a three-year period of financial distress, 64% of workers leave the firm, with 3.8% joining a new firm (or 6% out of the workers that leave). While our numbers are instead significantly smaller, they are likely driven by differences in the data sample. First, Babina [2015] primarily focuses on joining a new firm, while we focus on founding a new firm. Secondly, we include reported new firm creation, while Babina [2015] includes employer firms and excludes Schedule C self-employed activity. 19

6.2.

High and Low Skilled Occupations

Thus far, we have documented the role of financial distress in transfering labor from unproductive firms to the entrepreneurial sector; we now focus more specifically on the transfer of human capital. Human capital may minimize the cost of displacement assuming these workers have easilytransferable skills of interest to alternative employers. However, the risks of entrepreneurship are particularly large for workers with human capital and a high wage, suggesting displacement might have a particular impact on these individuals. To identify human capital, we first consider the role of occupation, distinguishing between high and low skilled roles. We define High Skilled Labor as: Management, Business and Financial Operations, Computer and Mathematical, and Architecture and Engineering. With our definition of occupation, we then determine how skilled labor interacts with worker characteristics, specifically, education and age. Education First, we evaluate the role of education in self-employment and new firm creation in Table 10. Education is associated with higher income [Grogger and Eide, 1995] as well as greater access to family wealth [Belley and Lochner, 2007]. Starting with self-employment, we break workers into four groups: low-skilled and no college, low-skilled and college, high-skilled and no college, high-skilled and college. Contrary to our expectation, we find no evidence that workers with low-skilled backgrounds and without a college education enter into self-employment after displacement. In comparison, low-skilled occupations with college education enter at a rate of 4.3%. Turning to high-skilled occupations, we actually find a decline in the rate of self-employment: 1.9% for no-college workers and 3.1% for college workers. Next, moving to the employer firms, the results are striking: displaced workers found a new firm only when they have both a college education and a background in a high-skilled occupation. In all other categories, we find no evidence of starting an employer firm. Age and Experience Next, we conduct a similar analysis using worker age. Prior evidence has found the start-ups disproportionately employ younger workers due to both their skillsset and risk tolerance[Ouimet and Zarutskie, 2014] .

20

We split age by three groups: under 30, 30-40, and 40 or above. We find that self-employment is more common among the older workers, with little self-employment among young workers. Yet, we find the opposite result for employer firms: employer firms are only started by younger workers in high-skilled occupations. The results is strongest for workers under 30 years of age. One concern with this analysis is that we do always have an accurate measure of age, requiring us to cut the sample. Therefore, we instead use work experience as a proxy for age: under 5 years, 5-10 year, and over 10 years. Again, we find similar results as self-employment is increasing with experience and is greatest for low-skilled workers. However, we only find evidence of new firm creation among high-skilled workers with under five years work experience.

6.3.

Robustness

In this section we highlight the robustness of our results. This is necessary given the novely of both our primary data source and our measure of employment loss. Emergence from Bankruptcy

We extend the analysis to control for differences across the dis-

tressed public firms in our sample. In Table 14 we first consider our primary regression, but control for differences in profitlability and productivity, leverage, size, and employment. We find the results are robust. Next, our key variable of interest is the percent employment loss (from the online networking data) during the two years following the bankruptcy filing. Given the uniqueness of our measure, it is useful to confirm that the results hold considering alternative measures of financial distress and the likelihood of displacement. In Table 14 we instead consider a binary variable that take a positive value when the firm is liquidated and does not emerge from the bankruptcy. As mentioned earlier, a firm emerges bankrupcy when it is acquired by another firm at confirmation, even if the acquiror contributes capital or credit enahancements to the company. Secondly, given evidence that liquidated firms have smaller number of plants, employ fewer workers, and are slightly younger, we include our firm controls Bernstein et al. [2015].

21

Location and Financing Thus far, we are comparing workers across US areas, even though rates of entrepreneurship are highly dependent on the location [Glaeser, 2007]. We next control for the locational differences in Table 14. First, numerous studies have documented the role of real estate as a source of collateral for entrepreneurs including Adelino, Schoar, and Severino [2013], Schmalz, Sraer, and Thesmar [2013a], Kleiner [2014], Corradin and Popov. These studies show that real estate growth impacts both the rate of entrepreneurship and the subsequent growth of young firms. Therefore, displaced workers in areas with high recent house price growth may have additional access to financing through home equity lines of credit. In the second test, we split the sample between locations with house price growth below and above five percent during the two years prior. Given there are significant differences across locations, we also include MSA fixed effects and population growth and income growth controls Second, we consider the role of access to financial services. Evidence from Petersen and Rajan [1994, 2002] suggests that local banks are more likely lend to small and young firms. Intuitively, more established firms require less screening and monitoring, and so can access financial service from non-local institutions. In this third test, we split the sample by access to local banks. Similar to before, we also include controls on income and population within each MSA. According to Table 14 we first determine that controlling for location differences- by including population and income measures and/or including MSA fixed effects-does not impact the estimated results. Next, we find little evidence that bank concentration results in more new firm creation. While house price growth is positively correlated with entrepreneurship, we find robust evidence that displacement results in new firm creation. Subsample Checks Finally, we determine if our results depend on the particular time period, industry, or geography and present the results in Table 13. Time Period

First, the results are likely to depend on the time period. First, the literature has

established long-term trends of entrepreneurship. Secondly, new firm creation, particuarly upon displacement, may depend highly on future employment prospects and so differ over the business cycle.

22

We first break the sample before after 2000. This is due to recent evidence that the rate of business start-ups in the US economy has fallen by half in recent decades, accelerating after 2000 Decker, Haltiwanger, Jarmin, and Miranda [2014]. This decline compliments a similar decline in the pace of job reallocation, where young workers now involved fewer employment switches before choosing a career path 4 . We present the results in Table ?? and find no statistical difference in the sample before and after 2000. Next, new firm creation is highly dependent on the business cycle as illustrated in Tables 1 and Table 2. Separating our results by recession and non-recession periods may help distinguish the exact mechanism at play. On the one hand, it may be more difficult to find full-time employment during a recession, increasing the number of entrepreneurs in the sample; a similar case could be made for periods of industry-specific distress. Alternatively, recessions periods may also denote a drop in profitability, decreasing the returns to starting a new firm5 . To test these arguments we define recessions as the years: 1981-1982, 1990-1991, 2000-2001, and 2007-2011. Note that we extend beyond the recent recessions beyond the NBER Recession Date of 2009 due to the high unemployment and slow growth during this period; however, the results hold even after limiting to 2009. We find evidence that new firm creation is more prevalent during recession periods, but still large and significant over the rest of the business cycle. The results suggest that obstacles to new employment increases new firm creation after initial displacement. Industry

We also illustrate that the results are not unique to our particular database; this is a

concern given that social network service is concentrated both by industry and geography. First, as Discussed in Figure 1 our online database overstates the relative size of a number of sectors (particularly Information, Professional and Technical Services, Finance, and Real Estate), while understaing the size of the Public Administration and Retail/Wholesale Trade Industries. As we are limited to the self-reported resumes in the sample, this may cause us to potentially oversample certain industries over others. Therefore, we separate there four overrepresented sectors from other industries and find no significant difference in the estimation/ 4 As a separate issue, given the data is self-reported from resumes, it is possible that current employment is more accurate and less noisy. Breaking down the sample helps overcome this concern. 5 Of course, non-entrepreneurial workers may also experience slow or even declining wage growth during recessions.

23

In Table 13 we consider all industries with at least 2500 hundred observations: Manufacturing (Codes 31-33), Trade (Codes 44-45), Information (Code 51), and Professional, Technical, and Scientific Services (Code 54). Again, the results do not depend on a particular industry. Geography

Similarly, our public firm sample includes technology firms that went bankrupt

following Tech Bubble. To confirm our results are not driven by this subset of the population, we exclude Californian firms from the sample. Again, the results hold outside California. Controlling for Industry Definition The current identification strategy exclusively compares the employees of bankrupt firms; the identification assumption is that three years prior to filing, it is difficult for employees to determine which bankrupt firm with experience the greatest employment loss. The primary downside with this framework is that we significantly limit the number of firms in the sample, and so limiting our ability to closely match firms, specifically on industry. Therefore, we run a similar estimation by matching each employees in bankrupt firms to employees in non-bankrupt firms using a nearest-neighbor algorithm as discussed in Abadie and Imbens [2006, 2011] and incorporated in Malmendier and Tate [2009], Almeida, Campello, Laranjeira, and Weisbenner [2009]. The method is capable of matching on both multiple criteria, both discrete and continuous variables, and can require exact matches for certain requirements. We conduct the analysis with replacement in the control group. Our matching process is two-step. In the first step we determine the three firms nearest to each bankrupt firm; in the second step we match each displaced worker to her closest equivalent at one of the other three firms. We match on firm distress given the possibility that individuals choose their employer based on the potential of future job displacement. We match on worker resume to develop the best possible control for each displacement. To match firms each neighbor is required to be in the same industry according to the fivedigit NAICS code. In addition, we require that one year prior to delisting each neighbor has a similar probability of default. We estimate the probability of default using variables similar to Altman [1968].6 Specifically, we match firms based on: (i) size, (ii) liquidity, (iii) profitability, (iv) productivity, (v) market leverage, and (vi) book leverage.The worker match is similar to our baseline strategy using age/work experience, education, and occupation. The results are similar to our 6

We do not incorporate the Altman-Z score directly as it is estimated only for manufacturing firms.

24

earlier estimates, suggesting our number are not driven by incorrectly matching firms in different industries.

7.

Conclusion

This paper uses new firm creation as a measure of productive labor reallocation due to corporate bankruptcy. Using a large panel dataset from a business networking service, we determine that displaced individuals are 2% more likely to engage in self employment and 0.03% more likely to start an employer firm. The results are robust to alternative explanations; however, they highly depend on the skillset of the worker. We find that starting an employer firm is more common among young and college-educated workers, but only those from high-skilled occupations. More broadly, this paper highlights the current data limitations researchers face when studying both self-employment and new firm creation. Current datasets lack detailed information on both worker characteristics (especially occupation) as well as small firm attributes. Therefore, we struggle to fully evaluate human capital at an individual level, especially for workers in the young firms with highest innovation and net employment growth. Yet this data is neccessary to understand the causes and consequences of human capital transfer across occupations, firms, industries, etc. By developing and extending both unique dataset and methodology we plan to continue this line of research and better understand how the transfer of human capital from older, unproductive firms to new firms impacts the greater economy.

25

References A. Abadie and G. W. Imbens. Large sample properties of matching estimators for average treatment effects. Econometrica, 74(1):235–267, 2006. A. Abadie and G. W. Imbens. Bias-corrected matching estimators for average treatment effects. Journal of Business & Economic Statistics, 29(1), 2011. M. Adelino, A. Schoar, and F. Severino. House prices, collateral and self-employment. Technical report, National Bureau of Economic Research, 2013. M. Adelino, S. Ma, and D. T. Robinson. Firm age, investment opportunities, and job creation. Technical report, National Bureau of Economic Research, 2014. P. Aghion, J. Moore, and O. Hart. The economics of bankruptcy reform. The Journal of Law, Economics, & Organization, 8:N3, 1992. A. K. Agrawal and P. Tambe. Private equity, technological investment, and labor outcomes. Technological Investment, and Labor Outcomes (July 17, 2014), 2014. M. Aguiar, E. Hurst, and L. Karabarbounis. Time use during the great recession. The American Economic Review, 103(5):1664–1696, 2013. A. Alba-Ramirez. Self-employment in the midst of unemployment: the case of spain and the united states. Applied Economics, 26(3):189–204, 1994. H. Almeida, M. Campello, B. Laranjeira, and S. Weisbenner. Corporate debt maturity and the real effects of the 2007 credit crisis. Technical report, National Bureau of Economic Research, 2009. E. I. Altman. Financial ratios, discriminant analysis and the prediction of corporate bankruptcy. The journal of finance, 23(4):589–609, 1968. D. B. Audretsch and M. Fritsch. Creative destruction: turbulence and economic growth in germany. Behavioral norms, technological progress, and economic dynamics: Studies in Schumpeterian economics, pages 137–150, 1996.

26

T. Babina. Destructive creation at work: How financial distress spurs entrepreneurship. Available at SSRN, 2015. K.-H. Bae, J.-K. Kang, and J. Wang. Employee treatment and firm leverage: A test of the stakeholder theory of capital structure. Journal of Financial Economics, 100(1):130–153, 2011. P. Belley and L. Lochner. The changing role of family income and ability in determining educational achievement. Technical report, National Bureau of Economic Research, 2007. S. Bernstein, E. Colonnelli, and B. Iverson. Asset reallocation in bankruptcy. Available at SSRN, 2015. D. G. Birch. The job generation process. 1979. D. G. Birch. Job creation in america: How our smallest companies put the most people to work. University of Illinois at Urbana-Champaign’s Academy for Entrepreneurial Leadership Historical Research Reference in Entrepreneurship, 1987. J. Brown and D. A. Matsa. Boarding a sinking ship? an investigation of job applications to distressed firms. Technical report, National Bureau of Economic Research, 2012a. J. Brown and D. A. Matsa. Boarding a sinking ship? an investigation of job applications to distressed firms. Technical report, National Bureau of Economic Research, 2012b. N. Cetorelli and P. E. Strahan. Finance as a barrier to entry: Bank competition and industry structure in local us markets. The Journal of Finance, 61(1):437–461, 2006. S. Corradin and A. Popov. Financing constraints, housing, and new business creation. K. R. Cortes. New firms, job creation and access to local finance. Federal Reserve Bank of Cleveland Working Paper, 2013. K. A. Couch. Late life job displacement. The Gerontologist, 38(1):7–17, 1998. S. J. Davis and J. Haltiwanger. Gross job creation, gross job destruction, and employment reallocation. The Quarterly Journal of Economics, pages 819–863, 1992. S. J. Davis and T. Von Wachter. Recessions and the costs of job loss. Brookings Papers on Economic Activity, 43(2 (Fall)):1–72, 2011. 27

R. Decker, J. Haltiwanger, R. Jarmin, and J. Miranda. The role of entrepreneurship in us job creation and economic dynamism. The Journal of Economic Perspectives, pages 3–24, 2014. F. Delmar and P. Davidsson. Where do they come from? prevalence and characteristics of nascent entrepreneurs. Entrepreneurship & regional development, 12(1):1–23, 2000. D. S. Evans and L. S. Leighton. Some empirical aspects of entrepreneurship. The American Economic Review, pages 519–535, 1989. D. S. Evans and L. S. Leighton. Small business formation by unemployed and employed workers. Small business economics, 2(4):319–330, 1990. R. W. Fairlie et al. The great recession and entrepreneurship. University of California at Santa Cruz working paper, 2010. H. S. Farber. Alternative and part-time employment arrangements as a response to job loss. Technical report, National Bureau of Economic Research, 1999. H. S. Farber, R. Hall, and J. Pencavel. The incidence and costs of job loss: 1982-91. Brookings papers on economic activity. Microeconomics, pages 73–132, 1993. M. Fritsch. New firms and regional employment change. Small business economics, 9(5):437–448, 1997. D. Garcia and Ø. Norli. Geographic dispersion and stock returns. Journal of Financial Economics, 106 (3):547–565, 2012. E. L. Glaeser. Entrepreneurship and the city. Technical report, National Bureau of Economic Research, 2007. J. R. Graham, H. Kim, S. Li, and J. Qiu. Human capital loss in corporate bankruptcy. Available at SSRN 2276753, 2013a. J. R. Graham, H. Kim, S. Li, and J. Qiu. Human capital loss in corporate bankruptcy. Available at SSRN 2276753, 2013b. J. Grogger and E. Eide. Changes in college skills and the rise in the college wage premium. Journal of Human Resources, pages 280–310, 1995. 28

J. Haltiwanger, R. S. Jarmin, and J. Miranda. Who creates jobs? small versus large versus young. Review of Economics and Statistics, 95(2):347–361, 2013. B. H. Hamilton. Does entrepreneurship pay? an empirical analysis of the returns to self-employment. Journal of Political economy, 108(3):604–631, 2000. J. Hombert, A. Schoar, D. Sraer, and D. Thesmar. Can unemployment insurance change the selection into entrepreneurship? In Measuring Entrepreneurial Businesses: Current Knowledge and Challenges. University of Chicago Press, 2014. C.-T. Hsieh and P. J. Klenow. Misallocation and manufacturing tfp in china and india. The Quarterly Journal of Economics, 124(4):1403–1448, 2009. E. Hurst and A. Lusardi. Liquidity constraints, household wealth, and entrepreneurship. Journal of political Economy, 112(2):319–347, 2004. E. Hurst and B. W. Pugsley. What do small businesses do? Brookings Papers on Economic Activity, 43 (2 (Fall)):73–142, 2011. L. S. Jacobson, R. J. LaLonde, and D. G. Sullivan. Earnings losses of displaced workers. The American Economic Review, pages 685–709, 1993. T. L. Jensen, S. Leth-Petersen, and R. Nanda.

Housing collateral, credit constraints and

entrepreneurship-evidence from a mortgage reform. Technical report, National Bureau of Economic Research, 2014. W. R. Kerr and R. Nanda. Democratizing entry: Banking deregulations, financing constraints, and entrepreneurship. Journal of Financial Economics, 94(1):124–149, 2009. K. Kleiner. How real estate drives the economy: An investigation of small firm collateral shocks on employment. 2014. L. G. Kletzer and R. W. Fairlie. The long-term costs of job displacement for young adult workers. Industrial & Labor Relations Review, 56(4):682–698, 2003. P. D. Koellinger and A. Roy Thurik. Entrepreneurship and the business cycle. Review of Economics and Statistics, 94(4):1143–1156, 2012. 29

V. Maksimovic and G. Phillips. Asset efficiency and reallocation decisions of bankrupt firms. Journal of Finance, pages 1495–1532, 1998. V. Maksimovic and G. Phillips. The market for corporate assets: Who engages in mergers and asset sales and are there efficiency gains? Journal of finance, pages 2019–2065, 2001. U. Malmendier and G. Tate. Superstar ceos. The Quarterly Journal of Economics, 124(4):1593–1638, 2009. R. Nanda. Cost of external finance and selection into entrepreneurship. Harvard Business School Entrepreneurial Management Working Paper, (08-047), 2008. D. Neal. Industry-specific human capital: Evidence from displaced workers. Journal of labor Economics, pages 653–677, 1995. D. Neumark, B. Wall, and J. Zhang. Do small businesses create more jobs? new evidence for the united states from the national establishment time series. The Review of Economics and Statistics, 93 (1):16–29, 2011. P. Ouimet and R. Zarutskie. Who works for startups? the relation between firm age, employee age, and growth. Journal of financial Economics, 112(3):386–407, 2014. S. C. Parker. The economics of entrepreneurship. Cambridge University Press, 2009. M. A. Petersen and R. G. Rajan. The benefits of lending relationships: Evidence from small business data. The journal of finance, 49(1):3–37, 1994. M. A. Petersen and R. G. Rajan. Does distance still matter? the information revolution in small business lending. The Journal of Finance, 57(6):2533–2570, 2002. C. v. Praag and H. V. Ophem. Determinants of willingness and opportunity to start as an entrepreneur. Kyklos, 48(4):513–540, 1995. C. J. Ruhm. Are workers permanently scarred by job displacements? The American Economic Review, pages 319–324, 1991. A. Sahin, J. Song, G. Topa, and G. L. Violante. Mismatch unemployment. 2012.

30

M. C. Schmalz, D. A. Sraer, and D. Thesmar. Housing collateral and entrepreneurship. 2013a. M. C. Schmalz, D. A. Sraer, and D. Thesmar. Housing collateral and entrepreneurship. 2013b. A. Schoar. Effects of corporate diversification on productivity. The Journal of Finance, 57(6):2379–2403, 2002. A. Schoar. The divide between subsistence and transformational entrepreneurship. NBER Innovation Policy and the Economy, 2009. E. L. Scott, P. Shu, R. M. Lubynsky, et al. Are ŞbetterŤ ideas more likely to succeed? an empirical analysis of startup evaluation. Harvard Business School Technology & Operations Mgt. Unit Working Paper, 45(16-013), 2015. A. Shleifer and R. W. Vishny. Liquidation values and debt capacity: A market equilibrium approach. The Journal of Finance, 47(4):1343–1366, 1992. A. R. Thurik, M. A. Carree, A. Van Stel, and D. B. Audretsch. Does self-employment reduce unemployment? Journal of Business Venturing, 23(6):673–686, 2008. R. Topel. Specific capital, mobility, and wages: Wages rise with job seniority. Journal of Political Economy, pages 145–176, 1991. P. Verwijmeren and J. Derwall. Employee well-being, firm leverage, and bankruptcy risk. Journal of Banking & Finance, 34(5):956–964, 2010. O. E. Williamson. Corporate finance and corporate governance. The journal of finance, 43(3):567–591, 1988.

31

Table 1: Regression of Unemployment on Small Firm Births from the Census Statistics of US Business for 1999-2011. Columns (i) considers all firms in the analysis. The (ii)-(v) compares the smallest firms (1-5 employees) to all other firm sizes. Column (iii) include fixed effects for each Industry-State-Size Group, (iv) also includes Year fixed effects, and (v) instead use Industry-Year fixed effects. All specifications cluster observations at the State level. T-Statistics are included below the coefficient. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the State Level.

New Establishments

Unemployment

(i)

(ii)

(iii)

(iv)

(v)

-1.590∗∗∗

-1.922∗∗∗

-2.152∗∗∗

0.472

0.557

(-6.15)

(-6.15)

(-6.23)

(0.80)

(0.97)

1.813∗∗∗

2.062∗∗∗

2.074∗∗∗

1.939∗∗∗

(5.42)

(5.16)

(5.20)

(4.91)

Small × Unemp

Year FE

No

No

No

Yes

Yes

Size-Industry-State FE

No

No

No

No

Yes

N

66496

66496

66496

66496

66496

R-squared

0.001

0.004

0.097

0.137

0.169

Table 2: Regression of Unemployment on Small Firm Births based on NAICS Classification Small Firm Establishment Proportion. Proportion is the percentage of small firm establishment in each industry in the initial year of our sample. All specifications include fixed effects for each IndustryState-Size Group and also includes Year fixed effects. We cluster observations at the State level. T-Statistics are included below the coefficient. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the State Level. New Establishments by Industry Concentration

Small × Unemp

Under 25

25-50

50-75

Over 75

-0.065

2.609∗∗∗

1.048∗∗

5.515∗∗∗

(-0.03)

(6.22)

(2.05)

(3.01)

Year Fixed Effects

Yes

Yes

Yes

Yes

N

5815

42079

16587

2015

R-squared

0.146

0.142

0.149

0.239

32

Table 3: Data Summary of Workers

N

Mean

Std

10th

90th

Firm Creation within Same Year

211678

0.0044

0.067

0

0

Firm Creation within Two Years

211678

0.016

0.13

0

0

Firm Creation within Five Years

211678

0.035

0.18

0

0

Self Employment within Same Year

211678

0.0037

0.061

0

0

Self Employment within Two Years

211678

0.014

0.12

0

0

Self Employment within Five Years

211678

0.029

0.17

0

0

Transition Firm within Same Year

211678

0.000066

0.0081

0

0

Transition Firm within Two Years

211678

0.00022

0.015

0

0

Transition Firm within Five Years

211678

0.00052

0.023

0

0

Employer Firm within Same Year

211678

0.00032

0.018

0

0

Employer Firm within Two Years

211678

0.0011

0.033

0

0

Employer Firm within Five Years

211678

0.0022

0.047

0

0

N

Mean

Std

10th

90th

Worker Age

124320

36.6

8.87

26

49

Work Experience

211678

12.1

8.11

3

24

Associate’s Degree

211678

0.069

0.25

0

0

3-Year College Degree

211678

0.090

0.29

0

0

Bachelor’s Degree

211678

0.52

0.50

0

1

Master’s Degree

211678

0.085

0.28

0

0

Law Degree

211678

0.013

0.11

0

0

MBA Degree

211678

0.090

0.29

0

0

PhD Degree

211678

0.011

0.10

0

0

MSA Per Capita Income

134102

36557.5

9620.7

24312

49906

MSA Population

134102

5148665.1

5384685.5

511391

12844070

Local Bank Concentration

122483

0.31

0.23

0.058

0.62

Two Year House Price Growth

68217

0.079

0.11

-0.038

0.19

Year

211678

6.31

1995

2010

33 2002.6

Table 4: Data Summary of Bankrupt Firms. The Top Table summarizes the variables while the Low Table estimates the correlation between our employment loss measure and other more standard measures of financial distress.

N

Mean

Std

10th

90th

%∆Emp

211678

0.58

0.27

0.24

0.92

Emerge from Bankruptcy

211072

0.70

0.46

0

1

Firm Sold in Bankruptcy

204745

0.51

0.50

0

1

%∆Profitability

202633

1.04

3.97

-1.16

4.68

%∆Productivity

202299

-0.14

4.19

-3.61

1.88

%∆Market Leverage

159241

6.88

10.9

-0.035

23.2

%∆Book Leverage

179455

0.19

0.43

-0.075

0.50

%∆Log(Assets)

205382

-0.015

0.037

-0.062

0.014

%∆Employees

195882

-0.045

0.60

-0.16

0.10

%∆Employees/Assets

196077

0.076

0.27

-0.20

0.49

N

Correlation

Emerge from Bankruptcy

211072

-0.65

Firm Sold in Bankruptcy

204745

0.085

%∆Profitability

202633

-0.092

%∆Productivity

202299

-0.16

%∆Market Leverage

159241

0.096

%∆Book Leverage

179455

0.065

%∆Log(Assets)

205382

-0.011

%∆Employees

195882

-0.075

%∆Employees/Assets

196077

-0.085

34

Table 5: Data Summary of Newly Created Firms. The Top Table summarizes all new firms, the Second Table summarizes self-employment, and the Third Table summarizes employer firms.

N

Mean

Std

10th

90th

Self Employment within Five Years

3211

0.91

0.29

1

1

Self Emp to Employer Firm within Five Years

3211

0.019

0.14

0

0

Employer Firm within Five Years

3211

0.074

0.26

0

0

Firm Creation in Prior Industry

567

0.24

0.43

0

1

Firm Creation in Future Industry

339

0.35

0.48

0

1

Manufacturing

3211

0.18

0.39

0

1

Finance and Insurance

3211

0.13

0.34

0

1

Information

3211

0.12

0.32

0

1

Retail Trade

3211

0.075

0.26

0

0

Professional and Technical Service

3211

0.046

0.21

0

0

N

Mean

Std

10th

90th

Survival after one Year

2878

0.98

0.13

1

1

Survival after Three Years

2695

0.82

0.39

0

1

Survival after Five Years

2414

0.69

0.46

0

1

Survival after Ten Years

1340

0.54

0.50

0

1

N

Mean

Std

10th

90th

3+ Employees

292

0.72

0.45

0

1

5+ Employees

292

0.52

0.50

0

1

10+ Employees

292

0.33

0.47

0

1

Firm Acquired

292

0.017

0.13

0

0

Survival after one Year

289

1.00

0.059

1

1

Survival after Three Years

275

0.86

0.35

0

1

Survival after Five Years

257

0.72

0.45

0

1

Survival after Ten Years

159

0.57

0.50

0

1

35

Table 6: Regression of Employment Decline around Bankrupty Filing. The First Row estimates the impact of displacement on new firm creation within the same calendar year. The Second Row includes a fixed effect for each occupation. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the Firm Level.

Baseline 0 Years

1 Year

2 Years

3 Years

4 Years

5 Years

0.010∗∗∗

0.021∗∗∗

0.028∗∗∗

0.031∗∗∗

0.035∗∗∗

0.037∗∗∗

(8.21)

(11.20)

(10.90)

(9.80)

(9.03)

(8.32)

0.000

0.002∗∗

0.004∗∗∗

0.007∗∗∗

0.010∗∗∗

0.013∗∗∗

(0.76)

(2.03)

(2.90)

(3.98)

(4.38)

(4.97)

N

73532

73532

73532

73532

73532

73532

R-squared

0.001

0.002

0.003

0.003

0.003

0.003

%∆Emp

Constant

Match Controls 0 Years

1 Year

2 Years

3 Years

4 Years

5 Years

0.012∗∗∗

0.022∗∗∗

0.029∗∗∗

0.031∗∗∗

0.033∗∗∗

0.036∗∗∗

(6.82)

(7.75)

(7.85)

(6.93)

(6.78)

(6.91)

N

73532

73532

73532

73532

73532

73532

R-squared

0.275

0.330

0.351

0.358

0.364

0.366

%∆Emp

36

Table 7: Regression of Employment Decline on Entry to Entrepreneurship by Year of Bankrupty Filing. We define T as the Year the Bankruptcy in Filed. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the Firm Level.

Immediate Firm Creation

%∆Emp

Constant

T-4

T-3

T-2

T-1

T

T+1

T+2

0.000

0.005∗∗∗

0.004∗∗

0.007∗∗∗

0.013∗∗∗

0.014∗∗∗

0.007

(0.15)

(2.77)

(2.18)

(3.17)

(3.84)

(2.79)

(1.21)

0.003∗∗∗

0.000

0.002

0.001

-0.000

-0.001

0.001

(3.82)

(0.03)

(1.56)

(0.70)

(-0.08)

(-0.30)

(0.35)

N

46760

42087

36622

31051

25417

17064

12677

R-squared

0.407

0.412

0.402

0.386

0.463

0.573

0.577

Firm Creation within One Year T-4

T-3

T-2

T-1

T

T+1

T+2

0.006∗∗

0.008∗∗∗

0.010∗∗∗

0.017∗∗∗

0.027∗∗∗

0.016∗∗

0.015∗

(2.42)

(2.78)

(3.57)

(4.62)

(5.73)

(2.58)

(1.89)

0.003∗∗

0.003∗

0.004∗∗

0.003

0.000

0.003

0.002

(2.56)

(1.83)

(2.08)

(1.38)

(0.13)

(1.13)

(0.51)

N

46760

42087

36622

31051

25417

17064

12677

R-squared

0.396

0.390

0.386

0.396

0.440

0.526

0.630

%∆Emp

Constant

37

Table 8: Regression of Employment Decline on Entry to Entrepreneurship by New Firm Type. The Table considers nonemployer firms, Nonemployer-to-Employer, and Initial Employer Firms. We focus on the Period T-1 to T+1. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the Firm Level.

Self-Employment

Self-Emp to Employer

Employer

Base

Match

Base

Match

Base

Match

b/t

b/t

b/t

b/t

b/t

b/t

0.0226∗∗∗

0.0242∗∗∗

0.0001

-0.0001

0.0027∗∗∗

0.0022∗∗∗

(9.8132)

(7.1771)

(0.3317)

(-0.2681)

(3.1138)

(2.9509)

0.0042∗∗∗

0.0033∗

0.0002

0.0002

-0.0001

0.0002

(3.1482)

(1.7901)

(1.5296)

(1.3448)

(-0.3180)

(0.4861)

N

73532

73532

73532

73532

73532

73532

R-squared

0.002

0.350

0.000

0.354

0.000

0.382

%∆Emp

Constant

38

Table 9: Regression on Firm Exit and Entry to Entrepreneurship. The First Table measures the likelihood of exit at each period surrounding the bankruptcy filing. The Second Table measures Self-Employment conditional on exit, the Third Table measures Employer Firm Creation conditional on exit. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the Firm Level.

Likelihood of Exiting Prior Firm

Constant

T-4

T-3

T-2

T-1

T

T+1

T+2

0.100∗∗∗

0.130∗∗∗

0.152∗∗∗

0.181∗∗∗

0.329∗∗∗

0.257∗∗∗

0.161∗∗∗

(31.94)

(33.23)

(25.14)

(34.15)

(18.31)

(15.10)

(19.70)

N

46760

42087

36622

31051

25417

17064

12677

R-squared

0.333

0.334

0.340

0.362

0.443

0.517

0.571

Self-Employment within Two Years Conditional on Exiting Prior Firm

Constant

T-4

T-3

T-2

T-1

T

T+1

T+2

0.028∗∗∗

0.030∗∗∗

0.034∗∗∗

0.033∗∗∗

0.030∗∗∗

0.031∗∗∗

0.024∗∗∗

(13.41)

(13.96)

(17.78)

(15.48)

(18.80)

(12.39)

(7.93)

N

4685

5468

5569

5635

8352

4386

2043

R-squared

0.748

0.692

0.684

0.657

0.588

0.708

0.800

Employer Firm within Two Year Conditional on Exiting Prior Firm T-4

T-3

T-2

T-1

T

T+1

T+2

b/t

b/t

b/t

b/t

b/t

b/t

b/t

0.0017∗∗∗

0.0031∗∗∗

0.0031∗∗∗

0.0028∗∗∗

0.0029∗∗∗

0.0023∗∗∗

0.0015

(2.7438)

(3.7613)

(4.3135)

(4.5856)

(4.7979)

(4.0861)

(.)

N

4685

5468

5569

5635

8352

4386

2043

R-squared

0.587

0.628

0.536

0.514

0.542

0.742

1.000

Constant

39

Table 10: Regression of Employment Decline on Self-Employment over Measures of of Occupational Skill and College Education. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the Firm Level.

Low Skill to Self

High Skill to Self

Low Skill to Employer

High Skill to Employer

No College

College

No College

College

No College

College

No College

College

0.008

0.043∗∗∗

0.019∗∗∗

0.031∗∗∗

-0.000

0.002

0.001

0.005∗∗∗

(1.07)

(3.97)

(3.45)

(4.74)

(-0.28)

(1.05)

(1.20)

(3.06)

0.010∗∗

-0.003

0.005

-0.000

0.001∗

-0.000

0.000

-0.000

%∆Emp

Constant

(2.47)

(-0.42)

(1.44)

(-0.04)

(1.95)

(-0.01)

(0.85)

(-0.50)

N

17781

14266

18878

22607

17781

14266

18878

22607

R-squared

0.539

0.476

0.364

0.286

0.712

0.592

0.548

0.282

Table 11: Regression of Employment Decline on Self-Employment over Measures of of Occupational Skill and College Age. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the Firm Level.

Low Skill to Self

%∆Emp

Constant

Under 30

30-40

0.000

0.044∗∗∗

(0.03)

(2.80)

0.014∗

-0.007

(1.70)

(-0.74)

High Skill to Self 40+

Low Skill to Employer Under 30

30-40

High Skill to Employer

Under 30

30-40

40+

0.032

0.005

0.027∗∗∗

0.034∗∗∗

40+

Under 30

30-40

0.005

-0.002

0.008∗∗

0.005∗∗

40+

0.001

(1.17)

(0.86)

(2.79)

(3.51)

(1.24)

0.007

(1.16)

(-0.77)

(2.21)

(2.11)

(1.60)

0.009

0.005

-0.000

0.001

-0.000

-0.001

0.003∗∗

-0.002

0.000

-0.002

(0.58)

(1.36)

(-0.08)

(0.18)

(-0.95)

(-0.50)

(2.19)

(-0.82)

(0.12)

(-1.00)

N

3026

7630

6740

3439

10657

10787

3026

7630

6740

3439

10657

10787

R-squared

0.380

0.421

0.720

0.328

0.273

0.484

0.027

0.402

0.933

0.199

0.360

0.477

40

Table 12: Regression of Employment Decline on Self-Employment over Measures of of Occupational Skill and Work Experiencel. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the Firm Level.

Low Skill to Self

High Skill to Self

Low Skill to Employer

Under 5

5-10

10+

Under 5

5-10

10+

0.010

0.034∗∗∗

0.049∗∗

0.020∗∗∗

0.022∗∗∗

0.046∗∗∗

-0.000

(1.35)

(3.75)

(2.10)

(3.98)

(3.85)

(3.59)

Constant

0.008∗

-0.000

-0.003

0.002

0.003

-0.003

(1.72)

(-0.07)

(-0.26)

(0.65)

(1.17)

(-0.50)

(1.14)

(0.18)

(.)

N

11974

13052

7021

14312

17314

9859

11974

13052

7021

R-squared

0.244

0.486

0.718

0.122

0.202

0.481

0.237

0.561

1.000

0.155

%∆Emp

Under 5

5-10

High Skill to Employer

10+

Under 5

5-10

10+

0.003

0.000

0.004∗∗∗

0.002

0.001

(-0.02)

(1.13)

(.)

(3.20)

(1.22)

(0.35)

0.000

0.000

0.001

0.000

0.001

0.001

(0.05)

(0.59)

(0.55)

14312

17314

9859

0.389

0.595

Table 13: Regression of Employment Decline over Subsamples of the Data. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the Firm Level.

Time

%∆Emp

Constant

Geography

Industry

Non-Recession

Recession

Prior 2002

After 2002

California

Outside CA

Pro Industry

Other Industries

0.024∗∗∗

0.032∗∗∗

0.027∗∗∗

0.030∗∗∗

0.012

0.034∗∗∗

0.027∗∗∗

0.031∗∗∗

(3.90)

(6.82)

(3.86)

(6.83)

(0.36)

(5.27)

(5.91)

(2.90)

0.007∗∗

0.001

0.005

0.003

0.017

0.000

0.004∗

0.004

(2.24)

(0.54)

(1.29)

(1.18)

(0.76)

(0.11)

(1.79)

(0.51)

N

30620

42912

18889

54643

4834

42083

49559

23973

R-squared

0.499

0.328

0.459

0.339

0.711

0.422

0.450

0.376

41

Table 14: Regression of Employment Decline on Entry to Entrepreneurship controlling for Public Firm Variables and Location Variables. We use * to denote significance at the 10% level, ** to denote significance at the 5% level, and *** to denote significance at the 1% level. Results are clustered at the Firm Level.

Financial Variables Base Measure %∆Emp

Emerge

Bank Concentration

House Price Growth

Bank Subset

Bank Con.

House Subset

House Price Growth

0.029∗∗∗

0.036∗∗∗

0.036∗∗∗

0.026∗∗∗

0.026∗∗∗

(5.23)

(5.71)

(5.71)

(2.82)

(2.80)

0.000

0.000

0.000∗∗∗

0.000∗∗∗

(0.38)

(0.36)

(2.90)

(2.68)

-0.000∗

-0.000∗

0.000

0.000

(-1.67)

(-1.67)

(0.04)

(0.06)

-0.010∗∗

Emerge from Bankruptcy

(-2.35) Liquidity

Profitability

Productivity

Market Leverage

Book Leverage

Log(Assets)

Employees

Employees/Assets

-0.004

-0.002

(-0.81)

(-0.43)

0.001

0.000

(0.35)

(0.07)

-0.015

-0.024

(-0.82)

(-1.33)

0.000∗

0.000

(1.72)

(1.44)

-0.006

-0.003

(-0.80)

(-0.40)

0.001

0.002

(0.61)

(0.70)

-0.002

-0.003

(-1.07)

(-1.17)

0.214

0.346

(0.75)

(1.11)

MSA Per Capita Income

MSA Population

Local Bank Concentration

0.001 (0.20)

Two Year House Price Growth

0.025 (1.60)

42 Constant

MSA Fixed Effects N

0.001

0.019

-0.002

-0.002

-0.023

-0.025

(0.08)

(1.02)

(-0.21)

(-0.23)

(-0.98)

(-1.03)

No

No

No

No

Yes

Yes

44273

44137

42435

42435

24029

24029

43

Figure 1: Comparison of US Employment Share from our Online Business Networking Database and BEA Employment Data