INSEAD-Wharton Alliance Center for Global Research & Development

Transcription

1 Performance of Private Equity Funds: Another Puzzle? by O. Gottschalg L. Phalippou and M. Zollo 2004/82/SM/ACGRD 6 (Revised Version of 2003/93/SM/ACGRD 3) Working Paper Series INSEAD-Wharton Alliance Center for Global Research & Development

2 Performance of Private Equity Funds: Another Puzzle? Oliver Gottschalg, Ludovic Phalippou and Maurizio Zollo September 2004 Please address all correspondence to: Ludovic Phalippou, University of Amsterdam, Faculty of Economics and Econometrics, Finance group, 11 Roerterstraat, 1018 Amsterdam, The Netherlands. Tel: , Oliver Gottschalg and Maurizio Zollo are affiliated to INSEAD. Financial support from the R&D Department at INSEAD, the INSEAD-Wharton Alliance Center for Global Research and Development, and the Gesellschaft Scholarship (Gottschalg) is gratefully acknowledged. The authors would like to thank Jesse Reyes and Thomson Venture Economics for making this project possible through generous access to their databases. We also thank Bernard Dumas, Alexander Groh, Robert Kosowski, Dima Leshchinskii, Eric Nowak, Andrew Metrick and seminar participants at INSEAD and the EFMA meeting in Basle for their constructive comments.

3 Abstract Using a novel and comprehensive database on the performance of US and EU private equity (PE) funds and their underlying investments, we find that the performance of PE funds is below that of public stock-markets. Given the illiquidity of such investments and the risk that they involve, such a result is puzzling. We document that fund performance co-varies positively with both business cycles and stock-market cycles, an unattractive property. In particular, we find that PE funds are exposed to substantial left tail risk and returns are sensitive to the valuation levels that prevail on public stock-markets when investments are exited. Although we cannot compute a satisfactory equivalent of the Jensen s alpha in our context, our analysis indicates that private equity funds underperform under conservative assumptions about the risk they carry. Such performance is perplexing and we cannot reject the possibility of mispricing by so-thought sophisticated institutional investors. JEL: G23, G24 Keywords: Private equity funds, risk-return trade-off 1

4 The activities of private equity (PE) funds have recently received increasing attention from both the investor and the academic community due, primarily, to two factors. First, the amount of capital committed to this asset class grew from US$2 billion in 1980 to US$140 billion in 2000, totaling over US$800 billion over the last 25 years (see Graph 1, as well as Kaplan and Schoar, 2003). 1 Second, it has been argued that fund managers (general partners, GPs) play an active and important strategic role in the companies they finance. This role has been studied extensively both theoretically (Admati and Pfleiderer, 1994, Hellmann, 1998) and empirically (e.g. Gompers, 1995, Lerner, 1995, Hellmann and Puri, 2002). Given the importance of private equity both as an investment vehicle and as a catalyst for economic growth, the need for a comprehensive assessment of the performance and risk profile of this industry is apparent. In a recent review paper, Gompers and Lerner (2001) classify the understanding of risk and return as what we don t know about venture capital, a statement which is equally true for buyout investments (the other and largest sub-category of private equity). 2 Graph 1 Using a unique and comprehensive dataset containing information on cash flows to investors (limited partners, LPs) and the investments made by more than 500 mature private equity funds, this paper first reports an estimate of the net (of fees) performance of private equity investments. It is important to note that such a task is rendered particularly difficult in that performance is definitive only for fully liquidated funds. Using a sample of largely liquidated funds, however, introduces a selection bias as the decision to liquidate is endogenous and is likely to be influenced by the success of the investments. When we include funds with investments that are not yet liquidated, the question then becomes how to value these on-going investments (residual value, RV). We estimate the value of RVs based on an analysis of how RVs historically convert into subsequent cash flows for now liquidated funds. The resulting performance estimate is optimistic and does not suffer from a selection bias. Despite our optimistic evaluation, we find that PE funds raised between 1980 and 1995 underperform public stock-markets. Consequently, the second question that we address in this paper is why is this so? In response, we first assume that the CAPM holds, estimate quarterly fund performance and show that - despite a very conservative estimate of beta (between 0.5 and 0.65) obtained by construction - Jensen s alphas are not statistically different from zero for 92% of the funds. 3 We also report that more reasonable estimate of the beta of funds is about 1.6 given both the industry 2

5 in which PE funds invest and the typical amount of leverage for buyout deals. These average betas are, furthermore, found to be significantly positively related to performance. This stresses that PE investments are exposed to non-negligible risks that should command a premium over public market investments rather than a discount. Second, we investigate the hedging properties of PE investments to assess whether they have desirable properties that could justify a relatively low compensation. In particular, we evaluate the influence of business cycles and stock-market cycles on PE performance and find two dominant effects. First, PE performance is higher when investments are exited in periods of high valuation levels on public stock-markets, as proxied by the overall earning to price ratio. Second, PE funds are exposed to substantial left tail risk. That is, they deliver significantly higher losses during large market downturns but are not as sensitive to economic conditions in good times. In addition, the average level of credit spreads, public market performance and GDP during the investment phase are also all significantly related to PE performance. The relationship between fund performance and each of these variables indicates that PE performance is significantly pro-cyclical, which is an unattractive property. In addition, we document the main drivers of underperformance and find that these are funds that are small, European and run by inexperienced GPs. Hence, in the future, more experienced and skilled PE firms may survive and offer higher returns. Nonetheless, funds that were raised more recently, particularly in 1998, 1999 and 2000, exhibit preliminary indicators of performance that are significantly lower than what their peers exhibited at the same stage. Performance is thus likely to remain low for at least a few more years. The fact that funds raised from 1998 to 2000 raised very large amounts of capital and will probably offer even lower performance than funds raised before 1995 also indicates that our estimate of average performance is optimistic. In addition, we already mentioned that our treatment of residual values is likely to lead to a higher performance than what will be realized. Moreover, certain LPs face additional costs when investing in private equity that we do not account for, such as fees paid to financial intermediaries (fund of funds) and transaction costs paid when selling shares. 4 These findings on the risk of PE funds, the difficulties faced by recently raised funds and our optimistic assumptions when evaluating performance, therefore, make the documented performance of PE funds puzzlingly low. Is it thus possible that LPs, who are mainly sophisticated institutional investors (large pension funds and university endowments), have mispriced this asset class and if so what errors have they made? 3

6 It is difficult to provide a definitive answer to this question. We argue that one element to consider is that numerous LPs invest in private equity for reasons other than performance. Also, the fee structure of PE funds is such that LPs may underestimate the impact of fees when deciding to invest in private equity. Finally, this asset class is relatively new, uncertainty is very high, and payoffs so skewed that even sophisticated investors may be prone to over-optimism or evaluation mistakes. Certain errors might also be common to those behind the results of Moskowitz and Vissing-Jorgensen (2002), who find that the return of entrepreneurial investment in non-public companies another private equity class is surprisingly close to public market returns, despite the lack of diversification. These results are important because private equity is now a major class of financial assets and very little is known about the net performance of private equity funds and the type of risk they carry. In addition, understanding why LPs invest so much in this asset class given such a low performance is an interesting area for further research. It may shed light on other finance puzzles such as the IPO puzzle (e.g. if certain LPs underwrite IPOs) and contribute to the understanding of the behavior of institutional investors and, in particular, how their growing importance modifies the financial markets and the real economy. Moreover, there is currently a debate as to whether Hedge funds and PE funds should be regulated. The reason why, in the US, they are not regulated is because they cater exclusively to sophisticated investors. If, as suggested by this paper, these sophisticated investors can be mislead by current reporting practices then regulations from authorities such as the SEC may be warranted (providing that the total benefit outweigh the total cost). The paper continues as follows: section 1 describes our data and methodology, section 2 offers a literature review, section 3 discusses and documents the pricing of risk, section 4 proposes various explanations for the observed low performance and section 5 briefly concludes. 4

7 I. Performance Measurement In this section, after detailing our data sources and our sample selection scheme, we offer descriptive statistics of the characteristics of both the funds and their investments. We then describe our approach to account for the residual value of ongoing investments. Finally, we discuss how to measure performance and report aggregate estimates. A. Data sources In this study we use several sources of data. Data on both Treasury bill rates and stock performance are from CRSP (via WRDS). Data on accounting are from Compustat (via WRDS). Data on real GDP and corporate bond yields are from the Federal Reserve Bank of Saint Louis. Data on private equity funds have been obtained from two datasets maintained by Thomson Venture Economics. These datasets cover funds raised from 1980 to Venture Economics (VE) records the amount and date of all cash flows as well as the aggregate quarterly book value of all unrealized investments for each fund until December Cash flows are net of fees as they include all fee payments, and distributions are reduced by the carried interest and other charges. VE also collects information on some of the private equity investments that funds undertake through its VentureXpert database. The basic information we use from this dataset is the company s industry, amount invested, and entry/exit dates of each investment. Details of these databases as well as certain corrections that we undertake are provided in Appendix A.II. B. Sample selection For each fund, we observe a series of cash flows to investors and the residual value (RV) of ongoing investments. Until a fund is entirely liquidated, the existence of a positive RV prevents a proper estimation of performance as neither the fund nor its underlying investments are publicly traded. The unique assessment of RV that we observe is the accounting value that is reported quarterly by funds. However, these accounting valuations are rather unreliable. 5 A solution here is to focus on a sample of fully liquidated funds, denoted as Sample 3. It contains 427 funds that have a ratio of residual value to total capital invested (RV/CI) below 1% and are either raised before 1995 or officially liquidated. In such a sample, the treatment of accounting values has a minimal impact but the sample might be biased toward winners. Indeed, funds that are not fully liquidated (and hence excluded from the sample) may be finding it difficult to 5

8 sell their current investments or may simply be waiting before realizing, and officially acknowledging, a poor performance. In addition, as we will detail in section IV.E, poorly performing funds have a strong incentive to postpone liquidation to artificially increase their IRR (which is their commonly used performance measure). The liquidation decision, therefore, may be endogenous and partly influenced by the success of investments. Our concerns in this case are twofold. First, the average performance of funds may be upwardly biased. Second, funds in this sample may have different characteristics from the universe which would influence our findings about the pricing of risk for instance. As a consequence, we predominantly consider a sample of all funds raised before 1995 (Sample 1). These 1208 funds are said to be mature as they are close to typical liquidation age (see Appendix A.I.). Their RV/CI is thus relatively low (32%) and there is no selection bias for this sample. The funds in Sample 1 are further divided in two groups. One group, denoted as Sample 2, consists of the funds for which the total amount of available investments (in VentureXpert) is more than half of the capital invested according to VE. It contains 485 venture funds and 157 buyout funds. Sample 2 will be used only when we need the data about the investment characteristics (part of the analysis of risk in section III). To assess the differences between these two samples that we use in this paper, we report their respective characteristics in Tables 1 and 2. Table 1 Funds in Sample 1 are typically larger and are more likely to be VC funds, US funds, and experienced funds then funds in Sample 2. This reflects the fact that VentureXpert is specialized in US venture capital deals. Nonetheless, we note that we have a large number of European funds for which more than half of their investments is available, especially so for the European BO funds. In addition, VC funds in Sample 2 also have a lower RV/CI and are more likely to be officially liquidated than VC funds in Sample 1. Generally, the descriptive statistics are similar to those reported in the literature (see Kaplan and Schoar, 2003). Venture funds are significantly smaller than buyout funds, with an average committed capital of $93 million for VC funds compared to $392 million for BO funds. It is also important to note the significance of residual values. RV/CI is on average 26% for VC funds and 34% for BO funds. Moreover, reflecting the fact that the PE industry is young and 6

9 rather inexperienced, over one third of the funds in our database are first-time funds and 20% are 4 th time and beyond, i.e. the parent firm has already raised more than three funds as of Table 2 We now turn to the description of investments as reported in Table 2. We find that VC funds have almost twice as many investments as BO funds. VC funds invest in 32 (28) companies on average (median) and BOs in 16 (12) companies for an average period of about four years and a half for VC deals and four years for BO deals. These figures are consistent with the findings of Gompers and Lerner (1999) for a different sample: they note that VC funds typically invest in two dozen firms over about three years. Several important elements are worth mentioning. First, we find that venture funds have 12% of their investments in buyouts and that as much as 26% of the investments of buyout funds are in venture deals. Nonetheless, the medians are smaller. This indicates that a few funds in each category (more so in BO funds) are diversified across deal types. Consequently, care should be taken when interpreting differences in the behavior of funds according to their objectives. Second, the industry focus is very similar for both VC funds and BO funds as they both concentrate 40% of their portfolio in one of the 48 industries defined by Fama and French (1997). Third, and finally, over 50% of the deals undertaken by VC funds are in high-technology industries. C. Valuing non-liquidated investments To avoid sample selection biases, we include all funds raised up to This implies that certain funds still have some ongoing investments. As already mentioned, neither funds nor their underlying investments are publicly traded. The unique assessment of residual values (RVs) that we observe is the accounting value that is reported by funds in December The problem is then to translate these RVs into expected discounted net cash flows. Given a general distrust of accounting valuations we may want to write off RVs. 6 This is a conservative choice in that performance is systematically understated and may bias the results about the determinants of performance. It is nonetheless a useful lower bound for overall performance that we will report in our sensitivity analysis (Section IV.B., Table 9). To best assess performance, however, we privilege the following approach. 7

10 The objective is to estimate a model that predicts the future cash flows of a fund, based on the accounting estimation of unrealized investments and fund characteristics. We use a set of pseudo-liquidated funds, for which we observe at each end of year beginning at their 8 th birthday their characteristics and what the discounted net cash-flow that they offer to investors is from that point up to December Appendix A.III. provides additional details about the construction of the variables and the estimation of the model. The predictions we make are considered optimistic since we use pseudo-liquidated funds to estimate the model; and these pseudo-liquidated funds are likely to have been successful at exiting their investments. Our treatment of residual values is, therefore, relatively aggressive. Table 3 Table 3 reports how the ratio of future cash flows to residual value is related to fund characteristics. This ratio can be interpreted as accounting conservatism. A high ratio indicates a very conservative accounting as a given dollar of residual value reported is associated with a relatively high value of cash inflows. We find six significant explanatory variables. Fund age, RV/CI, and the time elapsed since the last distribution are all negatively related to the ratio of future cash flows to residual values. In contrast, fund size, percentage of committed capital invested and being a venture fund are positively related to the ratio. Hence, funds that are either larger, venture, or younger are more conservative in their accounting valuations. The most significant explanatory variable is the time elapsed since the last distribution. If the last distribution occurred a long time ago, residual values are much too optimistic as, probably, the fund faces difficulties to exit investments. The next most important variable is the amount of the residual value compared to the capital invested (RV/CI). If a fund reports a relatively high residual value, the future cash inflow is expected to be relatively low. Finally, larger funds may be conservative because they have an established reputation. However, experienced funds do not appear to bias their reported residual value. The average ratio of the present value of future cash flows to residual value is Hence, overall, accounting is conservative. Conservatism is actually encouraged by the US National Venture Capital Association which provides guidelines that in most cases result in carrying investments at cost in the book if there is no alternative unambiguous valuation, especially in the early years (that may explain the above finding about fund age being negatively related to our ratio). Furthermore, these figures are audited and thus may give GPs an incentive 8

11 to remain conservative in order to avoid lawsuits. Nonetheless, when we use the selected model, the average extrapolated ratio is 43% for the funds raised before 1995 that have an RV/CI above 5% (as of December 2003); the minimum is 5% and the maximum is 132%. This figure seems fair as these funds are at least 8 years old and have a substantial amount of unrealized investments. In addition, the recent difficult economic environment causes residual values to be optimistic as they are typically reported at cost. In addition, the expected value of future cash flows is computed assuming a CAPM-beta of (only) one and the conversion model was derived from funds that did well (since they exited most of their investments). Hence, our treatment of accounting values is rather optimistic. D. Performance Measures Having selected a sample and found a way to treat accounting values, we need to decide on an appropriate measure of performance. The principal performance measurement in the industry is the internal rate of return (IRR), which has also been used (in addition to alternative performance measures) by Kaplan and Schoar (2003) and Jones and Rhodes-Kropf (2003). Here, three important remarks are in order. First, IRR implicitly assumes that capital distribution occurring before liquidation can be reinvested at the fund s IRR. This is problematic as even if another PE fund raises capital when the distribution occurs, investment is not immediate. Second, the profile of investments in the industry allows GPs to manipulate their IRR by strategically reporting their residual values and timing their cash flows. 7 Third, inflows and outflows are treated as two flows with the same risk, an assumption that seems unwarranted as discussed below. 8 Due to the shortcomings of the IRR, we compute a profitability index (PI). PI equals the present value of cash inflows divided by the present value of cash outflows. 9 The main issue is thus the choice of appropriate discount rates. 10 In practice, the take-down schedule is unknown and when a call is made, cash has to be delivered in days (see appendix A.I.). It is, therefore, likely that investors have to rely on short-term bonds to invest the capital committed. In practice, LPs often invest the committed capital in public equity but it is reasonable to assume that capital calls are independent of stock-market performance. Hence, the beta of the cash outflows is zero and the appropriate cost of capital to be used is the risk free rate even though the investment is made in public equity (assuming a CAPM setting). Consequently, we use the 30-day Treasury- 9

12 bill yield to discount outflows. Finally, inflows are discounted using the public stock-market portfolio as proxied by the CRSP value-weighted index. 11 This way, PI will indicate whether private equity investments offer a higher return than public equity investments, implicitly assuming a beta of one in a CAPM setting. Table 4 Table 4 displays statistics about the performance of PE funds for three samples and for VC and BO funds separately. The 1208 funds raised before 1995 (Sample 1) have significantly underperformed the public market. The average PI is 78% (if equally weighted) and 86% (if value weighted). Both indicate underperformance as they are below 100%. Such a performance means that less than 90% of the money invested has been returned to investors in present value terms (assuming a beta of 1 with the market portfolio). Furthermore, this overall performance hides a great heterogeneity and skewness. While a quarter of the funds have returned less than a third of the capital invested (in present value terms) another quarter of the funds have outperformed the public market portfolio. Funds in Sample 2 have a higher performance. This is important for studies that compute returns on PE deals in VentureXpert as deals reported in this database appear to have above average performances. Finally, we want to point out that the average performance estimate for both sample 1 and sample 2 are not definitive as a quarter of the investments (in terms of value) are still not exited as of December However, we note that although liquidated funds have a higher performance, they still underperform on an equally-weighted basis and match exactly the public market performance on a value-weighted basis. These results are puzzling overall as private equity investments are often seen as risky and display the unattractive feature of being highly illiquid. In addition, our performance estimate is optimistic as it assumes a beta of one and converts accounting values optimistically. In the next section, we review the literature about both the systematic risk of PE investments and their performance. 10

13 II. Literature review on risk and return of private equity investments Several papers document the gross performance of VC investments and estimate their correlation with public stock markets. Peng (2001), Quigley and Woodward (QW, 2003), and Woodward and Hall (WH, 2003) compute a VC index and deduce the correlation between this index and a public market index. The index is built from discretely observed valuations (new financing round, IPOs, acquisitions, or liquidation). With similar observations, Cochrane (2003) proposes another approach. He assumes that the change in the log of the company valuation follows a log-capm and models selection bias explicitly as he assumes that the probability of observing a new round follows a logistic function. Accounting for selection bias is indeed an important task as performance is more often observed for those investments that performed well. However, it is very difficult to reliably adjust for this bias as a return is observed for less than a quarter of the investments (see Cochrane, 2003). 12 Results vary quite substantially. 13 QW find gross real returns on VC investments of about 5% per semester, which is less than the S&P 500 and the Nasdaq over the same period. They find a beta close to zero with the S&P 500 and 0.4 with the Nasdaq. WH s index estimates that the average performance is 20% per year. Taking the Nasdaq as the market portfolio, they find a CAPM-abnormal performance of 8.5% per year and beta of Peng (2001) finds an average return of 55% per year ( ) and a beta between 0.8 (monthly frequency) and 4.7 (yearly frequency). The annualized CAPM-alpha varies from -3.8% to 3.2% depending on the frequency of the estimation and on the chosen market portfolio. Finally, Cochrane (2003) offers the most optimistic picture: 59% annual average (arithmetic) gross return and a corresponding alpha of 32%. He, nonetheless, points out that the idiosyncratic volatility is very high. These studies focus exclusively on the investment level and not on the fund level as we do in this paper. At the fund level, the selection bias is substantially reduced as cash flows are more likely to reflect both successful and unsuccessful investments. Hence, this study as other fund-level studies, does not tackle directly sample bias issues. An additional advantage of fundlevel studies is that they include buyout funds; doing so is economically important as BO funds have a total amount of capital committed that is more than twice as much as that of VC funds (Table 1). It is also important to bear in mind that the anonymity of our dataset prevents us to merge the investment data with external sources such as SDC, as done by Cochrane (2003) for example. Also, we do not know which cash flow corresponds to which investment. 11

14 Consequently, we cannot compute the performance at the investment level like the above studies and cannot compute an alpha for each investment. The investment data that we have are, nonetheless, useful as they provide us with certain indicators of the degree of pro-cyclicality of the fund performance; see next section. At the fund level, Kaplan and Schoar (2003) and Jones and Rhodes-Kropf (2003) also report aggregate performance. They have the same dataset for both cash flows and residual values as we have but do not have access to the investments that funds made over time. Consequently, Kaplan and Schoar (2003) focus both on the persistence of returns and on the performance-flow relationship, and do not document or adjust performance for risk. In contrast, Jones and Rhodes-Kropf (2003) estimate the systematic risk of PE funds, but rely on the timeseries of residual values. The focus of their study is on the pricing of idiosyncratic risk and is more theoretical. Also at the fund level, Ljungqvist and Richardson (2003) analyze the investment behavior of a subset of private equity fund managers. In particular, they report the determinants of draw downs and capital distribution. They have a smaller sample in comparison to us but do not have any noise at the investment level. That is, for part of our sample -20% in terms of value and 10% in terms of number- Ljungqvist and Richardson (2003) know the cash flows and each firm in which funds have invested. The work closest to that presented in this paper is that of Gompers and Lerner (1997), who study the risk-adjusted performance of a single fund group (Warburg Pincus). Their idea is to mark-to-market each investment, and in doing so they achieve a quarterly market value of the fund. In quarters when there are no cash flows or change in the book value, the investment value is changed by the same amount as the change in the valuation of publicly traded companies in the same industry. The resulting time-series of portfolio value is regressed on asset pricing factors, thereby obtaining an alpha that indicates whether there is out-performance; a similar methodology is used in the next section. 12

15 III. Risk based explanations Many institutions invest heavily in PE based on the belief that PE returns are largely uncorrelated with market returns. If this is so, then the above performance may be justified by the attractive hedging properties of private equity investments. In the previous section, we report that there is little documentation about how various sources of systematic risk influence returns, and existing estimates are quite conflicting. Relying on a richer dataset and a different approach, we attempt, in sub-sections A, B and C, to investigate how business cycles and stock-market cycles influence PE performance. Then, in sub-section D, we report how certain fund characteristics relate to performance in order to understand the primary drivers of underperformance. A. Systematic risk: Alphas à la Gompers and Lerner (1997) We begin by implementing a methodology similar to what Gompers and Lerner (1997) propose in a similar context. 14 For each fund, we compute quarterly returns as follows: R q = D q T q + BV BV t t 1 BV t 1 1 Where R q is the return during quarter q, D q and T q are respectively the distribution and takedown that occurred during quarter q, and BV t is the book value of the fund at time t (which is the end of quarter q). The book value equals what is reported by the fund if and only if this value is different from the value reported at the end of the previous quarter. The idea is that we trust the accounting estimation only when it is revised. If the reported book value is the same as the book value reported for the previous quarter, then we estimate the current book value to be: BV BV (1 + PR t = t 1 q ) Where PR q is the return offered by publicly traded companies. This return is determined as follows. If we have information about the content of the fund portfolio during quarter q, 48 then PRq = ωi, qri, q, otherwise PR q = R m, q ; Rm,q is the return of the public market portfolio i= 1 (CRSP-equally weighted), ω i, q is the fraction of the fund portfolio that is invested in industry i, and R i,q is the average (equally-weighted) return of publicly traded companies in industry i. The idea is to mark-to-market the value of the fund. This time-series of returns can then be regressed on the time-series of the public stock-market portfolio returns, thereby obtaining the 13

16 (1997). 16 Table 5 equivalent of a Jensen s alpha and CAPM-beta. 15 Results for both Sample 1 and Sample 2 are reported in Table 5. Two models are used: the CAPM and the four-factor model of Cahart We find that the average CAPM-beta is low. The (equally-weighted) average beta varies between 0.5 and 0.65 across samples and specifications. This low value comes certainly from the fact that our marking-to-market strategy is very conservative at the fund level. It is likely that, in reality, the market value of a fund varies more than what we assume and certainly covaries more with the market than what we estimate. The outcome of this exercise is that despite such a low estimated beta, we find that the alpha is not statistically significantly different from zero for 92% of the funds. This means that even with a low estimate of beta, there is no significant overperformance or underperformance of PE funds. This is a first indication that taking systematic risk into account exacerbates the low performance puzzle instead of solving it. Nonetheless, we note that the economic magnitude of alphas is reasonably high (3% per quarter), hence the non-statistical significance comes from the high volatility of alphas; a result that is similar to what Cochrane (2003) observes. In addition, alphas are high because we do not adjust the residual values reported in December 2003 as we do in section I. There, we found that residual values reported in December 2003 are optimistic given the characteristics of the funds at that date. The estimated CAPM-betas are surprisingly low. Typically, small firms are expected to have a beta higher than one, especially so high-tech firms; and levered firms are also expected to have high betas. This suggests that a private equity portfolio beta should be above one. In the next sub-section, we compute the level of betas that we would expect in a CAPM-world. B. CAPM Beta à la Kaplan and Ruback (1995) Were private equity investments continuously traded, it is likely they would have relatively high betas; the reason being that LBO firms have a high level of debt and VC firms are small and typically belong to high-beta industries. To keep a certain order of magnitude in mind, we compute, for each fund, the average beta of its investments. Importantly, this average investment beta is not the same as the fund beta. They would be equal if all investments were 14

17 taking place at funds inception and exited when funds are liquidated. We think of this average investment beta as a proxy for fund beta. We assume that the CAPM holds and that the betas on asset are the same inside a given industry. 17 By making further assumptions regarding the leverage of the deals we can assess the beta of each investment made by GPs. The methodology that we follow is basically what is proposed by Kaplan and Ruback (1995) and construction details are presented in the appendix. The fundamental idea is to compute the beta on asset that prevails in each industry, assign it to each PE investment made in this industry, and lever it up as a function of the deal type (VC or BO). The betas are then averaged for each fund. Given our assumptions, note that variations in betas reflect a combination of the riskyness of the industry in which funds invest at the time of investment, the type of investments made and the duration of each investment. To compute betas we assume that the leverage of BOs decreases linearly over time from a debt-to-equity ratio of 3 (at entry) down to the leverage that prevails in the industry (at exit). The beta on debt is assumed to be 0.25 and the beta for VC deals is assumed to be the same as the beta of the 20%-smallest stocks in the corresponding industry. Importantly, betas are computed using Dimson s correction. The choice of the beta on debt is based on Cornell and Green (1991), who evaluate the beta for high-grade debt (from 1977 to 1989). The choice of equity-to-debt ratio is based on Cotter and Peck (2001) and on Lerner and Hardymon (2002, p112), who argue that a 3:1 debt-toequity ratio for buyouts is rather conservative. Table 6 The median and average estimated betas for various sub-samples are reported in Table 6, Panel B. We also document how betas differ among venture vs. buyout funds and large vs. small funds. The average beta of BO funds is 1.7 and is a similar 1.6 for VC funds. To the extent that beta is a measure of aggressiveness, we find that the 25% smallest funds (average beta is 2.2) are more aggressive than the 25% largest funds (average beta is 1.6). These betas are, therefore, significantly higher than what we obtained in the previous sub-section. To compare these two sets of betas, the Gompers-Lerner betas described in the previous sub-section are reported in Table 6, Panel A. We then test whether fund average betas are related to performance. Table 7-Panel A reports a significant and positive univariate relationship between beta and performance. Again, this beta is not a fund beta in the CAPM sense and the intercept does not correspond to the 15

18 Jensen s alpha. This is nonetheless an interesting finding. It is also important to note that this relationship stays significant when we control for fund characteristics (Panel C). In contrast (unreported result) Gompers-lerner betas are not found to be related to fund performance. To conclude, the performance evaluation in section I appears optimistic as we implicitly assumed a beta of one. Moreover, even when beta is estimated in a conservative manner, alphas are relatively high but the volatility of returns is such that alphas are not statistically different from zero. To further understand the interaction between both stock-market cycles and business cycles, and fund performance, we now investigate the sensitivity of fund performance to their exposure to various proxies of these cycles. C. Fund exposure to risk factors Using available data about the underlying investments of funds, we can compute the exposure of each fund to several factors. Construction details are given in the appendix but the idea is straightforward. Assume that we want to know the influence of credit spreads during the life of the investments on performance. As we mentioned in section II, we cannot document it at the investment level. Nonetheless, we can have a sense of the influence of credit spreads on the net fund performance. We know when certain investments took place for a fund. If we assume these investments to be representative enough, then we can compute the average level of credit spreads during the life of each investment and average them at the fund level (see appendix A.V.). We label the output as fund loading on credit spreads. Indeed, if a fund invests mainly during times when credit spreads are high then its loading on credit spreads will be high. Finally, by running a cross-sectional regression of fund performance on their loadings, we assess the exposure/sensitivity of net fund performance to credit spreads or other variables. Nonetheless, the intercept of this regression cannot be interpreted as a Jensen s alpha as this exercise is not a risk correction. The goal is simply to assess the conservativeness of the assumption of a beta equals to one by investigating the potential hedging properties of private equity. If we find that performance is significantly counter-cyclical, then that will be a step toward explaining why performance was found to be relatively low. If we do not, then the puzzle is still standing and, potentially, even greater. The variables that we choose are related to both business cycles and public stock-market cycles, and are expected to play an important role for performance. We construct the loading on 16

19 the following variables: unexpected real GDP growth rate, BAA-bond yields, credit spreads, market portfolio performance, option-based risk factors, aggregate level of earning-to-price ratio and amount of equity issued. All the variables are taken at a quarterly frequency, except the latter two (aggregate level of earning-to-price ratio and amount of equity issued) that are computed at a yearly frequency. The unexpected GDP growth rate is obtained by modeling real GDP growth as an AR(1) process and is aimed at capturing the influence of business cycles during the life of the investments. 18 Yields of Corporate-BAA bonds as reported by Moody s reflect the cost of debt for buyout investments. Credit spread is the difference between the yield on the corporate-baa bonds and the 10-year treasury bonds. This variable captures both the probability of default and the expected recovery in case of default in the economy, which is important for the cost of financing buyouts. It also captures certain dimensions of business cycles that are relevant for both venture and buyout investments. Times of high credit spread typically correspond to down economic times, especially for small and innovative firms. The market performance aims at assessing the impact of investing in periods of high stock-market performance. Finally, the option-based risk factors consist of out-of-the-money European call and put options on the S&P 500 composite index traded on the Chicago Mercantile Exchange. These option-based factors are proposed by Agarwal and Naik (2004), who argue that it is important to take into account optionlike features to evaluate the risk of hedge funds. 19 In the context of private equity, both venture and buyout investments are likely to have an option-like payoff: buyout investments as they are highly levered and venture investments as they are carried up by stages with, each time, a target market value beyond which there is a new round of financing. Consequently, Cochrane (2003) points out that venture capital investments are like options; they have a small chance of a huge payoff. This type of payoff is expected to resemble the payoff of a call option. In addition, PE investments pay-off if and only if a market for PE exit is active. Indeed, in bad economic times, both the M&A and the IPO market are inexistent. This represents a substantial risk for PE funds as, if they are close to their liquidation date, they will be forced to sell many investments at a substantially depressed price. This resembles the situation of an investor who is short on a put option on the S&P composite index. Finally, we report both the average earning-to-price ratio (price multiples) and the state of the IPO market that prevails when investments are exited. These indicators being related to the valuation of firms that are exited, they are very likely to play an important role for performance. 17

20 Table 7 In Panels A and B of Table 7, we report how these loadings relate to performance. We first report univariate evidence. In order to compare the relative influence of each loading, variables are standardized by subtracting their sample mean and dividing by their sample standard deviation (i.e. each variable is expressed in terms of z-score). PE funds that entered investments at time when interest rates and credit spreads were high underperform. The results for the loadings during investment life are all with the expected sign and with a high statistical significance. PE performance increases substantially when investments are realized during times of high GDP growth, high public market performance, high return for call options, low returns on put options, low credit spreads and low interest rates. To assess the economic importance of the public stock-markets, let us see what the expected PE performance should be in two scenarios: when the stock-markets return Rm=12% per year on average and when they return Rm=24% per year on average. The mean and standard deviation of the dependent variable (ln(1+pi)) are 0.55 and 0.34 and the mean and standard deviation of the independent variable (gross average quarterly return of the CRSP-VW index) are 1.03 and In the univariate regression, the exposure to the market loading is Hence, in the first scenario (Rm=12%), the expected performance is 0.73 and in the second scenario (Rm=24%) the expected performance is 0.86, which is larger but not economically sizeable. Note, however, that a profitability index (our performance measure) is a measure of performance relative to the public market. This suggests that when the public stock-market does badly, PE funds do slightly worse than the market and vice versa when the market does well. A similar result is obtained when we use the fund IRR as a performance measure, the relationship between fund performance and market performance is basically one to one. In the first scenario (Rm=12%), the expected IRR is 9.4% and in the second scenario (Rm=24%), the expected IRR is 19.2%. 20 The effect of the public stock-market performance, put option performance and of credit spreads are the three most important during-investment variables. These three variables have a similar impact on PE performance: a decrease in the credit spread, a decrease in the return on put options and an increase in the market performance by one standard deviation translate in about 0.15 standard deviation increase in PE performance (relative to the market) for each of the three variables. 18

21 For the exit conditions, both loadings are statistically significant and of the same sign as anticipated. When the IPO market is active, or when earnings are low relative to stock prices, PE performance is significantly higher. The importance of price multiple reflects the fact that when investments are exited, the public market valuation levels are used at the time of the IPO or M&A. Hence, PE performance highly depends on the valuation levels that prevail in public stock-markets. Finally, we investigate which effects dominate in a multiple regression. First, we show that the level of credit spreads at entry is more important than the level of credit spreads during the investment life. Second, the return of put options is more important than the return of call options. It means that the two option-like effects that we outlined above exist but the second one dominates. That is the payoff of PE funds resembles more the payoff of a short position on a put option. Third, when the most significant variables are included then two stay significant: the put option and price-multiples at exit. This result is very strong, robust to the inclusion of fund characteristics (see Panel C) and very interesting. It means that the foremost risk that is involved in private equity is twofold: whether an exit market is active and if it does what is the valuation levels that prevail. Hence PE funds, like hedge funds (see Agarwal and Naik, 2004), bear significant left-tail risk. In bad economic times, there is no exit market hence there is no cash distribution to investors and this is likely to be times when they would value it most. When the economy is a normal or a high state, then the uncertainty is about the valuation levels. This indicates that PE funds are exposed to significant risks that are likely to be undiversifiable and also different from the standard market risk that we assessed in sub-sections A and B above. D. Fund characteristics and performance We now focus on the characteristics that are related to performance. The main goal here is to characterize the type of funds that underperform in order to better understand the source of the low average performance. The results are displayed in Table 7 Panel C. To begin with, we include fund size and fund size squared. Size is an important characteristic and captures several dimensions that should positively relate to performance such as reputation, economies of scale and learning. Indeed, a larger fund may learn faster as it makes more deals than a smaller fund. On the other hand, larger funds have more scope for opportunistic behavior that may not benefit LPs (see Gompers and Lerner, 2002). For example, 19

22 large US venture funds are more likely to invest in certain buyout deals or in Europe to obtain a track record for these types of investment. In case of success, a GP would be able to raise extra funds that focus on buyout or on PE investments in Europe. This brings both diversification and additional income to the GP at the cost of LPs. Two proxies for this behavior are constructed: Herfindahl indices for both countries and type of investments. We find no robust and consistent relationship between these variables and fund performance. An additional downside of running a large fund is that it increases the difficulty of finding good deals. To conclude on the effect of size, we expect a concave relationship that may be different for VC and BO funds. Results show that there is indeed a concave relationship for all funds but when we separate the effect of size for VC and BO fund, we find that the concavity is exclusive to BO funds. An explanation is that large VC funds, as exemplified above, have a higher incentive to engage in an opportunistic behavior. As discussed in section I, we find that liquidated funds have a significantly higher performance than non liquidated funds and funds that keep their investments longer underperform. This is consistent with GPs timing the exit of their investments and the closure of funds to mask (stress) bad (good) performance. The highest statistical significance is found for the proportion invested in Europe. These investments have strongly underperformed. Kaplan et al. (2003) and Hege et al. (2003) also document that certain European private equity investments have lead to very poor performance and advance various reasons for this, such as a wrong type of contract used and institutional differences. Typically, the European PE industry is also seen as younger and at a lower point in the learning curve. We also find that first-time funds underperform while experienced funds significantly outperform. Being experienced increased the expected profitability index by 0.15 (from 0.78 to 0.93). This effect is of considerable economic magnitude and in line with what Gompers and Lerner (1999) and Kaplan and Schoar (2003) report for different samples. Kaplan and Schoar find that first-time funds are raised after periods of high returns for the PE industry. It is possible that unskilled GPs enter the market at such times so as to attract naive investors. Investments in high-technology industries paid off. 21 Likely, funds raised before 1995 that have invested in high-tech industries have benefited from the buoyant IPO market of the late 90s. Also of interest is the fact that when a GP is often the main investor, its fund outperforms. If 20