Inside Rounds and VC Returns

Transcription

1 Inside Rounds and VC Returns Michael Ewens, Matthew Rhodes-Kropf and Ilya Strebulaev First version: June 2014 This version: May 4, 2015 Preliminary, please do not cite. Abstract We study sequential investment decisions in the venture capital (VC) industry. VC-backed companies typically need to raise several rounds of funding from VC funds. The decision whether to provide further funding to the company and the terms of the new funding determine the returns of VC funds and their ability to back successful companies. We show that investment outcomes in the VC industry can be predicted by whether the existing VC investors can attract new outside investors to participate in the next round. Inside rounds, in which only existing investors participate, lead to a higher likelihood of failure, lower probability of IPOs, and lower cash on cash multiples than outside rounds. We explore a number of possible explanations for this result including escalation of commitment, mismeasurement of returns and too little capital. The strong relationship between inside rounds and outcomes remains. Ewens: California Institute of Technology; Rhodes-Kropf: Harvard University and NBER; Strebulaev: Stanford University and NBER. 1

2 1 Introduction Innovation in technology and business practices is widely believed to drive long-term economic growth. In the Unites States, a sizeable fraction of innovation activity is done by young high-growth companies financed by the venture capital industry. For example, recent evidence shows that the companies that had been backed by venture capital generate revenue of about $3 trillion per annum or 20% of the U.S. GDP and employ about 12 million people in the United States. Understanding the economics of the venture capital industry is therefore an important pursuit. Recently, significant progress has been made in exploring this industry. A number of stylized facts about venture capital (VC) has emerged that suggests that the way the VC industry operates is di erent from many other financial sectors. First, VC funds expect to lose on many, if not most, of their investments, because predicting the success of highgrowth technologically innovative companies is challenging. Second, VC funds expect to make money (and cover their losses) from a small number of successful and very profitable investments. Third, because success is so di cult to predict, VC funds exercise real options by investing in stages. For example, a typical $100 million early stage VC fund may invest a relatively small amount of money in up to companies. While it will see its investment evaporate on most of these companies, it will then invest further in remaining companies. For successful VC-backed companies the number of investment rounds can easily reach 4 or 5. An important determinant of the VC industry returns as well as its ability to channel capital to successful innovation is therefore the sequential allocation of capital (e.g. Gompers (1995)). To be successful, VC funds need to allocate su ciently small amount of capital to arelativelylargenumberofstart-ups,observetheevolutionofthesestart-upsclosely,make e cient and timely calls to liquidate those start-ups that don t perform well and invest an increasing amount of capital in those start-ups that show great potential. 2

3 In this paper we address an important gap about our current knowledge of the VC industry by studying to what extent these sequential decisions are made optimally and how it a ects outcomes and returns of VC investments. While the success of individual funds as well as the whole industry hinges on the ability to separate lemons from peaches as early as possible, there are a number of factors that may prevent venture capitalists from making these decisions optimally. Consider for example, a situation when a VC fund invested in astart-upthatdoesnotperformwell. Itmaybeoptimaltoliquidatethisstart-upby freeing VC fund s capital and its partners time to devote to better start-ups in its portfolio. However, unlike investors in public equity markets, venture capitalists participate actively in the corporate governance of their portfolio companies. Specifically, they sit on start-up boards and interact frequently with the founders and executives. This interaction improves their knowledge of start-up prospects. They may also develop a personal attachment to the start-up and its employees. A large psychological literature shows that these interactions frequently lead to the so-called escalation of commitment. E ectively, in making future decisions an agent cannot ignore the past decisions that were made and the costs that were sunk. In dynamic situations, escalation of commitment can lead to throwing good money after bad, in industry s parlance. Escalation of commitment is an important problem recognized by the VC industry. As asolution,ineachsubsequentroundsoffunding,theexistingvcinvestorsactivelyseekto attract those VC investors that have not invested in the same start-up in the past (here, outsiders ). By investing, as well as negotiating the terms of the investment round, new investors signal to the market the fair value of the company as well as unbiased opinion about its potential. In fact, many limited partners will ignore any marking up of an entrepreneurial firm that does not include at least one outside investor. The perceived wisdom in the industry is that sequential transactions that involve outsiders leads to the resolution of the escalation of commitment problem, while the deals that involve only those investors that 3

4 already invested in the start-up previously may lead to lower returns and are thus actively avoided. In this paper, we explore whether the perceived wisdom has any grounds in the data. We start by showing that for many VC investments only those VC funds that invested in the start-up before take part in the next round. We call these round of investments inside rounds. In our sample that covers the period of and has a total of 21,596 investment rounds in 8,990 entrepreneurial firms (excluding the initial investment rounds), there are 6,459 inside rounds. These rounds are such that all investors in that round already invested in that start-up before. Such a high fraction of inside rounds is surprising given anecdotal evidence on how the industry works. But does it matter? We show that it does. Inside rounds are more likely to lead to failures, are less likely to lead to IPOs, and generate lower cash on cash multiples than outside rounds. These findings persist when we control for a number of company-specific and time-series variables. We also show suggestive evidence that suggests that those VC funds that participate in inside rounds may have lower returns on those inside investments than those VC funds that participate in outside rounds. We also find that several features of financings, investors, and firm history are good predictors of the outside investor participation. Smaller rounds of financing and those that take place later in a firm s life are more likely to be inside rounds. Moreover, if the current syndicate lacks a successful exit in their portfolio in the recent past, we find an 18% higher probability of an inside round. The analysis considers several possible explanations for the relationship between inside rounds and investment outcomes. VC firm fixed e ects analysis shows that the results are not driven by the quality of participating investors and the quality of their investments. It is possible that the underperformance of inside rounds stems from a few insiders failing to rationally walk away. We find there is no di erence in returns and outcomes for inside rounds with full or little participation by existing investors. Inside rounds have many observable 4

5 di erences with outside rounds, of which total capital invested is primary. Perhaps inside rounds are VCs attempts to increase bargaining power by holding back capital. We find that inside rounds with dramatically smaller financing amounts relative to firm capital stock have the same return performance. Finally, it is possible that inside investors do not consider the current investment in isolation, but rather attempt to maximize the returns on their total equity position. Our preliminary attempts at measuring the total return earned by insiders the sum of returns across all investments still show strong negative correlations. Overall, the relationship between inside rounds and investment returns is quite robust to a host of rational and behavioral explanations. Our results are important because they shed light on the determinants of VC returns and show that there is a certain degree of cross-sectional predictability in this industry. While the perceived wisdom would suggest that the escalation of commitment is the leading mechanism at play, a host of other mechanisms can explain some of our findings. Inside deals suggest that there is no outside interest and the objective prospects of the company are poor. The insider investors, however, face agency problems vis-a-vie investors in their funds (limited partners or LPs). For example, in the absence of another successful deal in their current fund, they may have incentives to gamble for resurrection. If they are raising another fund at the same time, they may want to prevent their existing companies to go bust as this may scare potential LPs. The VC environment arguably constitutes one of the best grounds to test various facets of agency problems. While we don t have access to full contracts signed between VCs and their portfolio companies (or VCs and their LPs), some of our tests suggest that agency problems can be at play. Furthermore, an inside round presents a unique view of two opposing hold-up problems that are the center of several theoretical analyses of staged financing (e.g. Admati and Pfleiderer (1994) and Fluck, Garrison, and Myers (2006)). The entrepreneur can walk away from the startup and likely lead to failure of the investment. The existing investors have the 5

6 capital unavailable to the entrepreneur and can threaten to abandon. As most startups are cash flow negative and rarely have large capital stocks, this threat is real. It is not clear from a theoretical standpoint which hold-up problem will dominate in inside rounds. For example, avcinvestormayhaveincentivesstemmingfromtheirtotalportfolioperformancethatlead to bargaining power shifted to the entrepreneur. The VC may have the bargaining power if the inability to raise outside capital stems from a negative shock to the entrepreneurial firm. An analysis of pricing and returns can shed light on typical hold-up scenario. While the escalation of commitment and agency costs explanations lead to VCs taking negative NPV decisions, continuing to invest in the presence of negative information can also be optimal for a number of reasons. Most importantly, the VC capital allocation is a multi-stage investment process and negative information at each stage may be insu ciently negative to abandon a project. In other words, even though NPV is negative, the adjusted NPV (taking into account embedded real optionality) can still be positive. A classical real option model of McDonald and Siegel (1986) can be adjusted to demonstrate this result. In itself, this reasoning is not su cient to preclude raising funds from the outside investors, to the extent that outside investors also understand the option nature of the decision making process. One potential mechanism is the existence of soft information available to insiders that is di cult for outsiders to evaluate. (Obviously, it also has to be taken into account that the interactions between VC investors is dynamic, with the inside-outside roles reverse many times over the course of life of the VC career, and reputation e ects are very strong.) Likewise, down rounds can be consistent with the real option model as well, if the nature of information (i.e., the shock that agents receive) is within a certain range. Importantly though, the true NPV of the project should still be positive to the VC insiders. If our proxy for VC returns captures the ex-post realization of these VC expectations well, then it is less likely that the real options rationale can explain our stylized facts. Note that VC funds can get higher returns also through better contracting terms with start-up founders. 6

7 Our paper builds on both theoretical and empirical research in finance and psychology. Escalation of commitment is defined in psychological and organizational literature as agreatertendencytocontinueanendeavoronceaninvestmentinmoney,e ort,ortime has been made, even if circumstances should dictate otherwise. It is closely related to the sunk cost fallacy and the endowment e ect. See Arkes and Blumer (1985) for experimental evidence and Staw and Hoang (1995) for empirical evidence of sunk costs (in the context of NBA games). There are also two interesting studies of organizational escalation that use the case studies of very large non-profit projects: Expo86 World Fair and the Shoreham Nuclear Power Plant (Ross and Staw (1986), Ross and Staw (1993)). In that literature, escalation e ects have been typically explained by prospect theory, self-justification concerns, inside view, the desire to avoid wasting resources, etc. From the economic viewpoint, escalation e ects are irrational if they lead people to take negative NPV projects. Escalation of commitment can reveal itself in many ways. For example, investors may prefer to modify project goals or standards instead of abandoning projects, in an e ort to create a more favorable outcome. In the VC context, a lack of clearly defined milestones naturally exacerbates the problem. It also could be due to rational overcommitment (Adner (2007)), the tendency of individual managers to continue projects with the hope of improving the outcomes, especially when their personal interests are at stake. Two related papers, Guler (2007a) and Guler (2007b), show that VCs, as a group, tend to make sequential investments in deals even if objective criteria suggest the deals need to be abandoned. Guler (2007b) lists the following reasons why escalation of commitment in face of negative information may be prevalent in the VC setting, based on her reading of the organizational and psychological literatures: (1) inability to update prior beliefs with new information; (2) failure to treat prior investments as sunk costs; (3) framing subsequent investments as opportunities to recover prior losses; (4) avoiding cognitive dissonance and saving face by further committing to earlier decisions. In addition, the organizational 7

8 structure of the VC funds may be important for how the escalation of commitment works. Typically, a deal has a pioneer in a VC partnership and the identity of that partner (seniority, previous success) as well as the structure of the company (size, hierarchical nature) may play a role. In experimental paper, Tan and Yates (2002) study how financial budgets a ect termination decisions. They find that escalation of commitment declines as financial budget gets binding. In relation to the VC industry, the extent of the escalation of commitment can depend on dry powder, other investments in the same fund, their current perception of success (future success can also be used as a proxy, assuming that the VC knows better about the future). There is also now a broad recognition of emotional biases and the importance of a ect in decision-making (see Loewenstein and Lerner (2003) and Lucey and Dowling (2006)). This is related to the endowment e ect and can lead to escalation. A related e ect is gambling in the presence of losses if there is a chance to break even (Thaler and Johnson (1990)). The paper proceeds as follows. Section 2 describes the data and variables. Section 3 presents some preliminary empirical tests. Section 4 follows with a set of robustness tests. Section 5 studies the decision to participate in an inside round and Section 6 concludes. 2 Data and variables This section describes our data and introduces main variables of interest. 2.1 Data Sources and Definitions of Variables We use the venture capital database VentureSource from Dow Jones. The data includes information about financing rounds, entrepreneurial firms, and investors over the period from 1990 to VentureSource is augmented and improved with data from Correlation 8

9 Ventures, a quantitative venture capital fund. 1 The additional data includes financing-level and exit valuations, which are important for returns calculation. The analysis in the paper considers a subset of all available equity financings. We start constructing the main dataset by considering only non-first round financings where we can begin to track the reinvestment decisions of insiders. An entrepreneurial firm is in the sample if it raised its first round of financing between 1990 and The upper bound allows ample time for an exit event for the investment. We exclude all firms that were founded prior to Next, each financing event must follow a previous equity financing where at least one of the investors is a traditional venture capital investor. Such an investor raises the fixed-life fund (10-12 years) from institutional investors such as endowments, pension funds, and trusts. In VentureSource, this includes institutional venture capital, diversified private equity, and SBIC (government grant-backed) funds. Hedge funds, mutual funds, investment banks, and corporate VC arms are excluded from this definition. Finally, we drop financings with small investment amounts to avoid potential hidden bridge financings or incorrectly split equity tranches. Small is defined by industry classification: financings in the lower 5% of investment size over the sample period are dropped. The main specification includes 21,596 financing rounds in 8,890 entrepreneurial firms. Acriticalvariableofinterestisthedegreeofround s insidedness. Thenotionofan inside round concerns the dynamic investment of investors and the staging of financings. Our sample includes only financing rounds that are preceded by at least one VC financing, where we can observe the set of participating investors. In each financing round t, t 2, investors that contributed capital in at least one prior round, are called Inside investors or simply insiders. If an investor in round t did not participate in any of the prior financings of this company, that investor is called an Outside investor at round t. All equity financing rounds can be classified as either inside or outside rounds. A dummy variable Inside 1 Ewens and Rhodes-Kropf are advisors to and investors in the fund. 9

10 VC Round equals one for financing round t if all VC investors contributing capital in this round are insiders. If there is at least one outside VC investor, then the round is Outside VC Round. We can also use more stringent definitions: Inside Round equals one if all (VC and non-vc alike) investors in this round are insiders. if at least one investor is new, it is then called an Outside Round. Clearly, all rounds, for which Inside Round equals one, will also have Inside VC Round equal one, but the reverse is not necessarily true. We also consider two continuous counterparts to this dummy variables. First, we specify the fraction of dollars provided by inside investors in financing round t of entrepreneurial firm i, regardlessofthecharacteristicsofnewinvestors,ki it,whichis: KI it = P j2i it 1 K ijt K it (1) where I it 1 is the set of inside investors (i.e., all investors that contributed in at least one round prior to round t), K it is the total capital raised in the financing t and K ijt is the capital contributed by investor j (which can be zero). Second, we also specify a similar fraction based on the number of investors. Let Inv it be the set of all investors in financing round t. Then, FracInside it = P j2i it 1 1[j 2 Inv it ] #Inv it (2) is the fraction of those investors that are insiders. The indicator 1[j 2 Inv it ]isoneifinvestor j is an investor in round t. To estimate the value of companies, we use the so-called Pre-money and Post-money valuation of entrepreneurial firms. The post-money valuation, Vit Post,valuescompanyi in financing round t taking into account the capital K it contributed in that round by investors on an as-converted-common equity basis. In other words, the valuation assumes that all the securities issued to investors will be converted to common equity upon exit. Although 10

11 it is a standard valuation technique in the VC industry, it is important to note that the valuation on an as-converted basis assumes that the eventual exit will be successful and VC investors that typically receive convertible preferred securities have an incentives to convert their claims into common equity. This measure gives a good indication of the change in valuation between rounds. Pre-money valuation, Vit Pre,isthevaluationofthecompanyin round t before the capital injection is taken into account. In other words, V Post it = V Pre it + K it. (3) We define the financing round t to be an up ( down, flat ) round, if the pre-money valuation in round t increases (decreases, is not changed) relative to the post-money valuation in round t 1(thatis,ifV Pre it >Vit Post 1, Vit Pre <Vit Post 1, Vit Pre empirical analysis, we often will combine the down and flat rounds. = Vit Post 1,respectively). In To calculate the return on a deal for non-failed investments, we require the final exit valuation (e.g., IPO valuation, acquisition price, etc.) and the dilution of equity due to subsequent financing events. We focus on the gross multiple variable M it for firm i in round t, whichisdefinedas: M it = V i V it TY s=t+1 where V i is the exit valuation, T is the total number of equity financing rounds, and the sequence D is are the dilutive factors. Any time an entrepreneurial firm raises outside equity, the positions of previous investors are diluted by one minus the fraction of equity sold. That D is, is, D is is calculated as 1 I is /V is,wherei is is the total capital raised in financing s. Thegross multiple considers a hypothetical dollar invested in round t, so these new financings directly impact the equity stake of the investor. For example, suppose a second round financing is followed one year later by a new equity round that raises capital in return for 30% of the 11

12 equity. The equity stakes of the investors in the second round financing will all fall by 30%, so that D it+1 is.7. Korteweg and Sorensen (2010) provide a detailed description of returns calculation and the discussion of sample selection issues. 2 Another important variable is the eventual outcome of the entrepreneurial firm. In the VC industry jargon, those are called exits. The three main exits for VC-backed companies are an initial public o ering, an acquisition or a shutdown/failure. For a subset of nonfailures, we explicitly observe the exit valuation V i.foripos,thecoverageisnearly100%, for acquisitions, it is 54%. As in other research using investment-level returns, these reported valuations are positively selected. A company without a valuation implies we cannot calculate any financing returns. There is an over-representation of quality exits IPOs along with low coverage of historically lower valued acquisition exits. A counterbalancing measure is that if the firm fails outright then returns are low and can assumed to be zero. Because of the selection issues, we consider both the log of exit valuation and dummy variables for exit types. 2.2 Descriptive statistics Table 2 provides descriptive statistics on some basic characteristics of the sample. The general variables, such as financing year, industry distribution, and total capital raised are consistent with other research. In terms of final outcomes, 10% of start-ups eventually become publicly listed companies, 44% are acquired, 21% are listed as outright failures, and 26% are still private (as of 2014). The latter category consists mostly of failures, too, given that they must have received their first equity financing in 2008 or earlier. The distribution of life spans confirms the failure hypothesis for many of these private firms. Unreported 2 For entrepreneurial firms that were acquired with a known exit valuation, we assume investors receive their capital back by order of investment stage and when we are unable to calculate returns. That is, if the total exit valuation is first split so that the last investors are made whole, followed by the previous investors. This approach assumes that each investor has senior 1X liquidation preference. We gain approximately 1,900 returns from this, however, the results are similar if we continue to treat them as missing. 12

13 in the table, out of 2,382 firms that are still private, 1,119 have not exited for more than 10 years and 823 for more than 12 years. In addition, as well known in the VC industry, acquisitions span a wide range in terms of valuations, from outright failures (penny sales) to huge successes. We have the exit valuation for 2,360 acquired firms, that constitutes about 60% of the total acquired sample. Of those, 1,414 have an exit valuation that exceeds capital invested. The average firm is founded in 1998, with the funding year and the company age widely distributed. More than 40% of the firms in the sample are from California, consistent with Silicon Valley being the world-wide hub of venture capital activity. The average firm in the sample raises 4.2 rounds of funding, between the minimum required 2 (by sample construction) and the maximum of % of the sample feature syndicated investment, in which at least two investors provide capital in the round. On average, each round of financing features a participation of almost 4 investors and a total capital injection of $13.4. While the median round size is $8.5 million, the amounts vary from insignificant rounds to more than $1 billion. It takes on average 1.3 years for the firm to raise its next round of funding, consistent with the prevailing industry notion that most firms raise subsequent funding in about 12 to 18 months. Valuations are known only for about 55% of the sample and gross multiple for 44%. For the subsample with valuations, the average capital raised is $15.2 million, similar to the $13.4 million value reported for the total sample. The average (median) post-money valuation of $87.7 ($38.2) million suggests that this subsample is biased, as expected, towards more successful companies. However, while the average gross multiple is a healthy 2.72, the median gross multiple is 0.22, indicating that in at least 50% of deals, for which this variable is reported, investors lose most of their money. The positive selection bias in this subsample thus ensures that investors are likely to lose on an overwhelming fraction of their investments. 13

14 Only the presence of very successful outliers enables some VC funds to make up for losses and deliver positive returns on their portfolios. Table 3 provides descriptive statistics on our main insidedness variables for the main sample (Panel A) and for the subsample, in which the amount of financing at the previous round is known (Panel B). In Panel A, all the investors are equally-weighted, while in Panel Binvestorsarevalue-weightedbytheircapitalcontributions. Insideinvestorsconstitute 64% of all the investors and contribute 61% of the capital in the next round of funding in an average deal, while the outside VCs constitute on average around 21% of the financing round. In around 30% of cases, however, there are no new investors in the round and in almost half of the cases there are no single new outside VC investor (to remind, there could be non-vc new investors, such as corporations, angel investors, individuals or investment banks, leading to the discrepancy in the values of these two variables). These findings suggest that the rounds in which only insiders participate are more common in the VC industry than some previous anecdotal evidence may have led researchers to conclude. As a stronger alternative to the presence of outside VCs, we also consider the identity of lead investors in the financing round. Lead investors are those that contribute the most capital to the round and are in reality the drivers of creating the funding syndicate. We identify leads in two ways. First, VentureSource may flag an investor or multiple investors as leads. Second, if such a flag is not available, then we identify the lead as the investor who provides the most capital. The division of capital within a round and across investors is often missing, so we split capital equally to those with unknown contributions. A round may be considered an outside round even though new investors may have contributed a small fraction to the financing. The presence of an inside lead VC investor is thus a strong indicator of insidedness. With this lead investor flag, a round is considered having a new lead investor if there is at least one new investor (i.e. outsider) who is also identified as a 14

15 lead in the round. We find that in almost 30% of cases, the round does not feature a new lead investor. To explore further the prevalence of insideness, Table 4 reports the insidedness and financing variables by the funding round. Panel A considers the traditional second financing round, while Panels B and C report the descriptive statistics of 3rd and 4th rounds, respectively. (The financing rounds are known in the industry by letters, e.g. the 2nd round would typically be called a Series B round.) All the measures of insidedness show a secular increase over financing rounds, indicating lower outside participation over the lifespan of the firm. For example, while 40% of 2nd rounds feature no outside VCs, 47% and 51% do not have outside VCs in the 3rd and 4th rounds, respectively. There are also more 4th rounds, at 36%, that are full inside rounds. Even though the latter panels are subject to significant survivorship bias, because many deals do not reach another round either because they failed or because they exited, the results are quantitatively similar for the sample that requires firms to survive at least N rounds, N being 3 or 4 (see Internet Appendix). 2.3 Inside vs non-inside rounds Are inside rounds truly di erent? In this section, we provide some preliminary comparisons between inside and outside rounds. Tables 5 and 6 and Figures 1 3 consider di erences in valuation, capital raised, returns, and other financing-related variables. Figure 1 compares the change in valuation between two subsequent rounds by showing the kernel densities of the ratio of pre-money valuation in the round to the post-money valuation in the previous round, Vit Pre /Vit Post 1, for inside and outside rounds. This ratio captures the relative gain in value for existing investors and is a good indicator of whether the entrepreneurial firm is moving in the right direction. A value of 1 is known as the flat round (also singled out by the vertical line), less than one is down, and greater than 1 15

16 is up. E ectively, for firms experiencing down or flat rounds the news is negative, either because of the idiosyncratic shocks or changes in the macroeconomic situation. The figure shows that inside and outside rounds experience di erent valuation dynamics. Inside rounds are much more likely to be down rounds and also have a larger mass at around 1. Moreover, the inside density attenuates much faster, implying lower likelihod of a round with a large increase in valuation. Taken together, the figure clearly shows that inside rounds are relatively worse in terms of valuation dynamics than outside rounds. Note that the sample of financings with a known valuation is selected positively in that firms fully reveal valuations when they go public and tend to reveal them if they are growing quickly. Figure 1, and the analysis below, however, suggests that our data capture a large chunk of negative valuation outcomes. Moreover, there is no ground to believe there is a strong selection mechanism that works di erently for outside and inside rounds (especially the one that would lead to inside rounds be more negatively selected). To check that the di erence between inside and outside rounds persevere, Figure 2 presents kernel densities by the log of capital raised, the financing variable that is available for most of the rounds in our dataset. Capital injection plays an important role in the VC industry, because corporate governance considerations make many VC investors demand a certain fraction of ownership in the startup firm, leading to a relationship between valuation and capital injection. The vertical line shows the mean capital invested, which is approximately $7.3m. The figure shows that the two round types di er dramatically. Inside rounds raise significantly smaller amounts of money than outside rounds. Small amounts of capital invested in inside rounds could imply that these rounds are quickly completed and may in fact be hidden bridge rounds. For example, such rounds could be used by insiders to quickly provide capital to startups when facing both internal and external shocks. Unreported results show, however, that when we repeat the analysis that excludes what appear to be bridge rounds (tenth of all financings 16

17 that raise the least capital and a quarter of all financings that are the earliest to be followed by the next financing), the results remain unchanged. 3 Successful VC backed start-ups tend to raise more capital at each subsequent round of financing. A slower or flat ramp up may indicate a firm struggling to raise capital or reach milestones. Figure 3 shows the di erences between inside and outside rounds in terms of relative capital raised across rounds. The leftmost red line in the figure shows a ratio of 1, while the rightmost line shows the mean across the whole sample. The average financing is twice as large as the previous. The inside rounds, however, are significantly more likely to raise less capital than the previous financing. Table 5 compares inside rounds to outside rounds and to the full sample along a number of other dimensions. Several results that stand out provide further di erentiation between inside and outside rounds. Inside rounds raise on average more rounds of funding, backed by VCs for a longer period of time, and are older. However, the span of time it takes for startups to raise the next round of funding is the same for inside and outside rounds, suggesting that while firms with inside rounds survive longer they do not raise funding at di erent time intervals. While the round syndicate size is significantly smaller for inside rounds (as expected, because inside rounds by definition are constrained to have a smaller set of investors), the total number of investors (that includes all previous investors) is virtually the same for inside and outside rounds. Inside rounds are much more likely to be a down or a flat round. For the subsample, for which we have the valuation data, these valuedecreasing rounds comprise 38% of inside rounds but only 20% of outside rounds. Finally, inside rounds feature significantly lower returns than outside rounds. The mean (median) gross multiple for inside rounds is 2.2 (0) versus 2.9 (0.3) for outside rounds. Given that it takes longer for inside rounds to exit, the actual di erence in returns is even larger. 3 See the Internet Appendix for the figure and summary statistics. 17

18 Conventional wisdom often mentioned by venture capitalists and those familiar with the VC industry is that entrepreneurial firms should do everything possible to avoid down rounds. Table 6 compares inside and outside rounds depending on whether the round was an up round or a down round ( flat rounds are incorporated in down rounds). The table shows that for the subsample of down rounds, inside rounds underperform across most dimensions. They are more likely to lead to zero returns as well as to fail, less likely to become public, and show lower gross multiples. In contrast, when valuation increases, inside rounds have weakly higher returns and better exit rates. These results suggest that the change in valuation is an important control variable. They also provide a first glimpse into the economics of inside rounds, suggesting that di erent economic mechanisms can be at play in down and up rounds, the issue that the next sections will explore in detail. 3 Basic empirical tests In this section we will explore the impact of insidedness on the economically important variables, including entrepreneurial firm outcomes, valuation, and investment returns. The upshot of our findings is that insidedness is an important variable explaining and predicting all of these variables that are of great interest in understanding better the economics of VC backed firms. 3.1 Inside rounds and entrepreneurial firm outcomes The exits of entrepreneurial firms are readily classified and success in the VC industry comes either as an IPO, widely considered to be the major objective in the VC industry, or as a combination of IPO and a highly valued M&A. The first question we tackle therefore is whether insidedness matter for explaining the eventual exit outcome. If inside rounds are a selection of lower quality investments, then we should expect to observe worse outcomes. 18

19 Alternatively, if the average inside round represents the insiders keeping the best deals to themselves, then we should find the opposite. Finally, if the observable features at the time of financing have little bearing on investment success, we should observe no relationship between inside rounds and eventual outcomes. Table 7 reports the results of this analysis, where we regress the eventual outcome of the entrepreneurial firm on insidedness and many control variables. We consider three outcome variables. First, in line with the VC conventional wisdom, we define an IPO dummy as one proxy for success. While the IPO dummy is very popular among researchers and is the most frequent measure of success used in the VC literature, an important problem with this dummy variable is that many M&A exits, in which the start-up is sold, are also great successes in terms of the valuation. Indeed, arguably, there are more successful M&A outcomes than IPO outcomes in the VC industry, especially recently. We therefore define another dummy, Good exit, which equals 1 if the outcome is either an IPO or an acquisition, in which a valuation is known and at least two times all the capital invested. While the exact threshold used in the definition of M&A success is necessarily ad hoc, our results are robust to using a higher threshold. Finally, we also use the eventual exit valuation as an indicator of success. We have the value of this variable only in cases of IPO or M&A, i.e. the results are conditioned on non-failure. We also use the log of exit valuation to control for the presence of many outliers. Specifically, Log exit valuation is the log of the reported valuation at sale or IPO. For the latter, the valuation if the market capitalization of the firm at the o ering price. The first three columns of Table 7 show our benchmark results. The first two columns show a strong, negative relationship between inside rounds and quality exit outcomes. The estimates imply a 14% lower probability of an IPO and a 20% lower probability of a Good exit if the firm had at least one insidedness round. The results for the log value are similar, but economically larger: firms with at least one inside round have a 34% lower exit valuation 19

20 conditional on not failing. From Table 5 we know that the average firm with an inside round fails more often, so the total di erence in exit valuation should in fact be larger than 34%. The final three columns introduce the interaction of the dummy Had down round and the indicator for an inside round. This interaction asks whether the worse outcomes observed in inside rounds can be explained by the higher propensity for down rounds in inside rounds. The inclusion of the down round control is motivated by the finding that inside down rounds seem to be economically di erent from inside up rounds. This inclusion reveals two interesting results. First, the correlation between a down round in the firm s history and outcomes is strong and negative. Second, the relationship between past inside rounds and outcomes remains after the inclusion of this interaction. Taken together, these results lead to a conclusion that the predictive power of inside rounds for outcomes is strong and is not accounted for by by past valuation changes. 3.2 Inside rounds and interim valuation changes One distinctive feature of the VC financing is that, because entrepreneurial firms raise many rounds of funding, they are valued repeatedly over its life cycle prior to the exit. The change in valuation across financing rounds reveals the interim success of the firm and the return earned by insiders from the previous round. While in most cases this return is paper money return in that inside investors holdings are illiquid and they cannot cash out, to the extent that prices reveal all available information about the eventual prospect of the firm, interim returns are useful at gauging the progress of the firm. In this section, we explore to what extent insidedness can explain changes in interim valuation. To compute the change in valuation between two subsequent financing events, we first compute the ratio of the current price over the past price paid. The current price is the pre-money valuation and represents a market capitalization prior to the new infusion of capital. The past price is the post-money valuation in the previous round of financing. 20

21 The log of this ratio thus represents a return to a hypothetical investment of one dollar in the previous round that was realized at the next round s price. 4 Again, although the venture capital market typically does not have such a secondary market, these returns are not directly earned by investors, they represent an implied return to investment. This dependent variable is especially informative in inside rounds because it represents the di erences in prices paid by the same investors over two rounds of financing. A relatively larger change in price for inside rounds indicates that either the insiders pay relatively too high a price at the previous round or they invest at relatively lower prices at the current round. The latter could stem from either negative shocks to the firm or from increased bargaining power for the insiders at the current round. In Section 4 we show that, at least for the sample we have collected, inside rounds do not exhibit di erent contractual provisions both over time series and in cross-section. Table 8 shows that inside rounds have a significant impact on the changes in valuations. To understand better the economic intuition of the result and its quantitative significance, note that the unit of observation in the table is an entrepreneurial firm financing, where one can observe the current round and previous round firm valuations. The estimates imply that inside rounds have a 15%-25% lower changes in valuation (or round-to-round return) than outside rounds. This relationship holds after controlling for financing observables such as year, industry, round sequence number, and investor count. Column (6) also shows that the relationship retains its significance, if we used no new lead dummy as a stronger proxy for insidedness. As emphasized above, this relationship is consistent with either the bargaining power or negative shock story. An analysis of the ultimate returns earned in these financings is necessary to help distinguish these explanations. 4 This variable is similar to the round-to-round return in Korteweg and Sorensen (2010) and Cochrane (2005) 21

22 3.3 Inside round and investment returns So far, we have shown that inside rounds are characterized by lower success rates, exit valuations, small capital raises and higher probabilities of down rounds. In this section, we explore whether the final returns to investment depends on insidedness. This is by far not a foregone conclusion. For example, the higher failure rate and lower IPO probabilities could be reflected in lower prices at the time of the inside round. Similarly, the down round events could be the right price that results in lower success, but later exhibiting identical returns than comparable outside round financings. Table 9 reports the results on financing-level returns. The dependent variable is the log of the gross multiple which measures the return of a hypothetical dollar invested in the financing round accounting for dilution. While this measure ignores the time value of money, it is a good first step to study returns. We again use two measures of insidedness, the dummy indicating whether the round is a full inside round and a dummy whether the round did not have a new outside lead investor. The first four columns presents the main results. In the first two regressions, we include the full sample of observed gross multiples. Failure outcomes are set to the log of 25% of capital invested. Controls include capital raised, time since last financings, and fixed e ects for financing year, industry, and the state of entrepreneurial firm headquarters. The insidedness coe cients imply a 12.8%-15.1% lower gross multiple return in inside rounds. Because our assumption about the recovery rate of investors in the failure cases is inevitably ad hoc, we also consider, in Columns (3) and (4), the subsample that includes only those firms that exited. For the inside round dummy, the coe cient is still negative, but is about half the size and loses significance; it is still strongly negative and economically significant for no new lead investor. The final four columns in Table 9 extends the analysis by including VC firm fixed e ects. Introducing VC firm fixed e ects allows us to address two major and related concerns about the previous results. First, inside rounds may simply 22

23 be conducted by relatively worse investors who have more behavioral biases or su er from worse incentive frictions. Second, VC firm quality is highly correlated with investment quality (e.g. Sørensen (2007) and the importance of deal flow). Introducing a VC firm fixed e ect will control for any time-invariant investment quality at the VC-level that is driving the results. Simply, we now compare inside and outside rounds within the VC firm portfolio. The unit of observation is a VC firm in a specific financing. All the patterns found in the first set of results remain. Thus, VC firm time-invariant quality and, in turn, some unobserved entrepreneurial firm qualities cannot explain the results. More importantly, the fixed e ects estimates imply that within the average VC firm portfolio, inside rounds are relative underperformers in terms of returns. 4 Robustness and Discussion Insidedness plays an important role in the outcomes of VC backed companies. In this section we explore a number of potential explanations for the observed di erences, as well as discuss various robustness tests that deal with measurement issues with investment returns and sample selection. 4.1 Inside participation The dummy variable Inside round simply requires that there be no new outside investors. There are no conditions on the number or composition of the insiders who do invest in the round. One possible explanation for the results above is that a small subset of insiders keep the entrepreneurial firm alive, while the majority walk away and write o the investment. If this is the case, then we are potentially underestimating the real returns earned by investors because so few participate in the financing. Columns (1) and (2) in Table 10 explore this explanation. The first has the dummy variable Full participation, which is equal to one 23

24 if all past investors invest in the current round. The second column uses a similar variable that is one if the fraction of insiders that participate is greater than the 75th percentile of participation across all rounds. If the underperforming inside round is confined to the sample with low inside participation, than the two interaction terms in columns (1) and (2) should explain the main results. The coe cients on Inside rounds are e ectively unchanged with the inclusion of these controls and the controls themselves are insignificant. The results indicate that the composition of insiders is not important for the insidedness mechanism. 4.2 Too little capital? The amount of capital invested is a strong predictor of outcomes. As Figure 2 shows, inside rounds are on average significantly smaller in terms of capital raised than outside rounds. To better understand if the size of the inside round is driving the lower returns, we construct adummyvariable Capitalrampdown. Thevariabletracksthesequenceofcapitalraised between two sequential financings of an entrepreneurial firm and equals to one if the change in capital raised is in the bottom quartile of changes across the sample. A small or negative change in capital raised across financings could signal worse prospects or strong financing constraints for the firm. If within-firm capital ramp downs drive poor results, the interaction of this dummy with inside round should be negative. As Column (3) of Table 10 shows, while the level of the variable loads negatively, as expected, there is no di erence within the sample of inside rounds. Thus, within-firm capital restrictions for inside rounds is unlikely to drive the returns di erences. 4.3 Return to total equity position The returns results indicate that a dollar invested in inside rounds returns less than dollar invested in similar outside rounds. For existing insiders, however, this may not be the appropriate comparison. An insider who faces the decision to continue investing, should 24

25 consider the return to the total equity position she has accumulated in the company after the financing closes. For example, consider a VC who invested in round A. In Round B, there are no other investors who are willing to invest in the company. The insider might be losing on any additional investment, but the total return would include a recovery rate on her original investment as well. Therefore, it is possible that the insider is willing to invest at relatively higher prices than an outsider. In other words, while from the perspective of the single investment, the insider may appear to be investing irrationally, the portfolio consideration provides a rational justification. To explore this possibility, we construct the Total return variable. The total return to an investment in a given financing is defined as the dollar-weighted return to the current return and all previous equity financings. This construction accounts for all previous investments made by the investor in this company. If insiders earn a lower return on the current financing with the goal of supporting their total equity position, the returns to the previous financings should in expectation to o set the relative losses in the current round. If this is the main mechanism, once controlled for, the inside round should be statistically indistinguishable from outside rounds. The last four columns of Table 10 test this conjecture. the results suggest that the average insider is not earning a total return that o sets the relative losses in the current financing. The coe cients on each specification are negative. For the full sample, the total return is 21 23% lower for inside rounds. For the smaller sample of non-failed companies the results are less significant. The lack of power here stems from the lack of variation of total return and return to the current financing for failed investments. Simply, the dollar-weighted mean of zero returns is zero. Overall, there is little evidence that the predictive power of insidedness for returns is driven by insider s total equity position. 25

26 4.4 Can preferred stock explain the results? The gross multiple used above assumes that the investor purchases a share of common equity or, equivalently, the investor s security is converted into common equity upon exit. As is well known, most VCs purchase preferred shares which include participation and liquidation rights. As inside rounds are more likely to be acquired and have lower exit valuations, it is possible that these contract features come into play when an inside round occurs. If this is the case, then we could be underestimating the true returns earned by investors in inside rounds. The next two tables address whether inside rounds are more likely to have non-common contract features. Table 11 exploits the database of VC contracts provided by VC Experts. This data set includes a sample of VC-backed companies primarily financed after 2005 and provides information on the contractual provisions of securities held by VCs. The data set comes from extraction of terms from articles of incorporation filed in the state of Delaware. For our analysis, we are concerned that the common equity assumption underestimates the true returns earned by the VCs. Several contract features would improve the VC returns. Liquidation preference provides downside protection for a preferred shareholder than can guarantee at least one times capital invested. These preferences can reach two or three times capital invested. Next, preferred stock seniority provides an investor additional downside coverage and priority in lower quality liquidation events. Third, the insider can purchase Participating preferred stock that provides both the liquidation preference and the ability to participate in the upside without conversion. Compared to a common equity position or preferred stock after conversion participating preferred can increase returns significantly. Finally, a preferred shareholder can have redemption rights which act as a put on their equity investment after some period of time. This right can provide bargaining power in liquidation events. 26

27 Each column of Table 11 asks whether there is a correlation between observed contract features and inside rounds. If insiders are shifting to the contracts that are more friendly to VCs, then the common equity assumption is an underestimate of average returns earned. The merge of VC Experts with VentureSource results in over 1,400 financing events with known contracts. Columns (1) - (4) show that insidedness is not statistically related to these contract features. For example, in Column (2) a dependent variable is a dummy that equals to one if the liquidation preference of the financing is greater than one times (known in the industry as 1x ). There is a weak positive relationship between insidedness and liquidation preference, which is not significant. The table also reports the mean of each dependent variable unconditionally and conditional on inside rounds. Again, the inside rounds tend to have slightly friendlier terms than outside rounds, but the di erence is small. The final column combines each contract feature with a dependent variable equal to one if any of these first four column variables equals to one. In this case, the relationship is significant, although given that over 80% of the contracts contain at least one of these rights, the evidence is weak. Given a small sample of contractual obligations, it is possible that there is a weak suggestive evidence that common equity may be an underestimate of returns for inside rounds. In the absence of better data on contracts, we proceed with the following exercise to explore in more detail the impact of better cash flow rights of VCs. Would these contracts matter if inside rounds were to have these provisions? Table 12 imposes these contracts on returns to understand if they can explain earlier results. In column (2), all inside rounds that have a return greater than 0 and less than two time capital invested are given a two times capital return. Similarly, outside rounds with observed multiples between zero and one are assigned the latter. This adjustment improves the returns on the average non-failed inside round. This is an extreme assumption, because the observed fraction of 2X liquidation in VC Experts is less than 15%, while this adjustment results in over 35% of inside rounds 27

28 having such a feature. As expected, the coe cient on the inside round dummy decreases; however, we still find a statistically significant negative relationship with returns. Column (3) takes an even stronger view. It treats all missing acquisition returns for inside rounds as a 2X gross multiple if the capital is available (i.e. the exit valuation is at least two times capital invested). We replace outside round missing returns similarly with 1X returns. After this adjustment, over 50% of the inside round sample have these features. Only after this strong assumption do inside rounds have no relationship with log returns. We conclude from this exercise that it is unlikely that strong preferred contract features would explain away the main di erences in financing returns. 4.5 Sample selection The analysis above that uses exit valuations, financing valuations and returns requires significant disclosure of information. As discussed and addressed in work by Korteweg and Sorensen (2010) and Cochrane (2005), this sample selection can dramatically bias estimates of returns. The bias is positive because the disclosure of valuation information often follows success and quality entrepreneurial firms have more investment events in the data. This bias should attenuate our findings on the relative underperformance of inside rounds. 5 Nonetheless, it is important to address the selection issues to ensure our results cannot be explained by missing data patterns. Columns (4) and (5) take two approaches to address the issue. Each are motivated by the techniques in Korteweg and Sorensen (2010). The selection bias in returns results in an over-sampling of publicly-held VC-backed firms and an underrepresentation of acquired firms (we observe all failed outcomes). We compute the fraction of each exit type in the full sample IPOs, acquisitions and failed. These fractions are then imposed on the regression analysis through a weighted regression and resampling. Column (4) conducts a weighted least squares where the weights are the inverse of the true 5 The omitted variable can be entrepreneurial firm quality, which is positively correlated with the probability of being in the data and negatively correlated with inside rounds. 28

29 exit probabilities. Column (5) creates a random sample of the main sample where the final proportion of exit types matches the full sample. In both columns, the negative relationship between inside rounds and returns holds. 5 What drives inside round financings? The tables discussed above make clear that inside rounds and insider participation correlates with financing- and firm-level observables. We next ask whether we can predict the prevalence of insiders in a financing. Table 13 considers the dummy for a financing having all insiders, however, the results are robust with a dependent variable that is the fraction of inside capital. Each specification includes the interaction of industry and year fixed e ects. Most of the controls have strong predictive power for inside rounds. Less capital raised in the previous round and fewer investors implies a higher probability of inside round or fraction of insiders. Larger previous syndicates weakly predicts more insider participation, perhaps because large snydicates have relatively more uncommitted investors. Perhaps surprisingly, the longer one waits for new round of financing, the higher the probability it has outsiders. This correlation could simply follow from a higher chance of finding a new capital provider over time. Although the results do not point to one story, the ability to predict insideness of financing suggests that the variable captures something meaningful about financings. Columns (2) - (5) of Table 13 ask whether investor characteristics predict inside participation. Column (2) of Table 13 introduces the dummy variable Syndicate fund raising that is equal to one if we can identify at least one investor in the process of fund-raising. An investor is fund-raising if they are within two years of their next fund closing. We discuss above how LPs discourage inside rounds as means of marking up investments, so we would predict less inside round events. Column (2) bears out this prediction. Column (3) of Table 29

30 13 asks where the current investors recent success can predict outside investor participation. The variable Big exit last 2 years is equal to one if at least one of the active investors had an IPO or large acquisition in the previous two years. Recent success could indicate both ahigherqualityinvestmentbutalsoaweakertendencyfortheinvestortoputgoodmoney after bad. The coe cient on this variable suggests success lowers the probability of an inside round. The next column (4) introduces a measure of the experience of the existing investors of the entrepreneurial firm at each financing event. The variable Log syndicate experience is the log of the total investments done by all insiders prior to the financing event. The coe cient on the variable is positive and significant. It appears that more experienced investors have a higher probability of participating in an inside round. Such a relationship is not driven by any changes in investors stage of investment as we include round number fixed e ects. The final column of Table 13 asks whether the VC fundraising environment can explain the prevalence of inside rounds. Perhaps we observe more inside rounds when there are less funds available to all VC-backed firm, which also coincides with weaker market conditions overall. This variable varies within-year, so attempts to capture time variation that is unavailable with financing year fixed e ects. The results indicate that when there are more funds raised in the previous two years, there are relatively more inside rounds. This result and that of column (5) are conflict with some of the results about returns and outcomes, so warrant further analysis. 6 Conclusion The results above present a preliminary view on the relationship between outside investor participation and entrepreneurial firm investment outcomes. We study sequential investment decisions in the venture capital (VC) industry. VC-backed companies typically need to raise 30

31 several rounds of funding from VC funds, and the decision whether to provide further funding to the company as well as the terms of the new funding determine to a large extent the returns of VC funds and their ability to back successful companies. We show that investment outcomes in the VC industry can be predicted by whether the existing VC investors can attract new outside investors to participate in the next round. Inside rounds, in which only existing investors participate, lead to a higher likelihood of failure, lower probability of IPO, and lower cash on cash multiples than outside rounds. We explore to mechanisms that explain these stylized facts: the escalation of commitment that leads VC investors to irrationally make negative NPV decisions, and agency costs between VC funds their investors that make VC investors gamble with their investors money. Ongoing work will introduce new measures of agency frictions and investment heterogeneity. 31

32 7 Figures and Tables Figure 1: Change in valuation: inside vs. outside rounds Notes: The figure reports the ratio of the current financing pre-money valuation over the previous financing s post-money valuation. We can only calculate this ratio for two financings where valuation is revealed. This sample is positively selected towards high-quality entrepreneurial firms (e.g. those that go public). The red vertical line is for the ratio value of one, which is called a flat round. 32

33 Figure 2: Log capital invested: inside vs. outside rounds Notes: The figure reports the kernel densities of log of capital invested for a financing event in two samples. Outside are those financings with at least one VC investor who is new to the pool of investors. Inside are those financings where all investors were previously invested in the firm. 33

34 Figure 3: Capital ramp up: inside vs. outside rounds Notes: The graph compares Raised t /raisedt be less than the 95th percentile. 1whenthevariableisavaiable.Thevariableiswinsorizedto 34

35 Table 1: Main variable description Notes: The table defines the major variables used throughout the analysis. % insiders (relative last syndicate) % insider investors % VC outsiders (relative last syndicate) %investorsthat are outsider VC % insider dollars (relative last syndicate) %insidedollars % VC outsider dollars (relative last syndicate) Full inside round Full VC inside (no new non-vcs) % insiders participate No new lead? Years since last financing Round number Capital raised Revenues / profitable? Fraction of investors in financing that are old, where any investor that was not in the previous syndicate is considered new. Fraction of investors in financings that are old, where insiders are defined as any previous investor. A new investor has never invested in the company. Fraction of investors that are new VC investors (rather than corporate and other) who did not invest in the previous syndicate. Fraction of the investors in the financing that are new and VCs, where new investors must not have invested at any point before. Fraction of the dollars in a financing provided by the investors in the previous financing syndicate. Fraction of the dollars in the financing provided by any existing investor. Fraction of dollars in the financing that were provided by new VC investors where new is defined as any investor not in the previous financing and is a VC. Equals one if the financing had all existing investors who were from the previous financing. Equals one if the financing had no new VC investor in the round. The fraction of inside investors that are investing in the current financing. Adummyvariableequaltooneifthefinancinghasonlyalead investor that was a previous investor. Years from the last financing to the current financing. The sequence number of the financing event. The total capital invested at the time of the financing (in millions of USD). Adummyvariablethatisoneifthefirmreportshavingrevenues or profits at the time of the financing. 35

36 Table 2: Summary statistics Notes: Table reports the summary statistics of the firms and financings in our sample. The main criteria for inclusion are post-first financing equity rounds for entrepreneurial firms where they have at least one traditional VC investor and received capital from 1990 to Panel A summarizes the set of entrepreneurial firms. Information Technology and Healthcare are firm industry categories. Public? is a dummy equal to one if the firm had an IPO by the end of the sample. Acquired? and Failed? are dummies for firms that were acquired or failed, respectively, by the end of the sample. Year founded is the year the entrepreneurial firm was founded. First capita raised is the capital raised in the entrepreneurial firm s first financing (in millions). Total financings is the total number of financings raised by the entrepreneurial firm prior to exit. Panel B summarizes the financing event characterstics, valuation measures and returns. Round number is the sequence of the financing event. Years since last fin. is the number of years between the current and previous financing. Syndicate size is the count of the number of unique investors in the current financing. Capital raised is the total capital investors in the current financing. Firm age is the age in years of the entrepreneurial firm by the financing event. Financing year is the year of the investment. The valuation measures pre-money and post-money are often missing, so the summary considers only a subsample of the data. The returns are also missing for many financings, so Gross multiple summarizes the set of financing returns where we observe the exit valuation and financing valuations. Gross multiple is the multiple of money earned by a hypothetical dollar invested in the financing accounting for any future dilution (zero for failed firms). Panel A Firm characteristics mean sd min p25 p50 p75 max count Information Technology Healthcare Public? Acquired? Failed? Still private? Year founded First capital raised (m) Total financings California New York Illinois Texas Panel B Financing characteristics mean sd min p25 p50 p75 max count Round number Syndicate size Years since last fin Capital raised (m USD) Firm age (as of financing) Revenues/Profitable? Financing year Valuations (when known) mean sd min p25 p50 p75 max count Post-money valuation Pre-money valuation Investment returns (when known) mean sd min p25 p50 p75 max count Gross multiple

37 Table 3: Characteristics of the various inside variables Notes: Table reports the characteristics of the main insideness variables. Panel A includes all financings where we could determine the insideness. Panel B restricts the sample to those where we have a complete history of capital invested by investors. It is often unknown how much capital was provided by past investors because they are general buckets (e.g. Individual Investors ) or lack a VC firm identifier in VentureSource. Variables are defined in Table 1. Panel A All financings mean sd min p10 p25 p50 p75 p90 max Full inside round %insiderinvestors No new outside VC %investorsthatareoutsidervc % insiders participate No new lead Observations Panel B Financings with known past dollars mean sd min p10 p25 p50 p75 p90 max Full inside round %insidedollars No new outside VC %insidevcdollars %investorsthatareoutsidervc % insiders participate No new lead Observations

38 Table 4: Characteristics of the various inside variables: by round type Notes: Table reports the characteristics of the main insideness variables. Panel A includes all financings where we could determine the insideness. Panel Brestrictsthesampletothosewherewehaveacompletehistoryofcapitalinvestedbyinvestors.Itisoftenunknownhowmuchcapitalwasprovidedby past investors because they are general buckets (e.g. Individual Investors ) or lack a VC firm identifier in VentureSource. Panel A: 2nd round mean sd min p10 p25 p50 p75 p90 max %insiderinvestors %investorsthatareoutsidervc Full inside round No new outside VC Year founded Years since last fin Capital raised (m USD) No new lead Observations 7602 Panel B: 3rd round mean sd min p10 p25 p50 p75 p90 max %insiderinvestors %investorsthatareoutsidervc Full inside round No new outside VC Year founded Years since last fin Capital raised (m USD) No new lead Observations 4704 Panel C: 4th round mean sd min p10 p25 p50 p75 p90 max %insiderinvestors %investorsthatareoutsidervc Full inside round No new outside VC Year founded Years since last fin Capital raised (m USD) No new lead Observations

39 Table 5: Inside vs. non-inside financing rounds Notes: Table compares financings where there the only investors are those with an existing equity stake to those with at least one outside investor of any kind. The numbers are the mean and median respectively, by sub-sample and for the full sample (i.e. Total ). Sample includes all entrepreneurial firms that were founded prior to 2008 to give ample time for an exit event and who had at least two financing events. The first panel considers characteristics of financings independent of whether we can calculate a return or observe a valuation. The remaining panels require non-missing data on each of these dimensions for sample inclusion. Years VCbacked is the number of years from the first observed VC financing to the current. Revenues/Profitable is adummyequaltooneifthefirmreportssomerevenuesorprofitasofthefinancing. #VCinvestors(all) is acountofthetotalnumberinpastinvestors. Pre$ t /Post$ t 1 isthechangeinvaluationfrompreviousto the current financing. Up round is a dummy variable equal to one if this value is greater than one (less than one for Down round and 1 for Flat round ). Other variables are as defined in Table 2. Full sample Inside round Outside round Total Capital raised (m USD) Round number Years VC-backed Firm age (as of financing) Years since last fin Revenues/Profitable? Syndicate size #VCinvestors(t 1) Number Financings Financing valuations Inside round Outside round Total Pre$ t /Post$t Up round Down or flat round Number observations Financing returns Inside round Outside round Total Gross multiple Gross multiple (95% winsorized) Zero multiple Number observations

40 Table 6: Inside vs. non-inside returns and outcomes: up vs. down financings Notes: Table compares financings where there the only investors are those with an existing equity stake to those with at least one outside investor of any kind. Down round is when the valuation falls or is unchanged between financing events. Sample includes all entrepreneurial firms that were founded prior to 2008 to give ample time for an exit event. The sample is also constrained by our ability to calculate the return, so comparing Table 5 s first panel to this one will reveal some sample selection issues. Gross multiple is the assumed cash-on-cash return for a hypothetical common equity purchase in the financing in question that accounts for future dilution and the final exit value. Zero return is a dummy equal to one if the multiple was zero (i.e. return -100%). Other variables are as defined in Table 2. Inside down Outside down Di./s.e. Gross multiple (0.387) Gross multiple (95% winsorized) (0.223) Zero return (0.0299) Round number (0.108) Years since last fin (0.0525) Capital raised (m USD) (1.123) Public? (0.0281) Failed? (0.0279) Acquired? (0.0307) #VCinvestors(t 1) (0.250) Number Financings Inside up Outside up Di./s.e. Gross multiple (0.811) Gross multiple (95% winsorized) (0.176) Zero return (0.0182) Round number (0.0557) Years since last fin (0.0278) Capital raised (m USD) (0.770) Public? (0.0196) Failed? (0.0167) Acquired? (0.0191) #VCinvestors(t 1) (0.130) Number Financings

41 Table 7: Entrepreneurial firm outcomes: inside vs. outside rounds Notes: Table reports probit and OLS regressions of entrepreneurial firm outcomes on a set of observables. Column (1) uses the dependent variable IPO that is equal to one if the firm went public by the end of the sample. Column (2) considers Good exit which is equal to one if the firm had an IPO or an acquisition with areportedvaluationgreaterthantwotimescapitalinvested. Logexitvalue incolumn(3)isthelogofthe valuation (log of 25% of capital raised if a failure). The variable Had inside round is a dummy variable equal to one if the entrepreneurial firm ever had an inside round in its financings. Columns (3) - (6) introduce controls for whether the firm had a down round in its history. Robust standard errors clustered at the financing year reported in parentheses.,, represent significance at the 10%, 5% and 1% level respectively. IPO Good exit Log exit value IPO Good exit Log exit value Non-failed Non-failed (1) (2) (3) (4) (5) (6) Had inside round ( ) ( ) (0.0610) (0.0164) (0.0192) (0.101) Had down round (0.0195) (0.0239) (0.113) Had inside X Had down round (0.0256) (0.0308) (0.156) Log total capital ( ) ( ) (0.0333) ( ) ( ) (0.0458) Total unique VCs ( ) ( ) ( ) ( ) ( ) (0.0120) Constant (0.0536) (0.0757) (0.495) (0.264) (0.254) (0.861) Observations R Founding year FE? Y Y Y Y Y Y Industry FE? Y Y Y Y Y Y State FE? Y Y Y Y Y Y 41

42 Table 8: Change in valuation: inside vs. outside rounds Notes: Table reports linear regressions of the log change in valuation between an entrepreneurial firm s two finaning events on set of observables. The dependent variable Log(Pre t/$post$ t 1 )takesthelogoftheratio of the current pre-money valuation over the previous post-money valuation (if both are reported). The value is also called a round-to-round returns and captures the return of investing at the t 1roundandsellingat the t round. Inside round is a dummy variable equal to one if the current financing has no outside investors. Column (6) introduces the No new lead dummy which is one if a lead investor is also not a new outside. $ raised t / $ raised t 1 is the ratio of capital raised in the current financing over the previous. Log raised is the log of the current capital invested. Years since last fin. is the number of years since the firm s previous financing event. # VC investors (t 1) is the count of the unique number of investors who have invested prior to this financings. Log total raised is the sum of non-first round capital raised by the firm. YearxIndustry FE are fixed e ects of interactions for the industry and financing year. Round # FE are fixed e ects for the financing sequence number. Standard errors clustered at the entrepreneurial firm are reported in parentheses.,, represent significance at the 10%, 5% and 1% level respectively. Log(Pre $ t / Post $ t 1) (1) (2) (3) (4) (5) (6) Inside round (0.0192) (0.0190) (0.0201) (0.0198) (0.0197) No new lead (0.0158) $raisedt /$raisedt ( ) ( ) ( ) ( ) ( ) Log raised ( ) ( ) ( ) ( ) Years since last fin (0.0135) (0.0134) (0.0134) #VCsinvestors(t 1) ( ) ( ) Constant (0.197) (0.197) (0.193) (0.194) (0.197) (0.171) Observations R Num. firms Year FE? Y Y Y Y Y Y Industry FE? Y Y Y Y Y Y YearxIndustry FE? Y Y Y Y Y Y Round # FE? Y Y Y Y Y Y 42

43 Table 9: Di erences in returns: log gross multiple Notes: Table reports regressions of a the log of the gross multiple when known on a set of observables. The dependendent variable is the financing-level return (logged) assuming common equity for a dollar invested in the round. If the company exits via an acquisition and the reported price is less than total capital raised, we assume that the last investors receive all their capital back before earlier investors share in the return. For failures, we assume that the final investors received 25% of the capital invested back, with all previous financings recieving 15%. The first four columns are OLS regressions at the financing-level. The last four columns use VC firm fixed e ects. Full inside round is a dummy equal to to one if all the investors were insiders. Columns (1)-(2) and (5)-(6) includes all financing returns. The remaining columns condition on non-failed companies. Log total capital is the log of the sum of total capital raised. Log raised (m USD) is the log of capital raised in the current round. All fixed e ects are as defined in Table 8. Robust standard errors clustered at the financing year reported in parentheses.,, represent significance at the 10%, 5% and 1% level respectively. 43 Log return Log return Log return Log return Log return Log Return Log return Log return All returns All returns Non-fail Non-Fail All returns All returns Non-fail Non-Fail (1) (2) (3) (4) (5) (6) (7) (8) Full inside round (0.0385) (0.0453) (0.0216) (0.0283) No new lead (0.0384) (0.0441) (0.0168) (0.0216) Log raised (0.0422) (0.0391) (0.0374) (0.0326) (0.0123) (0.0116) (0.0165) (0.0155) Years since last fin (0.0361) (0.0357) (0.0382) (0.0380) (0.0140) (0.0140) (0.0183) (0.0183) Total VC investors ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Log total capital (0.0476) (0.0479) (0.0609) (0.0613) (0.0128) (0.0128) (0.0202) (0.0202) Constant (0.471) (0.465) (0.518) (0.508) (0.566) (0.565) (0.509) (0.506) Observations R Number firms Number VCs VC FE? N N N N Y Y Y Y Industry FE? Y Y Y Y Y Y Y Y Year FE? Y Y Y Y Y Y Y Y Round # FE? Y Y Y Y Y Y Y Y State FE? Y Y Y Y Y Y Y Y

44 Table 10: Di erences in gross multiple returns: participation, capital and measurement Notes: Table reports regressions as describe in Table 9 with additional controls and interactions. The first three columns use the log gross multiple for all returns as the dependent variable. Columns (1) and (2) introduce two variables that summarize insider participation. The variable Full participation is a dummy equal to one if all the past investors have invested in the current financing. High inside participation is a dummy for whether the fraction of insiders currently participating is above the sample median. Column (3) introduces the dummy Capital ramp down which is equal to one if the ratio of amount of capital in the current financing to the previous is below the 25th percentile observed in the sample (zero otherwise). The final three columns use the Total return dependent variable defined as the dollar-weighted observe multiple for the current and past investments. Robust standard errors clustered at the financing year reported in parentheses.,, represent significance at the 10%, 5% and 1% level respectively. Log return Log return Log return Total return Total return Total return Total return Non-failed Non-failed (1) (2) (3) (4) (5) (6) (7) Full inside round (0.0402) (0.0495) (0.0319) (0.0469) (0.0861) No new lead Full participation (0.0528) Inside X Full participation (0.0789) High inside part (0.0352) Inside X high inside part (0.0619) Capital ramp down X Inside (0.0606) Capital ramp down (0.0531) (0.0478) (0.0683) Log raised (0.0420) (0.0415) (0.0459) (0.0548) (0.0442) (0.0535) (0.0419) Years since last fin (0.0361) (0.0361) (0.0370) (0.0457) (0.0461) (0.0449) (0.0461) Total VC investors ( ) ( ) ( ) (0.0126) (0.0113) (0.0127) (0.0115) Log total capital (0.0476) (0.0477) (0.0492) (0.0509) (0.0700) (0.0515) (0.0706) Constant (0.473) (0.472) (0.468) (0.519) (0.467) (0.496) (0.469) Observations R Number VCs Industry FE? Y Y Y Y Y Y Y Round # FE? Y Y Y Y Y Y Y State FE? Y Y Y Y Y Y Y 44

45 Table 11: Inside rounds and current contract features Notes: The table reports the correlations between inside rounds and strong contract features. Contracts data from VC Experts was merged onto VentureSource financing when possible. Column (1) has a dependent variable equal to one if the financing had senior equity. Column (2) has a dependent variable equal to one if the contract has a liquidation preference greater than 1X. Column (3) has a dependent variable equal to 1 if the contract had participating preferred. Column (4) has a dependent variable equal to one of the contract had redemption rights. The last column has a dependent variable equal to one if any of these contract types exist. All variables are as defined in above tables. Robust standard errors clustered at the financing year reported in parentheses.,, represent significance at the 10%, 5% and 1% level respectively. Senior > 1X? Part. Pref.? Redemption? Any? (1) (2) (3) (4) (5) Full inside round (0.0840) (0.111) (0.0856) (0.0838) (0.102) Log raised (0.0418) (0.0570) (0.0430) (0.0415) (0.0497) Years since last fin (0.0412) (0.0554) (0.0427) (0.0427) (0.0495) Observations Pseudo-R Num. firms Mean dep. var Mean dep. var. Inside Founding Year Y Y Y Y Y Industry FE? Y Y Y Y Y Round # FE? Y Y Y Y Y 45

46 Table 12: Di erences in returns: gross multiple robustness Notes: Table reports regressions of a the log of the gross multiple or dollar-weighted multiple when known on asetofobservables. Thedependendentvariableisthefinancing-levelreturn(logged)assumingcommonequity for a dollar invested in the round. If the company exits via an acquisition and the reported price is less than total capital raised, we assume that the last investors receive all their capital back before earlier investors share in the return. For failures, we assume that the final investors received 25% of the capital invested back, with all previous financings recieving 15%. Full inside round is a dummy equal to to one if all the investors were insiders. Column (1) repeats the estimate from column (1) of Table 9. Column (2) replaces the gross multiple of known inside returns that are less than 2X capital with 2. That is we assume that inside rounds have at least a2-timesliquidationpreference. Outsideroundsareassumedtohave1X for non-failures. Column (3) replaces missing inside round returns with 2X and missing outside returns with a 1-times capital invested. Column (4) estimates a weighted OLS regression that corrects for the over-representation of IPOs. In this regression, acquisition returns are weighted so that they have importance equal to the full sample distribution of acquisition rates. The final column resamples the full sample of returns data to create a final sample with the full-sample exit proportions for IPOs, acquisitions and failures (see Korteweg and Sorensen (2010)). Log total capital is the log of the sum of total capital raised. Log raised (m USD) is the log of capital raised in the current round. YearXIndustry FE and the interaction of the firm s industry and current financing year FE. Robust standard errors clustered at the financing year reported in parentheses.,, represent significance at the 10%, 5% and 1% level respectively. All returns Inside 2X Missing is 2X Exit-Weighted Re-sampled Outside 1X Outside 1X (1) (2) (3) (4) (5) Full inside round (0.0385) (0.0390) (0.0408) (0.0488) (0.0647) Log raised (0.0422) (0.0434) (0.0388) (0.0467) (0.0510) Years since last fin (0.0361) (0.0362) (0.0313) (0.0558) (0.0555) Total VC investors ( ) ( ) ( ) (0.0109) (0.0140) Log total capital (0.0476) (0.0476) (0.0453) (0.0485) (0.0404) Constant (0.471) (0.475) (0.395) (0.716) (0.972) Observations R Founding YearxIndustry FE? Y Y Y Y Y Industry FE? Y Y Y Y Y Round # FE? Y Y Y Y Y State FE? Y Y Y Y Y 46

47 Table 13: Prediciting insideness Notes: Table reports regressions of a financings fraction of inside investors on a set of controls. The dependent variable is equal to one if all the current investors are insiders or previous investors. The controls are all lagged to the previous financing event (all financings are at least the second round). Last log raised is the log capital invested in the firm s previous round. Years since last financing is the years since the previous financing event. First capital raised is the total capital raised in the entrepreneurial firm s first financing event. Last #investors isthecountofthetotalnumberofactiveinvestorsinallpreviousrounds. Lastsyndicatesize is the total number of investors in the previous round. Syndicate fund raising is a dummy equal to one if at least one participating investor is within two years of raising their next fund. Big exit last 2 years is a dummy variable equal to one if at last one syndicate member have a large exit (IPO or big acquisition) in the previous two years. Log syndicate exper. (# deals) is the log of the total investments done by the insiders prior to this round. Log new funds last 2 years is the log of the total new funds raised over the previous two years (lagged one year prior to the financing). YearXIndustry FE and the interaction of the firm s industry and current financing year FE. Robust standard errors clustered at the entrepeneurial firm reported in parentheses.,, represent significance at the 10%, 5% and 1% level respectively. Syndicate fund raising Big exit last 2 years Log syndicate exper. (# deals) Log new funds last 2 years Inside? Inside? Inside? Inside? (1) (2) (3) (4) (0.0134) (0.0131) ( ) (0.0108) Last log raised ( ) ( ) ( ) ( ) Years since last fin (0.0106) ( ) ( ) ( ) Last # investors ( ) ( ) ( ) ( ) First capital raised ( ) ( ) ( ) ( ) Last syndicate size ( ) ( ) ( ) ( ) Constant (0.165) (0.242) (0.304) (0.208) Observations R Num. firms YearxIndustry FE? Y Y Y N Round # FE? Y Y Y Y State FE? Y Y Y Y 47

48 References Admati, Anat R, and Paul Pfleiderer, 1994, Robust financial contracting and the role of venture capitalists, The Journal of Finance 49, Adner, Ron, 2007, Real options and resource reallocation processes, Advances in Strategic Management 24, Arkes, Hal R., and Catherine Blumer, 1985, The psychology of sunk cost, Organizational Behavior and Human Decision Processes 35, Cochrane, John H, 2005, The risk and return of venture capital, Journal of financial economics 75, Fluck, Zsuzsanna, Kedran Garrison, and Stewart C Myers, 2006, Venture capital contracting: Gompers, Paul A, 1995, Optimal investment, monitoring, and the staging of venture capital, The journal of finance 50, Guler, Isin, 2007a, An empirical examination of management of real options in the u.s. venture capital industry, Advances in Strategic Management 24, Staged financing and syndication of later-stage investments, Unverö entlichtes Arbeitspapier.,2007b,Throwinggoodmoneyafterbad?politicaladinstitutionalinfluencesonsequential decision making in the venture capital industry, Administrative Science Quarterly 52, Korteweg, Arthur, and Morten Sorensen, 2010, Risk and return characteristics of venture capital-backed entrepreneurial companies, Review of Financial Studies p. hhq

49 Loewenstein, George, and Jennifer S. Lerner, 2003, The role of a ect in decision making, in R.J. Davidson, H.H. Goldsmith, and K.R. Scherer, ed.: Handbook of A ective Science (Oxford University Press: Oxford, England). Lucey, Brian M., and Michael Dowling, 2006, The role of feeling in investor decision-making, Journal of Economic Surveys 19, McDonald, Robert, and Daniel Siegel, 1986, The value of waiting to invest, Quarterly Journal of Economics 101, 707. Ross, Jerry, and Barry M Staw, 1986, Expo 86: An escalation prototype, Administrative Science Quarterly pp ,1993,Organizationalescalationandexit:Lessonsfromtheshorehamnuclearpower plant, Academy of Management Journal 36, Sørensen, Morten, 2007, How smart is smart money? a two-sided matching model of venture capital, The Journal of Finance 62, Staw, Barry M., and Ha Hoang, 1995, Sunk costs in the nba: Why draft order a ects playing time and survival in professional basketball, Administrative Science Quarterly 40, Tan, Hun-Tong, and J. Frank Yates, 2002, Financial budgets and escalation e ects, Organizational Behavior and Human Decision Processes 87, Thaler, Richard H., and Eric J. Johnson, 1990, Gambling with the house money and trying to break even: The e ects of prior outcomes on risky choice, Management Science 36,