ESSAYS IN VENTURE CAPITAL, ENTREPRENEURSHIP, AND MANAGERIAL SUCCESS

Transcription

1 ESSAYS IN VENTURE CAPITAL, ENTREPRENEURSHIP, AND MANAGERIAL SUCCESS by QIANQIAN DU M.Sc., The University of Oxford, 2004 B.A., Shandong University, 2003 THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE FACULTY OF GRADUATE STUDIES (Business Administration) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver) June 2009 Qianqian Du, 2009

2 ABSTRACT The first chapter of my dissertation examines the preferences of venture capitalists for syndication partners. Heterogeneity among syndication partners may cause efficiency loss and increase transaction costs but offer syndication partners valuable learning opportunities in the long run, suggesting a tradeoff between the short-term costs versus long-term benefits. Using data on U.S. venture capital investments, I find that venture capital firms are less likely to syndicate with partners who are different from them. The preferences for syndication partners, however, have different implications for the portfolio companies and the venture capital firms. Companies funded by heterogeneous syndicates are less likely to go public or get acquired by other companies. However, venture capital firms that co-invest with more heterogeneous partners are more likely to survive. This paper develops a new method for empirically examining the formation of syndication among multiple firms. It also addresses issues of endogeneity. In the second chapter, we develop an economic framework which articulates the impact of the quality of legal protection offered to investors on the incentives of start-up founders to recruit partners or opt for sole ownership. The theoretical analysis predicts that a positive relationship is likely to exist between the quality of the legal system and ownership concentration of start-ups. This prediction is supported by the data obtained from the Adult Population Survey of the Global Entrepreneurship Monitor project between 2001 and The third chapter finds that the number of CEOs born in summer is disproportionately small, and firms with summer born CEOs have higher market valuation. Our evidence is consistent with the relative-age effect due to school admissions grouping together children with age differences up to one year, with summer-born children disadvantaged throughout life by being younger than non-summer-born classmates. Those younger children who nevertheless succeed have to be particularly capable. ii

3 TABLE OF CONTENTS ABSTRACT... ii TABLE OF CONTENTS... iii LIST OF TABLES... iv LIST OF FIGURES...v ACKNOWLEDGEMENTS... vi DEDICATION... vii CO-AUTHORSHIP STATEMENT... viii 1 INTRODUCTION BIRDS OF A FEATHER OR CELEBRATING DIFFERENCES? THE FORMATION AND IMPACT OF VENTURE CAPITAL SYNDICATION Introduction Theoretical Considerations Data and Variables Empirical Analysis Conclusions References INTERNATIONAL PATTERNS OF OWNERSHIP STRUCTURE CHOICES OF START-UPS: DOES THE QUALITY OF LAW MATTER? Introduction Does Law Matter? Law and Ownership Decisions Data and Methodology Econometric Analysis Discussions and Conclusions References BORN LEADERS: THE RELATIVE-AGE EFFECT AND MANAGERIAL SUCCESS Introduction The Relative-age or Birth-date Effect The Relevant Seasons of Birth Data Employed: CEO Birth-dates and Firm Characteristics The Prevalence and Performance of CEOs by Birth Season Relative Age, CEO Birth-dates and Birthday-related Performance Conclusions References CONCLUDING CHAPTER Conclusion and Discussions References iii

4 LIST OF TABLES Table 1 The Definition of Variables...23 Table 2 Descriptive Statistics of Samples...27 Table 3 Correlation Matrix of Key Variables...27 Table 4 The Base Model...29 Table 5 Network Centralities and Geographic Distances...33 Table 6 Prior Relations...34 Table 7 Sample for Portfolio Companies Performance...36 Table 8 Sample for the Performance of Syndication...38 Table 9 The Performance of Syndication...39 Table 10 Selection Issues of the Performance of Syndication...43 Table 11 Sample for VCs Survival...44 Table 12 Correlation Matrix for VCs Survival...45 Table 13 VCs Survival...50 Table 14 Selection Issues of VCs Survival...51 Table 15 Country Composition and Legal Origin...65 Table 16 Description of the Legal Variables...68 Table 17 Description of Other Variables...69 Table 18 Descriptive Statistics of Variables...71 Table 19 Regressions on Legal Origins...74 Table 20 Regressions on Legal Enforcement...76 Table 21 Robustness Checks...77 Table 22 Country Level Regressions...80 Table 23 Descriptive Statistics...92 Table 24 Correlation Matrix...93 Table 25 Season of CEO Birth and Firms Valuation...98 Table 26 Season of CEO Birth and Firms Stock Performance Table 27 U.S. School Cutoff Dates Table 28 Relative Age and Firms Valuation Table 29 The Latest Possible Quarter and Firms Valuation Table 30 Relative Quarter and Firms Valuation iv

5 LIST OF FIGURES Figure 1 Number of CEOs by Birth Season...94 Figure 2 Season of Birth: CEO Sample versus U.S. Population...95 Figure 3 Performance of Portfolios by CEO Birth Season Figure 4 Number of CEOs by Relative Age v

6 ACKNOWLEDGEMENTS I feel grateful for all of my advisors who have offered invaluable help and advice during my doctoral study at UBC, and all of my co-authors for their great inspirations and support. I am greatly indebted to Professor Thomas Hellmann, who led me to an exciting research area and trained me to think more deeply and critically. I also want to thank Professor Ilan Vertinsky in particular for his excellent guidance. I owe special thanks to my parents, who always try their best to support me in any possible way at anytime. vi

7 DEDICATION To my parents vii

8 CO-AUTHORSHIP STATEMENT My second chapter is co-authored with Ilan Vertinsky. I developed the research idea. I obtained the data from Global Entrepreneurship Monitor and performed econometric analysis. I also prepared for the first draft of the paper and made my contribution to the subsequent revisions of the paper. My third chapter is co-authored with Huasheng Gao and Maurice Levi. I developed the research idea and discussed it with my co-authors. I also collected data from various sources with my co-authors. I performed econometric analysis independently and then discussed and compared the results with my co-authors. I also contributed to the manuscript preparation. viii

9 1. INTRODUCTION 1.1 Venture Capital Syndication Venture capital firms (VCs) have played an important role in providing equity financing to technology intensive start-up companies. A common feature of venture capital investments is syndication among venture capitalists, which means different VCs co-investing in a financing round of their backed portfolio company. Whether or not venture capital syndication can reduce investment risks and achieve better performance than standalone investments crucially depends on what kind of syndication partners a venture capital firm is able to attract. My first chapter focuses on the heterogeneity among syndication partners and studies how heterogeneity among syndication partners affects the formation and performance of syndication. Heterogeneity entails costs. Based on positive assortative matching on VCs experience, which is a key dimension of heterogeneity, VCs with different levels of experience are less likely to form syndication in the first place. After the syndication is formed, heterogeneous syndicates may suffer from higher transaction costs, ex post. There are also benefits of heterogeneity. Heterogeneous syndicates can provide valuable learning opportunities to syndication partners. We also expect that the benefits from learning may not be realized immediately but more likely to be harvested by VCs in the long-term. To empirically examine the impact of investor heterogeneity, I use the venture capital investments made to U.S. companies between 1990 and 2005 from the Thomson Financial s VentureXpert database. I also developed a new algorism to predict the formation of syndication among a group of VCs while prior research mostly predicts alliance formation between two firms. After addressing endogeneity problems, I obtain the following findings. First, VCs have strong preferences for syndication partners with similar levels of experience and performance. The 1

10 findings remain robust after controlling for other characteristics of VCs. Second, companies funded by heterogeneous syndicates, in which VCs have different levels of performance, are less likely to have IPOs and sales to other companies, which proxy for the performance of venture capital investments. Third, VCs, whose partners are more heterogeneous, are more likely to make new investments and diversify their investment portfolios, and eventually survive in the future. This paper makes multiple contributions to the literature. First, it captures heterogeneity among VCs in a syndicate for the first time, and shows that such investor heterogeneity can partially explain both the formation and impact of syndication. Second, this paper develops a new matching algorism and predicts the syndication formation among a group of investors for the first time. Third, although this paper uses venture capital as research context, it has general implications for alliances among firms, teams, and social networks. 1.2 Legal System and Entrepreneurs Ownership Choices Prior research suggests that high quality legal systems, which offer effective investor protection, can lead to better developed capital markets and more dispersed ownership (La Porta et al., 1998; Glaeser, Johnson, and Shleifer, 2001; Djankov et al., 2003; and Demirguc-Kunt and Levine 2001). Despite the important contributions of small and medium sized enterprises (SMEs) to economic growth (Berger and Udell, 1998), to our knowledge, no study has considered the impact of legal systems on the ownership structures at the founding stage of small and medium firms that are not backed by private equity firms. This segment of new enterprises contains the majority of start-ups, both by value and number. 1 In the theoretical framework, we assume that there are two sources of capital to finance start-ups at founding stage: internal capital and external capital. Internal capital mainly refers to the equity 1 For example, 94.5% of U.S. nonfarm, nonfinancial, nonreal-estate small businesses or $ billion in monetary value belong to this segment (Berger and Udeall, 1998). 2

11 capital obtained from start-up founders and their co-owners while external capital mainly refers to the debt capital raised by entrepreneurs from banks. We argue that the quality of a legal system can have different impacts on internal and external investors. If the legal system fails to provide adequate protection, increases in internal investors ownership and control rights can work as a substitute for the inadequate legal protection. External investors, however, do not have such substitute and thus become less willing to invest in start-ups. Therefore, in a poor legal system, the relative cost of external capital to internal capital tends to be higher, resulting in less external debt financing and more dispersed ownership structures. Using data from the Adult Population Survey of the Global Entrepreneurship Monitor project between 2001 and 2004, our predications are supported. 1.3 Relative-age and Managerial Success Who are more likely to become CEOs of S&P 500 companies? What makes a successful CEO? My third chapter suggests one important factor that partially explains the probability of becoming a CEO and the subsequent success of a CEO. There is mounting empirical evidence that summer born children are at a disadvantage as a result of being up to a year younger than other classmates in their school grade, due to the fact that the cutoff dates for admission into school generally fall at the end of summer. The disadvantage faced by summer born children has been shown to exist throughout school, and even to affect the success at entering university. This well-documented condition has become known as the relative-age effect or the birth-date effect. My third chapter examines whether a relative-age or birth-date effect extends to the selection and performance of CEOs of S&P 500 companies. We argue that as non-summer born children are relatively older in the class, they have a better chance of gaining leadership-related experience (e.g. becoming a team captain, a school monitor, 3

12 etc.). Some summer-born children who are exceptional in ability within their age cohort manage to gain the same experience. To be selected for leadership related activities that will help become CEOs later, a threshold of demonstrated performance is required. The threshold can be met by a combination of development which is related to age, and innate ability which is evenly distributed across birth seasons. Via their greater development, more non-summer born is given such leadership related experience and therefore more of them become CEOs. Since non-summer born achieve the threshold of demonstrated performance more through development than through innate ability while the summer born achieve the threshold more through innate ability than through development, the average ability of summer born CEOs is likely to be higher than that of the non-summer born CEOs. As a result, firms headed by summer born CEOs outperform those headed by non-summer born CEOs. We construct a birth-date dataset for the CEOs of S&P 500 companies between 1992 and 2006 and supplement it with other characteristics of CEOs and company characteristics from ExecuComp, CRSP, and Compustat. We find that non-summer born individuals have a significantly higher chance of becoming a CEO of an S&P 500 company. Conditional on becoming a CEO, those who were born in summer add higher value to their company whether this is considered via Tobin s Q involving market and book value of assets, or the market to book value of equity, M/B. We also show the return from a policy of forming a portfolio based on buying companies with summer born CEOs and selling short companies with non-summer born CEOs. This generates an annual abnormal return of 8.32 percent. 4

13 1.4 References Berger, A., and G. Udell The Economics of Small Business Finance: The Roles of Private Equity and Debt Markets in the Financial Growth Cycles. Journal of Banking & Finance, 22: Demirguc-Kunt, A. and R. Levine Financial Structure and Economic Growth: A Cross-Country Comparison of Banks, Markets, and Development. Cambridge, MA: MIT Press. Djankov, S., R. La Porta, F. Lopez-De-Silanes, and A. Shleifer Courts. The Quarterly Journal of Economics, 118 (2): Glaeser, E., S. Johnson, and A. Shleifer Coase versus the Coasinas. The Quarterly Journal of Economics, 116 (3): La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny Law and Finance. Journal of Political Economy, 106 (6):

14 2 BIRDS OF A FEATHER OR CELEBRATING DIFFERENCES? THE FORMATION AND IMPACT OF VENTURE CAPITAL SYNDICATION Introduction Venture capital firms (VCs) provide equity financing to technology intensive start-up companies and realize returns if their funded companies go public or get acquired by other companies. To reduce the high risks associated with investing in private companies, venture capital firms may choose to syndicate their investments. Syndication is formed when different VCs co-invest in a financing round of their backed company. Syndication is very common among VCs. For example, around 60% of venture capital backed start-up companies have received at least one syndicated financing rounds in the past two decades in the U.S. It is the syndication that helps to create a national network of VCs (Kogut, Urso, and Walker, 2007). Compared with standalone investments, venture capital syndication is formed to achieve different goals, for example, to obtain a second opinion from partners and improve the assessment of the investment opportunities (Lerner, 1994), to access complementary management skills of syndication partners (Brander, Amit, and Antweiler, 2002), to invest in geographically distant companies (Sorenson and Stuart, 2001), or to gain future reciprocity (Hochberg et al., 2007). Syndication also matters for performance. For example, Brander et al. (2002) found that syndicated investments tend to outperform standalone investments. Hochberg et al. (2007) showed that venture capital firms that are more connected in a network are associated with better performance. Whether syndicates can achieve the above goals and meet performance targets crucially depend on what kind of syndication partners a venture capital firm is able to attract. This paper focuses 2 A version of this chapter will be submitted for publication. Du, Q. Birds of a Feather or Celebrating Differences? The Formation and Impact of Venture Capital Syndication. 6

15 on the heterogeneity among syndication partners and studies how heterogeneity among syndication partners affects the formation and performance of syndication. 3 Prior research has documented a long list of the costs and benefits of heterogeneity. On the one hand, heterogeneity may make communication and coordination less effective among group members (Van den Steen, 2004), resulting in slower actions and responses in the competitive environments (Hambrick, Cho, and Chen, 1996). On the other hand, heterogeneous groups provide valuable learning opportunities for the group members in the long term. Heterogeneity may encourage group members to collect new information (Van den Steen, 2004), improve the group s problem solving ability (Hoffman and Maier, 1961), and may lead to increased innovation within groups. The tradeoff between the costs and the benefits of heterogeneity, therefore, has important implications for firms preferences for alliance partners and the performance of the formed alliances. Specifically, this paper studies the following three research questions: (1) Do VCs prefer partners that are similar to or different from them? (2) Will VCs preferences for partners affect the performance of the syndicated investments? (3) Will VCs preferences for partners have any impact on the VCs? One important characteristic that differentiates VCs from each other is their experience. Through deal selection and value added services, VCs experience has a positive impact on venture success (Gorman and Sahlman, 1989; Hellmann and Puri, 2002; Sorensen, 2007). As returns to private equity persist (Kaplan and Schoar, 2005), a VC s prior performance is an appropriate indicator of its future performance. Therefore, experience and performance are two key attributes VCs use to select syndication partners. Heterogeneity entails costs. First, given the importance of VCs experience for venture success, it is reasonable to assume that VCs experience is complementary to the returns of the syndicated 3 As this chapter studies how VCs choose syndication partners, it focuses on VCs that prefer to syndicate their investments to invest alone. Therefore, both the theoretical and empirical analysis is based on VC syndicates instead of standalone investments. 7

16 investments. Therefore, VCs try to syndicate with partners that are at least as experienced as they are in order to maximize returns. In equilibrium, the levels of VCs experience in a syndicate are positively correlated. Second, heterogeneous syndicates may incur higher transaction costs ex post. Communication and coordination costs can be higher, leading to more conflicts, slower decision making, and delayed execution. Different knowledge of one market can increase the information asymmetry among VCs, leading to less effort from the less experienced investors. Third, similarity attraction makes heterogeneous syndicates less attractive. The costs of heterogeneity imply that VCs are less likely to form syndicates with partners who are different from them. After the syndication is formed, companies funded by heterogeneous syndicates may underperform those funded by homogeneous syndicates. There are also benefits of heterogeneity. First of all, heterogeneous partners can bring in different insights and perspectives which enable them to make better investment decisions in the future. The benefits of heterogeneity may be greater for VCs which operate in highly risky environments, in which decision making can be subjective. Through co-investing with partners with diverse experience backgrounds, VCs can observe how their partners make investment decisions and expose themselves to more options to solve problems. Although learning can be beneficial, we would not expect the benefits of learning to realize immediately. Instead, we expect that VCs learn to invest better gradually and harvest the benefits of learning in the long-term. The benefits of heterogeneity suggest that VCs, whose partners are more heterogeneous, may be able to make more and better investments in the long term. To test the above empirical implications, I use the venture capital investments made to U.S. companies between 1990 and 2005 from the Thomson Financial s VentureXpert database. To predict the syndication formation, I need to collect realized syndicates and construct hypothetical syndicates that are not realized. I develop the Single Deviation Method, in which I construct hypothetical syndicates by replacing each VC in a realized syndicate with each potential investor 8

17 one at a time. While prior research mostly predicts the alliance formation among two firms, the Single Deviation Method allows me to predict syndication formation among a group of investors, and measure heterogeneity among these investors. Results remain the same after I perform robustness checks on this method. When I test the second implication, I base the analysis on the realized syndicates only. To examine the long-term benefits of learning for VCs, I construct a panel data with VC-year pairs as units of analysis and include VCs with at least two syndication partners in any year between 1995 and To measure the heterogeneity among VCs, I apply the coefficient of variation measure to continuous variables and the entropy measure to categorical variables, both of which are widely used in the sociology literature. I focus on two dimensions of heterogeneity in VCs: VCs prior experience, and performance in the portfolio companies industries. 4 The empirical findings are summarized in three parts: First, VCs have strong preferences for syndication partners that are similar to themselves. Specifically, syndicates are more likely to be formed among VCs with similar levels of experience and performance. The findings remain robust after controlling for VCs network centrality scores that capture how connected VCs are in networks, geographic distances among VCs, geographic distances between VCs and their portfolio companies, and prior syndication among VCs. Second, companies funded by heterogeneous syndicates, in which VCs have different levels of performance, are less likely to have IPOs and sales to other companies, which capture performance of venture capital investments. Two types of selection may bias the results. The first type is the selection among VCs which tend to form less heterogeneous syndicates. To study how heterogeneity among VCs affects the performance of their syndication, we need to base the 4 Due to the limited space, I use VCs overall experience and performance, and their experience and performance in the portfolio companies geographic location (states) as robustness checks. I obtain similar results, which are available upon request. 9

18 analysis on the realized syndicates that are in general less heterogeneous than those not realized. To address the first type of selection, I apply the Heckman two-step procedure which includes a selection equation and a performance equation. The identification of the Heckman two-step model comes from the geographic distances among VCs, and the heterogeneity in distances between each VC and the portfolio company. Such geographic distances have a negative impact on the formation of syndication but should not have a direct impact on the performance of the syndication once it is formed. The second type is the more well-known selection between VCs and start-ups (Sorensen, 2007), in which top tier VCs invest in top tier start-ups. To address this type of selection, I adopt an instrumental variable approach based on Ackerberg and Botticini (2002). The instruments are dummy variables of companies local markets, constructed as an interaction of companies industries with their geographic location (states). The predictions from the Heckman two-step procedure and the Ackerberg and Botticini Approach are consistent with the original results. Third, since it takes years for VCs to successfully exit their investments, my sample of relatively recent transactions limits my ability to study the long-term performance of VC funds, which are typically organized as ten-year close end funds. Instead, I focus on another important feature of VCs their survival. I find that VCs, whose partners are more heterogeneous, are more likely to make new investments and diversify their investment portfolios, and eventually survive in the future. The results of survival remain robust to different definitions of VCs survival, and in a sub-sample of U.S. independent VCs. Anticipating heterogeneous partners may help survival in the future, VCs may select into more heterogeneous syndicates. To address this selection, I construct the local availability of heterogeneous partners for each VC as an instrument for the heterogeneity in VCs syndication partners. The local availability of heterogeneous partners can affect a VC s choice of syndication partners but should not affect the survival of that particular VC. The instrument variable approach confirms the original predictions. 10

19 This paper makes multiple contributions to the literature. First, it captures heterogeneity among VCs in a syndicate, which has not been studied in the literature, and shows that such investor heterogeneity can partially explain both the formation and impact of syndication. Second, this paper develops a new matching algorism and predicts the syndication formation among a group of investors for the first time, to the best of my knowledge. Third, although this paper uses venture capital as research context, it has general implications for alliances among firms, teams, and social networks. For example, firms and people are more likely to be attracted to those that are similar and easy to find. The preference for similarity may generate some immediate benefits. In the long run, however, it may limit learning and future business opportunities. The remainder of this paper is organized as follows: Section 2.2 presents the theoretical considerations and reviews the related literature. Data and variables are described in Section 2.3 and empirical results are presented in Section 2.4. Section 2.5 concludes the paper Theoretical Considerations Dimensions of heterogeneity One important characteristic that differentiates VCs from each other is their experience. VCs experience can improve the performance of their portfolio companies through two channels: First, more experienced VCs can add more value to their funded companies. Unlike traditional financial intermediaries, VCs closely monitor and advise their portfolio companies, including sitting on the boards of portfolio companies (Lerner, 1995; Baker and Gompers, 2003), recruiting management teams (Gorman and Sahlman, 1989), advising portfolio companies to adopt stock options and hiring outside CEOs (Hellmann and Puri, 2002), reducing the costs of going public (Megginson and Weiss, 1991), and facilitating strategic alliances among their portfolio companies (Lindsey, 2008). Second, more experienced VCs invest in higher quality deals (Sorensen, 2007). More experienced VCs may have access to some proprietary deal flow (Kaplan and Schoar, 2005) and are more able to select promising start-ups (Casamatta and 11

20 Haritchabalet, 2007). Entrepreneurs also tend to accept financing offers from more reputable and experienced VCs (Hsu, 2004). It is the heterogeneity in VCs skills that may cause the heterogeneous performance of private equity funds (Kaplan and Schoar, 2005). Another important characteristic is VCs prior performance. Kaplan and Schoar (2005) find that the returns to private equity investments persist. Therefore, VCs prior performance is a good indicator of their future success. Both prior experience and performance are two important attributes VCs use to select syndication partners. VCs are usually specialized in certain industries and geographic location. VCs industry specialization is attributed to their general partners (GPs) business experience before entering the VC industry. Such business experience enables them to be more actively involved with their portfolio companies (Bottazzi, Da Rin, and Hellmann, 2007). VCs may prefer closely located portfolio companies because the geographic proximity helps them to monitor the investees more effectively (Sorenson and Stuart, 2001; Tian, 2008). Therefore, when forming syndication, VCs will also consider as important factors partners relevant experience and performance in the target companies industries and geographic location The costs of heterogeneity In a two-sided matching framework (Roth and Sotomayor, 1990), heterogeneous syndicates may incur efficiency loss. 5 Given the importance of VCs experience for venture success, it is reasonable to assume that VCs experience is complementary to the returns of the syndicated investments, which means the return to a VC s experience will increase with its partners experience. Therefore, VCs try to syndicate with more experienced partners to maximize their returns. In equilibrium, the levels of VCs experience in a syndicate are positively associated. The following symbolic representation illustrates the idea. 5 I borrowed the idea from two-sided matching but the problem we have here is not a standard two sided matching problem as we are matching within VCs not matching VCs and companies. 12

21 The economic problem is how VCs select syndication partners to maximize the returns of the syndicated investments. 6 The incumbent VC that already commits to investing in the target company needs to recruit another VC to form a syndicate. The incumbent is denoted by i and the entrant by j. The matching is based on the levels of VCs experience denoted by E ( EH for high level of experience while EL for low level of experience), to maximize the returns denoted by U ij. I also assume complete information so that VCs experience is common knowledge. As the problem focuses on the selection of syndication partners after VCs have chosen to syndicate, every VC will be matched with a partner in equilibrium. The complementarities of VCs experience are given by: U 2 ij ( i i E E, j j E E ) 0 or U i, j ) U ( i, j ) U ( i, j ) U ( i, j ) ( EH EH EL EH EH EL EL EL Although inexperienced VCs prefer more experienced partners, they are not chosen by more experienced VCs and have to syndicate with other inexperienced VCs. In equilibrium, only VCs with similar levels of experience form syndication. The total payoff of the positive assortative matching is superior to that of the negative assortative matching: U ( ieh, jeh ) U ( iel, jel) U ( ieh, jel) U ( iel, jeh ) Anticipating that heterogeneous syndicates may incur higher transaction costs ex post, VCs may consider syndication partners with different levels of experience less attractive. In a game-theoretical model, Van den Steen (2004) shows that agents with heterogeneous beliefs and preferences may have more information distortion during their communication, and less alignment of actions. In venture capital syndication, VCs with heterogeneous venture experience may have more disagreements and conflicts when making investment decisions. In addition, concerns with agency costs may prevent experienced VCs from co-investing with inexperienced VCs. The less experienced VCs may exert less effort due to the fact that they may not be able to 6 This example takes the matching between VCs and companies (Sorensen, 2007) as given. It only considers the matching among VCs to simplify the problem. 13

22 provide high quality advice to the portfolio companies or they simply free ride the services provided by more experienced partners. Homophily exists in various relationships, including friendship, co-membership, and marriage relationship (McPherson, Smith-Lovin, and Cook, 2001). The similarity attraction among VCs may be caused by lower costs of searching for similar partners, because the information of investment opportunities tends to be circulated in the markets where VCs with similar investment preferences participate. A VC s experience may also signal its reputation. More experienced VCs may not be willing to be associated with partners that are less reputable. The costs of heterogeneity have implications on both the formation and performance of syndication. VCs may be less likely to form syndicates with partners who are different from them. Conditional on syndicates which are already formed, companies, which are funded by heterogeneous syndicates and thus suffer more from transaction costs, may underperform those funded by homogeneous syndicates The benefits of heterogeneity Syndicating with heterogeneous partners can create valuable learning opportunities for VCs in the long run. First of all, heterogeneous partners bring in different insights and perspectives that can improve firms decision making in the future. The evidence of learning can be drawn from both the lab experiments, in which heterogeneous groups generate higher quality solutions to complex and non-routine problems (Hoffman and Maier, 1961), and the real world firm actions, in which firms can sample various decision making processes and learn how different decisions lead to different outcomes (Beckman and Haunschild, 2002). Although heterogeneity in partners experience may cause conflicts, firms may thus have incentives to present evidence and collect new information in support of their own actions. If all disagreeing parties try to do so, there will be more useful information available for good decisions. This type of learning can be an 14

23 application of a more general economic model developed by Rotemberg and Saloner (1995), in which the overt conflicts between different function areas in a firm can be beneficial for top management s decision making. The benefits of heterogeneity may be greater for VCs which face many complex and non-routine problems when investing in highly risky private companies. 7 Decision making in venture capital investments, such as opportunity assessment, recruiting the right management team, and the timing of exit, can be subjective. Co-investing with partners with diverse experience can be an effective way of learning different management skills and collecting more useful information. Such learning can improve VCs abilities of deal selection and enable VCs to capture more investment opportunities in their specialized industries. However, we do not expect the benefits of learning to realize immediately. Instead, we would expect VCs learning from the current investments will benefit their investments in the future so that VCs may be able to make more successful investments in the long run. The benefits of heterogeneity suggest that VCs, whose partners are more heterogeneous, may be able to make more and better investments in the long term Related literature The first paper studying venture capital syndication is Lerner (1994), based on 271 biotechnology firms that received VC financing between 1978 and Lerner (1994) shows that VCs have more syndication partners with similar fund sizes in the first financing rounds but the paper does not intend to predict the formation of a particular syndicate, which is one focus of this paper. After analyzing detailed financing contracts between venture capitalists and start-up companies, Kaplan and Stromberg (2004) found that the size of the syndicate, measured by the number of venture capital funds forming the syndicate, is positively associated with monitoring and support provided to the start-up companies. 7 Please see Sorensen (2008) on how VCs mitigate the investment uncertainties through exploitative and exploratory learning. 15

24 Casamatta and Haritchabalet (2007) provided a theoretical model explaining why venture capital firms syndicate. They argued that since inexperienced VCs receive a noisy signal of the true quality of an investment project, they do not fear to disclose their own evaluation of the project to attract syndication partners. However, due to the competition among VCs, more experienced VCs, who receive an accurate signal of the true quality of the project, are reluctant to reveal their own evaluation and hence syndicate less frequently unless they have even more experienced partners. Cestone, Lerner, and White (2006) focused on a contract design to induce VCs to truthfully disclose their signals of the investment opportunities. Depending on whether or not the lead VCs can manipulate their signals, the gains from syndication may or may not be maximized if syndication partners are too experienced. The two theoretical models have similar empirical implication on the syndication pattern for experienced VCs, that is, experienced VCs tend to syndicate with other experienced VCs. This paper lends empirical support to their theoretical models. Prior work also studied the formation and performance of venture syndication but with very different perspectives from this paper. For example, a recent paper by Sorenson and Stuart (2008) studied the formation of distant dies among VCs when there is heat in target companies industries and location. This paper differentiates from Sorenson and Stuart (2008) by focusing on the formation and impact of syndication based on the tradeoff between the costs and the benefits of heterogeneity, and treating syndication as a group concept instead of dyadic relations. When studying the performance of venture capital syndicates, Brander et al. (2002) compared the performance of syndicated deals with that of standalone deals. This paper focuses on the syndicated deals and investigates the impact of heterogeneity among syndication partners on the performance of syndicates. Hochberg et al. (2007, 2009) have studied how venture capital firms network centralities influenced venture capital funds performance and deterred entry. Their key variable of a VC s network centrality, which is a function of the quantity and the quality of 16

25 syndication partners, is very different from the heterogeneity measure of syndicating VCs used in this paper Data and Variables Sources of data The data are complied from the Thomson Financial s VentureXpert database, which provides information on the characteristics of VCs, portfolio companies, and the deals. I also merge the VentureXpert database with Thomson Financial s Global New Issues and Mergers and Acquisitions databases to identify more IPOs and acquisitions. I use the matching software 8 to take care of the inconsistent records of company names across these databases (Egan, 2007). The VentureXpert database tends to record parent VC firms and their affiliates as different firms. It also tends to report an international VC which has overseas offices in the U.S. as a U.S. firm. I visit VCs websites whenever available and check with the Hoover s company records to identify changes in firms names, acquisitions, firms affiliates, firms countries of origin, firms types, and their branch offices. I also exclude the financing rounds at the stages of Buyout/Acquisition, Other, Unknown, and focus on the MoneyTree 9 deals. Another well recognized problem with VentureXpert is the round-splitting problem (Gompers and Lerner, 1999), in which the same financing rounds were recorded more than once. To identify these splitted rounds, I analyzed a sample of financing rounds if the duration between any two of them is less than or equal to 3 months. If two financing rounds had the same VCs and raised the same amount of money within 3 months, I eliminate the later round which is likely to 8 Egan (2007) describes different techniques of normalizing companies names. The technique used here is to remove the redundant suffix such as Inc., Corp., Ltd., etc. 9 MoneyTree report is a collaborated report between the PricewaterhouseCoopers and the National Venture Capital Association in the U.S. It produces quarterly reports on venture capital activities in the U.S. VentureXpert has a variable to indicate if a deal is a MoneyTree deal. 17

26 be a repetition of an earlier round and was included in VentureXpert by mistake. If two financing rounds had different VCs and raised different amounts of money within three months, I combined these two rounds as a single round and added up VCs and the amount of financing. This exercise is helpful to identify some suspicious splitted rounds and test whether they cause any biases. Econometric analysis performed in a sample without suspicious splitted rounds generated consistent predictions with those obtained in the main sample used in this paper. Detailed data of VCs financial returns is not publicly available because the disclosure of their returns is not required by law. I therefore adopt a widely used proxy for VCs returns in the literature the number of successful exits, which is measured by the number of portfolio companies which have IPOs or acquisitions by other companies (Hochberg et al., 2007). Based on a sample of VC investments with real financial returns, Phalippou and Gottschalg (2007) find a positive and significant correlation between the real financial returns and the number of VCs successful exits through IPOs or acquisitions. To study the formation of syndication, I focus on the first round syndicated VC investments made to U.S. companies between 1995 and Syndication is defined as different VCs co-investing in the first financing round. I use the first round deals because the syndication in later financing rounds largely depends on who has participated in the first rounds (Admati and Pfeiderer, 1994), and the risks of investments are greatly reduced at later stages. Therefore the demand for risk sharing and the criteria for partners are most salient in the first rounds. 11 After excluding syndicates with undisclosed VCs and companies, there are 3,385 U.S. companies that received their first round financing from VC syndicates between 1995 and The entire sample covers VC investments between 1990 and As I calculate VCs experience in a year as the number of companies it funded in the past five years, the analysis of the syndication formation therefore starts with Sorensen (2007) also uses the first rounds to study how VCs experience affects the chance of IPOs. 18

27 Examining the survival of VCs requires a different data structure, in which I can trace the activities of each VC over time. Therefore, I construct a panel data with VC-year pairs as units of analysis and include VCs if they have at least two syndication partners in any year between 1995 and The final sample includes 3,380 VCs with 12,434 VC-year observations The Single Deviation Method To predict whether the syndicates are formed or not, I need to have realized syndicates that are already recorded in the dataset and hypothetical syndicates that are not realized. The key task of constructing hypothetical syndicates is to find VCs who are potential investors in the realized deals but do not make actual investments. The potential investors then form hypothetical syndicates for each realized deal. The following two steps describe how I construct the hypothetical syndicates: Step 1: I first match each realized syndicate with a pool of first round syndicated deals that meet the three inclusion criteria, that the realized syndicate and its matched syndicates should be in the same industry, state, and funded in the same quarter of the year. I then form a set of potential investors by including VCs participating in the matched syndicates and excluding the actual investors from the realized syndicate. As VCs are specialized in certain industries and geographic areas, the criteria of the same industries and states include VCs who are actually interested in investing in such industries and states. The criterion of having investments in the same quarter of the year makes sure that VCs still have funds available for investments. The final sample has 2,451 portfolio companies that 12 A syndication partner is defined as a co-investor in a financing round. For each VC-year observation, I include all of the syndication partners a VC had in the year. The requirement of at least two partners is to make sure a meaningful measure of heterogeneity in VCs syndication partners. 19

28 are matched with potential investors based on the three inclusion criteria. 13 Such criteria will be relaxed in the later robustness checks. Step 2: I then apply the Single Deviation Method, which replaces each VC in a realized syndicate by each potential investor one at a time. It looks for a local maximum in a neighborhood as there is only one different investor between a realized syndicate and a hypothetical syndicate. Alternatively, the Multiple Deviation Method searches for a global maximum, in which multiple VCs in a realized syndicate are replaced by other potential investors. As the Multiple Deviation Method will create a sample with too many observations for a computer to handle efficiently, I only apply the Multiple Deviation Method to syndicates formed by three VCs as a robustness check and focus on the Single Deviation Method. The two steps are summarized in the following example: Star Inc. is a biotechnology company in California and was funded in the first quarter of 2002 by VCs A, B, and C. Meanwhile, VCs D and E funded another biotechnology company in California in the first quarter of Therefore, D and E are potential investors that could have funded Star Inc. The following table illustrates all of the hypothetical syndicates constructed by the Single Deviation Method and the Multiple Deviation Method, respectively. Each column represents a syndicate of VCs. VCs who are potential investors of Star Inc. are in italic. 13 There are about 900 companies which are not matched with other deals because they are the only companies funded by VCs in a state, or in an industry, or in a quarter of the year. To check whether excluding these unpopular companies affects the empirical results, I add them back in an extended dataset. I also relax the three inclusion criteria by focusing on any two of them at a time. The number of observations increases dramatically but the empirical results remain the same. To make the later analysis more efficient, I only focus on the sample of 2,451 companies. 20

29 The Formation of Hypothetical Syndicates Real Hypothetical Syndicates Single Deviation Method Multiple Deviation Method A D E A A A A D A D B B B D E B B E D B C C C C C D E C E E Total 3 x 2 + 1=7 3 x 1 +1 =4 There are several advantages of using the Single Deviation Method. It allows me to study the syndication formed by a group of investors. It also enables me to focus on the time-varying characteristics of VCs. As regards the type of syndicates VCs form, it is determined by their characteristics at the time of syndication formation. Third, constructing hypothetical syndicates based on portfolio companies allows me to study company specific characteristics. For example, I can control for company specific variables and construct investor characteristics that are the most relevant for the portfolio companies Variables Dependent variables To predict syndication formation, I construct a dummy variable REALIZED as the dependent variable, which is equal to 1 if the syndication is realized. To predict the performance of syndication, I use EXIT as the dependent variable, which is equal to 1 if the portfolio company funded by the syndication has an IPO or sale to another company, and 0 otherwise. 14 To study 14 It is possible that some acquisitions of VC-backed companies do not generate profits for their investors. Therefore, I constructed a variable as a proxy for the return of acquisition. It is the difference between the amount paid by the acquirers and the total amount of money raised by the startup from VCs, divided by the total amount of money raised by the startup. I replicated the regression analysis in samples where the successful exits include IPOs and acquisitions with returns bigger than 10% or 20%. I obtained similar results. Given the risks and illiquidity associated with VC investments, the returns of acquisition of 10% or 20% only serve as a lower bound of a profitable acquisition. 21

30 VCs survival, I use a dummy variable DEATH as the dependent variable. As VCs are organized as limited partnerships, accurate information on their death is not publicly available. I treat a VC as dead if it no longer makes investments based on two reasons: the core business of VCs is to invest in private companies; and VCs are able to raise new funds for future investments if their previous funds perform well. I also check the online business news (e.g. the Company Insight Center from the BusinessWeek) to see if the discontinuation of the investments is caused by name changes or acquisitions. Although the definition of death used here is not perfect, it serves as a valid proxy for the real death. 15 Main independent variables The two dimensions of heterogeneity experience and performance can be either general or relevant for the portfolio company s industry or location. Due to the high correlations among general experience (performance), experience (performance) in the portfolio company s industry, and state, and the similar impact they have on syndication formation, I only report the heterogeneity in VCs experience (performance) in the portfolio company s industry. The following paragraphs describe how I construct the main independent variables in greater detail. The definition of other independent variables can be found in Table 1. I apply the coefficient of variation measure to continuous variables to capture the degree of heterogeneity (Beckman and Haunschild, 2002). For each VC in a syndicate, I calculate as its experience the number of companies it funded in the portfolio company s industry in the past five years. 16 For each syndicate, EXP HETER is equal to the standard deviation of VCs 15 The sample for analysis starts from 1995 to To check the death of a VC, I extend the sample to include VC investments made in 2006 and 2007, based on which I calculate the last year of investment for each VC. If a VC s year of investment is its last year of investment, I record DEATH as 1 and the VC firm drops out of the sample. 16 Alternatively, the experience can be captured by the number of companies a general partner, who sits on the company s board, has funded in the past five years. Unfortunately, information on individual partners in VentureXpert is highly incomplete. The variable here is a VC firm s experience in the portfolio company s industry. If a VC firm only allocates partners who are experienced in the portfolio companies industry as directors of the 22

31 Table 1 The Definition of Variables Variable Name REALIZED TOTAL SYNDICATES EXIT EXP MEAN EXP HETER PERF MEAN PERF HETER CTRY HETER TYPE HETER DEGREE MEAN DEGREE HETER EIGENVEC MEAN EIGENVEC HETER DIST VC Definition A dummy variable equal to 1 if the syndicate is realized. The number of realized and hypothetical syndicates each company has. A dummy variable equal to 1 if the portfolio company has an IPO or gets acquired by other companies. For a syndicate, it is the average of VCs relevant industry experience. The relevant industry experience is calculated as the number of companies a VC funded in the portfolio company s industry in the past five years. For a syndicate, it is the standard deviation of VCs relevant industry experience divided by their average relevant industry experience. For a syndicate, it is the average of VCs relevant industry performance. The relevant industry performance is calculated as the number of successful exits a VC had divided by the number of companies the VC funded in the portfolio company s industry in the past five years. For a syndicate, it is the standard deviation of VCs relevant industry performance divided by their average relevant industry performance. For a syndicate, it is the entropy measure of the diversification of VCs countries of origin. It is calculated as i proportion of VCs in country i in each syndicate. pi ln p where i p is the i For a syndicate, it is the entropy measure of the diversification of VCs types. It is calculated as of type i in each syndicate. i pi ln p where i pi is the proportion of VCs For a syndicate, it is the average of VCs normalized degree centrality scores in a 5-year window. For a syndicate, it is the standard deviation of VCs normalized degree centrality scores divided by their average normalized degree centrality scores in a 5-year window. For a syndicate, it is the average of VCs normalized eigenvector centrality scores in a 5-year window. For a syndicate, it is the standard deviation of VCs normalized eigenvector centrality scores divided by their average normalized eigenvector centrality scores in a 5-year window. The average distance between any two VCs in a syndicate. portfolio company, which is highly possible, the measures of a VC firm s industry-relevant experience should be highly correlated with the experience of partners sitting on the board. 23

32 Variable Name DIST HETER PRIOR RELATION RDAMT SYNDICATION SIZE DEATH NEW DEAL INDUDIV PARTEXP MEAN PARTEXP HETER PARTPERF MEAN PARTPERF HETER PARTCTRY HETER PARTTYPE HETER VCEXP Definition For a syndicate, it is the standard deviation of the distances between each VC and the portfolio company divided by the average VC-company distances. The number of pairs of VCs that syndicated in the past five years divided by the total number of pairs of VCs in a syndicate. The amount of money (in thousand dollars) a company received in the first financing round. The number of VCs in a syndicate. A dummy variable equal to 1 if a VC no longer makes investments in the future. The number of first round deals VCs make in the following year. For each VC, it is the entropy measure of its own industry diversification in the following year. It is calculated as i pi ln p where i pi proportion of its investments in each industry i in the following year. is the For a VC in a given year, it is the average experience of its syndication partners. The experience is calculated as the number of companies each partner funded in the past five years. For a VC in a given year, it is the standard deviation of its syndication partners experience divided by their average experience. For a VC in a given year, it is the average performance of its syndication partners. The performance is calculated as the number of successful exits each partner had divided by the number of companies it funded in the past five years. For a VC in a given year, it is the standard deviation of its syndication partners performance divided by their average performance. For a VC in a given year, it is the entropy measure of the diversification of the VC s syndication partners countries of origin. It is calculated as i pi ln p i where p i is the proportion of syndication partners in country i. For a VC in a given year, it is the entropy measure of the diversification of the VC s syndication partners types. It is calculated as i pi ln p i where p i is the proportion of syndication partners of type i. For a VC in a given year, it counts the number of companies the VC funded in the past five years. experience divided by their average experience. Similarly, to obtain the heterogeneity in VCs performance, I first calculate as a VC s performance the ratio of the number of successful exits a 24

33 VC has in the portfolio company s industry to the total number of companies the VC funded in the portfolio company s industry in the past five years (e.g. for a VC that funded 5 companies in biotechnology with 2 of them either listed in the stock market or acquired by other companies, its exit ratio in biotechnology is 0.4). For each syndicate, PERF HETER is equal to the standard deviation of VCs performance divided by their average performance. To capture the degree of heterogeneity of categorical variables, I follow Jacquemin and Berry (1979) and apply the entropy measure to VCs types 17 and countries of origin 18. Jacquemin and Berry (1979) show a numerical example to compare two measures of corporate diversification in different industries, entropy measure and the Herfindahl Index. They suggest that for a small change in industry diversification, there will be an increase in the entropy measure while the Herfindahl Index largely ignores the change i.e. the entropy measure is more sensitive to changes in diversification than the Herfindahl Index. As the entropy measure weights the proportion of each category, denoted by p i, by the logarithm of 1 / p, the entropy measure of VCs types and countries of origin therefore increases with the diversification in VCs types and countries of origin within a syndicate. I calculate CTRY HETER as i i pi ln p i, in which pi is the proportion of VCs in country i in a syndicate. I calculate TYPE HETER as i pi ln p i where pi is the proportion of VCs of type i within a syndicate. 17 Based on the information from VentureXpert and VCs websites, I classify VCs into seven types: (1) Private equity firms; (2) Corporations; (3) Banks, insurance companies, asset management companies, and real estate companies; (4) Consulting firms, marketing firms, law firms, and accounting firms; (5) Angel networks and; (6) Governments, university endowments, nonprofit organizations, and incubators. 18 Based on the information from VCs websites, I record the 48 countries (or regions) of VCs. They are: Argentina, Australia, Bahrain, Belgium, Brazil, Canada, Cayman Islands, China, Chile, Czech Republic, Denmark, Finland, France, Germany, Greece, Hong Kong (China), Iceland, India, Indonesia, Ireland, Israel, Italy, Japan, Kuwait, Luxembourg, Malaysia, Mauritius, Mexico, Monaco, Netherlands, New Zealand, Norway, Philippines, Poland, Portugal, Puerto Rico, Russia, Singapore, Slovenia, South Africa, South Korea, Spain, Swaziland, Sweden, Switzerland, Taiwan, United Kingdom, and United States. 25

34 2.4 Empirical Analysis Syndication formation Table 2 reports summary statistics of the sample for syndication formation. The unit of analysis is a syndicate. There are 156, 407 syndicates with 2% of them being realized. The average number of syndicates for each company is 64. On average, each VC in a syndicate funded 17 companies in the portfolio company s industry in the past five years. The average heterogeneity in VCs relevant industry experience in a syndicate is On average, 11% of a VC s investments had either IPOs or sales to other companies in the past five years. The heterogeneity in VCs performance is very close to that in VCs experience in a syndicate, due to the normalization of the standard deviation by the means. The correlation coefficient between heterogeneity in VCs performance and heterogeneity in VCs experience is about 0.39 as shown in Table 3, suggesting that they are not completely mutually exclusive. The entropy measures show that there is higher diversification in VCs types than that in VCs countries of origin in a syndicate. The positive correlation between the heterogeneity in VCs types and VCs countries of origin shows an interesting phenomenon: international VCs that invest in U.S. private companies tend to be multinational corporations and financial institutions. 26

35 Table 2 Descriptive Statistics of Samples This sample has 156,407 realized and hypothetical VC syndicates. The syndicated investments are made to 2,451 U.S. companies that received the first round financing from 1,669 VCs between 1995 and Variable No. of Obs. Mean Std. Dev. Min Max REALIZED TOTAL SYNDICATES EXP MEAN EXP SD EXP HETER PERF MEAN PERF SD PERF HETER CTRY HETER TYPE HETER DEGREE HETER EIGENVEC HETER DIST VC DIST HETER PRIOR RELATION Table 3 Correlation Matrix of Key Variables The matrix is abased on the sample of 156,407 realized and hypothetical VC syndicates. All of the correlation coefficients are significant at the 1% level. Index Variable Name REALIZED 1 2 EXP HETER PERF HETER CTRY HETER TYPE HETER DEGREE HETER EIGENVEC HETER DIST HETER DIST VC PRIOR RELATION

36 I use the Conditional Logit model as the main econometric model, which groups the realized and hypothetical syndicates for each portfolio company. This is similar to performing portfolio company fixed effects in logistic regressions when the data has a panel structure. As the Conditional Logit model calculates the likelihood of syndication formation for each company, the interpretation of the results should be: within a company, how heterogeneity in VCs affects the formation of syndication for the company. Formally, the Conditional Logit model is shown as follows: X denotes key explanatory variables and Z denotes control variables; portfolio company is indexed by i and each syndicate is indexed by j: exp( i X ij Zij ) Pr( REALIZED 1 X ij, Zij ) 1 exp( X Z ) i ij ij Table 4 reports the baseline results of syndication formation. Column (1) presents the cross sectional evidence based on a regular logistic regression on four main independent variables only. Heterogeneity in VCs experience, performance, types, and countries of origin has a negative and significant impact on syndication formation. Column (2) presents regression results after controlling for the fixed effects of the size of the syndication, the industry, state, and year of the first round financing of the portfolio company. The predictions in Column (2) are consistent with those in Column (1). Column (3) applies the Conditional Logit model and obtains the within company evidence: for a company, the heterogeneity in VCs experience, performance, types, and countries of origin decreases with the likelihood of syndication formation. Due to the lack of variation in the syndication size, industry, state, and year of the first round financing within a portfolio company, I cannot control for the fixed effects of these variables. I use Column (3) as the main model specification to study syndication formation The reason for not including in the regression the average of the experience (performance) of VCs in a syndicate is that the heterogeneity measure is already a de-mean measure. In unreported regressions, I control for the averages and obtain the same predictions from the heterogeneity variables. 28

37 Table 4 The Base Model The regressions in this table are based on the sample of 156,407 realized and hypothetical VC syndicates. The unit of analysis is a syndicate. The first two columns report regression results from the Logit model. The last column reports regression results based on the Conditional Logit model, in which observations are grouped at the portfolio company level. Robust and clustered standard errors at the portfolio company level are reported in parentheses. I use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively. (1) (2) (3) VARIABLES REALIZED REALIZED REALIZED Cross sectional Cross sectional Panel FE EXP HETER *** *** *** (0.0463) (0.0474) (0.0484) PERF HETER *** *** *** (0.0350) (0.0356) (0.0363) CTRY HETER *** *** *** (0.0769) (0.0806) (0.0887) TYPE HETER *** * ** (0.0602) (0.0656) (0.0592) LN RDAMT (0.0245) SYNDICATION SIZE FE NO YES NA YEAR FE NO YES NA INDUSTRY FE NO YES NA STATE FE NO YES NA COMPANY FE NO NO YES CONSTANT *** *** (0.0470) (0.221) Log likelihood No. of Obs To check whether or not the empirical predications are mainly driven by an artifact of a particular matching method, I perform the following robustness checks on the Single Deviation Method : First, to select potential investors, I relax the inclusion criteria by focusing on any two of the three criteria at a time. 29

38 Second, I apply the Multiple Deviation Method to syndicates formed by three VCs, in which I replace any two VCs in a realized syndicate with any other two potential investors. Third, some companies have more hypothetical syndicates than others due to the popularity of their industries, states, or the time of financing. To check whether different numbers of hypothetical syndicates for each company affect the results, I form two sub-samples one of syndicates with two investors, and the other with three investors. I then replace each VC in the realized syndicate with one randomly selected VC from the pool of potential investors, leading to the same number of hypothetical syndicates for each company in the two respective sub-samples. I apply the Rare Event Logit model (King and Zeng, 1999a, 1999b), which corrects for the oversampling of the true events (i.e. 33% or 25% of the syndicates are realized in the two respective sub-samples), and the Conditional Logit model to replicate the baseline analysis shown in Column (3) of Table 4. Fourth, the Single Deviation Method treats each VC in a syndicate equally and replaces every VC by a potential investor. This is consistent with the focus of this paper, which studies how VCs select each other to form a syndicate. However, if there exists a lead VC who initiates the deal and selects its co-investors, the lead VC should not be replaced in the hypothetical syndicates. VentureXpert does not report who are the lead investors and prior research provides a few ways of identifying the lead VCs. 20 I first identify the VC with the biggest equity share in a 20 Bottazzi et al. (2007) used the survey instrument to directly ask whether or not the VC is the syndicate lead. Others have used different criteria to identify lead investors if they do not have a direct measure. For example, Sorensen (2007) assigns VCs with the largest total investment in the company as the lead VCs. Similarly, Hochberg et al. (2007) define a lead VC as the one making the largest investments in each financing round and change to VCs cumulative investment in the company if there are ties. Gompers (1996) classifies lead VCs as those sitting on the board for the longest time and turns to the biggest equity holder if there are ties. Brander et al. (2002) equals the first investor in the company as the lead investor. Sorenson and Stuart (2001) take the first investor in a company as the lead investor. If there are multiple investors in the first round, they will treat the VC that invests in every subsequent round as the lead VC. 30

39 financing round as the lead VC. If the information on VCs amount of investment is missing or there are ties of shares, I then identify the VC that sits on the board of the portfolio company, and prefers to be the deal originator as the lead VC. Altogether, I am able to identify lead VCs for about 60% of companies in the whole sample. When constructing hypothetical syndicates for these companies, I keep the lead VC in every syndicate. The results I obtain from the above samples are consistent with those from the base model. The regression tables are not reported due to the limited space and are available upon request Other control variables The management literature suggests that firms prefer alliance partners with similar social status (Brass et al, 2004; Chung, Singh, and Lee, 2000; Podolny, 1994). A proxy used in the literature for social status is a firm s network centrality, which captures how connected a firm is in a market. There are different measures of network centralities and I focus on two measures here: the degree centrality (Freeman, 1979) which directly measures the number of connections each firm has, and the eigenvector centrality (Bonacich, 1972) which may have the biggest economic impact on VCs performance (Hochberg et al., 2007). To measure a VC s network centrality, I first define the connection as the syndication among VCs. Using the social network analysis software Ucinet, I obtain a VC s degree centrality and eigenvector centrality in a five-year window. The degree centrality counts the number of unique VCs a VC has syndicated with. To ensure the comparability of VCs centralities across years, the degree centrality is further normalized by the maximum possible degree centrality a VC can have in the network. The eigenvector centrality (Bonacich, 1972) not only counts the number of syndication partners a VC has, but puts more weights on the syndication with those partners who are well connected in the network. It is also normalized by the maximum possible eigenvector centrality a VC can have in the network. Conceptually, the degree centrality captures the 31

40 quantity of connections a VC has, while the eigenvector centrality captures the quality of connections a VC has. For each syndicate, the heterogeneity in VCs degree centrality defined as DEGREE HETER is the standard deviation of the normalized degree centrality scores of VCs divided by the average normalized degree centrality scores in a 5-year window. Similarly, the heterogeneity in VCs eigenvector centrality defined as EIGENVEC HETER is calculated as the standard deviation of the normalized eigenvector centrality scores of VCs divided by the average normalized eigenvector centrality scores in a 5-year window. Columns (1) and (2) in Table 5 report the empirical results of degree centrality and eigenvector centrality, respectively. Heterogeneity in VCs network centrality, which is an indicator of their social status, is negatively associated with the syndication formation. After controlling for the heterogeneity in VCs network centralities, the heterogeneity in VCs experience and performance still have similar predictions to those in the base model, although their coefficients become smaller due to their high correlations with VCs network centralities. Column (3) in Table 5 further explores how geographic distances affect the syndication formation. I consider two types of distances here: the distances among VCs in a syndicate, and the distances between each VC and the portfolio company. I match the zip codes with their corresponding longitude and latitude scores, based on which I calculate the distances. I only have the longitude and latitude scores of U.S. zip codes, so the analysis in Column (3) is based on the syndicates formed by U.S. VCs and international VCs with U.S. offices. To calculate the distances among VCs denoted by DIST VC, I first calculate the distances between any two VCs in a syndicate and then take the average of them. For VCs with multiple zip codes, I use the shortest distances between them. For the second type of VC-company distances, I first calculate the distances between each VC and the portfolio company and use the shortest VC-company 32

41 distances if a VC has multiple zip codes. The heterogeneity in VC-company distances DIST HETER is the standard deviation of the VC-company distances in a syndicate divided by the average VC-company distances in the syndicate. Column (3) reports the regression results that syndication is more likely to be formed among VCs located close to each other. Meanwhile, similar VC-company distances can facilitate the formation of syndication among these VCs. Table 5 Network Centralities and Geographic Distances The regressions in this table are based on the sample of 156,407 realized and hypothetical VC syndicates. The unit of analysis is a syndicate. All of the columns report regression results based on the Conditional Logit model, in which observations are grouped at the portfolio company level. Robust and clustered standard errors at the portfolio company level are reported in parentheses. I use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively. (1) (2) (3) VARIABLES REALIZED REALIZED REALIZED EXP HETER ** * *** (0.0586) (0.0578) (0.0528) PERF HETER *** *** *** (0.0362) (0.0362) (0.0398) DEGREE HETER *** (0.0598) EIGENVEC HETER *** (0.0583) DIST VC *** (2.48e-05) DIST HETER *** (0.0510) CTRY HETER *** *** *** (0.0886) (0.0887) (0.103) TYPE HETER * * (0.0592) (0.0592) (0.0664) Log likelihood No. of Obs

42 Table 6 Prior Relations Columns (1), (2), and (3) report regression results based on the sample of 156,407 realized and hypothetical VC syndicates. Column (4) is based on the sample of syndicates in which VCs did not syndicate with each other before (PRIOR RELATION equal to 0), while Column (5) is based on the sample of syndicates in which VCs syndicated before (PRIOR RELATION greater than 0). All of the regressions are based on the Conditional Logit model, in which observations are grouped at the portfolio company level. Robust and clustered standard errors at the portfolio company level are reported in parentheses. I use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively. (1) (2) (3) (4) (5) VARIABLES REALIZED REALIZED REALIZED REALIZED REALIZED Full Full Full PRIOR RELATION=0 PRIOR RELATION>0 EXP HETER *** ** 0.184* (0.0526) (0.0507) (0.0612) (0.104) PERF HETER *** *** *** (0.0373) (0.0357) (0.0443) (0.0768) PRIOR 1.150*** 1.166*** 1.171*** 1.674*** RELATION (0.0629) (0.0627) (0.0612) (0.197) CTRY HETER *** *** *** *** (0.0910) (0.0910) (0.0909) (0.130) (0.136) TYPE HETER ** ** ** *** (0.0605) (0.0604) (0.0603) (0.0840) (0.0982) Log likelihood No. of Obs Another important explanation for tie formation is the prior ties among firms (Gulati and Gargiulo, 1999; Podolny, 1994). Although prior ties have strong explanatory power for future tie formation, it cannot explain: how is the tie formed at the first time when firms do not have any prior ties? To explore this question, I construct PRIOR RELATION for each syndicate, which is the number of pairs of VCs that syndicated before divided by the maximum number of pairs of VCs in the syndicate. Base on this variable, I form two sub-samples of syndicates one sample of new relations including syndicates formed by VCs who did not syndicate with each other before (PRIOR RELATION equal to 0), and the other sample of repeated relations including 34

43 syndicates formed by VCs who syndicated before (PRIOR RELATION greater than 0). I then replicate the baseline analysis in the two sub-samples, respectively. Prior syndication among VCs has a positive and significant impact on their future syndication formation as shown in Column (1) of Table 6. After controlling for the prior syndication, the coefficient of heterogeneity in VCs experience becomes insignificant due to its correlation with heterogeneity in performance. Therefore, when heterogeneity in experience and performance are introduced separately to regressions in Column (2) and Column (3), both of them remain statistically significant. Columns (4) and (5) report the regression results based on the sub-samples of new relations and repeated relations, respectively. VCs are less likely to form syndication with strangers who have different levels of experience, performance, and come from different countries. Given that VCs syndicated with each other before, heterogeneity in experience can no longer prevent different VCs from forming syndication together. The different impacts of heterogeneity in VCs experience in these two sub-samples suggest that: First, the criteria for choosing alliance partners from those with which a VC co-invested before can be quite different from the criteria used for choosing alliance partners from the strangers. Second, VCs can build trust among each other through prior co-investments, which help reduce transaction costs incurred in heterogeneous syndicates, making it possible for VCs with different levels of experience to form future syndication. Third, the positive assortative matching is most salient among VCs who form syndication for the first time, compared with a selected sample in which VCs choose syndication partners from those they already syndicated with before The performance of syndication To study the performance of syndication, I focus on the realized syndicates only. The unit of analysis is a portfolio company (or a realized syndicate). Table 7 describes the sample of 2,451 companies that received the first round financing from the VC syndicates between 1995 and % of them provided VCs with successful exits either through IPOs or sales to other 35

44 companies. On average, each VC in a realized syndicate funded 15 companies in the portfolio company s industry in five years before the syndicate was formed. The average heterogeneity in the experience of VCs in a realized syndicate is 0.9, which is 14% lower than that in the whole sample of realized and hypothetical syndicates. The average exit rate for each VC in a realized syndicate is 11%, which is the same as that in the whole sample. The heterogeneity in VCs performance and other VCs characteristics is also smaller than that in the whole sample. Companies received on average $7.89 million in the first rounds from three VCs. In 1995, 128 companies received their first round financing from VC syndicates. The number increased steadily and peaked at 587 companies in After reaching its lowest point in 2002, the number of companies funded by VC syndicates began to increase again. Computer related industry attracted more than half of the total VC investments, followed by Communications and Media, and Medical/Health/Life Science. More than half of the VC investments were received by companies located in California, followed by Massachusetts, New York and Texas. Table 8 reports the correlation matrix of the key variables. Table 7 Sample for Portfolio Companies Performance This sample has 2,451 U.S. companies that received the first round financing from 1,669 VCs between 1995 and This table also reports the number of companies funded in each industry and the four largest states with the most VC investments. Variable No. of Obs. Mean Std. Dev. Min Max EXIT EXP MEAN EXP SD EXP HETER PERF MEAN PERF SD PERF HETER CTRY HETER TYPE HETER DEGREE HETER EIGENVEC HETER

45 Variable No. of Obs. Mean Std. Dev. Min Max DIST VC DIST HETER PRIOR RELATION RDAMT SYNDICATION SIZE Categorical Variable Frequency Percent Industry of companies: Non-High-Technology Biotechnology Communications and Media Computer Related Medical/Health/Life Science Semiconductors/Other Electronics State of companies: California Massachusetts New York Texas To study the performance of syndication denoted by EXIT, which equals 1 if the portfolio company financed by the syndication has an IPO or sale to another company, I first perform the baseline analysis of EXIT and then address the selection issues. The empirical results in Table 9 are based on the regular logistic regressions. Since the unit of analysis is a company, I cannot control for company fixed effects any more. I therefore control for dummy variables of companies states, industries, sizes of syndication, and years of the first financing rounds. Column (1) shows that companies funded by VCs with different levels of performance are less likely to have IPOs or sales to other companies. I then explore whether the control variables that affect the syndication formation have any implications for the performance of syndication. Interestingly, I find that heterogeneity in VCs network centrality (in Columns (2) and (3)), heterogeneity in VC-company distances (in Column (4)), average distance among VCs (in Column (4)), and prior relations among syndication partners (in Column (5)) do not have a statistically significant impact on the performance of syndication, once the syndication is 37

46 formed. 21 Another consistent finding in these regressions is the positive impact of the amount of financing a company received in the first round on the likelihood of having an IPO or sale in the future. One possible explanation is that companies with sufficient funds to start with may develop faster and are more likely to survive unexpected financial difficulties. Another explanation can be the amount of financing a company can obtain simply proxies for its quality which leads to better future performance. Table 8 Sample for the Performance of Syndication The matrix is abased on the sample of 2,451 U.S. companies. I use * to denote significance at the 10% level. Index * 0.36* * 0.06* * 0.11* 0.16* * 0.31* 0.13* 0.17* * 0.30* 0.13* 0.17* 0.96* * 0.09* 0.05* 0.19* 0.13* 0.13* * 0.04* 0.05* 0.15* 0.09* 0.10* 0.43* * -0.09* -0.11* -0.15* -0.39* -0.43* -0.13* -0.18* 1 Index Variable Name Index Variable Name 1 EXIT 6 DEGREE HETER 2 EXP HETER 7 EIGENVEC HETER 3 PERF HETER 8 DIST HETER 4 CTRY HETER 9 DIST VC 5 TYPE HETER 10 PRIOR RELATION 21 I also run regressions in which I only include these control variables and obtain statistically insignificant results. 38

47 Table 9 The Performance of Syndication This sample has 2,451 U.S. companies that received the first round financing between 1995 and The unit of analysis is a portfolio company (or a realized syndicate). All regressions are based on the Logit model, which controls for fixed effects of portfolio company s state, industry, syndicate size, and year of the first round financing. I use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively. (1) (2) (3) (4) (5) VARIABLES EXIT EXIT EXIT EXIT EXIT EXP HETER (0.116) (0.146) (0.143) (0.129) (0.119) PERF HETER ** ** ** *** ** (0.0896) (0.0907) (0.0910) (0.0991) (0.0896) EXP MEAN (0.004) (0.0051) (0.0054) (0.0043) (0.0043) PERF MEAN (0.495) (0.513) (0.520) (0.540) (0.502) DEGREE HETER (0.158) DEGREE MEAN (0.0204) EIGENVEC HETER (0.155) EIGENVEC MEAN (0.0202) DIST VC -1.33e-05 (5.90e-05) DIST HETER (0.114) PRIOR RELATION (0.155) CTRY HETER (0.196) (0.196) (0.196) (0.287) (0.196) TYPE HETER (0.161) (0.161) (0.161) (0.181) (0.162) LOG RDAMT 0.160** 0.158** 0.157** 0.161** 0.160** (0.0626) (0.0627) (0.0629) (0.0679) (0.0626) SYNDICATION SIZE YES YES YES YES YES FE YEAR FE YES YES YES YES YES INDUSTRY FE YES YES YES YES YES 39

48 (1) (2) (3) (4) (5) VARIABLES EXIT EXIT EXIT EXIT EXIT STATE FE YES YES YES YES YES CONSTANT *** *** *** *** *** (0.564) (0.570) (0.571) (0.615) (0.566) Log likelihood No. of Obs Two types of selection may bias the results of performance. The first type is the selection among VCs who tend to syndicate with partners who are similar. To study how heterogeneity among VCs affects the performance of their syndication, we need to base the analysis on the realized syndicates that are in general less heterogeneous than those not realized. The second type is the more well-known selection between VCs and start-ups (Sorensen, 2007), in which top tier VCs invest in top tier start-ups. To address the selection among VCs, I apply the Heckman two-step procedure: In the selection equation, I study how the heterogeneity among VCs affects the syndication formation; In the EXIT equation, given that the syndication is formed, I study how the heterogeneity among VCs affects the performance of syndication. The model identification comes from two measures of geographic distances: the average distances among VCs in a syndicate measured by DIST VC, and the heterogeneity in VC-company distances in a syndicate measured by DIST HETER. Geographic proximity among investors can increase the likelihood of forming syndication among them. After the syndication is formed, simply locating close to each other should not directly affect the success of their investments. Although companies located close to their investors can have better performance (Tian, 2008), the heterogeneity in the VC-company distances should not have a direct impact on the company s performance after the syndication is formed. I apply the STATA s Heckprob model which implements the Heckman two-step procedure when the dependent variable of the EXIT equation is a dummy variable. Formally, the Heckprob model combines the syndication formation model in its first stage and the following Probit model in its second stage: 40

49 Pr( EXIT 1 X ij, Z ij ) 1 ( X Z ij ) ij The likelihood function of the full model is shown as follows: Let 1 H denote all of the independent variables in the selection equation and 2 H denote all of the independent variables in the EXIT equation. DIST VC and DIST HETER are included in the function identifiable. L i S i i S ; yi 0 i i S ; yi 0 ln{ ln{ ( H offset 1 ln{1 ( H offset i 2 2 ( H offset i 2 i 2 i i i i )} 1, H offset 1, H offset i i 1 H i i, )}, )} but not in 2 H to make Column (1) in Table 10 reports the results of the selection equation. Both DIST VC and DIST HETER are negatively associated with the syndication formation. Column (2) shows that after controlling for the selection among VCs, heterogeneity in VCs performance remains negative and statistically significant. Rho, the correlation coefficient of the error terms from the two equations, is not significant, suggesting that the results of the performance of syndication are unlikely to be driven by the selection among VCs. To address the selection between VCs and companies, I apply the instrumental variable approach proposed by Ackerberg and Botticini (2002). The theoretical motivation is that VC investments are usually concentrated in certain industries and location. As a result, characteristics of such industries and location can capture the local availability of venture capital, entrepreneurial talents, and other unobservable local characteristics. In fact, the local availability of certain firm characteristics has served as a popular instrument in finance literature (see Berger et al., 2005 and Bottazzi et al., 2007). 22 Such local characteristics have a direct impact on the availability of syndication partners and portfolio companies, and therefore the matching between a portfolio company and a syndicate of VCs. After the syndicates are formed and investments made, such 22 For example, Berger et al. (2005) use the median size of the bank in the local market to instrument for the size of a bank. Bottazzi et al. (2007) instrument a partner s business experience by the number deals made by a VC firm relative to the total number of deals made in the company s country 41

50 local characteristics should not directly affect the success of a particular company. To implement Instrument Variable (IV) estimation, I construct portfolio companies local markets as an interaction between companies industries and states, resulting in 75 local markets for the entire sample. I then test which local markets have predictive power to explain the endogenous variable (i.e. heterogeneity in VCs past performance). In a regression where the endogenous variable is regressed upon 74 dummy variables of local markets as well as all relevant exogenous variables, 14 out of 74 local markets obtain significant coefficients. Based on a LM test, dummy variables with non-significant coefficients are redundant and they do not improve the asymptotic efficiency of the estimation. Furthermore, including all 74 dummy variables as instruments for one endogenous variable can not satisfy the overidentifying restrictions. Therefore, in the first step of IV estimation, I regress the endogenous variable on 14 dummy variables while control for all relevant exogenous variables. To check whether the instruments are valid, I perform three types of tests. First, an F test of joint significance of 14 instruments generates an F statistic of F (14, 2257) = 65.23, suggesting that these 14 variables are jointly significant. Second, including the 14 instruments improve the R-square of the first step IV estimation from to (i.e. 12%). A Kleibergen-Paap test for weak identification rejects the null hypothesis that these instruments are only weakly correlated with the endogenous regressor. Third, to test whether the instruments are correlated with the error terms, I perform the overidentifying restrictions test. The Hansen J statistic (J statistic= and Chi-sq (13) P-value= ) suggests that we cannot reject the null hypothesis that the instruments are exogenous to the dependent variable in the main regression (i.e. exit). In the second step of the IV estimation, I regress the performance of a portfolio company on the predicated value of heterogeneity in VCs performance and all other relevant variables, shown in Column (3) of Table 10. Consistent with the original findings, the IV estimation reports a 42

51 negative impact of heterogeneity on the performance of syndication. 23 Table 10 Selection Issues of the Performance of Syndication This sample has 2,451 U.S. companies that received the first round financing between 1995 and The unit of analysis is a portfolio company (or a realized syndicate). Columns (1) and (2) report regression results from the Heckporb model which implements the Heckman two-step procedure when dependent variables are dummy variables. Column (1) reports the regression results of the selection equation with REALIZED as its dependent variable. Column (2) reports the regression results of the EXIT equation with EXIT as its dependent variable. Column (3) reports the EXIT equation based on the Ackerberg and Botticini Approach. The selection equation regresses PERF HETER on 14 dummy variables, which are constructed as the interaction of the portfolio companies states and industries, and all of the control variables used in Column (3). I use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively. (1) (2) (3) VARIABLES REALIZED EXIT EXIT HECKPROB: SELECTION Eq. HECKPROB: EXIT Eq. A & B IV: EXIT Eq. EXP HETER *** (0.0224) (0.0827) PERF HETER *** *** ** (0.0167) (0.0569) (0.1143) EXP MEAN (0.0006) (0.0023) (0.0005) PERF MEAN (0.0946) (0.299) (0.1166) CTRY HETER *** (0.0464) (0.167) (0.0338) TYPE HETER (0.0301) (0.104) (0.0273) LN RDAMT * ** ** (0.0113) (0.0407) (0.0103) DIST VC -3.98e-05*** (9.68e-06) DIST HETER *** (0.0195) SYNDICATION SIZE FE YES YES YES YEAR FE YES YES YES 23 If EXP HETER is included in the regression, it obtains a positive sign due to its correlation with PERF HETER. Whether or not EXP HETER is included, PERF HETER has a negative and statistically significant coefficient. 43

52 (1) (2) (3) VARIABLES REALIZED EXIT EXIT HECKPROB: SELECTION Eq. HECKPROB: EXIT Eq. A & B IV: EXIT Eq. INDUSTRY FE YES YES NO STATE FE YES YES NO CONSTANT *** ** (0.102) (0.986) (0.0823) Rho (0.4133) Log likelihood R-squared No. of Obs Table 11 Sample for VCs Survival This sample has 3,380 VCs that co-invested with at least two other VCs in any year between 1995 and There are 12,434 VC-year observations. This table also reports the top three countries with the most VC investments in the U.S., and the three most popular types of VCs. Variable No. of Obs. Mean Std. Dev. Min Max VCs Characteristics: DEATH NEW DEAL INDUDIV VCEXP Syndication Partners Characteristics: PARTEXP MEAN PARTEXP HETER PARTPERF MEAN PARTPERF HETER LOCAL HETER LOCAL EXP PARTCTRY HETER PARTTYPE HETER Categorical Variable Frequency Percent Year of syndication:

53 Variable No. of Obs. Mean Std. Dev. Min Max VCs country of origin: United States United Kingdom Canada VCs type: Private Equity Firm Corporations Financial Institutions Table 12 Correlation Matrix for VCs Survival The matrix is based on the sample of 12,434 VC-year observations. I use * to denote significance at the 10% level. Index Variable Name DEATH 1 2 NEW DEAL -0.18* 1 3 INDUDIV -0.38* 0.45* 1 4 VCEXP -0.16* 0.43* 0.43* 1 5 PARTEXP HETER -0.06* 0.05* 0.11* 0.07* 1 6 PARTPERF HETER -0.02* 0.04* * 0.39* 1 7 PARTCTRY HETER -0.07* 0.04* 0.11* 0.17* 0.24* 0.11* 1 8 PARTTYPE HETER -0.07* 0.09* 0.13* 0.15* 0.34* 0.15* 0.30* The survival of VCs The analysis of VCs survival is based on the panel data of 12,434 VC-year observations between 1995 and Table 11 describes the sample. About half of the VCs in the sample went out of business. On average, VCs invest in one new deal in the following year and 15 companies on average in the past five years. In each year, each syndication partner funded an average 43 companies in the past five years. The big difference between the average of VCs 45

54 own experience and the average of partners experience is mainly due to the oversampling of experienced VCs as syndication partners, given the fact that more experienced VCs syndicate more often. There are more observations during the internet bubble years where many new VC firms were established, thus creating more syndication opportunities. About 77% of the VCs are from the U.S., followed by U.K. and Canada. More than half of the VCs are independent VCs and another 30% are captive VCs affiliated with corporations or financial institutions. Table 12 reports the correlation matrix of the key variables. It is interesting to examine how heterogeneous syndication partners affect VC funds performance. However, VC funds are usually organized as 10 year close-end funds, and the sample period is not long enough to study their performance. Therefore, I only focus on the survival of VCs. Taking advantage of the panel structure, I apply the fixed effects OLS model 24 where I control for VC fixed effects, and year fixed effects to reduce the autocorrelations of error terms over time. The interpretation of the coefficients after controlling for VC fixed effects is: within a VC firm, how changes in its partners heterogeneity affect the VC s survival. The econometric model is illustrated as follows: Y it X v it i it Yit is DEATH for VC i at year t. X it includes the measures of heterogeneity among VC i s syndication partners, as well as other time-varying characteristics of VC i. vi is VC i s individual effect. Its correlation with model should be used. it determines whether fixed effects or random effects Before studying VCs survival, Columns (1) and (2) in Table 13 suggest two possible 24 The fixed effects logistic model does not achieve convergence. 46

55 mechanisms through which heterogeneous partners may help VCs to survive. First, learning from partners with diverse experience, VCs may be able to make more investments. I construct NEW DEAL as the number of the first round deals (i.e. new investments instead of following on rounds) VCs make in the following year. The regression results in Column (1) suggest that the heterogeneity in partners experience (performance) 25 is positively associated with the number of the first round deals in the following year. Second, learning from partners with diverse experience may help VCs to diversify their investments in their preferred industries. I construct the entropy measure INDUDIV to capture the extent to which VCs diversify their investments into certain industries. Technically, the entropy measure of diversification does not simply count the number of industries VCs invest in, but focuses more on if there is an even distribution of VCs investments in their preferred industries. Practically, it is difficult for VCs to invest in industries where they do not have prior business experience. However, keeping a balanced investment portfolio in their preferred industries may make VCs less vulnerable to the shocks to a particular industry. Column (2) shows that heterogeneity in syndication partners experience is positively associated with VCs industry diversification. The results in Columns (1) and (2) are based on the cross sectional evidence only. The coefficients of heterogeneity no longer reach statistical significance when the within firm effects are examined. Column (3) reports the positive impact of heterogeneity in partners experience on a VC s survival based on the cross sectional evidence, while Column (4) shows the within firm evidence that for a VC firm, increase in heterogeneity in its syndication partners experience has a positive impact on the VC s survival. In addition to the heterogeneity in partners experience, VCs own experience also helps their future survival. I perform several robustness checks to mitigate the concerns with the measurement of VCs death. First, I focus on a sample of VC-year observations from 1995 to As the VC activities are available until the end of 2007, the death is defined as no future investments in at 25 The insignificant coefficient of PARTEXP HETER is due to its correlation with PARTPERF HETER. When included separately in regressions, both of them are positive and significant. 47

56 least five years. The positive impact of partners heterogeneity on VCs survival remains the same. Second, captive VCs, for which VC investments are not the core business, may only make VC deals occasionally. The discontinuation of their VC investments may be caused by factors unrelated to their VC activities. Therefore, I replicate the analysis in Column (4) in a sample of independent U.S. VCs in Column (6) and obtain similar results. Third, the number of syndication partners may be inflated during the internet bubble years. VCs learning from partners that they syndicated with during the bubble period may be different from those they co-invested with in other years. To check whether the results are affected by the syndication in the bubble years, I replicate the analysis on a sample that excludes VC-year observations in 1999 and I obtain similar results (in unreported regressions). 26 Some selection issues may bias the above results. First, there is positive assortative matching among VCs, as shown in the first part of this paper. Implementing the Heckman two-step procedure is inappropriate due to the different data structure used for the syndication formation and for the VCs survival. I therefore control for VCs own experience and apply VC fixed effects to mitigate such concern. Second, anticipating heterogeneous partners may help survival in the future, VCs may self-select into more heterogeneous syndicates. If such selection is based on VCs experience over time or time-invariant characteristics, controlling for VCs experience and VC fixed effects should reduce such concern. If the selection is based on VCs time-varying 26 As the dataset used in this section has VC-year as a unit of analysis, VC s experience in portfolio companies industry is difficult to identify. I then tried four measures to capture the industry-specific heterogeneity: 1) I construct a dataset with VC-year-deal as the unit of analysis, based on which I first calculate the heterogeneity of syndication partners experience in the deal s industry and then take the average of heterogeneity across deals for a VC in a given year; 2) I assign each VC a core industry, in which the VC make the most investments in the sample period, and then calculate the heterogeneity in partners experience in the VC s core industry; 3) I calculate the distribution of each partner s investments in each industry and then measure the extent to which these partners share similar industry preferences; and 4) I construct the heterogeneity in partners experience in each of the six industries. I use heterogeneity in partners experience in each industry and the sum of the heterogeneity across industries. These measures have quite similar predictions to the baseline analysis in Column (4). 48

57 characteristics, I need some extra exogenous variation to identify the impact of heterogeneity. 27 To construct instrumental variables that capture the local characteristics of VCs, I first define local available VCs as all of the live VCs located in the VC s state whenever the location of the VC is known but exclude the VC itself and its syndication partners in that year. The first instrument, denoted by LOCAL HETER, measures the heterogeneity of local available VCs for each VC in a given year. The second instrument, denoted by LOCAL EXP, measures the average experience of local available VCs. The theoretical motivation of the instruments is that the local availability of VCs can affect VCs choices of syndication partners. However, the local VCs that the VC does not syndicate with should not directly affect the survival of this VC. I have tried the following tests to check the validity of the two instruments. First, both instruments should have significant impacts on heterogeneity in syndication partners experience. Column (1) of Table 14 reports the regression results, showing a negative and significant correlation between the heterogeneity in other local VCs and the endogenous variable. An F test of the joint significance of the instruments generates an F statistic of F (2, 7279) = Second, to check whether they are strong instruments, I perform a Kleibergen-Paap test, which rejects the null hypothesis that the instruments are only weakly correlated with the endogenous regressor. Third, to test whether the instruments are correlated with the error terms, I perform the overidentifying restrictions test. The Hansen J statistic (J statistic= and Chi-sq (1) P-value= ) suggests that we cannot reject the null hypothesis that the instruments are exogenous to the dependent variable in the main regression (i.e. survival of VCs). In the second step of the IV estimation, I regress the VCs survival on the predicated value of heterogeneity in syndication partners experience and all other relevant variables, shown in Column (2) of Table 14. Consistent with the original findings, the IV estimation reports a positive impact of partner 27 This type of selection can also be interpreted as active learning that VCs actively seek partners with diverse experience for more learning opportunities, as opposed to the random learning that VCs learn from the available partners who happen to have diverse experience. 49

58 heterogeneity on VCs survival. 28 Table 13 VCs Survival This table is based on the sample of 12,434 VC-year observations. I use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively. (1) (2) (3) (4) (5) (6) VARIABLES NEW INDUDIV DEATH DEATH DEATH DEATH DEAL FULL FULL FULL FULL YEAR<2003 U.S. & IVC PARTEXP *** *** *** *** *** HETER (0.0632) (0.0106) (0.0074) (0.0086) (0.0099) (0.0098) PARTPERF 0.219*** HETER (0.0583) (0.0097) (0.0069) (0.0077) (0.0084) (0.009) PARTEXP e e e e-05 MEAN (0.0009) (0.0002) (0.0001) (0.0001) (0.0002) (0.0001) PARTPERF *** * 0.109** 0.211*** MEAN (0.382) (0.0638) (0.0456) (0.0508) (0.0549) (0.0608) VCEXP *** *** *** *** *** *** (0.0008) (0.0002) (0.0002) (0.0002) (0.0002) (0.0003) PARTCTRY * ** HETER (0.0095) (0.0110) (0.0115) PARTTYPE *** *** ** HETER (0.0097) (0.0110) (0.0110) VC FE NO NO NO YES YES YES YEAR FE YES YES YES YES YES YES 28 I also tried another set of instrumental variables, following Ackerberg and Botticini (2002). I construct each VC s local markets as an interaction of the VC s state (where the largest proportion of the VC s past investments is made) with the year of syndication. In the first step IV estimation, I regress the heterogeneity in syndication partners on these instruments as well as all relevant control variables. In the second step of IV estimation, I regress VCs survival on the predicated values of heterogeneity in syndication partners with other controls. The IV estimation generates a significant and negative coefficient on heterogeneity in syndication partners, consistent with those reported in the Table 13 and Table

59 (1) (2) (3) (4) (5) (6) VARIABLES NEW INDUDIV DEATH DEATH DEATH DEATH DEAL FULL FULL FULL FULL YEAR<2003 U.S. & IVC CONSTANT *** 0.358*** 0.212*** 0.258*** 0.146*** (0.116) (0.0193) (0.0146) (0.0159) (0.0176) (0.0188) R-squared No. of Obs Table 14 Selection Issues of VCs Survival This table is based on the sample of 12,434 VC-year observations. Column (1) reports the regression results of the selection equation with PARTEXP HETER as its dependent variable and LOCAL HETER and LOCAL MEAN as two instruments. Column (2) reports the regression results of the DEATH equation with DEATH as its dependent variable. All regressions apply fixed effects OLS models with robust standard errors. I use ***, **, and * to denote significance at the 1%, 5%, and 10% level (two-sided), respectively. (1) (2) VARIABLES PARTEXP HETER DEATH IV APPROACH: SELECTION Eq. IV APPROACH: DEATH Eq. LOCAL HETER *** (0.0187) LOCAL MEAN (0.0016) PARTEXP HETER ** (0.0781) PARTPERF HETER 0.315*** * 29 (0.0132) (0.026) PARTEXP MEAN *** * (0.0002) (0.0003) PARTPERF MEAN *** (0.0880) (0.092) VCEXP *** (0.0001) (0.0003) 29 The positive coefficient of PARTPERF HETER is purely driven by its correlation (the correlation coefficient is 0.39) with PARTEXP HETER. PARTPERF HETER is not significant when included alone and becomes significant once PARTEXP HETER is introduced to the regression. The coefficient of PARTEXP HETER remains negative and significant with or without PARTPERF HETER. 51

60 (1) (2) VARIABLES PARTEXP HETER DEATH IV APPROACH: SELECTION Eq. IV APPROACH: DEATH Eq. PARTTYPE HETER 0.252*** (0.015) (0.0221) VC FE YES YES YEAR FE YES YES R-squared No. of Obs Conclusions Syndication among firms is common. It has recently received more attention from researchers in economics and finance. This paper contributes to the literature by examining how VCs select syndication partners, and how their preferences for partners affect their successful exits from portfolio companies, and eventually, their survival. Based on a comprehensive dataset of venture capital investments made to U.S. companies from 1990 to 2005, this paper shows VCs have strong preferences for partners who are similar to themselves. Although the preferences for similarity can generate immediate benefits for VCs, syndication with more heterogeneous partners can help VCs to survive in the future. This paper introduces another type of selection among VCs when forming syndication, in addition to the more well-known selection between VCs and portfolio companies. Although VC industry is an attractive research context, the implications from this study are quite general. When firms form alliances, they tend to ally with others who are similar to them and easy to get along with. 30 The preferences for similarity may lead to successful alliances 30 For example, Fernando, Gatchev, and Spindt (2005) showed the existence of positive assortative matching between equity issuers and underwriters that higher ability underwriters are matched with higher quality issuers. Rhodes-Kropf and Robinson (2008) also showed evidence of assortative matching in the market of mergers and acquisitions that acquirers tend to buy targets with similar market-to-book ratios. 52

61 captured by their immediate financial returns. In contrast, firms allying with partners who are different need to bear additional costs and spend more effort to make the alliance work. However, the valuable learning opportunities and different perspectives brought by the diverse partners may help firms to survive in the long run. 53

62 2.6 References: Ackerberg, D. and M. Botticini (2002), Endogenous matching and the empirical determinants of contract form, Journal of Political Economy, Vol. 110, No. 3, pp Admati, A.R. and P. Pfeiderer (1994), Robust financial contracting and the role of venture capitalist, Journal of Finance, Vol. 49, No.2, pp Baker, M. and P. Gompers (2003), The determinants of board structure at the initial public offering, Journal of Law and Economics, Vol. 46, pp Beckman, C. M. and P. R. Haunschild (2002), Network learning: The effects of partners heterogeneity of experience on corporate acquisitions, Administrative Science Quarterly, Vol. 47, No. 1, pp Berger, A., N. Miller, M. Petersen, R. Rajan, and J. Stein (2005), Does function follow organizational form? Evidence from the lending practices of large and small banks, Journal of Financial Economics, Vol. 76, No. 2. pp Bonacich, P. (1972), Factoring and weighting approaches to status scores and clique identification, Journal of Mathematical Sociology, Vol. 2, pp Borgatti, S. P., M. G. Everett, and L. C. Freeman (2002), Ucinet 6 for Windows: Software for Social Network Analysis, Harvard: Analytic Technologies Bottazzi, L., M. Da Rin, and T. Hellmann (2008), Who are the active investors? Evidence from venture capital, Journal of Financial Economics, Vol. 89, No. 3, pp Brander, J., R. Amit, and W. Antweiler (2002), Venture capital syndication: Improved venture selection versus the value-added hypothesis, Journal of Economics and Management Strategy, Vol. 11, pp Brass, D. J., J. Galaskiewicz, H. R. Greve, and W. Tsai (2004), Taking stock of networks and organizations: A multilevel perspective, Academy of Management Journal, Vol. 47, No. 6, pp Casamatta, C. and C. Haritchabalet (2007), Experience, screening and syndication in venture capital investments, Journal of Financial Intermediation, Vol. 16, No.3, pp Cestone, G., J. Lerner, and L. White (2006), The design of syndicates in venture capital, working paper, Harvard Business School. 54

63 Chung, S., H. Singh, and K. Lee (2000), Complementarity, status similarity and social capital as drivers of alliance formation, Strategic Management Journal, Vol. 21, No. 1, pp Egan, E. J. (2007), Matching firm names for research in business, working paper, University of British Columbia Fernando, C. S., V. A. Gatchev, and P. A. Spindt (2005), Wanna dance? How firms and underwriters choose each other, Journal of Finance, Vol. 60, No. 5, pp Freeman, L. C. (1979), Centrality in social networks: conceptual clarification, Social Networks, Vol. 1, pp Gompers, P. (1996), Grandstanding in the venture capital industry, Journal of Financial Economics, Vol. 42, pp Gompers, P. and J. Lerner (1999), The Venture Capital Cycle, Cambridge: MIT Press, 1999 Gorman, M., and W. Sahlman (1989), What do venture capitalists do? Journal of Business Venturing, Vol. 4, No. 4, pp Greene, W. H. (2000), Econometric Analysis, Prentice-Hall, Inc. Gulati, R. and M. Gargiulo (1999), Where do interorganizational networks come from? American Journal of Sociology, Vol. 104, No. 5, pp Hambrick, D. C., T. S. Cho, and M. Chen (1996), The influence of top management team heterogeneity on firms competitive moves, Administrative Science Quarterly, Vol. 41, No. 4, pp Hellmann, T. and M. Puri (2002), Venture capital and the professionalization of start-up firms: Empirical evidence, Journal of Finance, Vol. 57, No. 1, pp Hochberg, Y, A. Ljungqvist, and Y. Lu (2007), Venture capital networks and investment performance, Journal of Finance, Vol. 62, No. 1, pp Hochberg, Y, A. Ljungqvist, and Y. Lu (2009), Networking as a barrier to entry and the competitive supply of venture capital, working paper Hoffman, L.R. and N.R.F. Maier (1961), Quality and acceptance of problem solutions by members of homogeneous and heterogeneous groups, Journal of Abnormal and Social 55

64 Psychology, Vol. 62, No. 2, pp Hsu, D. S. (2004), What do entrepreneurs pay for venture capital affiliation? Journal of Finance, Vol. 59, No. 4 pp Jacquemin, A.P. and C. H. Berry (1979), Entropy Measure of Diversification and Corporate Growth, Journal of Industrial Economics, Vol. 27, No. 4, pp Kaplan, S. and A. Schoar (2005), Private equity performance: Returns, persistence, and capital flows, Journal of Finance, Vol. 60, No.4, pp Kaplan, S. N. and P. Stromberg (2004), Characteristics, contracts, and actions: evidence from venture capitalist analyses, Journal of Finance, Vol. 59, No. 5, pp King, G. and L. Zeng (1999a), Logistic regression in rare events data, Department of Government, Harvard University, available from King, G. and L. Zeng (1999b), Estimating absolute, relative, and attributable risks in case-control studies, Department of Government, Harvard University, available from Kogut, B., P. Urso, and G. Walker (2007), Emergent Properties of a New Financial Market: American Venture Capital Syndication, , Management Science, Vol. 53, No. 7, pp Lerner, J. (1994), The syndication of venture capital investments, Financial Management, Vol. 23, No. 3, pp Lerner, J. (1995), Venture capitalists and the oversight of private firms, Journal of Finance, Vol. 50, No. 1, pp Lindsey, L. (2008), Blurring firm boundaries: the role of venture capital in strategic alliances, Journal of Finance, Vol. 63, No. 3, pp McPherson, M., L. Smith-Lovin, and J. M. Cook (2001), Birds of a feather: homophily in social networks, Annual Review of Sociology, Vol. 27, pp Megginson, W. and K. Weiss (1991), Venture capital certification in initial public offerings, Journal of Finance, Vol. 46, No. 3, pp

65 Phalippou, L. and O. Gottschalg (2007), The performance of private equity funds, working paper Podolny, J. M. (1994), Market uncertainty and the social character of economic exchange, Administrative Science Quarterly, Vol. 39, No. 3, pp Rhodes-Kropf, M. and D. T. Robinson (2008), The Market for Mergers and the Boundaries of the Firm, Journal of Finance, Vol. 63, No. 3, pp Roth, A.E. and M.A.O. Sotomayor (1990), Two-Sided Matching, Cambridge, UK and New York: Cambridge University Press Rotemberg, J. J. and G. Saloner (1995), Overt interfunctional conflict (and its reduction through business strategy), RAND Journal of Economics, Vol. 26, No. 4, Symposium on the Economics of Organization (Winter, 1995), pp Sorensen, M. (2007), How smart is smart money: A two-sided matching model of venture capital, Journal of Finance, Vol. 62, No. 6, pp Sorensen, M. (2008), Learning by investing: evidence from venture capital, working paper Sorenson, O. and T. E. Stuart (2001), Syndication networks and the spatial distribution of venture capital investments, American Journal of Sociology, Vol. 106, pp Sorenson, O. and T. E. Stuart (2008), Bringing the context back in: Settings and the search for syndicate partners in venture capital investment networks. Administrative Science Quarterly, Vol. 53, No. 2, pp Tian, X. (2008), Geography, Staging, and Venture Capital Financing, working paper Van den Steen, E. (2004), The costs and benefits of homogeneity, with an application to culture crash, working paper Wooldridge, J. M. (2002), Econometric Analysis of Cross Section and Panel Data, the MIT Press 57

66 3 INTERNATIONAL PATTERNS OF OWNERSHIP STRUCTURE CHOICES OF START-UPS: DOES THE QUALITY OF LAW MATTER? Introduction Ownership concentration has been an important subject in the corporate governance literature (Demsetz and Villalonga, 2001; La Porta, Lopez-de-Silanes, Shleifer, and Vishny, 1998; La Porta, Lopez-de-Silanes, Shleifer, and Vishny, 1999b). The determinants or optimal design of ownership structure and the impacts of various ownership structures on firm performance have been studied extensively in the context of publicly traded companies. The consequences of ownership concentration, however, are ambiguous. On the one hand, highly concentrated ownership provides the largest equity holders with more rights to deal with corporate matters, probably leading to an efficient governance structure. On the other hand, a high degree of ownership concentration may cause minority shareholders to fear expropriation of their investment or violation of their rights by the large shareholders and thus reduce their willingness to invest unless the legal system provides them with sufficient protection. Without effective legal protection of investors, external financing may be less available to firms. The macro-effects of having a high quality legal system have been explored by prior research. The general conclusion was that a high quality legal system, typically defined as a common law system with effective enforcement and norms of law and order in the population, can lead to more valuable capital markets and more dispersed ownership (see Glaeser, Johnson, and Shleifer, 2001; Djankov et al., 2003; and Demirguc-Kunt and Levine 2001). Less is known, however, about the effect that the quality of the legal system has on specific segments of the population of enterprises in a country. Exceptions were the work of La Porta et al. (1998) dealing with ownership concentration of large publicly traded companies and the work of Lerner and Schoar 31 A version of this chapter has been submitted for publication. Du, Q. and Vertinsky, I. International Patters of Ownership Structure Choices of Start-ups: Does the Quality of Law Matter? 58

67 (2005) on ownership structures of private equity investees. Despite the importance of the contributions of founding small and medium sized enterprises (SMEs) to economic growth (Berger and Udell, 1998), to our knowledge, no study has considered the ownership structures at the founding stage of small and medium firms that are not backed by private equity firms. This segment of new enterprises contains the majority of start-ups, both by value and number. For example, 94.5% of U.S. nonfarm, nonfinancial, nonreal-estate small businesses or $ billion in monetary value belong to this segment (Berger and Udeall, 1998). Indeed private equity backed investments constitute only a very modest share of the value of all investments in most countries. Only 1.85% of U.S. nonfarm, nonfinancial, nonreal-estate small businesses or $31 billion in monetary value are funded by venture capital firms (Berger and Udeall, 1998). Our paper fills this gap in the literature by developing and empirically testing an analytical framework which takes into consideration the specific differentiating characteristics of firms in this segment and the specific nature of their interactions with different types of funding sources. In their seminal paper La Porta et al. (1998) hypothesized that a higher quality legal system is likely to encourage a dispersed ownership structure. They argued that in a high quality legal system, minority shareholder s rights are well protected. They are, therefore, willing to invest. Consequently, we would expect a dispersed ownership structure of enterprises in countries with high quality legal system. They found support for their hypothesis using data from large public corporations. Lerner and Schoar (2005) studying private equity backed investees concluded that in a low quality legal system private equity firms will substitute for the lack of effective protection from the legal system by acquiring majority positions in the enterprises they invested in. Thus founders face higher costs (including loss of control) of securing external funding from private equity firms where the legal system in place offers lower protection to investors. The implication is that ownership structures of investees are likely to be more concentrated in lower quality legal 59

68 systems. They tested their framework with data obtained from private equity groups operating internationally. The implications of our theoretical framework suggest that in de novo founding of small and medium firms, most of which are not backed by established private equity firms, founders are more likely to attract co-owners and secure internal capital from all start-up owners in environments with lower legal protection. In environments with strong legal protection, they are more likely to raise external debt capital and retain full ownership at founding. We test our prediction using data from the Adult Population Survey of the Global Entrepreneurship Monitor project from 2001 to The empirical setting we use has several distinct advantages compared to the studies of La Porta et al. (1998) and Lerner and Schoar (2005). Studying ownership structure choices at founding allows for a less biased estimation of the impact of legal systems. The ownership structures of established firms can dramatically differ from their initial ownership choices as they evolve over time. Furthermore, ownership structure and firm s characteristics (firm size and performance for example) are endogenously determined (Bitler, Moskowitz, and Vissing-Jorgensen, 2005; Cassar, 2004). In addition, studying ownership structure decisions at founding reduces problems of survival bias, a serous problem given the high mortality rate of start-ups. Finally, we use in our econometric analysis micro level data that enables us to analyze ownership preferences of individual entrepreneurs and estimate more accurately the ownership concentration of start-ups in the country. We also are able to examine and control for the impacts of founder s and firm s characteristics on ownership structure choices. In addition, our study also finds that entrepreneurs demographic characteristics can affect their choices of ownership structures. This paper is organized as follows: Section 3.2 briefly reviews the literature on how the legal system plays a role in economic decisions. Section 3.3 outlines the theoretical framework. 60

69 Methodology and data are discussed in Section 3.4 and empirical findings are presented in Section 3.5. Section 3.6 concludes the paper. 3.2 Does Law Matter? This section briefly reviews the existing literature that establishes the association between the legal system and investors protection. Our theoretical framework examining the mechanism through which a legal system affects entrepreneurs choices of ownership will be presented in the next section. La Porta et al. (1998) argued that legal origins and legal enforcement are the two key measures of a legal system with respect to the protection of investors. According to La Porta et al. (1998, 1999, 2003), there are two broad origins of legal systems - common law and civil law systems. Common law originated in England. Civil law has its origins in the Roman Empire with three representative legal families: French, German and Scandinavian. 32 These legal families were then transplanted to many other countries through conquest, colonization or voluntary adoption. Although each country s legal system has developed over time and borrowing from other legal families is possible, the essential features of the legal origin remained intact. Two types of explanations were advanced to suggest why legal origins signal different levels of protection of minority investors. The judicial explanation is based on the fact that in the common law system, judges can make their decisions based on general rules or precedent judgment, but are not limited to them in making choices, while in civil law countries, judges cannot make decisions beyond what is prescribed by the legal rules (Glaeser and Shleifer, 2002). La Porta et al. (1999a) also provided a political explanation that the legal differences are explained by the relative power of the king and the property owners. As early as the 17 th 32 Russian law is originated from civil law but does not belong to the three common legal families. Therefore, it is treated as a separate category in the analysis. 61

70 century, the crown in England lost some control of the courts which were guided by parliament where the voice of property owners was dominant. As the power of parliament increased, the protection of investors gradually expanded. This was not the case in France or Germany where the government remained in control of the courts and legislators. The effectiveness of the legal system in protecting investors depends not only on the decisions of the courts but also on the effective enforcement of these decisions. The quality of legal implementation is reflected both in the resources and effectiveness of enforcement institutions, and the propensity of citizens to obey laws. For example, transparency in financial reporting and bureaucracy increase trust in the society. Higher voluntary compliance with laws and lower use of litigation to resolve private disputes reduce the burden on the legal system. 3.3 The Quality of the Legal System and Financing and Ownership Decision by Founders of Start-ups De novo start-ups have some distinct characteristics. They are typically small privately held firms. Information opacity and asymmetries in information held by founders and potential investors are acute as public records of performance (or indeed any records of performance) are not available. Thus obtaining external financing is a challenge irrespective of the level of protection offered by the legal system to investors. Indeed in most cases financing is one of the key factors affecting ownership structures and rights or control allocation within start-ups. In our framework we assume that there are two sources of capital to finance start-ups at founding stage: internal capital and external capital. Internal capital mainly refers to the equity capital obtained from start-up founders and their co-owners while external capital mainly refers to the debt capital raised by entrepreneurs from banks. The division of internal capital and external capital is reasonable because the major sources of financing for start-ups are from principal 62

71 owners and banks, and external equity financing, such as venture capital, is very limited at founding stage (Berger and Udell, 1998). In a good legal system, the law could protect well both internal and external capital investors. It can be true that start-up founders prefer internal capital financing because co-owners could add some extra value by advising, monitoring the business, and providing access to networks. It can also be true that start-up founders prefer external capital financing to maintain absolute control of start-ups. We can argue that in equilibrium, given a particular level of quality of legal protection, there is some fixed ratio of the amount of internal capital to the amount of external capital used by start-up founders to finance their ventures. In a poor legal system, the weak investor protection makes it more difficult to raise external capital than in a good legal system. External capital investors, i.e. lenders do not have access to information, monitoring, and control as much as equity holders and have to rely on bankruptcy laws and contract enforcement to protect their interests. Therefore, in a poor legal system, external capital investors anticipate a higher default rate from entrepreneurs, leading to a higher cost of external capital. On the other hand, start-up founders may succeed in securing internal capital from other possible owners to finance their ventures. Although the investor protection offered by the legal system is inadequate, once becoming co-owners, internal capital investors can rely on their ownership shares and control as a substitute for the ineffective legal system. Prior research has documented the existence of substitution for the poor legal protection in private companies. For example, Bergman and Nicolaievsky (2007) observed that in Mexico the law provides only scant protection to investors, leaving a need for investors to contractually opt out of the legal system and obtain protection provided by investees privately (pp. 739). Lerner and Schoar (2005) studying private equity investments in developing countries found that poor legal systems constrained the ability of private equity partners to write sophisticated contracts that can separate control rights from cash flow rights and offer adequate protection for investors. 63

72 In such cases private equity investors may seek majority ownership to gain control as a substitute for the lack of legal protection. In our case of start-ups at founding stage, acquiring a larger stake is less financially prohibitive, thus the substitution mechanisms can be more effective than that in large public companies. We predict that in equilibrium, start-ups may have more owners in a poor legal system than in a good legal system. Given the limited number of owners for start-ups at founding stage, we are more likely to observe partial ownership structure in a poor legal system than in a good one. 3.4 Data and Methodology The data used to test our hypotheses was derived from the GEM (Global Entrepreneurship Monitor) project database. GEM is an annual assessment of entrepreneurial activities at the country level. GEM adopts a broad definition of entrepreneurial activities that include any general start-up activities and are not limited to high technology sectors, although high technology start-ups and venture capital backed start-ups are important constituents of entrepreneurship. Our dataset included start-up founders from 19 countries or regions in 2001, which expanded to 31 countries or regions by Each country or region participating in GEM project conducted an Adult Population Survey to get a random sample of no less than 2,000 individuals. Individuals were asked to report their demographic characteristics, status of employment, and characteristics of the start-ups. This paper makes use of the sub-sample of respondents who identified themselves as independent founders of start-ups. To be included in our sample, only founders who have made significant commitments and took a significant action to develop their business (e.g. buying equipment and renting space) were included. The final sample has 9,561 founders of independent start-ups in the surveys conducted between 2001 and To check the robustness of our results, we also conducted country level analysis, in which we calculated the percentage of entrepreneurs who expected to fully own the business in each 64

73 country and explored the relationship between this percentage and legal and economic variables. The country level analysis is comparable to prior research that focused on large publicly traded companies (La Porta et al., 1998). Table 15 presents the country composition in our sample and their legal origins. Table 15 Country Composition and Legal Origin Country No. of Obs. Percentage Year Australia , 2003, 2004 Canada , 2002, 2003 Hong Kong , 2003, 2004 India , 2002 Ireland , 2004 Israel , 2002, 2004 New Zealand , 2004 Singapore , 2002, 2003, 2004 South Africa , 2004 Thailand Uganda United Kingdom , 2003, 2004 United States , 2002, 2003, 2004 English Origin Total Argentina , 2002, 2003, 2004 Belgium , 2003, 2004 Brazil , 2004 Chile , 2003 France , 2002, 2003, 2004 Greece , 2004 Italy , 2002, 2004 Jordan Mexico Netherlands , 2003, 2004 Peru Portugal , 2004 Spain , 2003, 2004 French Origin Total China , 2003 Croatia , 2003,

74 Country No. of Obs. Percentage Year Germany , 2002, 2003, 2004 Hungary , 2002, 2004 Japan , 2002, 2003, 2004 Korea , 2002 Poland , 2002, 2004 Slovenia , 2003, 2004 Switzerland , 2003 Taiwan German Origin Total Denmark , 2002, 2003, 2004 Finland , 2002, 2003, 2004 Iceland , 2003, 2004 Norway , 2002, 2003, 2004 Sweden , 2002, 2003, 2004 Scandinavian Origin Total Country No. of Obs. Percentage Year Russia Note: The sources of countries legal origins are La Porta et al (1998) and To characterize the enforcement quality in the legal system, we used several indexes developed by the World Bank between 2004 and 2005 based on the series of work from La Porta et al. Although the GEM surveys were conducted between 2001 and 2004, we do not expect the legal enforcement to change significantly over such a short horizon. The four legal enforcement variables Legal Rights of Borrowers and Lenders, Disclosure Index, Director Liability Index, and Shareholder s Suits Index capture different dimensions of legal enforcement of judicial decisions. Legal Rights of Borrowers and Lenders directly captures how collateral and bankruptcy laws facilitate lending and can be used to directly test whether better protection of creditors leads to sole ownership. Disclosure Index measures how transparent the transactions are and thus how the legal system mitigates the asymmetric information problem. Director Liability Index measures the liabilities of self-dealing while Shareholder s Suits Index captures the ease for shareholders to sue officers and directors for their misconducts. 66

75 The above four legal enforcement variables are not mutually exclusive. They are correlated to each other and to the legal origins as well. Common law countries usually have better legal enforcement in terms of higher Legal Rights of Borrowers and Lenders, Investor Protection Index, Disclosure Index, and Director Liability Index. In a good legal system, the expropriation of investors holdings is discouraged and investors are well protected. When the laws are not protective, different types of investors are affected to different extents. Our micro-level data allows us to examine the effects of entrepreneur level and company level characteristics on ownership preferences. AGE acts as a proxy for experience. The more experienced the entrepreneurs are, the more likely they are to choose full ownership because they perceive themselves more capable of managing their firms on their own. Old age may also mean the dislike of complexity brought by shared ownership. The effect of GENDER on ownership preference is more ambiguous. DeMartino and Barbato (2003) show that female and male entrepreneurs may have different career motivations. Female entrepreneurs prefer flexibility and balance between work and family while male entrepreneurs tend to choose careers where they can accumulate wealth. On one hand, female entrepreneurs may want to fully own their business because they enjoy the flexibility of being the sole owner of their businesses. On the other hand, female entrepreneurs may be more motivated to look for business partners to share the workload so that they can have more time for family obligations. The effect of EDUCATION on ownership preferences is also ambiguous. On one hand, higher education may act as a positive signal to secure loans so that sole ownership is made more possible (Bates, 1990). On the other hand, higher education also acts as a positive signal that attracts business partners, potentially facilitating a partial ownership structure. 67

76 Table 16 Description of the Legal Variables Variables Description & Sources COMMLAW dummy variable which equals 1 if the country s legal system has common law origin; equals 0 if the country s legal system has civil law origin. Source: La Porta et al (1998) Law and Finance and LEGALR continuous variable from 0 to 10 to measure how well collateral and bankruptcy laws facilitate lending. The bigger the value, the better the laws facilitate lending. Source: World Bank website, doing business section DISCLS This is the Disclosure Index from the World Bank. It is a continuous variable from 0 to 10 to measure how transparent are the transactions. The bigger the value, the more transparent the transactions. Source: World Bank website, doing business section DIRLIA This is the Director Liability Index from the World Bank. It is a continuous variable from 0 to 10 to measure the liability of selfdealing. The bigger the value, the more liabilities of self-dealing. Source: World Bank website, doing business section SUITS This is the Shareholder Suits Index from the World Bank. It is a continuous variable from 0 to 10 to measure the shareholders ability to sue officers and directors for misconduct. The bigger the value, the more ease the shareholders will have to suit directors. Source: World Bank website, doing business section Personal wealth has been regarded an important factor in entrepreneurial finance. Due to the lack of information on wealth, we use INCOME as a proxy. Again, wealth could also have ambiguous effects on ownership choice as wealthy entrepreneurs are able to attract both internal and external investors. The ambiguity on how gender, education, and income affect entrepreneurs ownership preferences will be resolved empirically in this study. The effect of risk preference on ownership choice is straightforward. If an entrepreneur is RISK AVERSE, the desire of risk-sharing will lead to partial ownership. 68

77 Table 17 Description of Other Variables Variables Description SOLEOWN dummy variable which equals 1 if an entrepreneur fully owns the business; equals 0 if the entrepreneur partly owns the business. OWNERS continuous variable which indicates how many people including the interviewed entrepreneurs will own and manage the new business. OWNER4 categorical variable which equals 1 if the expected number of owners of the new business is 1; equal 2 if the expected number of owners is 2; equals 3 if the expected number of owners between 3 and 5; equals 4 if the expected number of owners is above 6. AGE continuous variable between 18 and 64. GENDER dummy variable which equals 1 if male; equals 0 if female. INCOME categorical variable which equals 1 if the income belongs to the lowest 33 percentile; equals 2 if income belongs to the middle 33 percentile; equals 3 if income belongs to upper 33 percentile. EDUCATION categorical variable which equals 1 if the educational attainment is some secondary; equals 2 if secondary degree obtained; equals 3 if post secondary education and equals 4 if graduate education. NETWORK dummy variable which equals 1 if the interviewed entrepreneurs know someone personally who started a business since the past 2 years; equals 0 otherwise. RISK AVERSE dummy variable which equals 1 if fear of failure would prevent the interviewed entrepreneurs from starting a business; equals 0 otherwise. INDUSTRY categorical variable which equals 1000 if the business is in agriculture, forest, and hunting; 2000 if mining and construction; 3000 if manufacturing, 4000 if transportation, communication, and utilities; 5000 if wholesale and repair; 6000 if retail, hotel, and restaurant; 7000 if finance, insurance, real estate; 8000 if bus services; 9000 if education and social services; and if consumer services. NO. of JOBS continuous variable which is a log of the number of jobs in the next 5 years. LNGNI GNI per capita taken average from the past five years. GDPGR Annual GDP per capital growth rate taken average from the past five years. Well-connected entrepreneurs have better access to information and advice about the entrepreneurial process (Hoang and Antoncic, 2003). The entrepreneurial NETWORK also makes it easier for these entrepreneurs to find suitable business partners and increase the likelihood of a partial ownership structure. Apart from the entrepreneurs personal characteristics, INDUSTRY type is also assumed to have 69

78 an impact on ownership choices. Traditional manufacturing industry usually requires significant commitments of capital, making sole ownership for many infeasible. Some high-tech (software for example) or consulting firms that are human capital intensive may not require large investments, making sole ownership more likely. FIRM SIZE is also an important factor in ownership determination as the size of commitment affects the ability of a single founder to obtain debt financing or to self finance the start-up. Different measures are used to capture firm size, such as firm s equity, asset or number of employees. Nascent start-ups usually have not recruited enough employees yet and a measure of expected number of employees in the future can be used to proxy for firm size. The level and growth rate (GDPGR) of GNP per capita (LNGNP) are both introduced in the regressions. Higher GNP per capita is correlated with financial market sophistication that we showed to be negatively correlated with higher concentration of ownership. Table 18 describes our sample. Among these 9,561 start-up founders, 52% chose to fully own their businesses. About 40% of the entrepreneurs in our sample operate their businesses in countries with common law systems. Within the civil law family, French civil law systems have the biggest number of start-up entrepreneurs, followed by German, Scandinavian, and Russian civil law system. In our sample, start-up investors enjoy an average Legal rights rating of 6.22 (out of 10), an average Disclosure Index of 6.89 (out of 10), an average Director Liability of 5.57(out of 10), and an average Shareholder s suits of 6.48 (out of 10). Contrasting with the depiction of entrepreneurs of optimists, only 19% of them declare to be risk averse. Given the importance of networking, 67% of the entrepreneurs were connected to others with entrepreneurial experience. 70

79 Table 18 Descriptive Statistics of Variables Variable No. of Obs. Mean Standard Deviation Soleown Owners Owners Common Law French German Scandinavian Russia Legal rights Disclosure Index Director liability Shareholder's suits Risk averse Network Income Education Gender Age Log of GNI/Capita GDP growth rate Expected No. of jobs Agriculture, forest, hunting, fishing 420 Mining, construction 504 Manufacturing 765 Transportation, communication, utils 458 Wholesale repair 704 Retail, hotel, restaurant 2766 Finance, insurance, real estate 378 Bus services 1557 Education & social services 659 Consumer services 724 Year Year Year Year

80 The average start-up entrepreneur is in the middle-income group and has obtained a secondary degree. About 64% of entrepreneurs are males with the average age of 37. The average GNI per capita in our sample is about $9,240 with an average annual GDP growth rate of 2.16%. For the start-ups in our sample, the average number of employees they plan to hire within 5 years is 155. They are concentrated in Retail, hotel, restaurant, followed by Bus services, Manufacturing, and Consumer services. The main research question of this paper is to determine the relationship between the founders tendency to choose sole ownership rather than enter into partnership and the quality of protection offered by the legal system. We estimate a binary dependent variable model to predict the probability of using one choice against the other (Greene, 2002). Five sets of variables are used to explain ownership choices. The first set of variables contains legal origins and legal enforcement variables. The second set of variables includes personal characteristics of start-up entrepreneurs. The third set of variables is used to control for differences in macroeconomic environments. The fourth set of variables controls for firm specific characteristics. We also control for time fixed effects. The estimated function therefore becomes: Prob (sole ownership=1) = f (Legal system, Personal characteristics, Macroeconomics, Firm characteristics, Time fixed effects) Since the data includes samples of individuals from different countries, there are potential correlations of error terms within each country (Greene, 2002) so in our model standard errors are clustered on the country level. 72

81 3.5 Econometric Analysis Individual Level Analysis The results of Logit regressions are reported in Table 19 to show the effects of the explanatory variables on ownership choices. In Table 19, we use legal origin variables to capture the quality of the legal system while in Table 20 we report regression results using legal enforcement variables. The first two regressions in Table 19 focus only on legal origins and macroeconomic variables, controlling for time fixed effects and clustering standard errors on country level. We find that entrepreneurs in common law countries are more likely to have sole ownership. Individual specific variables are introduced in the third and fourth regressions. Again, legal origin variables remained significant. Entrepreneurs with more income, more education, who are more risk averse and having more access to entrepreneurial networks are more likely to choose partial ownership while older entrepreneurs prefer full ownership. Gender, however, does not have a significant coefficient. The fifth and sixth regressions introduce firm specific variables in addition to entrepreneurs characteristics. The effects of legal origins remained robust except for the coefficients of German and Russian legal systems. Almost all personal characteristics had the same impact on the choice of ownership except that gender has a strong and positive impact in this regression, indicating that male entrepreneurs are more likely to have sole ownership. Firm size measured by the expected number of jobs in the future had a significant positive effect on the tendencies to choose partial ownership. Industry fixed effects are also controlled. The log of GNI per capita and GDP are not significant after controlling for individual and firm characteristics. 73

82 Table 19 Regressions on Legal Origins (1) (2) (3) (4) (5) (6) VARIABLES SOLEOWN SOLEOWN SOLEOWN SOLEOWN SOLEOWN SOLEOWN Common Law 0.48*** 0.45** 0.44** (3.72) (3.38) (3.24) French -0.61*** -0.57*** -0.57*** (-4.38) (-4.12) (-4.13) German -0.26* (-2.14) (-1.62) (-1.20) Scandinavian -0.56** -0.56** -0.61*** (-3.42) (-3.38) (-3.82) Russia (-1.66) (0.96) (1.05) Risk averse -0.22*** -0.22*** -0.25*** -0.25*** (-3.88) (-3.87) (-4.31) (-4.35) Network -0.25*** -0.28*** -0.21** -0.23*** (-4.29) (-5.16) (-3.70) (-4.45) Income -0.10* -0.11** -0.09* -0.10* (-2.53) (-2.73) (-2.04) (-2.23) Education -0.12*** -0.12*** -0.13*** -0.12*** (-4.52) (-4.59) (-4.26) (-4.29) Gender * 0.12* (0.67) (0.76) (2.18) (2.35) Age 0.02*** 0.02*** 0.02*** 0.02*** (5.87) (5.97) (5.84) (5.90) Log of GNI/Cap (-1.82) (-1.85) (-1.15) (-1.36) (-1.16) (-1.15) GDP growth rate (0.56) (-0.67) (0.29) (-0.71) (0.66) (-0.28) Exp. No. of jobs -0.14*** -0.14*** (-8.34) (-8.50) Industry FE YES YES Year FE YES YES YES YES YES YES Constant ** ** * (1.66) (2.69) (1.10) (2.79) (0.84) (2.03) Log pseudolikelihood No. of obs

83 Table 20 reports the results of regressions on legal enforcement variables. Efficient legal systems measured by Legal rights of borrowers and lenders and Shareholders Suits are more likely to encourage full ownership of start-ups. The variable Legal rights of borrowers and lenders serves as a direct test of our theoretical framework: if the external investors, i.e. lenders are not well protected in a poor legal system, it becomes more difficult for entrepreneurs to borrow, leading to more internal capital financing and more dispersed ownership structures in a poor legal system. The predictive power of Legal rights of borrowers and lenders is the strongest among the four types of legal enforcement. As shown in the fifth column, after introducing Legal rights of borrowers and lenders the other three measures of legal enforcement become statistically insignificant. Access to networks, income and age display patterns consistent with the results reported in Table 19 showing that entrepreneurs with higher income and network access would prefer partial ownership while older entrepreneurs would choose sole ownership. Because some of our individual level variables could affect access to and ability to use the legal system to obtain protection of rights, we have also tested whether interactions of the quality of the legal system with personal characteristics have any effect on the ownership choice. We expected that the education and income would interact with the quality of the legal system as more educated and wealthier founders are able to use a good legal system more effectively but the interaction coefficients were insignificant. 75

84 Table 20 Regressions on Legal Enforcement (1) (2) (3) (4) (5) (6) VARIABLES SOLEOWN SOLEOWN SOLEOWN SOLEOWN SOLEOWN SOLEOWN Legal Rights 0.11** 0.11** 0.03 (2.61) (2.66) (0.70) Disclosure Index (0.71) (-0.71) (-1.45) Director Liability * (0.32) (-0.93) (-2.21) Shareholder s Suits 0.07* (1.96) (1.33) (-0.32) Common Law 0.66** (3.25) Risk Averse *** -0.25*** -0.24*** -0.24*** -0.25*** -0.27*** (-4.02) (-4.12) (-4.23) (-4.07) (-4.14) (-4.57) Network -0.23*** -0.24*** -0.25** -0.25*** -0.25*** -0.22*** (-4.07) (-3.80) (-3.78) (-4.08) (-4.44) (-4.01) Income -0.10* -0.09* -0.10* -0.09* -0.10* -0.11* (-2.20) (-2.02) (-2.21) (-2.03) (-2.50) (-2.58) Education -0.10** -0.10** -0.09** -0.10*** -0.10** -0.13*** (-3.14) (-3.19) (-2.82) (-3.37) (-3.44) (-4.74) Gender 0.11* 0.10* * 0.11* 0.11* (2.15) (2.03) (1.92) (1.98) (2.14) (2.16) Age 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** (5.70) (5.82) (5.83) (6.17) (5.84) (5.80) Log of GNI/Capita -0.16** ** (-2.61) (-0.51) (-0.60) (-1.04) (-3.33) (-0.54) GDP growth rate (-0.20) (0.76) (0.78) (0.79) (-0.21) (0.52) Expected No. of jobs -0.14*** -0.14*** -0.14*** -0.14*** -0.14*** -0.14*** (-8.92) (-8.60) (-8.61) (-8.73) (-9.31) (-8.61) Industry fixed effects YES YES YES YES YES YES Year fixed effects YES YES YES YES YES YES Constant 1.04* (2.01) (0.36) (0.62) (0.26) (1.84) (1.48) Log pseudolikelihood No. of obs

85 Table 21 Robustness Checks (1) (2) (3) (4) (5) (6) (7) VARIABLES SOLEOWN SOLEOWN OWNER4 OWNER4 OWNER4 OWNER4 SOLEOWN PROBIT PROBIT OLS OLS OPROBIT OPROBIT LOGIT Common Law 0.27** -0.14* -0.21* 0.51*** (3.27) (-2.28) (-2.45) (0.12) French -0.35*** 0.16* 0.23** (-4.17) (2.63) (2.83) German (-1.21) (0.52) (0.75) Scandinavian -0.38*** 0.28** 0.37** (-3.84) (3.21) (3.32) Russia * 0.19** (1.01) (2.52) (2.65) Trust 0.60 (0.57) Risk Averse -0.16*** -0.16*** 0.09** 0.09*** 0.13*** 0.13*** -0.25*** (-4.36) (-4.40) (3.38) (3.63) (3.63) (3.86) (.06) Network -0.13** -0.15*** 0.09*** 0.09*** 0.13*** 0.13*** -0.17*** (-3.72) (-4.47) (3.89) (4.06) (4.01) (4.22) (0.06) Income -0.06* -0.06* * ** (-2.05) (-2.24) (1.54) (1.70) (1.67) (1.84) (.05) Education -0.08*** -0.08*** 0.06*** 0.06*** 0.08*** 0.08*** -.014*** (-4.26) (-4.30) (4.61) (4.64) (4.65) (4.69) (0.03) Gender 0.07* 0.08* ** (2.18) (2.36) (0.20) (0.04) (-0.25) (-0.42) (0.05) Age 0.01*** 0.01*** -0.01*** -0.01*** -0.01*** -0.01*** 0.02** (5.92) (5.97) (-3.91) (-3.81) (-4.50) (-4.41) (0.003) Log of GNI/Cap (-1.18) (-1.17) (0.85) (0.31) (0.98) (0.51) (0.06) GDP growth rate (0.65) (-0.30) (-0.60) (-0.07) (-0.57) (0.01) (.05) Exp. No. of jobs -0.09*** -0.09*** 0.07*** 0.07*** 0.10*** 0.10*** -0.13*** (-8.49) (-8.65) (10.09) (9.92) (10.50) (10.45) (0.02) Industry FE YES YES YES YES YES YES YES Year FE YES YES YES YES YES YES YES Constant * 1.50*** 1.43*** 0.43 (0.84) (2.03) (7.54) (7.74) (0.50) Log pseudolikelihood No. of obs

86 A series of robustness checks have been done in Table 21. In Column (1) and (2) of Table 21, Probit instead of Logit regression was used and generate the same prediction. For other regressions from Column (3) to Column (6), the dependent variable changes from a binary variable to a categorical variable. The bigger the value of the categorical variable, the more owners there are in the start-ups. Both Ordinary Least Square models and Ordered Probit models are estimated for the categorical dependent variables. In the OLS and Ordered Probit regressions, entrepreneurs in countries belonging to French and Scandinavian law show a tendency to have more partners. Income, risk aversion, education and network access show consistent results as before but lose some significance while gender loses its explanatory power. In Column (7) of Table 21, we examine whether one type of social capital among people trust, has any impact on start-up entrepreneurs choices of ownership structures. We obtain data from the World Value Survey which reports people s values and cultural changes all over the world. As the surveys of start-up ownership structures used in this paper were conducted in 2001 and 2004, we choose the World Value Survey conducted in 1995 and 2000 to avoid the possible reverse causality bias in regressions. Survey respondents answer two similar questions about trust: Generally speaking, would you say that most people can be trusted or that you can t be too careful in dealing with people? in the 1995 survey, and Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people in the 2000 survey. We are able to obtain the data of trust for 31 countries in our original sample. For each country, we calculate the percentage of respondents who choose Most people can be trusted as an indicator of general trust among people. If a country participates in both the 1995 and 2000 surveys, we take the average of the scores of trust. Column (7) of Table 21 reports the regression of ownership structures on legal origins, trust, and other control variables. The regression results suggest that general trust among people do not have a direct impact on entrepreneurs choices of ownership structures. 78

87 Country Level Analysis The country level analysis can be seen as either a complementary analysis or robustness check of the individual level analysis. The country index of preference for sole ownership was defined as the percentage of entrepreneurs with sole ownership (Sole Ownership Preference Index). Table 22 shows the regressions explaining Sole Ownership Preference Index as a function of two legal origins and five legal families. Table 22 presents the results of the country level regressions. Column (1) shows the significant and positive relationship between common law countries and the percentage of sole owners in a country without controlling for country level variables. The coefficient of common law does not reach conventional statistical significance after control variables are introduced in Column (2). Column (3) and column (4) replicate the regressions in Column (1) and Column (2) except that the legal origin variable is replaced by memberships in the five legal families. Only the French law family and the Scandinavian family show significant results without the controls and only French law family remains significant with controls. The insignificance of legal variables is partly due to the limited number of observations. Our results should be interpreted with some caution. First, direct examination of contracts and fuller details about informal strategies used by both founders and investors to deal with risks is needed to validate our conclusions. Second, future detailed case studies of financing and ownership structure decisions under high and low quality legal systems as well as longitudinal studies of changes in behaviors that occurred in systems which have transitioned from low quality to high quality legal system will provide a fuller account of the role that protection offered by the legal system has on founders and investors behaviors. 79

88 Table 22 Country Level Regressions (1) (2) (3) (4) SOLEOWN RATIO SOLEOWN RATIO SOLEOWN RATIO SOLEOWN RATIO Common Law (1.74) (1.67) French (-1.92) (-1.84) German (-0.38) (-0.41) Scandinavian -0.13* (-2.33) (-1.59) Russia (0.23) (-0.20) Log of GNI/Cap (1.09) (0.63) GDP growth rate -0.03* (-2.33) (-1.94) Constant 0.49*** 0.73*** 0.55*** 0.77*** (24.92) (5.94) (19.22) (5.88) Adj. R-Square No. of Obs Discussions and Conclusions The focus of this paper was the relationship between the quality of the protection offered by the legal system to investors and the propensity of start-up founders to opt for sole ownership. The issue of concentration of ownership received significant attention in the law and economics, and finance literatures. La Porta et al. (1998) highlighted the importance of the quality of the legal protection offered by a country to minority shareholders to the development of its financial markets. They argued that ownership concentration is negatively related to effective legal protection. Without adequate protection minority shareholders are likely to be discouraged from investing. La Porta et al. (1998) found evidence from samples of large publicly owned firms to support their theoretical arguments. 80

89 Our theoretical framework suggests that the relatively low cost of external financing in a good legal system encourages start-up founders to finance their start-ups by external debt financing while retain full control of their ventures. Without means of effective protection, financial institutions are less willing to lend the required capital as the risks of recovering their investments are higher. As a result, start-up founders tend to attract possible co-owners from whom they secure internal capital, at the expense of giving up some ownership and control of the ventures. Our results shed a light on an important area of entrepreneurial research which has received relatively little attention the relationship between the choices of start-up founders with respect to modes of financing and ownership structures, and the quality of the legal system. Our results suggest that there is a significant relationship. The findings which imply that the costs of equity financing of start-ups are less sensitive to the quality of the law have interesting ramifications. They may imply that in financing start-ups protection mechanisms which do not depend on the legal system are available and may play a significant role in investment decisions. There is need for future research to better understand the nature and consequences of these substitutes for protection offered to investors by a legal system. Given the evidence that founders, especially of family businesses, prefer to retain ownership and control, improvement in the quality of the legal system and thus the availability of debt financing are likely to stimulate the emergence of entrepreneurial ventures. 81

90 3.7 References Bates, T Entrepreneur Human Capital Inputs and Small Business Longevity, The Review of Economics and Statistics, 72(4): Berger, A., and G. Udell The Economics of Small Business Finance: The Roles of Private Equity and Debt Markets in the Financial Growth Cycles. Journal of Banking & Finance, 22: Bergman, N. K. and Daniel Nicolaievsky Investor protection and the Coasian view. Journal of Financial Economics, 84(3): Bitler, M., T. Moskowitz, and A. Vissing-Jorgensen Testing Agency Theory with Entrepreneur Effort and Wealth. The Journal of Finance, 60 (2): Bottzzi, L., M. Da Rin, and T. Hellmann What Role of Legal System in Financial Intermediation? Theory and Evidence. Journal of Financial Intermediation, Forthcoming Burkart, M., and F. Panunzi Agency Conflicts, Ownership Concentration, and Legal Shareholder Protection. Journal of Financial Intermediation, 15 (1): 1 31 Carter, N., W. Gartner, and P. Reynolds Exploring Start-Up Event Sequences. Journal of Business Venturing, 11 (3): Cassar, G The financing of business start-ups. Journal of Business Venturing, 19 (2): DeMartino, R., and R. Barbato Differences between women and men MBA entrepreneurs: exploring family flexibility and wealth creation as career motivators. Journal of Business Venturing, 18 (6): Demsetz, H., and B. Villalonga Ownership structure and corporate performance. Journal of Corporate Finance, 7: Demirguc-Kunt, A. and R. Levine Financial Structure and Economic Growth: A Cross-Country Comparison of Banks, Markets, and Development. Cambridge, MA: MIT Press. Djankov, S., R. La Porta, F. Lopez-De-Silanes, and A. Shleifer Courts. The Quarterly Journal of Economics, 118 (2): Djankov, S., C. McLiesh, and A. Shleifer Private Credit in 129 Countries. Journal of 82

91 Financial Economics, 84 (2): Glaeser, E., S. Johnson, and A. Shleifer Coase versus the Coasinas. The Quarterly Journal of Economics, 116 (3): Glaeser, E., and A. Shleifer Legal Origins. The Quarterly Journal of Economics, 117 (4): Greene, W Econometric Analysis. 5th edition, Prentice Hall Hoang, H., and B. Antoncic Network-based research in entrepreneurship: A critical review. Journal of Business Venturing, 18 (2): Kaufmann, D., A. Kraay, and M. Mastruzzi Governance Matters III: Governance Indicators for working paper La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny Legal Determinants of External Finance. The Journal of Finance, 52 (3): La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny Law and Finance. Journal of Political Economy, 106 (6): La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny. 1999a. The quality of government. Journal of Law, Economics and Organization, 15 (1): La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny, 1999b. Corporate Ownership around the World. The Journal of Finance, 54 (2): La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny Investor Protection and Corporate Governance. The Journal of Financial Economics, 58 (1-2): Levi, M., Li, K., and Zhang, F Deal or No Deal: Hormones and the M&A Game, working paper Lerner, J. and A. Schoar, A Does Legal Enforcement Affect Financial Transactions? The Contractual Channel in Private Equity. The Quarterly Journal of Economics, 120 (1): Shleifer, A. and D. Wolfenzon Investor Protection and Equity Markets. Journal of Financial Economics, 66 (1): World Bank, World Development Indicators (Washington, DC: The World Bank, 2004) 83

92 4 BORN LEADERS: THE RELATIVE-AGE EFFECT AND MANAGERIAL SUCCESS Introduction There is mounting empirical evidence that summer-born children are at a disadvantage as a result of being up to a year younger than other classmates in their school grade. Summer-born children are in this position due to the fact that the cutoff dates for admission into school generally fall at the end of summer. Therefore, those just born before the grade entry cutoff date during the summer months are up to a year younger and less physically and intellectually developed than their classmates. The disadvantage faced by summer-born children has been shown to exist throughout school, and even to affect the success at entering university. This well-documented condition has become known as the relative-age effect or the birth-date effect. This paper investigates whether a relative-age or birth-date effect extends to the selection and performance of CEOs of S&P 500 companies. Given the high level of corporate achievement such positions represent, and the relatively fierce competition faced in reaching such positions, CEOs represent an ideal context for studying the evidence for such an effect, and to investigate whether it extends well beyond school and into adulthood. In order to investigate the possible presence of a relative-age effect among highly successful U.S. CEOs we construct a birth-date dataset for the CEOs of S&P 500 companies between 1992 and Based on the prevailing cutoffs of school entry in the United States, we distinguish four birth seasons with the summer season being from July to September. We obtain the names of the CEOs from ExecuComp, and their birth-date and educational background from the Biography Resource Center. We were able to identify birth-dates and education backgrounds of A version of this chapter has been submitted for publication. Du, Q., Gao, H., and Levi, M. Born Leaders: The Relative-Age Effect and Managerial Success. 84

93 CEOs. 34 Stock price data were obtained from CRSP and accounting data from Compustat. The principal findings are as follows. First, we find that non-summer born individuals have a significantly higher chance of becoming a CEO of an S&P 500 company. This occurs relative to both an equal division of births across seasons, and relative to the actual seasonal pattern of births. The effect is robust to measurement of relative age within school grade that allows for the different grade-entry cut-off dates in different US states. Second, conditional on becoming a CEO, those who were born in the summer add higher value to their company whether this be considered via Tobin s Q involving market and book value of assets, or the market to book value of equity, M/B. We find that having a summer-born CEO increases Tobin s Q and M/B at a p-value of less than one-percent. Third, we demonstrate the return from a policy of forming a portfolio based on buying companies with summer-born CEOs and selling short companies with non-summer-born CEOs. These portfolios are reset once each year. This is shown to generate an annual abnormal return of 8.32 percent. This indicates that financial markets are yet to realize the effect of CEO birth season on asset returns. We also investigate whether the relative-age effect shows up in CEO compensation. We find that summer-born CEOs receive unconditionally higher compensation, but after controlling for firm characteristics like firm performance, size, risk, and other unobservable firm characteristics, the season of CEO birth has no impact on compensation. This is consistent with a lack of recognition of the possible influence of birth-date on managerial performance. Some variation exists in the school-year cutoff dates applicable in different states. As a result, CEOs born in the summer may not have been the youngest students in their grade year. Therefore, to check the robustness of our results, we construct three additional variables to capture CEOs relative ages in their respective age cohorts. Based on a sub-sample of CEOs 34 We address the sample selection issue later. We do not expect our sample of birth-dates to be biased in any obvious way. 85

94 where their state of birth and their state s grade cutoff date are both available, we are able to evaluate: First, how many days older a CEO is relative to the hypothetically youngest classmate (who would be born 1 day before the cutoff day and therefore be the youngest possible student in the grade). Based on this we can determine whether or not a CEO was born in the latest possible quarter for enrollment in the school year; Third, we can compare CEO success according to quarter of birth based on state specific school cutoff dates. The robustness tests using these three alternative measures of CEO relative age generate consistent predictions. We provide an explanation of the empirical relevance of season of birth both for the disproportionately higher number of non-summer born CEOs, and for the better performance of CEOs born in summer. The explanation does not rely on there being any eventual difference in management skills in adulthood according to the season of a person s birth. The explanation relies only on the advantage that non-summer born have in being selected for activities that provide them with experiences which benefit them in the competition to become CEOs, and on the need for summer-born to distinguish themselves among their cohort. A simple model is used to demonstrate how both of these effects occur. The paper is organized as follows. Section 4.2 discusses the extensive literature on the relative-age or birth-date effect. Section 4.3 considers the relevant season of birth. Section 4.4 explains how birth-date data for S&P 500 CEOs were collected and organized. In Section 4.5 we consider the prevalence and performance of summer-born CEOs and perform robustness checks to support our findings. Section 4.6 turns to the explanation for what we have found. Section 4.7 of the paper concludes. 86

95 4.2 The Relative-age or Birth-date Effect It is probably fair to say that of the literally hundreds of research papers documenting the relative-age or birth-date effect, the vast majority focus on sport. 35 Sport is an activity in which children are grouped by age, usually in one-year categories, in order to control for age-related differences in physical and intellectual development. The relevance of relative-age for long-run success in sports has been documented in just about every sport played, from soccer, to basketball, to ice hockey, to baseball, to tennis. Taking sports in no particular order, we can begin with soccer where a study by Glamser and Vincent (2004) showed that a disproportionate number of elite youth soccer players in the United States were born early in the school year, making them generally older than their teammates. The effects of relative-age continue into professional soccer, as has been shown by Ashworth and Heyndels (2007) who examined wage levels and showed that players born late after the cutoff date making them older than teammates earned more than other professional soccer players. The importance of the relative-age effect in ice hockey has been studied by Baker and Logan (2007) who considered the selection order in the annual National Hockey League (NHL) draft. As in this paper, athletes were divided into four quartiles by the season of their birth. It was determined that relative age still plays a role at the time of the NHL draft. This study followed an earlier investigation of ice-hockey success by Barnsley, Thompson, and Barnsley (1985). In this study the birth months of Ontario Hockey League and Western Hockey League players were organized by month. The authors showed an almost straight line decline in numbers reaching these leagues, which are just below the NHL in prestige, from team members born in January 16 percent of the players to those born in December, about 4 percent. In English-speaking Canada, the school year cutoff is December 31, so children born in January are the oldest and those born in December are the youngest. 35 However, as we shall see, the birth-date effect has been documented in other endeavors, including academic success. See Bedard and Dhuey (2006). 87

96 With height being a highly favored characteristic in basketball it is little surprise that a relative-age effect has been observed. Esteva, Dobnic, and Puigdellivol (2008) examined the birth-dates of Spanish youth and professional players as well as National Basketball Association (NBA) players in the 2004/05 season, and the best 50 players in NBA history. By sorting birth-dates into four categories, or seasons, and applying a chi-quadrate test, a relative-age effect was identified which extended right up to the all-time very top NBA professionals. A more extensive study of the relative-age effect in basketball involving over 150,000 male players and 100,000 female players in France by Delorme and Raspaud (2008) confirmed the importance of the calendar quarter of the year of birth. The developmental benefit enjoyed by those born in the first few months of the selection year has been documented in baseball by Thompson, Barnsley, and Stebelsky (1991). The relative-age effect is confirmed in a population of 837 major league baseball players, again showing that season of birth can have a lifelong effect. Another sport where the relative-age effect has been shown to survive into adulthood is tennis. This involved investigation of birth-dates of nationally ranked junior tennis players in the United States by Giacomini (1999). Even in a non-team sport such as tennis, those born earlier in a year are older and enjoy advantages. The importance of the relative-age effect has been shown to extend well beyond sport, and to be found in academic performance throughout school and even in the likelihood of attending university. For example, Bedard and Dhuey (2006) have documented the effect across OECD countries: older children perform several percentiles better than younger class members. Other country-specific studies confirm these effects, including the impact on income. 36 Most importantly for the evidence presented later in this paper, it has also been shown that relative-age has a significant effect on high school leadership activities. To quote the abstract from one 36 For example, for the situation in Japan see Kawaguchi (2006). For Germany see Jurges and Schneider (2008). For Britain, see Hutchison and Sharp (1999). The issue of disadvantaged summer-born children recently gained so much attention in Britain that the Education Secretary launched a review in January 2008 to see what might be done to help them. 88

97 research paper in this area, school entry cutoffs induce systematic within grade variation in student maturity, which in turn generates differences in leadership activity. We find that the relatively oldest students are 4-11 percent more likely to be high school leaders, (Dhuey and Lipscomb, 2008). 4.3 The Relevant Seasons of Birth The predominant practice in the research cited above is to divide the year into the four calendar quarters. The youngest in class are taken as those born in the summer, July through September, the third quarter of the year. The oldest are taken as those born in the fourth quarter. This is based on the prevailing cutoffs for entry for either kindergarten or first grade. In this regard it is worthy to note that thirty-seven of the U.S. states plus Puerto Rico have kindergarten entry cutoff dates between August 31 and October This would make the children born in July and August among the youngest in class, with September born in many cases also being younger. Those born after mid-october through the end of the year would be among the oldest. However, the fourth quarter could contain some of the younger students. Nevertheless we would expect the number of younger children to be relatively higher in the third quarter than the fourth quarter. The ranking in terms of average age is not in doubt for the first and second quarters, with these containing the second and third youngest children. We follow the universal convention in the relative-age research of classifying the seasons as each of the four calendar quarters. However, as a robustness check we also present results based on excluding the month of possible age ambiguity, September. We do this by considering pairs of months rather than calendar quarters: the youngest in class have July and August birthdays. We again find the relative-age effect to be present. As a yet further robustness check, for those situations where we have CEO state of birth we also recalculate relative age according to the hypothetically youngest possible student to be admitted to the CEO s 37 Education Commission of the States, State Notes Kindergarten. The relevant, detailed data can be found at 89

98 grade. 38 That is, relative age is recalculated, where possible, to reflect heterogeneity of states cut-off ages for grade entry. Our conclusions are shown to be robust to these alternative measures of relative age. 4.4 Data Employed: CEO Birth-dates and Firm Characteristics In order to investigate the possible presence of a relative-age effect among highly successful U.S. CEOs we construct a birth-date dataset for the CEOs of S&P 500 companies between 1992 and Based on ExecuComp, we first identify the names of the CEOs, and then search for their birth-date and educational background in Biography Resource Center. Biography Resource Center is a database providing comprehensive biography information of notable individuals in business, art, government, and other endeavors. We were able to identify birth-dates and education backgrounds of 321 CEOs. Although the source we employ does not provide the birth information of all CEOs we do not expect our sample of birth-dates to be biased in any obvious way. Bias would require the likelihood of birth-date information being available to differ according to the quarter of the year of birth. We can think of no reason why this would happen. Stock price data were obtained from CRSP and accounting data from Compustat. The sample we use containing the 321 CEOs for which we have birth-dates provides 2168 firm-year observations. As for the financial data, following Gompers, Ishii and Metrick (2003) and others we measure Tobin s Q as the market value of assets divided by book value of assets. We also measure the market-to-book (M/B) ratio as the market value of equity over book value of equity. The market value of equity (MV Equity) is computed as the yearly closing share price multiplied by the number of shares outstanding. We define Volatility as the stock return standard deviation using 38 Our sample size is reduced in the construction of this more precise measure because we lack birth states for some CEOs, or because some CEOs were born outside the United States or in states with flexible admissions. 90

99 monthly returns of the preceding three years. Return on assets (ROA) is measured as the ratio of operational income before depreciation over total assets. We compute Leverage as the ratio of long-term debt over total assets. Capital Expenditure is the firm s yearly capital expenditure normalized by total assets. MBA is a dummy variable, taking value of one if the CEO holds an MBA degree and zero otherwise. Summer is a dummy variable which equals one if the CEO is born in the summer, consisting of July, August and September when calendar data are considered (Relative age allowing for state differences involves a more complex structure as we shall explain). All the monetary variables in the sample are measured in 2006-constant dollars. To ensure outliers in the data are not driving our results, we winsorize all the continuous variables at the 1 st and 99 th percentiles. Panel A of Table 23 reports the characteristics of the firms in our sample. The median firm has a Tobin s Q of 1.49 and its M/B is During the sample period the firms are performing well with a median Stock Return of 17.3% and ROA of 13.2%. Moreover, the median firm is moderately levered with a Leverage of 16.1%; its market value of equity is $12.71 billion; it makes Capital Expenditure of 4.3% and experiences Volatility of FirmSize is defined as the natural logarithm of market value of equity. The mean and median values for CEOage are respectively 56.7 and 57. The MBA dummy takes an average value of 0.39, indicating that 39% of the sample CEOs have an MBA. In Panel B of Table 23, we further describe firms characteristics classified by their CEO birth seasons. The firms with summer-born CEOs tend to have higher market valuation, better performance, bigger stock volatility and capital expenditure, but smaller market value of equity and lower financial leverage. Table 23 reports the correlation matrix of explanatory variables. All the correlation coefficients are below 0.5 in magnitude. 91

100 Table 23 Descriptive Statistics Panel A: Descriptive Statistic of Firm Characteristics The sample consists of 2168 firm-year observations managed by 321 CEOs from 1992 to All the sample firms are S&P 500 companies. We obtain stock price data from CRSP and accounting data from Compustat, and collect the CEO birth data from the Biography Resource Center. Following Gompers et al. (2003), Tobin s Q is computed as the market value of assets divided by book value of assets. M/B is the ratio of market value of equity over book value of equity. Summer is a dummy variable which is equal to 1 if the CEO was born in July, August, or September. RelativeAge is how many days a student is older than the hypothetical youngest student in the class. Latestqt is a dummy variable which is equal to 1 if RelativeAge is less than 90 days (roughly one quarter). Relativeqt is a categorical variable and is constructed based on quartiles of RelativeAge. Relativeqt is equal to 1 if the values of RelativeAge fall into the first quartile of RelativeAge, equal to 2 if the second quartile, equal to 3 if the third quartile, and equal to 4 if the last quartile. StockReturn is the annual stock return of the firm. Volatility is the stock return standard deviation based on the monthly return of past three years. ROA is the accounting return of assets, obtained as the ratio of earnings before interest and taxes to total assets. Leverage is the ratio of long-term debt (book value) over total assets. MV Equity ($billion) refers to the market value of equity computed as the yearly closing stock price multiplied by the number of shares outstanding. Capital Expenditure is the firm s yearly capital expenditure normalized by total assets. CEOage measures the age of the CEO. MBA is a dummy variable, taking the value of one if the CEO has an MBA degree and zero otherwise. All the dollar-value variables are measured in 2006-constant dollars. Mean Std 5 th Pct Median 95 th Pct No. of Obs. Tobin s Q M/B Summer RelativeAge Latestqt Relativeqt Stock Return 18.4% 30.9% -30.8% 17.3% 69.3% 2168 Volatility ROA 14.1% 8.3% 2.4% 13.2% 30.3% 2168 Leverage 18.3% 13.5% 0.1% 16.1% 42.1% 2168 MV Equity ($B) Capital Expenditure 5.2% 4.4% 0 4.3% 14.3% 2168 CEOage MBA

101 Panel B: Firm Characteristics Classified by CEO Birth Season The sample consists of 2168 firm-year observations managed by 321 CEOs from 1992 to All the sample firms are S&P 500 companies. In the sample, 384 firm-year observations are managed by 66 CEOs who are born in the summer season, which includes July, August and September. The middle column and final column give the difference of the two means and the two medians, respectively. The tests of means are based on t-statistics; the tests of medians are based on Wilcoxon signed tests. The notation ***, ** and * denote statistical significance at the 1%, 5% and 10% level, respectively. Mean Median Summer (1) Non Summer (2) Difference (1)-(2) Summer (4) Non Summer (5) Difference (4)-(5) Tobin s Q ** M/B Stock Return ** * Volatility *** *** ROA * Leverage *** *** MV Equity *** Capital Expenditure *** ** CEOage * MBA *** *** Table 24 Correlation Matrix The sample consists of 2168 firm-year observations managed by 321 CEOs from 1992 to Variables used in this matrix are defined in Panel A of Table 23. Summer takes the value of one if the CEO is born in the summer, and zero otherwise. FirmSize is defined as the natural logarithm of market value of equity. Correlations with absolute value greater than 0.1 are significant at the 1% level. (1) (2) (3) (4) (5) (6) (7) (8) (9) Summer (1) 1 ROA (2) FirmSize (3) CEOage (4) Volatility (5) MBA (6) Leverage (7) Capital Expenditure (8) Stock Return (9)

102 4.5 The Prevalence and Performance of CEOs by Birth Season A. Distribution of CEOs by Birth Seasons Figure 1 shows the number of CEOs sorted by birth season. For convenience, January to March, April to June, July to September, and October to December are respectively referred to as winter, spring, summer and fall. We see 66 of the 321 S&P 500 CEOs were born in the summer, the smallest number among the seasons. The season with the most CEOs is winter with 92, followed by fall with 82 CEOs and spring with 81 CEOs. This is close to what we would expect from a relative-age effect, particularly where there is some blurring between summer and fall around the variation across U.S. states with regard to September. Most importantly, we find that only around 20% of the sample CEOs were born in the summer. Figure 1 Number of CEOs by Birth Season This figure reports the number of CEOs classified by their birth seasons. The sample consists of 321 CEOs of S&P 500 companies from 1992 to Spring includes April, May, and June. Summer includes July, August, and September. Fall includes October, November, and December. Winter includes January, February, and March. 94

103 In determining the statistical significance of the number of summer born in the dataset of 321 CEOs we define a summer born dummy, summerceo, as one if the CEO is born in the summer and zero otherwise. With the null-hypothesis that the population birth number is uniformly distributed across the four seasons, we conduct a two-tail t-test versus the null that summerceo = The null is rejected with a corresponding p-value of Figure 2 Season of Birth: CEO Sample versus U.S. Population This figure compares the proportion of CEOs born in the four seasons to that of the U.S. population. The CEO sample consists of 321 CEOs of S&P 500 companies from 1992 to The U.S. population sample consists of all births from 2000 to Spring includes April, May, and June. Summer includes July, August, and September. Fall includes October, November, and December. Winter includes January, February, and March. The test above assumes that the distribution of births is uniform throughout the year. In actual fact, however, United States births follow a seasonal pattern. In particular, by examining birth information of the U.S. population from 1978 to 1987 Nunnikhoven (1992) showed that there is above-average birth frequency in the summer. This means that if we compare the seasonal summer-born CEO numbers with the actual summer-born numbers, the presence of a particularly low number will be more striking. 95

104 In order to judge the relevance of comparing CEO birth frequencies with the actual birth seasonality we consider birth information for the U.S. population between 2000 and 2005 from the National Center for Health Statistics (NCHS). 39 NCHS provides information on the number of births each month from In Figure 2, we compare the proportion of CEOs born in the four seasons with the proportions for the U.S. population. Consistent with Nunnikhoven (1992), summer has higher birth frequency for the population than other seasons, with 26.3% of births being in the summer. It is apparent that the result that fewer CEOs are born in the summer is not because of the seasonality of births. Indeed, when we re-do the t-test with the null hypothesis: summerceo=0.263 the null hypothesis is rejected and the corresponding p-value is This greater significance is no surprise since we are comparing 20% of CEOs to 26.3% of the population born in summer rather than 25%. B. Season of CEO Birth and Market Valuation In this section, we investigate the relationship between the season of CEO birth and firms valuations. We control for potentially important firm characteristics, including ROA, FirmSize, Volatility, Leverage, and Capital Expenditure. We also include MBA and CEOage to control for CEO education background and age: Betrand and Schoar (2003) have shown that these two variables significantly influence managerial style. Moreover, we use year dummies to capture macroeconomic and time trend effects, as well as firm fixed effect to control for unobserved heterogeneity and industry effects. Specifically, we estimate the following regression: Valuation a a Summer a ROA a FirmSize a CEOage a5volatilityit 1 a6mbait a7 Leverageit 1 a8captialexpenditureit 1 YearDummy Firm Fixed Effect it 0 1 it 2 it 1 3 it 1 4 it it (1) where i indexes firms and t indexes years. The dependent variable is the firm s valuation, 39 The birth information can be found at 96

105 measured by Tobin s Q or by the M/B ratio. The primary independent variable of interest is Summer, a dummy which equals one if the firm s CEO is born in the summer and zero otherwise. Estimating a significant positive coefficient for a1 would reject the null hypothesis that CEOs born in the summer are equally successful as CEOs born in the other seasons. Table 25 shows a positive relation between firms valuations and the Summer dummy; the relation is both statistically and economically significant. The dependent variable in Column (1) is Tobin s Q, where the coefficient of Summer is 0.36 and is significant at the 1% level. The interpretation of this coefficient is as follows: since we control for firm fixed effects, the result indicates that, within a firm, when the CEO is summer born so the dummy, Summer, goes from zero to one, Tobin s Q increases by 0.36 compared to the sample median of The median firm in our sample has a market value of assets (equity and total debt in 2006 dollars) of approximately $30.4 billion, and so the positive effect on Tobin s Q of 0.36 means an enhanced firm value of about $7.34 ( /1.49) billion. Clearly, this result is very economically significant. The M/B ratio is considered as an alternative measure of the effect of a CEO being summer born in Column (2). The variable Summer has a coefficient of 0.55 which is significant at the 1% level. This indicates that for a firm, if its CEO was born in the summer so that the dummy, Summer, goes from zero to one, there is an associated increase in M/B of Since our median sample firm has a market value of equity of $12.7 billion and M/B of 2.56, an increase in M/B of 0.55 is associated with an increase in firm value of approximately $2.73 ( /2.56) billion. Again, this is economically significant. The coefficients on the control variables in the regression are generally consistent with those in the existing literature. As shown in the table, firms tend to have higher market valuation 97

106 Table 25 Season of CEO Birth and Firms Valuation The sample consists of 2168 firm-year observations from 1992 to In the sample, 384 firm-year observations are managed by the CEOs who are born in the summer season. The summer season includes July, August and September. The dependent variables are Tobin s Q and M/B ratio. Tobin s Q is computed as the market value of assets divided by book value of assets. M/B is measured as the ratio of market value of equity over book value of equity. Summer takes the value of one if the CEO is born in the summer, and zero otherwise. Column (1) reports the results of the regression with Tobin s Q as the dependent variable, while Column (2) reports the regression results in which M/B is the dependent variable. Both regressions use an OLS model controlling for firm fixed effects. P-values are reported in brackets. We use the notation ***, ** and * to denote statistical significance at the 1%, 5% and 10% level, respectively. (1) (2) VARIABLES Tobin s Q M/B Summer 0.36*** 0.55** (0.003) (0.05) ROA 9.05*** 15.7*** (0.000) (0.000) FirmSize 0.12*** 0.33*** (0.01) (0.002) CEOage -0.02*** -0.04*** (0.001) (0.008) Volatility 4.02*** 4.25 (0.000) (0.12) MBA Degree -0.27*** -0.73*** (0.01) (0.004) Leverage (0.88) (0.16) Capital Expenditure (0.66) (0.24) Year Dummy Yes Yes Firm Fixed Effect Yes Yes Constant 0.31*** -0.23*** (0.000) (0.000) N Adjusted R2 15% 12% when they have better accounting performance, when they are bigger firms, and when they experience higher stock volatility. Older CEOs tend to lower market valuations. Surprisingly, MBA degrees are associated with lower market valuation. The negative coefficient of MBA may 98

107 share the same cause of the underperformance of non-summer born CEOs: Only exceptionally capable people can manage to become CEOs without an MBA degree and conditional on becoming CEOs, those without MBA degrees are better at increasing firms valuation. Most importantly, the results in Table 25 strongly support the relative-age effect in the context of senior level corporate management, with CEOs born in the summer creating higher market value than CEOs born in other seasons. C. Season of CEO Birth and Stock Returns This section investigates the relation between the season of CEO birth and stock returns. In order to do this, we construct portfolios of companies based on the CEOs birth seasons. In particular, we construct a Spring Portfolio, a Summer Portfolio, a Fall Portfolio and a Winter Portfolio. We also construct a non-summer Portfolio as a value-weighted portfolio of stocks of companies whose CEOs are born in the spring, fall, and winter. All of these portfolios are reset once every year according to the birth-dates of the CEOs. Figure 3 shows the stock performance of portfolios being selected by CEO birth season. We find that an investment of $1 in the Summer Portfolio on January 1, 1992, the beginning of our data period, would have grown to $27.4 by December 31, In contrast, a $1 investment in the non-summer Portfolio would have grown to $14.8 over this period. This is equivalent to annualized returns of 24.7% for the Summer Portfolio and 19.7% for the non-summer portfolio, a difference of approximately 5% per year. The Winter, Spring, and Fall Portfolios perform very similarly to each other; their annualized returns are 21.2%, 18.6%, and 19.2%, respectively. It is apparent that stocks of companies with CEOs born in the summer outperform stocks of companies with CEOs born in the other seasons. To ensure that differences in riskiness or style of the portfolios are not driving the performance 99

108 differences, we run the following four factor model of Carhart (1997) over the daily portfolio returns: R RMRF SMB HML Momentum (2) t 1 t 2 t 3 t 4 t t where Rt is the excess return of a certain portfolio relative to the risk-free rate on day t. The first factor RMRF is the day t value-weighted market return minus the risk-free rate. t Figure 3 Performance of Portfolios by CEO Birth Season This figure shows the stock performance of the portfolios classified by CEO birth seasons. The Y axis represents the dollar value by December 31, 2006 for a $1 investment in the portfolio on January 1, The corresponding equivalent annualized rates of return are reported in parentheses. The sample consists of 2168 firm-year observations from 1992 to In the sample, 384 firm-year observations are managed by the CEOs who are born in the summer season. We construct stock portfolios based on the CEO s birth season. In particular, Spring Portfolio, Summer Portfolio, Fall Portfolio, Winter Portfolio, and non-summer Portfolio are the value-weighted portfolios of stocks whose CEOs are born in the spring, summer, fall, winter, and non-summer respectively. The second and third factors, SMBt and HML t, are based on Fama and French (1993) and represent the difference in returns between portfolios of small versus large firms, and between portfolios of high and low book-to-market ratios, respectively. The momentum 100