Performance reversals and attitudes towards risk in the venture capital (VC) market

Performance reversals and attitudes towards risk in the venture capital (VC) market

Journal of Economics and Business 62 (2010) 537–561 Contents lists available at ScienceDirect Journal of Economics and Business Performance reversa...

324KB Sizes 0 Downloads 7 Views

Journal of Economics and Business 62 (2010) 537–561

Contents lists available at ScienceDirect

Journal of Economics and Business

Performance reversals and attitudes towards risk in the venture capital (VC) market Oghenovo A. Obrimah ∗, Puneet Prakash Virginia Commonwealth University, United States

a r t i c l e

i n f o

Article history: Received 31 December 2008 Received in revised form 23 March 2010 Accepted 10 June 2010 JEL classification: G24 G11 Keywords: Venture capital Performance reversals Portfolio risk Overconfidence Syndication Investment staging Reputation

a b s t r a c t In this paper, we find evidence of reversals in relative exit performance between the “short” and “long-run” in the VC market, with the short-run defined to be the first 5 years of business, and the long-run, the 6th year of business onwards. Using proxies for the risk of venture capital assets that are derived from VCs’ risk management strategies—investment staging and deal syndication—we find reversals in relative performance are explained by the coexistence of persistent and non-persistent attitudes towards risk along two complementary dimensions. First, while syndication strategy is persistent, attitudes towards portfolio risk reverse between the short- and long-run periods, such that syndication is associated with high risk portfolios in the short-run, and low risk portfolios in the long-run. Second, VCs that hold high quality portfolios in the short-run do not persist with this strategy; however, high quality portfolios are associated with superior long-run exit performance. The performance effects of the transitions in attitudes towards risk that we observe are consistent with theories of costly reputation building in markets characterized by adverse selection problems and empirical evidence that VCs’ deal screening skills are more important for success than advisory or monitoring skills. © 2010 Elsevier Inc. All rights reserved.

In markets that are at least weak-form efficient, past performance is not expected to have a significant impact on future performance. Hence, both performance persistence (momentum trading), as initially documented in Jegadeesh and Titman (1993), and performance reversals, as initially documented in De Bondt and Thaler (1985), are market anomalies that have been documented in public equity markets. ∗ Corresponding author at: Snead Hall, 301 West Main Street, P.O. Box 844000, Richmond, VA 23284-4000, United States. Tel.: +1 804 501 8195; fax: +1 804 828 3972. E-mail address: [email protected] (O.A. Obrimah). 0148-6195/$ – see front matter © 2010 Elsevier Inc. All rights reserved. doi:10.1016/j.jeconbus.2010.06.002

538

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

In the venture capital (VC) market, Kaplan and Schoar (2005) examine the relation between past and future performance, and find evidence for performance persistence. That is, follow-on funds raised by VCs whose initial funds perform well, continue to perform well, based on returns delivered to investors. In a complementary study that compares the returns delivered by VCs to different classes of investors, Lerner, Schoar, and Wong (2005) find some investors (“smart” institutions) are better at picking winning funds relative to other investors such as banks and investment advisors. Specifically, Lerner et al. (2005) find that smart institutions are less likely to invest in a follow-on fund raised by a VC in which they are currently invested, suggesting, in spite of the documented persistence in individual VCs’ performance, that relative performance is not persistent in the VC market. In this paper, we study the evolution of relative exit performance in the VC market. Using a VC firm’s excess performance with respect to exit success (exit performance relative to all firms that commenced business in the same year) as the measure of relative performance, we examine the relation between initial and future exit performance within the VC market. Given the empirical findings in Metrick and Yasuda (2007), and Gompers, Kovner, Lerner, and Scharfstein (2008), we consider the first 3 or 5 years of business as potential time frames for VCs to establish a venture capital portfolio or track record (for ease of exposition, we label this initial period the “short-run”), and find, by examining the relation between past performance and future fund flows, that the 5-year horizon provides a more robust definition of the short-run. The short-run is thus defined to be the first 5 years of a VC’s life, while the long-run is the 6th year onwards.1 Our empirical findings show a strong negative relation between excess long-run and excess shortrun exit performance, with the observed performance reversals occurring regardless of whether VCs commence business during relatively hot or cold IPO periods. These performance reversals are not explained by survival bias, the notion that superior short-run performers’ deal flow ran dry, or systematic changes in VCs’ exit opportunities. Performance reversals also persist when we exclude the oldest VCs from our venture capital sample. In order to better understand why performance reversals occur within the cross-section of the VC market, we draw on the extant literature, which has documented performance reversals in public equity markets (see for example, Chopra, Lakonishok, & Ritter, 1992; De Bondt & Thaler, 1985, 1987), and the mutual funds industry (see for example, Chevalier & Ellison, 1997; Ippolito, 1992; Sirri & Tufano, 1998). In the mutual funds industry, explanations of performance reversals are linked to the convex relation between future fund flows and past performance (see for example, Berk & Green, 2004; Lynch & Musto, 2003). As in Kaplan and Schoar (2005), however, we find the relation between future fund flows and past performance is concave within the VC market, with average past performers associated with larger fund flows relative to superior past performers. The shape of the flow-performance relation thus rules out explanations of performance reversals that are based on fund flows to superior short-run performers. Explanations of performance reversals observed in public equity markets are either risk based (see for example, Fama & French, 1996; Fama, 1998) or behavioral (see for example, Barberis, Shleifer, & Vishny, 1998; Lakonishok, Shleifer, & Vishny, 1994). Given the returns VCs deliver to investors are largely private information, the VC market need not be weak-form efficient, and explanations of performance reversals can either be risk-based or behavioral. In this study, we focus on a risk-based explanation of performance reversals that is based on the following premise. If all VCs maintain their short-run attitudes towards risk in the long-run, risk-based explanations are insufficient to explain performance reversals. Similarly, if all VCs change their attitudes towards risk in exactly the same manner, risk-based explanations are insufficient to explain performance reversals. If, however, some classes of VCs persist in their short-run attitudes towards risk, while some classes of VCs switch attitudes, the coexistence of persistent and non-persistent attitudes towards risk can be sufficient to explain performance reversals within the VC market.

1

Kaplan and Schoar (2005) find a positive, concave relation between past performance and future fund flows.

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

539

In order to implement risk-based explanations of performance reversals, we require proxies for the risk of venture capital assets. The risk proxy utilized in Kaplan and Schoar (2005) to explain performance persistence is the S&P 500 index. Cochrane (2005) finds, however, that the risk of venture capital is most similar to the risk of small Nasdaq stocks, suggesting the S&P 500 may not be the most appropriate risk proxy for venture capital assets. Consistent with the findings in Cochrane (2005), Kaplan and Schoar (2005) find no significant relation between venture capital returns and returns to the S&P 500 index. Given the dearth of well accepted proxies for the risk of venture capital assets, we adopt two measures of the risk of venture capital assets that are derived from VCs’ risk management strategies; that is, investment staging and deal syndication. As documented in Sahlman (1990) and Gompers (1995), both investment staging and deal syndication are important tools that VCs utilize in managing investment risk. By syndicating investments, VCs are able to take on riskier projects (the risk sharing benefit of syndication) and a larger number of projects, and thereby increase the probability of exit success. As a result, we expect exit performance will increase with syndicate size. Our second proxy for the risk of venture capital assets is derived from investment staging. Investment staging is a risk management tool that enables VCs to minimize losses on projects that go ‘bad’ ex post. Given the option to abandon a project is most valuable during the first round of financing, the capital amounts VCs disburse to projects during the first round of venture financing are proxies for subjective estimates of the value of the option to abandon a project, and by implication, the probability of a successful exit.2 Capital disbursements that are less than VCs’ mean disbursements to portfolio projects indicate lower than average probabilities of a successful exit, while capital disbursements that are larger than VCs’ mean disbursements indicate higher than average probabilities of a successful exit. The existence of demand for VCs (delegated monitors) guarantees that the supply of average quality projects is greater than the supply of high quality projects (see for example, Huang & Litzenberger, 1988), indicating portfolios characterized by similar first round investments are more likely to consist of low and/or average quality projects rather than high quality projects. As a result, low dispersion in capital disbursements is evidence of higher risk or lower quality portfolios, while widely dispersed capital disbursements are evidence of lower risk or higher quality portfolios. In the long-run, we expect exit performance will increase with portfolio quality. However, under the assumption that there are search costs associated with identifying or attracting high quality projects (see for example, Chemmanur & Fulghieri, 1994), exit performance may not increase with portfolio quality in the shortrun. We find reversals in relative exit performance are explained by the coexistence of persistent and non-persistent attitudes towards risk. First, while syndication strategy is persistent, we find attitudes towards portfolio risk reverse between the short and long-run periods, such that syndication is associated with high risk portfolios in the short-run, and low risk portfolios in the long-run. Second, rather surprisingly, we find VCs that hold high quality portfolios in the short-run do not persist with this strategy. In the long-run, however, both high quality portfolios and large syndicates are associated with superior long-run exit performance. Combined, the coexistence of persistent and non-persistent attitudes towards risk results in performance reversals within the cross-section of the VC market. Though surprising, our finding that syndicate size is associated with high risk portfolios in the short-run, and with low risk portfolios in the long-run is consistent with models of costly reputation building in markets characterized by adverse selection problems.3 In Diamond (1989), for instance, economic agents develop reputation by initially taking on risky portfolios. Those that are successful then transition to less risky or higher quality portfolios after establishing some level of market reputation. Within the context of our study and models of costly reputation building, syndication appears to be evidence of risk sharing among new entrant or young VCs, and evidence of market reputation among established VCs. Our reputation-based explanation of transitions in attitudes towards risk is

2 Cochrane (2005) finds the risk associated with venture capital is greatest at the first round of financing. Schwienbacher (2008) also argues that the value of the option to abandon a project is highest when the risk of default or failure is highest; that is, at the first round of venture financing. 3 Such models include Chemmanur and Fulghieri (1994).

540

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

consistent with the finding in Tian (2007) that ventures backed by large syndicates are associated with superior post-IPO performance relative to ventures backed by small syndicates and the association of large syndicates with superior exit performance in Hochberg, Ljungqvist, and Lu (2007). Using proprietary data on actual venture capital returns, Kaplan and Schoar (2005) provide evidence on performance persistence within the VC market. Our finding that exit performance is persistent in syndicate size provides additional evidence on performance persistence within the VC market. Our findings differ from the prior literature, however, in that we also find evidence of non-persistence in exit performance within the VC market, with non-persistence in exit performance explained by changing attitudes towards portfolio risk or quality. The combination of persistent attitudes towards syndication and non-persistent attitudes towards portfolio risk then results in performance reversals within the cross-section of the VC market. Our empirical results, which are based on publicly available data on exit success rates, complement the findings in Kaplan and Schoar (2005), Tian (2007), and Hochberg et al. (2007). Our finding that VC firms associated with high quality portfolios and superior performance during the short-run period do not persist with this strategy indicates these VCs switch from higher to lower quality portfolios between the short and long-run periods. This switch ends up being costly, however, because, in line with expectations, high quality portfolios are associated with superior long-run exit performance. The switch in attitudes towards portfolio quality that we observe is consistent with theories of investor overconfidence. In Post, van den Assem, Baltussen, and Thaler (2008), economic agents transition from risk averse to risk seeking behavior when prior expectations are surpassed by favorable outcomes. In Gervais and Odean (2001), traders become overconfident because they take too much credit for their successes, and are thus induced to take on more investment risk. Daniel, and Hirshleifer, Subrahmanyam (1998), Malmendier and Tate (2005, 2008), and Chuang and Lee (2006) also link overconfidence to risk seeking behavior. Within the context of the VC market, superior shortrun performers that attribute their short-run success to managerial and monitoring abilities, rather than the quality of their portfolio projects (deal screening ability), adopt riskier portfolio strategies in the long-run, resulting in inferior long-run performance. Under the assumption that VCs’ monitoring and managerial abilities cannot make ‘bad’ projects ‘good’, portfolio quality, or equivalently, the ability to identify and/or attract superior quality projects is more important for success in the VC market, in comparison with managerial and monitoring abilities. In this respect, our findings are consistent with the conclusion in Kaplan, Sensoy, and Stromberg (2007) that investors are better off backing horses (business plans) rather than jockeys (entrepreneurs or management teams). Our contributions to the literature on venture capital are twofold. First, using exit rates, we provide evidence of non-persistence in investment performance within the VC market. Second, we find that, over time, interactions between syndicate size and portfolio quality result in the coexistence of persistent and non-persistent attitudes towards risk, the combination of which results in performance reversals within the cross-section of the VC market. The performance effects of the transitions in attitudes towards risk that we observe within the VC market are consistent with theories of reputation building in markets characterized by adverse selection problems and empirical evidence that VCs’ deal screening skills are more important for success than advisory or monitoring skills. The rest of the paper proceeds as follows. Section 1 develops the empirical framework, while Section 2 discusses the data. Empirical results are reported in Section 3, and Section 4 concludes. 1. Empirical framework Standard finance theory predicts the cross-sectional variation in asset returns can be explained using covariance with the market portfolio. The standard version of the capital asset pricing model of Sharpe (1964) and Lintner (1965) can be stated as follows: R¯ i − Rf = −

 where

 dU

dU dU − d / dW W

dW



/

dU dW



ˇip p for all i

(1)

equals the investor’s marginal rates of substitution between expected wealth

and standard deviation (negative sign ensures marginal rates are positive), holding expected utility

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

541

of end of period wealth constant; R¯ i is the average return on risky assets; Rf is the return on the risk free asset; ˇip is the marginal contribution of risky asset i to portfolio standard deviation; and  P is portfolio standard deviation. Asset pricing tests that include past performance as a predictor of future performance modify Eq. (1) as follows: R¯ i,t+1 − Rf,t+1 = −

 dU dW

/

dU dW



ˇip p,t + ip rp,t

for all i

(2)

where period t + 1 is the future period; period t is the past period; rp,t is past performance; ip is the marginal contribution of past performance to future performance; and  p,t is portfolio variance constructed using data from period t. Given venture capital returns are not publicly available, VCs’ exit rates have been utilized as proxies for VCs’ investment performance (see for example, Hochberg et al., 2007; Knill, 2009). In line with the prior literature then, we substitute VCs’ exit success rates for returns in Eq. (2), resulting in the following equation: Ri,t+1 − R¯ t+1 = −

 dU dW

/

dU dW



ˇip p,t + ip rt + ip (p,t · rt ),

for all i

(3)

where R¯ t+1 is the average exit success rate achieved during the long-run period by all venture capital firms that commenced business in the same year as firm i; Ri,t+1 is the exit success rate achieved during the long-run period by firm i; rt are period t excess exit success rates (Ri,t − R¯ t ); and  p,t are three portfolios into which VCs are divided based on period t portfolio risk or quality; and ip is the marginal contribution of the interaction of past performance and past portfolio risk to future performance.

1.1. The construction of excess exit performance and risk proxies In this sub-section, we discuss the construction of VCs’ relative exit performance and risk proxies derived from investment staging or deal syndication. We also discuss the construction of alternative measures of portfolio risk or quality, such as investment stage and industry specialization. As discussed in the preceding sub-section, exit rates have been utilized as proxies for VCs’ investment performance. Our measure of VCs’ relative performance (excess exit rates) is derived from the conventional construction of exit success rates as follows: For each VC firm i, we determine the proportion of its portfolio companies successfully exited via an IPO or acquisition (firm i’s exit success rate) and denote this ri . We then determine the average exit success rate for VC firms that commenced investments during the same year as firm i and denote this r¯ i . Firm i’s excess exit performance, r˜i is then defined to be r˜i = ri − r¯ i . Given average exit rates fluctuate with events in public equity markets (see for example, Gompers et al., 2008), this construction of excess exit rates ensures an exit rate of 40% during a period where the average exit rate is 20% is comparable to an exit rate of 60% during a period where the average exit rate is 40%. This construction of excess exit rates also eliminates the need for year dummies in our empirical tests. Given excess exit rates only shift the conventional construction of exit rates by a constant, the higher (second, third, and fourth) moments of actual and excess exit rates are exactly the same, and our transformation results in no change in the variability of exit performance. Let Cj be capital disbursements to project j during first financing rounds, and let N be the number of projects that received financing from VC i. Then, VC i’s portfolio quality (PQ) is defined to be the standard deviation of Cj divided by average Cj . Within the context of portfolio risk, a low PQ reflects a high degree of project homogeneity, while a high PQ reflects a high degree of project heterogeneity. Given the prediction in Huang and Litzenberger (1988) that VCs will, in equilibrium, mostly attract average quality projects, a low PQ is evidence for higher risk or lower quality portfolios, while a high PQ is evidence for lower risk or higher quality portfolios. In order to construct PQ, all capital disbursements in a particular year are deflated with the appropriate GDP deflator so that each investment is in real not

542

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

nominal terms.4 Average syndicate size is constructed as the average number of VC firms participating in financing deals that involve VC i. Given the finding in Knill (2009) that both investment stage and industry specialization are associated with smaller fund flows and higher exit performance in future periods, we utilize investment stage and industry specialization Herfindahls as alternative measures of portfolio risk or quality. Investment stage specialization is constructed using the Herfindahl (Herfindahl-Hirschman) Index, and is defined to be the sum of squares of the percentage of total investments in small (start-ups or early stage) ventures and large (expansion or later stage) ventures. Industry specialization is also constructed using the Herfindahl Index, and is defined to be the sum of squares of the percentage of total investments in the following five industries: (1) computer related; (2) media, communications, semiconductors, and electrical industries; (3) consumer related; (4) industrial and energy; and (5) biotechnology and medical. Both industry and investment stage classifications are obtained from VentureXpert. Our empirical framework requires the construction of portfolios with different levels of portfolio risk or quality during the short and long-run periods. For each risk or quality proxy then, VCs are divided into three groups consisting of equal numbers of VC firms. The choice of three groups aligns with a framework where portfolio risk or quality is either low, medium, or high. Given the empirical evidence in Kaplan and Schoar (2005) that fund flows to VCs are concave in past performance, the choice of three groups also makes it relatively easy to observe concavity. 2. Data We impose two major restrictions on our data. First, only non-affiliated or independent VCs are included in our sample. Affiliated VCs have strategic reasons for participating in the VC market and their portfolio make-up need not be sensitive to changes in the risk of venture capital. The finding in Bottazzi, Da Rin, and Hellmann (2008) that independent VCs are more active than strategic VCs is consistent with this data restriction. Second, only first round financing transactions are utilized in characterizing the risk and return of venture capital. This restriction is predicated on the findings in Cochrane (2005) and Schwienbacher (2008) that the risk of venture capital is greatest at the first round of venture financing and our contention that the value of the option to abandon a project is greatest at the first round of financing. As a result, portfolio risk proxies are constructed using investment data from first financing rounds only. We also restrict the data to include: transactions by independent VC firms in the U.S. that commenced business between 1980 and 20005 ; transactions where the identities of the VC firms providing financing are disclosed along with the amounts invested; transactions within the U.S. by U.S.-based firms. The restriction to investments within the U.S. enables us to abstract away from differences in macroeconomic risk across countries. Since U.S. VCs’ investment decisions exhibit home bias (see for example, Lerner, 1995), the exclusion of international investments is not expected to bias study findings. The venture capital data are obtained from VentureXpert. 3. Empirical results In this section, we provide empirical evidence of performance reversals, and demonstrate that these reversals are explained by the coexistence of persistent and non-persistent attitudes towards risk within the cross-section of the VC market. Given empirical findings in Metrick and Yasuda (2007) and Gompers et al. (2008) indicate it takes between 3 and 5 years for VCs to become established, the first 3 or 5 years of business are regarded as “short-run” periods, while the 6th year and onwards is regarded as the “long-run”. First, we demonstrate that performance reversals occur in the VC market,

4

GDP deflator data are obtained from the World Development Indicators CD ROM. We exclude funds that commenced business prior to 1980 due to the relative thinness of the data with respect to the number of new funds started in each year, incompleteness and data bias issues prior to 1975 as discussed in Gompers and Lerner (2004). We exclude funds that commence business after 2000 to coincide with the end of the stock market boom of the late 1990s, and also to allow at least 5 years for VCs to achieve some success. 5

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

543

regardless of whether the short-run (the time it takes VCs to develop a portfolio or track record) is defined to be the first 3 or 5 years of a VC’s life. Next, we show that the observed performance reversals are explained by interactions between syndicate size and portfolio quality between the short and longrun periods. In robustness tests, we find no evidence for explanations of performance reversals that are based on survival bias or the notion that superior short-run performers’ deal flow ran dry after the first 5 years of business. We also demonstrate that alternative measures of portfolio risk, such as industry or investment stage specialization are not substitutes for either syndicate size or portfolio quality. 3.1. Exit rates by firm founding year Using the definition of the short-run as the first 5 years of business, in Table 1, we report average exit rates for all VCs that commenced business in a particular year, as well as average excess exit rates Table 1 Exit rates by firm founding year. Firm year (I)

# of firms (II)

1980 20 1981 38 1982 37 1983 40 1984 32 1985 22 1986 23 1987 28 1988 12 1989 16 1990 10 1991 7 1992 14 1993 23 1994 25 1995 47 1996 63 1997 69 1998 58 Mean for 1980–1991 Mean for 1992–1998 Overall mean

Average yearly 1st day returns to IPOs (III)

23.36 16.09 14.91 13.86 11.04 11.44 11.51 12.92 13.38 14.55 14.55 15.85 15.67 15.94 18.58 30.33 36.18 35.58 35.42 14.45 26.81 19.00

Short-run performance

Long-run performance

Mean success rates (%) by firm year (IV)

Mean excess exit rates (%) for “superior” performers (V)

Mean success rates (%) by firm year (VI)

Mean excess exit rates (%) for “superior” performers (VII)

0.2909 0.0901 0.2437 0.1326 0.1637 0.0385 0.0786 0.1104 0.0149 0.2690 0.0173 0.0211 0.0959 0.0683 0.1804 0.2634 0.2559 0.1767 0.0909 0.1225 0.1616 0.1369

0.2707 0.1439 0.1193 0.1140 0.1291 0.0694 0.1959 0.2307 0.0921 0.5695 0.0732 0.1336 0.0708 0.0831 N/Aa 0.1760 0.1184 0.0588 0.2108 0.1784 0.1196 0.1588

0.2403 0.3184 0.1763 0.3212 0.2305 0.3912 0.3039 0.3084 0.3271 0.1045 0.2603 0.1679 0.2049 0.1764 0.0539 0.0144 0.0379 0.0258 0.0848 0.2625 0.0854 0.1972

0.2408 0.2335 0.2065 0.2096 0.2811 0.3127 0.2691 0.2495 0.2431 0.1788 0.2777 0.1377 0.1998 0.0909 0.1088 0.1001 0.4093 0.5202 0.2045 0.2366 0.2333 0.2354

This table reports descriptive statistics for excess exit success rates computed for both short-run (first 5 years of a firm’s life) and long-run (6th year onwards) investment horizons. Excess exit success rates measure a firm’s excess exit performance relative to the average success achieved by firms that started business in the same year, and are computed as follows. For each firm we determine the proportion of portfolio companies successfully exited either via an IPO or third party sale as in Gompers et al. (2008), and denote this exit rate ri We also determine average exit success rates for VCs that started business during each sample year y, y ∈ [1980, 1998], and denote these r¯ i . For each firm i that commenced business in year y, the firm’s excess exit success rate relative to peers, r˜i is defined to be r˜i = ri − r˜i . VCs are sorted into terciles based on their short and long-run performance relative to peers. VCs in the bottom, middle, and top terciles of excess performance are termed inferior, average, and superior short-run performers, respectively. We report average exit success rates for each firm year (the year a VC firm commenced business), as well as excess exit success rates for VCs that started business in a particular year, and are classified as superior performers. Average yearly initial returns to IPOs are constructed using data on average monthly first day returns to IPOs obtained from Jay Ritter’s website. The data are obtained from VentureXpert, and sample VC firms commenced business between 1980 and 1998 (in order to facilitate the measurement of long-run performance). The total number of VC firms is 584. a No firm founded in 1994 is classified as a “superior” short-run performer.

544

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

for superior performers by firm founding year. The results in columns IV and V show VC firms that commenced business in 1980 achieved an average exit rate of 29.09% (3 out of 10 investments) in the short-run and 24.03% in the long-run. Four out of these 20 firms are classified as superior shortrun performers, and exceed the average exit rate of their cohort firms by 27.07%. The combination of average and excess exit rates thus results in a short-run exit rate of 56.16% (29.09% + 27.07%) for superior short-run performers that commenced business in 1980. For superior long-run performers that commenced business in 1980, the results in VII indicate these VCs exceeded the average exit rate of their cohort firms by 24.08%. Combined with an average long-run exit rate of 24.03% in column VI, superior long-run performers achieved a long-run exit rate of 48.11%. In the entire sample, average exit rates are 13.69% during the short-run period (column IV). Given excess exit rates of 15.90% for superior short-run performers, superior short-run performance is associated with an average exit rate of 30%. The combination of average and excess exit rates in columns VI and VII results in average exit rates of about 44% for superior performers during the long-run period. The finding in Gompers et al. (2008) that events in public equity markets have an impact on VCs’ exit opportunities and investment decisions suggests VCs that commence business during relatively cold or hot IPO periods face different exit opportunities and investment opportunity sets. It is possible then that the occurrence of performance persistence or reversals depends on the period during which VCs commence business. In order to determine whether performance persistence or reversals are cyclic, we divide VC firms into two groups, depending on the extent to which IPO markets were hot during their first year of business. Column II shows the number of new firms entering the VC market increases monotonically between 1991 and 1997, while column III shows average yearly initial returns to IPOs are significantly higher during the 1992–1998 period (average of 26.81%) relative to the 1980–1991 period (average of 14.45%). Given these patterns in the number of new VC firms and IPO underpricing, we classify the period, 1980 through 1991 as a relatively cold IPO period, and the period, 1992 through 2000 as a relatively hot IPO period. For VC firms that commenced business during the cold IPO period, average exit rates increase from 12.25% during the short-run period (column IV) to 26.25% during the long-run period (column VI). Given the cold IPO period is immediately followed by the hot IPO period, the increase in average exit rates is consistent with the expectation that exit rates are higher during hot IPO periods. For VC firms that commenced business during the hot IPO period, average exit rates decrease from 16.2% during the short-run period to 8.5% during the long-run period, in line with the observation that the hot IPO period of the late 1990s is immediately followed by the stock market bust during the year 2000. 3.2. Performance reversals in the VC market In Table 2, we report results from empirical tests that examine the relation between excess longrun and excess short-run exit performance within the VC market. In line with the findings in Gompers et al. (2008) and Kaplan and Schoar (2005) that it takes 3–5 years for new entrant VCs to become relatively established or begin raising new funds, we consider 3- and 5-year horizons as potential time frames for VCs to establish a venture capital portfolio or track record within the VC market (for ease of exposition, we label this initial period the “short-run”). The long-run is thus defined to be the fourth or 6th year of business onwards. In addition to the continuous version of excess short-run performance, VCs are divided into terciles (three groups) consisting of equal numbers of VCs, based on excess short-run exit performance. In ascending order from one to three, VCs in these groups are labelled “inferior”, “average”, and “superior” short-run performers, respectively. The two regression models utilized in Panels A and B of Table 2, respectively, are specified as, Performi,t+1 = ˛0 + ˛1. Performi,t

(4a)

Performi,t+1 = ˛0 + ˛1. PerformTerci,nt

(4b)

where Performi,t+1 is excess long-run exit performance, Performi,t is excess short-run exit performance, and PerformTerci,nt are the three terciles into which VC firms are divided based on excess short-run exit performance. The regressions are implemented using censored Tobit models.

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

545

Table 2 The relation between excess long-run and short-run exit performance in the venture capital (VC) market. Long-run performance Cold IPO period firms

Hot IPO period firms

I

III

IV

−0.1428c (−3.90) −0.1504c

−0.2489c

II

Panel A: using actual short-run exit performance Actual 5-year performance −0.3078c (−5.10) Actual 3-year performance −0.3766c (−5.57) Constant −0.0100 −0.0871c –0.0001 (−0.85) (−6.75) Model p-value 0.0000 0.0000 # of VC firms 180 188 Panel B: using terciles of short-run exit performance Average 5-year performers −0.0414 (−1.28) Superior 5-year performers −0.1183c (−4.06) Average 3-year performers 0.0401 (1.14) Superior 3-year performers −0.1005c (−3.11) Constant 0.0399 −0.0787c (1.61) (−2.77) Model p-value 0.0001 0.0000

−0.0665c –0.0081 (−0.02) 0.0001 184

Full sample VI

(−7.06) −0.2876c (−4.58) −0.0787c (−13.82) 0.0000 247

(−8.25) (−1.31) 0.0000 364

(−13.50) 0.0000 435

−0.0340b (−2.08) −0.0847c (−5.71)

−0.0318b (−2.29) −0.0444c (−3.42)

0.0263c (2.96) 0.0024

V

−0.0335c (−2.64) −0.0490c (−4.46) −0.0393c (−5.01) 0.0001

0.0308c (2.70) 0.0000

−0.0070 (−0.45) −0.0855c (−6.18) −0.0488c (−4.50) 0.0000

This table reports results from empirical tests that examine the relation between excess long-run and short-run exit performance in the VC market. The dependent variables are firm-level excess exit rates achieved from the 4th or 6th year of business onwards (the long-run), while the independent variables are excess exit rates achieved during the first 3 or 5 years of business (the short-run). Our regression models are specified as follows: Model 1 (Panel A): Performi,t+1 = ˛0 + ˛1. Performi,t Model 2 (Panel B): Performi,t+1 = ˛0 + ˛1. PerformTerci,nt where Performi,t+1 and Performi,t are excess long-run and short-run exit performance (continuous variables), and PerformTerci,nt (n = 1, 2, 3) are three terciles into which VCs are divided based on excess exit performance during the first 3 or 5 years of business. The tests are implemented using Tobit models. Coefficients reported in Panel B are relative to VCs in the first (bottom) tercile of excess exit performance, that is, inferior short-run performers. Excess exit success rates are constructed as follows: For each VC firm i, we determine the proportion of its portfolio projects successfully exited via an IPO or third party acquisition and denote this ri We then determine the average exit success rate for VC firms that commenced business during the same year as VC firm i and denote this r¯ i . Firm i’s excess performance, r˜i is then defined to be r˜i = ri − ri . Average 5-year and/or 3-year Performers belong to the second tercile of excess short-run exit performance, while Superior 5-year and/or 3-year Performers are in the third tercile of excess short-run exit performance. Columns I, III, and V report empirical results where the short-run is defined to be 5 years, while columns II, IV, and VI report results where the short-run is defined to be 3 years. The data are obtained from VentureXpert, and sample VC firms started business between 1980 and 1998. Cold IPO period firms commenced business between 1980 and 1991, while hot IPO period firms started business between 1992 and 1998. The segmentation of the data into cold and hot IPO periods is justified in Table 1. t-statistics are reported in parentheses. a Significance at 10% confidence level. b Significance at 5% confidence level. c Significance at 1% confidence level.

The results obtained using the full sample in Panel A are qualitatively and quantitatively similar, regardless of whether the short-run is defined to be the first 3 or 5 years of business. The results in columns V and VI show a marginal increase in superior short-run performance decreases excess long-run performance by 24.89% and 28.76%, respectively. Given the standard deviation of excess short-run exit rates in column V is 17.24%, a one standard deviation increase in excess short-run exit performance decreases excess long-run performance by 4.29%. With an average long-run exit performance at 26.25%, this decrease in excess long-run exit performance is economically significant. The comparison of the results in columns I through IV shows performance reversals are consistently more severe for firms that commenced business during cold IPO periods than for firms that commenced business during hot IPO periods. Using the 5-year horizon for the short-run, superior performers are associated with negative excess exit performance of −30.78 (−14.28)% in the long-run if they

546

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

commenced business during relatively cold (hot) IPO periods. These marginal effects translate into a decrease in excess long-run exit rates of 5.56% (2.15%) in response to a one standard deviation increase in excess short-run exit performance. Results obtained with the short-run defined to be 3 years yield inferences that are qualitatively identical to those obtained using the 5-year time frame. The results in Panel A provide evidence that relative firm-level exit performance is not persistent within the VC market. In Panel B, we utilize the regression model in Eq. (4b) to examine the relation between excess long-run and short-run exit performance. The results yield inferences that are qualitatively identical to those obtained in Panel A. Specifically, reversals in exit performance obtained using the 3 or 5year time frame continue to be very similar in columns V and VI, with superior short-run performers associated with negative excess exit performance of −8.47% and −8.55%, respectively in the long-run. The comparison of the results in columns I through IV also continues to indicate that performance reversals are worse for VC firms that commence business during relatively cold IPO periods. Combined, our findings in Table 2 provide strong evidence of reversals in relative firm-level performance within the cross-section of the VC market. Given our findings are based on relative firm-level performance, rather than absolute fund-level performance (as is the case in Kaplan & Schoar, 2005), our findings do not contradict evidence of performance persistence within the VC market. In order to determine the more robust definition of the short-run, we examine the relation between long-run fund flows and excess short-run exit performance. When we define the short-run to be the first 3 (5) years, future fund flows are the size(s) of funds raised by VCs from the 4th (6th) year of business onwards. The test of empirical fit is the concave relation between future fund flows and past performance documented in Kaplan and Schoar (2005). In order to observe concavity, VCs are divided into terciles (three groups) consisting of equal numbers of VCs, based on excess exit performance during the first 3 or 5 years of business. In ascending order from one to three, VCs in these groups are labelled “inferior”, “average”, and “superior” short-run performers, respectively. The regression model is implemented using Ordinary Least Squares (OLS) and specified as, FundSizeih,t+1 = ˛0 + ˛PerformTerci,nt + Controls(year, industry), n = 1, 2, and 3,

(5)

where FundSizeih,t+1 is the size of fund h raised by firm i during the long-run period, and PerformTerci,nt are the three terciles into which VCs are divided based on excess short-run exit performance. The results in Table 3 show the concave relation between long-run fund flows and excess short-run performance is more robust when the short-run is defined to be the first 5 years of business. In the full sample (column V), average short-run performers are associated with funds that are $57 million larger than those managed by inferior short-run performers. Given superior short-run performers are associated with fund flows that are $83 million larger than funds that accrue to inferior performers, relative to average short-run performers, incremental fund flows to superior performers turn out to be only $26 million ($83 million less $57 million); that is, less than half of the incremental fund flows associated with the transition from inferior to average short-run performance. The results in column III indicate that fund flows are also concave in past exit performance for VC firms that commenced business during hot IPO periods. The results in column VI, which are obtained using the 3-year definition of the short-run, neither refute nor support the expected concave relation between fund flows and past performance. The results in column II show, however, that fund flows are concave in past exit performance for VC firms that commenced business during cold IPO periods. Given the concave relation between fund flows and past performance only holds within the entire sample when the short-run is defined to be 5 years, the flow-performance relations in Table 3 indicate the 5-year time frame is the more robust definition of the short-run. 3.3. Portfolio quality, syndicate size, and alternative measures of portfolio risk In Tables 4 and 5, we examine the relation between our proxies of the risk of venture capital assets; that is, portfolio quality and average syndicate size We also examine the relation between syndicate size or portfolio quality and portfolio characteristics, such as such as average fund size, industry specialization, and investment stage specialization. In Table 4, relations between portfolio quality, syndicate

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

547

Table 3 Long-run fund flows and short-run exit performance. Long-run performance

Average 5-year performers Superior 5-year performers

Cold IPO period funds

Hot IPO period funds

I

III

Average 3-year performers Superior 3-year performers Constant Year controls Sector controls R-Squared # of VC fundsd

II

4.9633 (0.55) 1.6039 (0.19)

38.908 (1.30) Yes Yes 0.0447 280

22.979c (2.69) 31.053a (1.89) 35.183 (1.60) Yes Yes 0.0845 229

IV

59.954a (1.72) 115.36 (1.56) 59.718

Full sample V

(1.23)

(1.57) 87.460b

99.766 187.46b (2.33) Yes Yes 0.0903 506

VI

57.183b (2.16) 83.407a (1.95) 54.548

(1.61) 140.08c (2.63) Yes Yes 0.0952 611

80.427b (2.00) Yes Yes 0.1490 968

(2.15) 1.4096 (0.03) Yes Yes 0.1333 1084

This table reports results from empirical tests that examine the relation between future fund flows and past performance. The dependent variable in these tests is the capitalization of funds raised by VCs during the long-run period, which is defined to be year 4 or year 6 of business operations onwards. The independent variables are VC firms’ excess exit success rates during the first 3 or 5 years of business (the short-run period). The model specification is: FundSizeih,t+1 = ˛0 + ˛1n PerformTerci,nt + Controls(year, industry), n = 1, 2, and 3, where FundSizeih,t+1 is the size of fund h raised by firm i during the long-run (period t+1), and PerformTerci,nt (n = 1, 2, 3) are the three terciles into which VCs are divided based on excess exit performance during the short-run period. The tests are implemented using Ordinary Least Squares (OLS) and coefficients reported are relative to VCs in the first (bottom) tercile of excess performance, that is, inferior short-run performers. Average 5-year or 3-year Performers belong to the second tercile of short-run performance, while Superior 5-year or 3-year Performers are in the third tercile of short-run performance. Excess exit success rates are constructed as follows: For each VC firm i, we determine the proportion of its portfolio companies successfully exited via an IPO or third party acquisition and denote this ri We then determine the average exit success rate for VC firms that commenced business during the same year as VC firm i and denote this r¯ i . Firm i’s excess performance, r˜i is then defined to be r˜i = ri − ri . Columns I, III, and V report results where the short-run is defined to be 5 years, while columns II, IV, and VI report results where the short-run is defined to be 3 years. The data are obtained from VentureXpert, and sample firms commenced business between 1980 and 1998. Cold IPO period firms started business between 1980 and 1991, while hot IPO period firms commenced business between 1992 and 1998. The segmentation of the data into cold and hot IPO periods is justified in Table 1. t-stats are reported in parentheses. a Significance at 10% confidence level. b Significance at 5% confidence level. c Significance at 1% confidence level. d The number of funds in columns 1 (2) and 3 (4) do not sum up to the total number of funds in column 5 (6) because funds raised between 2001 and 2005 are also included in the full sample empirical tests.

size, investment stage or industry specialization, and average fund size during the short-run period are examined within a univariate framework. In Table 5, relations between portfolio quality, syndicate size, and other portfolio characteristics are examined within a multivariate framework. Results reported for the short-run period are independent of survival bias; that is, are not restricted to VCs that survive beyond the short-run period. Given we have demonstrated in Table 3 that the 5-year time frame is the more robust definition of the short-run, the definition of the short-run is restricted to the first 5 years of a VC’s life. In Panel A, sample VCs are divided into terciles based on portfolio quality during the short-run period. The results show syndicate size decreases with portfolio quality, with low portfolio quality associated with a syndicate size of 3.56, while high portfolio quality is associated with a syndicate size of 3.24. Differences in syndicate size of 0.32 between low and high quality portfolios and 0.25 between medium and high quality portfolios are statistically significant at the 10% confidence level. Given syndication is a risk sharing mechanism, it is expected that syndication is associated with riskier portfolios among new entrants into the VC market. The negative correlation between portfolio

548

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

Table 4 Fund size, investment staging and industry specialization by terciles of portfolio quality and syndicate size. Terciles of portfolio quality Tercile [1]

Tercile [2]

Difference-in-means Tercile [3]

[1]–[2]

[1]–[3]

Panel A: the relation between portfolio quality and syndicate size: using terciles of portfolio quality PQt −0.796c 0.463 0.808 1.260 −0.345c (0.014) (0.023) 3.558 3.492 3.238 0.066 0.320a Syndicate sizet (0.170) (0.168) Mean fund sizet 105.16 84.949 118.11 20.220 −12.941 (22.22) (23.61) 0.7724 0.7508 0.7323 0.0216 0.0401b Inv. stage specializationt (0.0199) (0.0198) 0.0049 0.6114 0.5622 0.6065 0.0492a Industry specializationt (0.0273) (0.0277) Terciles of syndicate size Tercile [1]

Tercile [2]

[2]–[3] −0.451c (0.021) 0.254a (0.149) −33.161b (16.85) 0.0186 (0.0190) −0.0443a (0.0259)

Difference-in-means Tercile [3]

[1]–[2]

[1]–[3]

Panel B: the relation between portfolio quality and syndicate size: using terciles of syndicate size 1.800 3.138 5.264 −1.338c −3.464c Syndicate sizet (0.037) (0.108) PQt 0.827 0.880 0.816 −0.053 0.010 (0.040) (0.040) Mean fund sizet 138.32 112.34 69.274 25.975 69.042c (22.67) (20.78) 0.8254 0.7684 0.8300 0.0570c −0.0047 Inv. stage specializationt (0.0176) (0.0175) 0.6323 0.6375 0.6857 −0.0052 −0.0534a Industry specializationt (0.0272) (0.0282)

[2]–[3] −2.126c (0.105) 0.063a (0.037) 43.068c (14.03) −0.0617c (0.0170) −0.0482a (0.0254)

In Panel A, we report averages of fund size, industry specialization or investment stage specialization Herfindahls (portfolio characteristics) and syndicate size within each of three terciles of portfolio quality during the first 5 years of a VC’s life (the “short-run”). In Panel B, we report averages of the different portfolio characteristics and portfolio quality within each of three terciles of syndicate size during the short-run period (period t). The choice of a 5-year time frame for the short-run is not arbitrary; rather, it is empirically determined in Tables 2 and 3. Firm i’s Portfolio Quality (PQ) is constructed as: PQ = (/); where  is the mean of real (inflation-deflated) capital disbursements to portfolio projects, and  is the standard deviation of real capital disbursements to portfolio projects during first financing rounds. Portfolio Quality increases with PQ. Firm i’s syndicate size is constructed as the mean size of financing syndicates that include firm i. VC firms are divided into terciles of PQ in Panel A, and into terciles of syndicate size in Panel B to reflect a framework where portfolio risk is low, medium, or high. Mean fund size is the average size of funds raised by VC firms during period t; investment stage specialization is constructed as the average of specialization Herfindahls computed as the sum of squared proportions of capital disbursements to projects in the (1) early or start-up phase and (2) later or expansion phase of a firm’s growth cycle. Investment stage specialization increases with specialization Herfindahls. Industry specialization Herfindahls are computed as the sum of squares of proportions of capital disbursements to five different industries: (1) computer related; (2) biotechnology/medical; (3) communications/electrical, (4) onsumer related; and (5) energy/industrials. Industry specialization increases with the specialization Herfindahls. The data (including investment stage and industry classifications) are from VentureXpert, and sample VC firms commenced business between 1980 and 1998. Standard errors obtained from tests for differences in means are reported in parentheses. a Significance at 10% confidence level. b Significance at 5% confidence level. c Significance at 1% confidence level.

quality and syndicate size thus constitutes evidence that our measure of portfolio quality does capture differences in portfolio risk or quality within the cross-section of the VC market. As is the case with average syndicate size, the results show investment stage specialization decreases across terciles of portfolio quality, with the difference of 0.04 in specialization Herfindahls between low and high quality portfolios significant at the 5% confidence level. Given investment stage specialization and syndicate size decrease with portfolio quality, the results in Panel A indicate that both syndication and specialization in small ventures are associated with riskier portfolios during the initial 5 years of a VC’s life. Average fund size and industry specialization show no distinct patterns across the terciles of portfolio quality.

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

549

Table 5 Portfolio quality, syndicate size, and portfolio characteristics.

Investment stage specializationt Industry specializationt Average fund sizet

PQt I

SyndSizet II

−0.3387c (−3.87) 0.0082 (0.14) 6.05e−05 (0.71)

0.5765 (1.39) −0.2455 (−0.82) −0.0019c (−7.00)

Investment stage specializationt+1 Industry specializationt+1 Average fund sizet+1 Massachusetts North East U.S. (excluding MA) All other states Constant R-Squared # of obs.

−0.0643 (−1.62) −0.1118b (−2.39) −0.1458c (−3.06) 1.1525c (14.03) 0.0450 577

−0.8239c (−4.56) −0.6570c (−3.03) −0.3757 (−1.56) 3.8852c (10.28) 0.1077 577

PQt+1 III

SyndSizet+1 IV

−0.4720c (−3.01) −0.1198 (−1.13) −1.90e−05 (−0.28) 0.0799 (1.25) 0.0193 (0.23) 0.0412 (0.53) 1.3026c (9.98) 0.0647 248

1.0349c (2.57) 0.9569c (3.49) −0.0001 (−0.54) −0.3320b (−2.27) −0.1852 (−1.10) −0.3429 (−1.53) 2.2954c (7.09) 0.1721 248

This table reports results from empirical tests that examine the relation between portfolio quality or syndicate size and portfolio characteristics, such as average fund size, and industry or investment stage specialization during the short-run (first 5 years of business) and long-run periods (year 6 of business, and onwards). The short-run is regarded as the time frame for new entrant VCs to establish a venture capital portfolio, and it’s length is empirically determined in Table 3. In implementing the empirical tests, we control for the geographical location of sample VC firms. The empirical tests are implemented using OLS regressions, and the model specifications for the short-run period are: PQit = f (InvStageit , IndSpecit , FundSizeit , Locationi ) SyndSizeit = f (InvStageit , IndSpecit , FundSizeit , Locationi ) where PQ is portfolio quality; SyndSize is the average number of firms participating in financing deals that involve firm i; IndSpec and InvStage are industry and investment stage specialization Herfindahls; FundSize is average fund size, and Location is a discrete variable which indicates a firm’s geographical location. The regression models are similarly specified for the long-run period (results in columns III and IV). With the exception of the Location variable, the construction of all variables is described in Table 4. Regression coefficients reported for the geographical location dummies are relative to VCs located in the state of California. Data on venture capital transactions are from VentureXpert, and sample firms started business between 1980 and 1998. t-statistics are reported in parentheses. a Significance at the 10% confidence level. b Significance at the 5% confidence level. c Significance at the 1% confidence level.

In Panel B, sample VCs are divided into terciles, based on syndicate size during the short-run period. The results show portfolio quality does not decrease with syndicate size. Given syndicate size decreases with portfolio quality (see Panel A), while portfolio quality does not decrease with syndicate size, the results indicate our measure of portfolio quality encompasses much more than portfolio risk. Consistent with our expectation that portfolio risk increases with syndicate size, the results show average fund size decreases, while industry specialization increases with syndicate size, resulting in the association of large syndicates with funds that are $43 million smaller than those associated with small syndicates. The results show no distinct patterns in the relation between syndicate size and investment stage specialization. Combined, our findings in Table 4 indicate, consistent with expectations, that portfolio risk decreases with portfolio quality, investment stage diversification, or average fund size, and increases with industry specialization or syndicate size. In Table 5, we examine the relation between portfolio quality, syndicate size, average fund size, and industry or investment stage specialization within a multivariate framework. As in Table 4, the

550

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

results in columns I and II show investment stage specialization decreases with portfolio quality, but is not significantly associated with syndicate size during the first 5 years of business. The results in column II also show that average fund size decreases with syndicate size during the short-run period. The coefficients of geographical location, which for the most part are negative, indicate VCs located outside of California are associated with lower portfolio quality and smaller syndicate size relative to VCs located in California. The results for the long-run period reported in columns III and IV show industry specialization is negatively correlated with portfolio quality and positively correlated with syndicate size among established VCs. Given syndication is evidence of risk sharing, this implied negative correlation between portfolio quality and syndicate size provides additional evidence that portfolio risk decreases with our measure of portfolio quality. As in Table 4, industry specialization is positively correlated with syndicate size, and insignificantly associated with portfolio quality. Combined, the results obtained from the multivariate framework in Table 5 are consistent with and further buttress results obtained within a univariate framework in Table 4. Specifically, the results provide evidence that portfolio risk decreases with portfolio quality. Consistent with the expectation that syndication is associated with higher portfolio risk, we obtain negative correlations between portfolio quality and syndicate size. We also find that portfolio risk increases with industry specialization and decreases with fund size. 3.4. Performance reversals and the risk of venture capital assets In Table 6, we attempt to explain performance reversals in the VC market using our two proxies for the risk of venture capital assets; that is, portfolio quality and syndicate size. We denote the two risk proxies aj , 1 ≤ j ≤ 2, in this order: (1) portfolio quality; and (2) syndicate size. If aj , 1 ≤ j ≤ 2 are risk proxies that explain performance reversals in the VC market, the interaction of these variables with short-run performance will help explain performance reversals reported in Table 2 in terms of the coexistence of persistent and non-persistent attitudes towards risk. The reasoning is as follows. If all VCs maintain their short-run attitudes towards risk in the long-run, reversals in relative performance cannot be explained using the risk of venture capital assets. Similarly, if all VCs change their attitudes towards risk in exactly the same manner, then again, reversals in relative performance cannot be explained using the risk of venture capital assets. If, however, some classes of VCs persist in their shortrun attitudes towards risk, while some classes of VCs switch attitudes, the coexistence of persistent and non-persistent attitudes towards risk can explain performance reversals. Two sets of empirical tests are conducted in order to explain performance reversals in the VC market. First, we examine the relation between short-run performance and short-run risk proxies. Next, using the framework discussed in Section 1, we examine the relation between long-run performance, short-run performance and the interaction of short-run performance and short-run risk proxies. The specifications of the two models are: Performi,t = ˛0 + ˛1 a1i,t + ˛2 a2i,t ,

(6)





Performi,t+1 = ˛0 + ˛1 Performi,t + ˛2 a1i,t+1 + ˛3 a2i,t+1 + ˛4n Tercajn,t · Performi,t ,

j = [1, 2] (7)

where Performi,t and Performi,t+1 are excess short-run and long-run performance; and Tercajn,t are three terciles of aj,t ; the subscript i denotes firm i, while ˛4n is a vector consisting of two regression coefficients for each risk measure. Coefficient ˛1 in Eq. (7) indicates the relation between excess long-run and short-run performance in the entire cross-section of the VC market independent of portfolio risk. The elements of vector ˛4n indicate the relation between excess long-run and shortrun performance within the second and third terciles of portfolio risk for each risk measure. In Eq. (7), specifications with a1,t are termed the portfolio quality model, while specifications with a2,t are termed the syndicate model. The results in column I and II show a marginal increase in portfolio quality decreases excess shortrun performance by 5.13%, while a marginal increase in syndicate size increases excess short-run

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

551

Table 6 Performance reversals and the risk of venture capital assets. Excess exit ratest I

Excess exit ratest+1

II

III

0.0187 (3.28)

−0.0451 (−1.88) 0.0178c (3.13)

−0.0513 (−2.12)

b

PQt

c

Syndicate sizet (SSt )

IV

V

VI

VII

−0.3231c (−6.83) 0.0360b (2.30)

−0.3337c (−7.09)

0.0162c (2.88)

−0.1463 (−1.49) 0.0364b (2.39) 0.0183c (3.271) −0.1812 (−1.38) −0.2701b (−2.48)

−0.0572c (−2.99) 0.0000 227

−0.0982c (−4.12) 0.0000 227

−0.0287 (−0.25) 0.0350b (2.32) 0.0174c (3.13) −0.1812 (−1.38) −0.2929b (−2.48) −0.1397 (−1.30) −0.2031a (−1.82) −0.0944c (−3.97) 0.0000 227

a

Excess exit ratet (EERt ) PQt+1 Syndicate sizet+1 Average SSt and EERt High SSt and EERt Average PQt and EERt High PQt and EERt Constant Model p-value Number of VC Firms

0.1122c (5.02) 0.0342 582

0.0048 (0.22) 0.0011 582

−0.0403b (−2.44) 0.0000 227

0.0458 (1.49) 0.0008 582

This table reports results from empirical tests that examine whether our proxies for the risk of venture capital assets are able to explain performance reversals observed in Table 3. The proxies we consider are: (1) portfolio quality, and (2) syndicate size. Portfolio Quality (PQ) is constructed as the standard deviation of VC i’s inflation-deflated capital disbursements to portfolio projects during first financing rounds, divided by the mean disbursement. Portfolio risk decreases with PQ. Syndicate size is defined to be the average size of financing syndicates that include VC i. Risk sharing increases with syndicate size. Sample firms are divided into terciles of portfolio quality (PQ) and syndicate size (SS). VC firms in the second and third terciles of portfolio quality or syndicate size have “average” and “high” portfolio quality or syndicate size, respectively. Let aj , j = [1,2] represent the continuous versions of portfolio quality and syndicate size, then the regression models in columns I through III are specified as: Performi,t = ˛0 + ˛1 a1i,t + ˛2 a2i,t ; while the regression models in columns IV, V, VI, and VII are specified as:





Performi,t+1 = ˛0 + ˛1 Performi,t + ˛2 a1i,t+1 + ˛3 a2i,t+1 + ˛4n Tercajn,t · Performi,t . Performi,t (EERi,t ) and Performi,t+1 (EERi,t+1 ) are excess short-run and long-run performance, respectively; and Tercajn,t are three terciles of portfolio risk that sample VCs are divided into for each risk proxy aj . The regressions are implemented using Tobit models, and the construction of excess exit rates is described in Tables 1 and 2. The venture capital data are obtained from VentureXpert, and sample VC firms commenced business between 1980 and 1998. t-statistics are reported in parentheses. a Significance at 10% confidence level. b Significance at 5% confidence level. c Significance at 1% confidence level.

performance by 1.87%. The results in column III show the relation between excess short-run performance and either of portfolio quality or syndicate size continues to hold when both risk measures are included in the same regression model. Comparatively, the results in column III indicate that a one standard deviation change in portfolio quality (0.38) decreases excess short-run performance by 1.71%, while a one standard deviation increase in syndicate size (1.60) increases excess shortrun performance by 2.85%. Given we have established in Tables 4 and 5 that portfolio risk increases with syndicate size, and decreases with portfolio quality during the short-run period, the negative performance correlation between portfolio quality and syndicate size is in line with expectations. In columns IV and V, we examine the relation between excess long-run performance, excess shortrun performance, and either of long-run portfolio quality or syndicate size. As in Table 2, the results in columns IV and V show excess long-run performance decreases with excess short-run performance. The results in column IV further show a marginal increase in portfolio quality increases excess longrun performance by 3.60%, while the results in column V show a marginal increase in syndicate size

552

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

increases excess long-run performance by 1.62%. Comparatively, these marginal effects indicate a one standard deviation increase in portfolio quality (0.45) increases excess long-run exit performance by 1.62%, while a one standard deviation increase in syndicate size (1.24) increases excess long-run exit performance by 2.01%. Given we find a negative performance correlation between portfolio quality and syndicate size during the short-run period (the results in columns I through III) and a positive performance correlation during the long-run period, our findings suggest either of portfolio quality or syndicate size is persistent between the short and long-run periods, while the other is non-persistent. As a consequence, it is likely that interactions between syndicate size or portfolio quality and excess short-run exit performance help explain performance reversals within the VC market. Since exit performance increases with syndicate size during the short and long-run periods, we first examine how interactions between short-run syndicate size and excess short-run exit performance affect performance reversals within the VC market. The results in column VI show the coefficient of excess short-run performance decreases in absolute value from 33.37% in column V to 14.63% in column VI. The coefficient also ceases to be significant. This finding indicates excess long-run performance does not decrease with excess short-run performance independent of portfolio risk. When we examine the coefficients of the interaction terms, we find excess long-run performance decreases with excess short-run performance among VCs in the third tercile of syndicate size; that is, among VCs with the highest risk portfolios. This finding suggests superior short-run performers that hold high risk portfolios during the short-run period switch portfolio strategies between the short and long-run periods. The results in column VII show the introduction of interactions between portfolio quality and excess short-run performance into the regression model in column VI further decreases the absolute value of the coefficient of excess short-run performance from 14.63% to 2.87%. The coefficient also continues to be statistically insignificant. As in column VI, this finding indicates excess long-run performance does not decrease with excess short-run performance independent of portfolio risk. When we examine the coefficients of the interaction terms, we find excess long-run performance decreases with excess short-run performance among VCs in the third tercile of portfolio quality or syndicate size; that is, among VCs that manage the highest risk or highest quality portfolios during the short-run period. Given risk-based explanations of performance reversals require the coexistence of persistent and non-persistent attitudes towards risk, our findings in column VII indicate that superior short-run performers that manage high risk portfolios during the short-run period either persist with or reverse this strategy. If superior short-run performers that manage high risk portfolios persist with this strategy in the long-run, it must be the case that superior short-run performers that manage high quality portfolios reverse this strategy in the long-run. Conversely, if superior short-run performers that manage high risk portfolios reverse their portfolio strategy in the long-run, it must be the case that superior short-run performers that manage high quality portfolios persist with this portfolio strategy in the long-run. Using the variance-covariance matrix obtained from the regression results in Table 6, we find no evidence that the results in columns VI and VII are driven either by noise or multicollinearity.6 In Table 7, we test for persistence and non-persistence in attitudes towards risk or portfolio quality between the short and long-run periods using ordered logit models. The short-run is the first 5 years of a VC’s life, while the long-run is the 6th year and onwards. First, we examine the relation between short-run portfolio quality and short-run syndicate size within a regression framework. Next, using the regression model specified in Eq. (8) below, we test for persistence or non-persistence in attitudes towards risk between the short and long-run periods. For each aj,t , 1 ≤ j ≤ 2 (1 = portfolio quality; 2 = syndicate size), Eq. (8) is specified as:





aij,t+1 = ˛0 + ˛1 Performi,t + ˛2 a1i,t+1 + ˛3 a2i,t+1 + ˛4n Tercajn,t · Performi,t + ˛5n Tercajn,t , j = [1, 2]

(8)

where ˛5n is a vector consisting of two coefficients for each aj , and all other variables are as previously defined (see Eq. (7)). 6

The matrix is shown in Table 11.

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

553

Table 7 Transitions in attitudes towards risk within the VC market. Portfolio quality (PQ) PQt I Syndicate size gt (SSt )

Syndicate size (SS)

PQt+1 II

PQt+1 III

SSt+1 IV

SSt+1 V

0.2031 (0.53)

0.3841 (1.26)

−0.2428 (−0.38) 0.1002 (0.98)

1.5031b (2.48) 0.1245 (0.98)

−0.0167 (−0.66) 0.0394 (0.48) 0.2031b (2.38)

−0.0036 (−0.15)

−0.0215a (−1.94)

Excess exit ratet (EERt ) PQt+1 SSt+1 Average syndicatet (ASt ) High syndicatet (HSt ) Average PQt (AQt )

0.0909 (1.17) 0.0536 (0.72)

High PQt (HQt ) −0.2057 (−0.34) −0.4466 (−0.95)

ASt and EERt HSt and EERt AQt and EERt HQt and EERt Pseudo R-squared # of observations

0.1978b (2.04) 0.4644c (4.32)

0.0031 582

0.0157 227

0.0194 (0.18) −0.0839 (−0.84) −0.1659 (−0.21) 1.4530 (1.51)

−0.6545 (−1.36) −0.5783 (1.26) 0.0063 227

0.0964 227

−1.9517b (−2.41) −1.0527 (−1.16) 0.0308 227

This table reports results from empirical tests that are designed to reveal transitions in VCs’ attitudes towards risk as they become more matured. The time it takes for new entrant VCs to develop a venture capital portfolio is labelled the “short-run” (the first 5 years of business). VCs 6 years and older are regarded as established VCs, and year 6 and onwards of a VC’s life is labelled the “long-run” period. The dependent variable in the empirical tests is either VC i’s syndicate size or VC i’s portfolio quality during the long-run period. Syndicate size is defined to be the average size of financing syndicates that include VC i. Portfolio quality (PQ) is constructed as the standard deviation of VC i’s inflation-deflated capital disbursements to portfolio projects during first financing rounds, divided by the average capital disbursement. Period t refers to short-run period, while period t+1 refers to the long-run period. Let aj , j = [1, 2] represent the continuous versions of portfolio quality and syndicate size, then the regression models (columns II through V) are specified as:





aij,t+1 = ˛0 + ˛1 Performi,t + ˛2 a1i,t+1 + ˛3 a2i,t+1 + ˛4n Tercajn,t · Performi,t + ˛5n Tercajn,t ,

j = [1, 2]

where Performi,t (EERi,t ) is excess short-run performance; and Tercajn,t are three terciles of portfolio risk that sample VCs are divided into with respect to each risk proxy aj The construction of Performi,t is described in Tables 1 and 2. VC firms in the second and third terciles of portfolio quality or syndicate size have “average” and “high” portfolio quality or syndicate size, respectively. The model in column I examines the relation between short-run portfolio quality and syndicate size, while the models in all other columns test for transitions in attitudes towards portfolio risk between the short-run and long-run periods. The regression coefficients reported for the second and third terciles of portfolio quality or syndicate size are relative to VCs in the first tercile of portfolio quality or syndicate size. The venture capital data are obtained from VentureXpert and sample VCs commenced business between 1980 and 1998. The empirical tests are implemented using ordered logit models and robust z-statistics are reported in parentheses. a Significance at 10% confidence level. b Significance at 5% confidence level. c Significance at 1% confidence level.

The results in column I show, consistent with our findings in Tables 4 and 5, that short-run portfolio quality decreases with short-run syndicate size. As in Tables 4 and 5 then, portfolio quality is negatively correlated with syndicate size during the short-run period. The results in column II show, however, that membership in large syndicates during the short-run period is 20% more likely to be associated with high portfolio quality in the long-run. Given membership in large syndicates is associated with

554

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

low portfolio quality during the short-run period, the results in column II provide evidence of a reversal in attitudes towards portfolio risk between the short and long-run periods. The results in column IV show membership in medium-sized (large) syndicates during the shortrun period is 19.78 (46.44)% more likely to be associated with membership in large syndicates during the long-run period. This finding indicates VCs’ syndication strategies are persistent between the short and long-run periods. Combined, the results in columns I, II, and IV show persistence in syndication strategies coexists with reversals in attitudes towards portfolio quality or risk between the short and long-run periods. Given risk-based explanations of performance reversals require the coexistence of persistent and non-persistent attitudes towards risk, our findings indicate performance reversals are explained by interactions between syndication strategy and portfolio quality between the short and long-run periods. The results in column V show long-run syndicate size increases with excess short-run performance among VCs in the lowest tercile of short-run portfolio quality. Long-run syndicate size also decreases with excess short-run performance among VCs in the middle tercile of short-run portfolio quality. These results neither refute nor support the coexistence of persistent and non-persistent attitudes towards risk within the cross-section of the VC market. The absence of significant relations between short-run and long-run portfolio quality in column III indicates, however, that attitudes towards portfolio quality do not persist between the short and long-run periods. Taken together, our findings in Table 7 are supportive of a risk-based explanation of performance reversals within the VC market. First, the results show syndicate size is persistent between the short and long-run periods. However, while large syndicates are associated with higher risk or lower quality portfolios during the short-run period, they are associated with higher quality or lower risk portfolios in the long-run. VCs that hold high quality portfolios during the short-run period also do not persist with this strategy in the long-run; however, high quality portfolios are associated with superior longrun exit performance. Combined, the coexistence of persistent attitudes towards syndication strategy and non-persistent attitudes towards portfolio quality results in performance reversals within the cross-section of the VC market. Though surprising, our finding that syndication is associated with risky portfolios during the initial 5 years of a VC’s life, but with less risky or higher quality portfolios subsequently can be rationalized by models of costly reputation building in markets characterized by adverse selection problems. In Diamond (1989), for instance, economic agents develop reputation by initially taking on risky portfolios. Those that are successful then transition to less risky or higher quality portfolios after establishing some level of market reputation. Within the context of models of costly reputation building then, syndication appears to be evidence of risk sharing among new entrant or young VCs, and evidence of market reputation among established VCs. Empirical evidence in Tian (2007) that projects backed by large syndicates achieve superior post-IPO performance relative to other VC backed projects are consistent with explanations of transitions in syndication strategy that are predicated on models of costly reputation building in markets characterized by severe adverse selection problems. A market reputation-based explanation of transitions in syndication strategy is also consistent with the findings in Hochberg et al. (2007) that large syndicates achieve superior exit performance relative to small syndicates. Our finding that VC firms associated with high quality portfolios and superior performance during the short-run period do not persist with this strategy indicates these VCs switch from higher to lower quality portfolios between the short and long-run periods. This switch ends up being costly, however, because, in line with expectations, high quality portfolios are associated with superior exit performance in the long-run. The switch in attitudes toward portfolio quality that we observe is consistent with theories of investor overconfidence. In Post et al. (2008), economic agents transition from risk averse to risk seeking behavior when prior expectations are surpassed by favorable outcomes. In Gervais and Odean (2001), traders become overconfident because they take too much credit for their successes, and are thus induced to take on more investment risk. Daniel et al. (1998), Malmendier and Tate (2005, 2008), and Chuang and Lee (2006) also link overconfidence to risk seeking behavior. Within the context of the VC market, superior short-run performers that attribute their short-run success to managerial and monitoring abilities, rather than the quality of their portfolio projects (deal screening ability), adopt riskier portfolio strategies in the long-run, resulting in inferior long-run performance.

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

555

Under the assumption that VCs’ monitoring and managerial abilities cannot make ‘bad’ projects ‘good’, portfolio quality, or equivalently, the ability to identify and/or attract superior quality projects is more important for success in the VC market, in comparison with managerial and monitoring abilities. In this respect, our findings are consistent with the conclusion in Kaplan et al. (2007) that investors are better off backing horses (business plans) rather than jockeys (entrepreneurs or management teams). 4. Robustness tests In this section, we examine whether our risk-based explanation of performance reversals is robust to alternative explanations. First, we examine whether performance reversals are robust to the exclusion of the oldest VCs; that is, VCs that commence business prior to 1986. This robustness test is predicated on the notion that VentureXpert’s coverage of VC firms that commenced business in the early 1980 s is not as exhaustive as the coverage in subsequent periods. The results in Panel A of Table 8 show performance reversals persist when VCs that commence business prior to 1986 are excluded from our sample. The comparison of the results in column I of Tables 2 and 8 shows, however, that the marginal effect of excess short-run performance on excess long-run performance decreases in absolute value from 30.78% to 16.55%. As a consequence, the marginal effect of excess short-run performance is nearly identical for firms that commence business during cold or hot IPO periods. These findings indicate that performance reversals are robust to the exclusion of the oldest VCs. However, the inclusion of the oldest VCs in our sample biases the negative relation between excess short-run and long-run performance upwards. In Panel B, we examine whether performance reversals observed within the cross-section of the VC market are explained by survival bias. In order to facilitate these empirical tests, we augment the model in Panel A with a dummy variable set equal to one if a VC firm is only associated with one fund. The results show firm failure is unable to explain performance reversals observed within the VC market. The comparison of model qualitative fit in Panels A and B using the Bayesian Information Criterion (BIC) further indicates the model in Panel A provides better model qualitative fit relative to the model that accounts for survival bias in Panel B (more negative numbers indicate better model qualitative fit). The results in Panel B provide no support for an explanation of performance reversals that stems from survival bias. In Panel C, we examine whether evidence of performance reversals are explained by the notion that superior short-run performers’ deal flow ran dry. In order to test for this possibility, we restrict the definition of the long-run to years 6 through 10 of business. We then examine whether performance reversals persist within the cross-section of the VC market. If performance reversals persist using our truncated definition of the long-run, this constitutes evidence that superior short-run performers’ deal flow ran dry subsequent to their initial success. The results for the full sample in column V of Table 8 shows performance reversals do not persist when excess long-run performance is based on our truncated definition of the long-run. Performance reversals also do not persist among VC firms that commenced business during relatively cold IPO periods (results in column I). The results in column III show, however, that performance reversals persist among VC firms that commenced business during the relatively hot IPO periods of the late 1990s. Given the hot IPO period of the late 1990s is immediately followed by the stock market crash of 2000, performance reversals associated with hot IPO markets can be induced by the fact that specialist VCs’ deal flow ran ran dry. Alternatively, the observed performance reversals may be explained by the fact that specialists’ exit opportunities ran dry consequent to the stock market crash. In order to distinguish between explanations of performance reversals that are based on changes in deal flow and changes in exit opportunities, we examine the relation between excess long-run and short-run performance, while controlling for interactions between industry specialization and excess short-run performance. The results in column IV of Panel C show performance reversals are no longer observed within the cross-section of the VC market when we account for interactions between industry specialization and excess short-run performance. The results also show that performance reversals only persist among industry specialists that achieve superior performance during the shortrun period. The results in column IV are consistent with the notion that performance reversals observed

556

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

in column IV occur because specialists’ exit opportunities dried up after the stock market crash of 2000. The results in column II and VI, which show performance is persistent within the cross-section of the VC market when interactions between industry specialization and excess short-run performance are accounted for continue to be inconsistent with explanations of performance reversals that are based on changes in deal flow. Taken together, our findings in Table 8 show evidence of performance reversals within the VC market are robust to the exclusion of the oldest VCs, survival bias, and explanations of performance

Table 8 Robustness tests—“thinness” of data, survivorship bias, and alternative explanations of performance reversals. Long-run performance (Performi,t+1 ) Cold IPO period firms Panel A: excluding VCs that commenced business prior to 1986 −0.1655b Performi,t (5-year) (1.96) Constant 0.0030 (0.16) Model BIC −53.360 # of VC firms 63 Panel B: restrictions in Panel A plus accounting for survivorship bias −0.1288 Performi,t (5-year) (−1.49) One fund −0.0622 (−1.57) Constant 0.0133 (0.67) Model BIC −51.647 # of VC firms 63 Cold period I

Hot IPO period firms

Full sample

−0.1428c (−3.90) −0.0001 (−0.02) −399.37 184

−0.1519c (−4.57) 0.0003 (0.06) −454.05 247

−0.1436c (−3.90) −0.0065 (−0.37) 0.0006 (0.10) −394.29 184

−0.1459c (−4.38) −0.0258 (−1.56) 0.0039 (0.63) −450.96 247

Hot period II

III

Full sample IV

V

Panel C: restrictions in Panel A plus truncation of long-run period to years 6 through 10 of business Performi,t (5-year) 0.1830 1.1179c −0.1698a 0.1306 −0.0462 (0.79) (3.21) (−1.76) (0.78) (−0.54) c −1.3991 −0.3219 MISt ·Performit (−3.24) (−1.39) −1.1214c −0.5723b HISt ·Performit (−2.65) (−2.42) Constant −0.0061 −0.0134 −0.0053 −0.0100 0.0035 (−0.12) (−0.29) (−0.37) (−0.69) (0.24) # of VC firms 48 48 172 172 220

VI 0.4247c (2.84) −0.6596c (−3.24) −0.6867c (−3.36) 0.0014 (0.10) 220

In Panel A, we restrict our sample to VC firms that commence businees from 1986 and onwards in order to determine whether our empirical results are robust to the exclusion of the oldest VCs and the relative thinness of the VentureXpert database during the early 1980s. In Panel B, we examine whether our results are robust to survivorship bias. In Panel C, we restrict the estimation of relative performance to financing deals that occur during the first 10 years of a VC’s life in order to examine whether performance reversals occur because superior short-run performers’ deal flow “ran dry”. The model specifications in Panels A, B, and C are, respectively: Performi,t+1 = f(Performi,t ); Performi,t+1 = f(Performi,t , Onefundi ); and Panel C: Performi,t+1 = f(Performi,t , TercISin,t · Performit ); where Performi,t+1 and Performi,t are excess long-run and short-run exit performance, with the long-run defined to be years 6 and onwards, while the short-run is the first 5 years of a VC’s life. In Panel C, the long-run is truncated to years 6 through 10 of a VC’s life. Onefund is a dummy variable that is equal to one if VC i is only associated with one fund. TercISin,t are three terciles (that take on values 1, 2, and 3) into which VCs are divided based on industry specialization during the short-run period. MIS and HIS are the second and third terciles of industry specialization, respectively. All three regression models are implemented using Tobit models. The construction of excess exit performance is described in Tables 1 and 2. The data are obtained from VentureXpert, and sample VC firms started business between 1980 and 1998. Cold IPO period firms commenced business between 1980 and 1991, while hot IPO period firms commenced business between 1992 and 1998. The segmentation of the data into cold and hot IPO periods is justified in Table 1. t-statistics are reported in parentheses. a b c

Significance at 10% confidence level. Significance at 5% confidence level. Significance at 1% confidence level.

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

557

Table 9 Robustness tests: performance reversals and alternative risk or quality proxies.

Fund sizet (FSt )

Excess exit ratest

Excess exit ratest+1

I

III

IV

V

VI

−0.3185c (−6.70) 4.69e−05b (2.06)

−0.3257c (−6.81)

−0.1559 (−1.58) 4.68e−05b (2.10)

−0.2362b (−2.17) 4.52e−05b (2.02)

0.0187c (3.33) −0.1668 (−1.29) −0.2473b (−2.09)

0.0192c (3.42) −0.1796 (−1.39) −0.2825b (−2.37) 0.1803a (1.76) 0.1218 (0.97) −0.0745c (−3.84) 0.0000 227

II

−0.0001 (−2.39)

b

−0.0023 (−0.05)

Inv. stage specializationt (TSt ) Excess exit ratet (EERt ) FSt+1

−0.0197 (−0.51)

TSt+1 Syndicate sizet+1 Average SSt and EERt High SSt and EERt Average FSt and EERt High FSt and EERt Constant Model p-value Number of VC firms

0.0803c (7.78) 0.0170 582

0.0707a (1.91) 0.9616 582

−0.0146a (−1.76) 0.0000 227

0.0080 (0.28) 0.0000 227

−0.0733c (−3.76) 0.0000 227

This table reports results from empirical tests that examine whether alternative proxies of the risk or quality of venture capital assets also explain performance reversals observed in Table 2. The alternate proxies we consider are investment stage specialization (TS) and Average Fund Size (FS). The models are specified as: ColumnsI(II) : Performit = f(FSit , TSit ); Columns III(IV) : Performit+1 = f(Performit , FSit+1 , TSit+1 ); Columns V(VI) : Performit+1 = f(Performit , FSit+1 , SSit+1 , TercSSin,t · Performit , TercFSin,t · Performit ); where Performi,t (EERi,t ) and Performi,t+1 (EERi,t+1 ) are excess short-run and long-run exit performance, respectively; TS are investment stage specialization Herfindahls; and FS is the average size of funds raised by VC i. TercFSin,t and TercSSin,t are three terciles of average fund size and syndicate size into which sample VCs are divided. VC firms in the second and third terciles of fund size or syndicate Size have “average” and “high” fund size or syndicate size, respectively. Period t (the “short-run” period) is the time it takes new entrant VCs to establish a venture capital portfolio and is defined to be the first 5 years of a VC’s life. As a result, period t + 1 (the “long-run”) is years 6 and onwards of a VC’s life. The robust definition of the short-run period is empirically determined in Tables 2 and 3, while the construction of excess exit performance is described in Tables 1 and 2. TS stands for investment stage specialization Herfindahls (a continuous variable). The regressions are implemented using Tobit models and regression coefficients reported for VCs in the average and high terciles of fund size and syndicate size are relative to VCs in the bottom terciles of fund size and syndicate size. t-statistics are reported in parentheses. a Significance at 10% confidence level. b Significance at 5% confidence level. c Significance at 1% confidence level.

reversals that are based on the notion that superior short-run performers’ deal flow ran dry. In what follows, we examine whether portfolio characteristics, which are shown to be correlated with portfolio quality or syndicate size in Tables 4 and 5, help explain performance reversals within the VC market. Both univariate and multivariate results reported in Tables 4 and 5 show investment stage specialization is negatively correlated with portfolio quality during the short and long-run periods. The results also demonstrate that average fund size is negatively correlated with syndicate size during the short-run period. In Table 9, we substitute investment stage specialization for portfolio quality, and average fund size for syndicate size in empirical tests reported in Table 6. The results in column I and III show excess short-run performance decreases with average fund size, while excess long-run performance increases with average fund size. The results in columns III and IV show, however, that investment stage specialization has no effect on excess short-run or excess

558

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

long-run performance. Combined, the results in columns I through IV show only average fund size is significantly associated with both excess short-run and excess long-run performance. The results also demonstrate that the performance effect of average fund size mimics the performance effect of portfolio quality. Given these findings, we reintroduce syndicate size into our regression model, and examine how the interplay of short-run syndicate size and average fund size affects excess long-run performance. The results in column V show, as in Table 6, that the introduction of interactions between syndicate size and excess short-run performance into the regression model in column V helps explain performance reversals within the VC market. The results in column VI show, however, that the introduction of average fund size into the regression model re-amplifies the negative relation between excess long-run and excess short-run performance. Interactions between short-run fund size and excess short-run performance also shed no light on performance reversals within the VC market. In aggregate, our findings in Table 9 demonstrate, in so far as it relates to the explanation of performance reversals within the VC market, that average fund size and investment stage specialization are not substitutes for syndicate size or our measure of portfolio quality. Though not reported, results obtained using industry specialization also demonstrate that industry specialization is not a substitute for syndicate size or portfolio quality.7 Hence, while performance reversals observed during hot IPO periods are explained by changes in VCs’ exit opportunities (see results in Panel C of Table 8), this explanation of performance reversals is not robust to performance reversals observed during cold IPO periods. In order to further demonstrate that our measure of portfolio quality really captures differences in portfolio risk within the cross-section of the VC market, we examine, within a multivariate framework, the relation between portfolio quality and the skewness of VCs’ investment distributions. Specifically, if portfolio risk decreases with our measure of portfolio quality, we expect that portfolio quality will increase with the skewness of investment distributions; that is, in line with predictions in Huang and Litzenberger (1988), portfolio quality will be associated with positively skewed investment distributions. As discussed in Kim and White (2004), the traditional measure of skewness, which is defined to be the third central moment normalized by the second central moment, is not robust to the presence of outliers.8 However, Groeneveld and Meeden (1984) develop a robust skewness measure where skewness is equal to zero for symmetric distributions and bounded by −1 and +1. This robust skewness measure is not affected by the presence of outliers in a data sample and skewness (SK) is defined as: SK =

−Q

2 ,  E yt − Q2 

(10)

where  and Q2 are the mean and second quantile (median) of some variable yt . This skewness measure is a generalization of the Bowley coefficient of skewness (Bowley (1920), and Hinkley (1975)). We utilize the robust formula for skewness proposed in Groeneveld and Meeden (1984) to construct our skewness measure. Hence,  is the mean dollar disbursement during first financing rounds to projects involving firm i, Q2 is the median dollar disbursement during first financing rounds to projects involving firm i, and yt is dollar disbursements during first financing rounds to projects involving firm i. In Table 10, we replicate empirical tests reported in Table 5, but with the addition of the skewness of investment distributions as an explanatory variable. In line with expectations, we find portfolio quality increases with the skewness of investment distributions in column I, while syndicate size decreases with the skewness of investment distributions in column II. This finding provides additional validation of the negative correlation between short-run portfolio quality and syndicate size documented in Tables 4 and 5, provides additional evidence that both portfolio quality and syndicate size capture differences in portfolio risk within the cross-section of the VC market.

7 8

These results are available from the authors upon request. Thanks to Gurdip Bakshi for bringing this to our notice.

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

559

Table 10 Robustness check- portfolio quality and the skewness of VCs’ investment distributions. PQt Investment stage specializationt Industry specializationt Portfolio skewnesst Average fund sizet

I

−0.2005c (−2.76) 0.0558 −1.08 0.5525c −14.03 5.71e−05 −0.72

SyndSizet II

Investment stage specializationt+1 Industry specializationt+1 Portfolio skewnesst+1 Average fund sizet+1 Massachusetts North East U.S. (excluding MA) All other states Constant R-Squared # of obs.

−0.021 (−0.65) −0.0742a (−1.75) −0.0832b (−2.11) 0.8218c −11.9 0.3485 577

PQt+1 III

SyndSizet+1 IV

−0.1114 (−0.64) 0.1973c −2.54 0.8130c −7.77 1.93e−05 −0.29 −0.0202 (−0.35) −0.0379 (−0.55) 0.0192 −0.18 0.6617c −4.48 0.2938 343

0.4008 −0.79 1.2034c −4.17 −0.0334 (−0.14) −0.0001 (−0.55) −0.2301a (−1.67) −0.2002 (−1.33) −0.5843b (−1.98) 2.5604c −5.66 0.1224 343

0.3423 −0.92 −0.0918 (−0.34) −0.4097b (−2.08) −0.0018c (−7.01)

−0.8543c (−4.72) −0.7091c (−3.21) −0.4200a (−1.74) 4.1017c −9.99 0.1165 577

This table reports results from empirical tests that examine the relation between portfolio quality or syndicate size and portfolio characteristics, such as average fund size, and industry or investment stage specialization during the short-run (first 5 years of business) and long-run periods (year 6 of business, and onwards). The short-run is regarded as the time frame for new entrant VCs to establish a venture capital portfolio, and it’s length is empirically determined in Table 3. Except for the inclusion of the skewness of investment distributions in the empirical tests reported in this table, the empirical tests are identical to those reported in Table 5. The empirical tests are implemented using OLS regressions, and the model specifications for the short-run period are: PQit = f (InvStageit , IndSpecit , Skewit , FundSizeit , Locationi ) SyndSizeit = f (InvStageit , IndSpecit , Skewit , FundSizeit , Locationi ) where PQ is portfolio quality; SyndSize is the average number of firms participating in financing deals that involve firm i; IndSpec −Q and InvStage are industry and investment stage specialization Herfindahls; Portfolio Skewness (SK) = E y −Q2 , where  and Q2 | t 2| are the mean and second quantile (median) of firm i’s capital disbursements to projects (yt ). FundSize is average fund size, and Location is a discrete variable which indicates a firm’s geographical location. The regression models are similarly specified for the long-run period (results in columns III and IV). The construction of all other variables is described in Table 4. Regression coefficients reported for the geographical location dummies are relative to VCs located in the state of California. Data on venture capital transactions are from VentureXpert, and sample firms started business between 1980 and 1998. t-statistics are reported in parentheses. a Significance at 10% confidence level. b Significance at 5% confidence level. c Significance at 1% confidence level.

The results in column III show portfolio quality continues to increase with the skewness of investment distributions during the long-run period. Consistent with our finding that portfolio risk does not increase with syndicate size during the long-run period, the results in column IV show syndicate size is not correlated with the skewness of investment distributions during the long-run period. The results in Table 10 provide strong corroborative evidence that both portfolio quality and syndicate size capture differences in portfolio risk within the cross-section of the VC market (Table 11).

560

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

Table 11 Robustness check—covariance matrix of variables in Table 6. Excess exit ratest Excess exit ratet (EERt ) PQt+1 Syndicate sizet+1 Average SSt and EERt High SSt and EERt Average PQt and EERt High PQt and EERt Constant

0.00224 2.01e−05

−9.49e−05

0.00222 −1.50e−05

0.00964 4.52e−05 4.28e−05 −0.00956 −0.00956

−2.76e−05

−0.0003

0.01331 −9.31e−07 1.42e−05 −0.00955 −0.0101 −0.0046 −0.0065 −0.0001

This table reports the variance-covarince matrix (relative to excess short-run exit performance only) that are obtained from empirical tests reported in Table 6 in order to demonstrate that the results documented in Table 6 are not unduly influenced by noise or multicollinearity in the data. All variables and empirical tests are as described in Table 6. The Venture capital data are obtained from VentureXpert, and sample VC firms commenced business between 1980 and 1998.

5. Conclusions This paper finds evidence of reversals in relative performance in the VC market. Using proxies for the risk of venture capital assets that are derived from VCs’ risk management strategies – investment staging and deal syndication – we find reversals in relative performance are explained by the coexistence of persistent and non-persistent attitudes towards risk along two complementary dimensions. First, while syndication strategy is persistent, attitudes towards portfolio risk reverse between the short- and long-run periods, such that syndication is associated with high risk portfolios in the short-run, and low risk portfolios in the long-run. Second, VCs that hold high quality portfolios in the short-run do not persist with this strategy; however, high quality portfolios are associated with superior long-run exit performance. The performance effects of transitions in attitudes towards risk that we observe within the VC market are consistent with theories of costly reputation building in markets characterized by adverse selection problems and empirical evidence that VCs’ deal screening skills are more important for success than advisory or monitoring skills. Acknowledgements Thanks to an anonymous referee for valuable comments. Thanks also to participants at the 2007 Northern Finance Association and the 2008 Eastern Finance Association annual meetings for helpful comments. We are grateful to Jiekun Huang, and Debarshi Nandy for helpful comments. All errors are the authors’ exclusive responsibility. References Barberis, N., Shleifer, A., & Vishny, R. W. (1998). A model of investor sentiment. Journal of Financial Economics, 49, 307–343. Berk, J. B., & Green, R. C. (2004). Mutual fund flows and performance in rational markets. Journal of Political Economy, 112, 1269–1295. Bottazzi, L., Da Rin, M., & Hellmann, T. (2008). Who are the active investors? Evidence from venture capital. Journal of Financial Economics, 89, 488–512. Bowley, A. L. (1920). Elements of statistics. New York: Charles Scribner and Sons. Chemmanur, T. J., & Fulghieri, P. (1994). Reputation, renegotiation, and the choice between bank loans and publicly traded debt. Review of Financial Studies, 7, 475–506. Chevalier, J., & Ellison, G. (1997). Risk taking by mutual funds as a response to incentives. Journal of Political Economy, 105, 1167–1200. Chopra, N., Lakonishok, J., & Ritter, J. (1992). Measuring abnormal performance: Do stocks overreact? Journal of Financial Economics, 31, 235–268. Chuang, W., & Lee, B. (2006). An empirical evaluation of the overconfidence hypothesis. Journal of Banking and Finance, 30, 2489–2515. Cochrane, J. H. (2005). The risk and return of venture capital. Journal of Financial Economics, 75, 3–52. Daniel, J., Hirshleifer, D., & Subrahmanyam, A. (1998). Investor psychology and security market under- and overreactions. Journal of Finance, 53, 1839–1885. De Bondt, W., & Thaler, R. (1985). Does the stock market overreact? Journal of Finance, 40, 793–805.

O.A. Obrimah, P. Prakash / Journal of Economics and Business 62 (2010) 537–561

561

De Bondt, W., & Thaler, R. (1987). Further evidence on investor overreaction and stock market seasonality. Journal of Finance, 42, 557–581. Diamond, D. (1989). Reputation acquisition in debt markets. Journal of Political Economy, 97, 828–862. Fama, E. F. (1998). Market efficiency, long-term returns, and behavioral finance. Journal of Financial Economics, 49, 283–306. Fama, E., & French, K. (1996). Multifactor explanations of asset pricing anomalies. Journal of Finance, 51, 55–84. Gervais, S., & Odean, T. (2001). Learning to be overconfident. Review of Financial Studies, 14, 1–27. Gompers, P. (1995). Optimal investment, monitoring, and the staging of venture capital. Journal of Finance, 50, 1461–1489. Gompers, P., & Lerner, J. (2004). The venture capital cycle (2nd edition). Cambridge: MIT Press. Gompers, P., Kovner, A., Lerner, J., & Scharfstein, D. (2008). Venture capital investment cycles: The impact of public markets. Journal of Financial Economics, 87, 1–23. Groeneveld, R. A., & Meeden, G. (1984). Measuring skewness and kurtosis. The Statistician, 33, 391–399. Hinkley, D. V. (1975). On power transformations to symmetry. Biometrica, 62, 101–111. Hochberg, Y., Ljungqvist, A., & Lu, Y. (2007). Whom you know matters: Venture capital networks and investment performance. Journal of Finance, 62, 251–301. Huang, C., & Litzenberger, R. H. (1988). Foundations for financial economics. New York: Elsevier Science Publishing. Ippolito, R. A. (1992). Consumer reaction to measures of poor quality: Evidence from the mutual fund industry. Journal of Law and Economics, 35, 45–70. Jegadeesh, N., & Titman, S. (1993). Returns to buying winners and selling losers: Implications for stock market efficiency. Journal of Finance, 48, 65–91. Kaplan, S. N., & Schoar, A. (2005). Private equity performance: Returns, persistence, and capital flows. Journal of Finance, 60, 1791–1823. Kaplan, S. N, Sensoy, B. A., & Stromberg, P. (2007). Should investors bet on the jockey or the horse? Evidence from the evolution of firms from early business plans to public companies. Journal of Finance, 64, 75–115. Kim, T., & White, H. (2004). On more robust estimation of skewness and kurtosis: Simulation and application to the S&P500 Index. Finance Research Letters, 1, 56–73. Knill, A. (2009). Should venture capitalists put all their eggs in one basket? Diversification versus pure play strategies in venture capital. Financial Management, 38, 441–486. Lakonishok, J., Shleifer, A., & Vishny, R. W. (1994). Contrarian investment, extrapolation, and risk. Journal of Finance, 49, 1541–1578. Lerner, J. (1995). Venture capitalists and the oversight of private firms. Journal of Finance, 50, 301–318. Lerner, J., Schoar, A., & Wong, W. (2005). Smart institutions, foolish choices? The limited partner performance puzzle. Working paper 4523-05; MIT Sloan School of Management. Lintner, J. (1965). The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets. Review of Economics and Statistics, 47, 13–37. Lynch, A. W., & Musto, D. K. (2003). How investors interpret past fund returns. Journal of Finance, 58, 2033–2058. Malmendier, U., & Tate, G. (2005). CEO overconfidence and corporate investment. Journal of Finance, 60, 2661–2700. Malmendier, U., & Tate, G. (2008). Who makes acquisitions? CEO overconfidence and the market’s reaction. Journal of Financial Economics, 89, 20–43. Metrick, A., & Yasuda, A. (2007). The economics of private equity funds. Working paper. Post, T., van den Assem, M. J., Baltussen, G., & Thaler, G. H. (2008). Deal or no deal? Decision making under risk in a large-payoff game show. American Economic Review, 98, 38–71. Sahlman, W. A. (1990). The structure and governance of venture capital organizations. Journal of Financial Economics, 27, 473–521. Schwienbacher, A. (2008). Venture capital investment practices in Europe and the United States. Financial Markets and Portfolio Management, 22, 195–217. Sharpe, W. F. (1964). Capital asset prices: A theory of market equilibrium under conditions of risk. Journal of Finance, 19, 425–442. Sirri, E., & Tufano, P. (1998). Costly search and mutual fund flows. Journal of Finance, 53, 1589–1622. Tian, X. (2007). The Role of Venture Capital Syndication in Value Creation for Entrepreneurial Firms. Working paper.