Raising Funds on Performance: Are Private Equity Returns Too Good to Be True? Niklas Hüther Indiana University May 13, 2016 Abstract This paper develops an identification strategy to analyze empirically whether agency problems impact reported returns at the time a new private equity fund is raised. I document that private equity funds do not only hold private deals but that they also hold shares in publicly traded stocks, in which they take an activist role. These stock investments provide observable performance throughout the fund life in contrast to prices of endogenous portfolio company exits. I apply this novel benchmark of private equity performance to explain reported performance run-ups when managers advertise a new private equity fund. Using quarterly cash flow and valuation data on 2,776 portfolio company investments by 138 U.S. buyout funds, I find in contrast to industry perception that fund managers time fundraising with strong current fund performance instead of manipulating interim performance estimates. I am very grateful for valuable advice from David Robinson, Alon Brav, Lukas Schmid, Michael Ewens, Ludovic Phalippou, Adriano Rampini, Roberto Steri and Jonathan Zandberg. I also thank an anonymous investor and seminar participants at Duke University, Indiana University, Erasmus University Rotterdam, University of Nebraska Lincoln and Southern Methodist University. Address: Kelley School of Business, Indiana University, 1275 E 10th St, Bloomington, IN 47405. Email: nhuether@indiana.edu.
1 Introduction Investments in private equity are typically structured as ten year limited partnerships in which fund managers act as general partners (GPs) and investors act as limited partners (LPs). Before private equity funds are liquidated, their returns rely heavily on reports of interim, unrealized performance of their investments. While these net asset values (NAVs) are subjective estimates of fund performance in privately held deals, in this paper I document and exploit the fact that private equity funds often also hold shares in publicly traded stocks as part of their portfolios. Just as with their private investments, I present evidence that private equity funds take activist roles in these companies. In contrast to endogenous portfolio company exits, returns of investments in publicly traded stocks are observable throughout the fund s life. Since performance is observable for these investments, they provide a simple empirical benchmark for private equity fund returns. Based on this benchmark I am able to provide empirical evidence on whether agency problems between GP and LPs impact reported fund performance, which concerns investors and policy makers alike. In particular, this benchmark allows me to shed new light on the fact that GPs new fundraising events seem to coincide with estimates of strong current fund performance in order to attract potential LPs. 1 This pattern is consistent with two hypotheses. The first is that GPs manipulate NAVs around fundraising events, which displays a hidden action to improve their quality signal that outsiders observe. The second is that fundraising events are endogenously timed to strong performance as the GPs intend to overcome hidden information about their quality. Both hypotheses are consistent with the conjecture that GPs try to signal high quality to attract potential LPs for a follow-on fund, but the implications and prescriptions for the industry hinge critically on GPs nature of grandstanding. 1 Important recent papers are: Barber and Yasuda (2015), Chakraborty and Ewens (2015), Brown et al. (2015) and Jenkinson et al. (2013). 1
I build a new hand-collected dataset of U.S. buyout funds with quarterly underlying asset valuations of deals, deal cash flows and information on their managing GPs. My dataset allows me to compare observable and unobservable performance due to deal-level information and to distinguish between funds that attempted to raise a new fund and those that did not try to fundraise. Previous work, that dealt with fund-level data, could not differentiate between observable and unobservable fund performance, which made it difficult to detect any discontinuities in reports of estimated performance. In addition, so far assumptions for fundraising events have been based on observable fundraising outcomes. With the use of this new dataset, my results are the following: I find that there is no systematic bias between private equity fund performance based on stock investments versus fund performance based on privately held deals. Cumulative portfolio excess returns in stock investments as well as in privately held deals peak around fundraising events. This is in line with the hypothesis that GPs time fundraising to true estimates of strong performance. My results provide no evidence of manipulated NAVs in order to advertise a new fund. While GPs that hold funds with interim performance peaks are, on average, unsuccessful to raise a follow-on fund, those that succeed should not be of concern to the industry. Being able to distinguish between GPs that attempted to raise and failed from those who did not try is a big advantage for my analysis of causation. Since fundraising events occur at different times in the private equity fund life (see e.g. Barber and Yasuda 2015), performance peaks at the same life time in funds with no successor might simply display the natural evolution of underlying asset valuations of the portfolio companies. While controlling for the fund life quarter, my results show that performance peaks are connected to actual fundraising events where GPs have an incentive to signal performance. The use of an actual event allows me to test in a second step based on my novel benchmark whether these strong NAVs around fundraising are subject to deception of 2
GPs. Of course, my benchmarking approach might appear questionable, since expected portfolio returns based on investments in publicly traded stocks could be severely different from private companies. My evidence however suggests that investments in stocks capture risk in excess of the public market consistent with findings for privately held deals (e.g. Korteweg and Sorensen 2010). Stock investments in my sample are accompanied by a Schedule 13D filing as an indication of activism and display abnormal returns around the announcement of activism. These excess market returns could reflect value added, selection abilities by the GP, higher systematic risk through leverage or other risk factors associated with private equity fund investments. In equilibrium the expected returns in publicly traded companies should be lower obviously than in private. Otherwise, GPs should not have an incentive to invest in private companies in the first place. I dont find a difference between private an publicly observable returns around fundraising which shows that managers dont only inflate valuation but seem to give a conservative private valuation. To back up this hypothesis, I actually observe that at liquidation the returns for the private investments are higher than for investments in publicly traded stocks. My results are also robust to alternative explanations. Since GPs only invest a small portion of their funds in publicly traded stocks that are not taken private, they might not have an incentive to time fundraising to strong observable performance. This would imply that strong fund performance in stock investments in contrast to privately held deals is potentially spuriously correlated to fundraising events and does not present a useful benchmark to analyze the impact of agency problems around fundraising. This possibility leaves scope for NAV manipulation. I address this concern with hazard models and find that observable as well as unobservable performance explain the temporal variation in the probability of raising a follow-on fund. 3
Another concern is that NAVs in privately held companies should reflect slower price adjustments to changes in expectations compared to stocks that are traded on a liquid market. As a result estimates of NAVs in privately held companies might be conservative before and quickly inflated around fundraising events. Based on deal-level data and information on GPs, my results however show that reported deal performance at heterogeneous fundraising times does not systematically differ between GPs that are invested in the same company. Overall, this evidence is consistent with timed fundraising to estimates of strong performance but not with agency problems impacting reported return estimates. The rest of the paper proceeds as follows. Section 2 describes the disclosure game between GPs and LPs. Section 3 discusses the sample. Section 4 provides descriptive statistics on the types of publicly traded portfolio companies and the stock market s reaction to invested buyout funds. Section 5 presents the main findings and Section 6 concludes. 2 Theoretical Prediction on Signaling Performance Although my analysis is purely empirical in nature -my aim in this paper is not to establish a testable theoretical model- it is nevertheless useful to embed GPs disclosures to LPs into interactions between the principals and agents of theory. Limited partners potentially face two types of problems associated with asymmetry of information in private equity. On the one hand, they could be concerned that GPs will not work hard to maximize value after LPs have committed capital to the fund. For such a case, when the GP s effort is unobservable to the LPs, compensation contracts are in place that depend on performance. Behavioral effects of compensation in private equity have been analyzed theoretically by Hüther (2014) and empirically by Robinson and Sensoy (2013) and Hüther et al. (2015). 4
On the other hand, LPs might also be concerned to detect good quality GPs before committing capital to a fund. If it is difficult for potential LPs to determine the quality of the GP so that they rely on performance signals, low-quality GPs without convincing track records, in particular, should have an incentive to signal strong current performance. The idea is that usually high-quality GPs, that have been in the market for a longer time, were able to establish a good verifiable performance signal. Thus, they should have less incentives to display high returns of their current fund at the time of a new fundraising event in order to signal their quality. If low-quality GPs also place less focus on current performance signals, the result might be a separating equilibrium in which LPs will completely stop investing in any further funds of low types. In a pooling equilibrium GPs avoid being classified as low types, while high types would still find it profitable to raise funds as long as fundraising times are heterogeneous (as shown by Barber and Yasuda 2015). In such a case, LPs are not only faced with hidden actions ex post, if effort is unobservable, but potentially with hidden action ex ante through low-quality GP s incentive of manipulating NAVs. The latter is focus of this paper. In the real world it is difficult for low types to mimic high types if they lack good verifiable performance. LPs may discount increased valuation before fundraising if the performance signal is noisy (Stein 1989). Sophisticated investors such as pension funds -that contribute the largest amount of capital to buyout funds- are successful fund pickers (Lerner, Schoar and Wongsunwai 2007) and provide even less scope for GPs to systematically alter their NAVs. Yet, evidence by Kaplan and Schoar (2005) has shown that low-quality GPs are able to raise new funds. Combined with previous evidence of spikes in reported returns for low-quality GPs around a new fundraising event, the findings are in line with GPs attempt to signal LPs they are of high caliber. Whether window-dressing or timing fundraising is the most efficient strategy for the GP is clearly conditional on her 5
evaluation of future compensation. Presumably, manipulating reported valuation only makes sense from an agency perspective if lower costs of managing performance signals outweigh potential reputational losses and a lower present value of future compensation. Otherwise, showing real improvements of NAVs by scoring with publicly observable and unobservable returns, gives low type GPs a competitive advantage in the fundraising game. Fundraising may not be based on raw performance but rather on peer comparisons (see Barber and Yasuda 2015) which might drive behavior. Timing fund-raising closer to strong true performance compared to high quality peers, requires low types to generate returns earlier in their fund life than their vintage year high type cohort funds. In other words, low type GPs need to generate true returns earlier than their high type competitors to signal strong relative performance without risking the loss of reputation. This is in line with the model by Shin (2003) who shows that it is optimal for a manager to disclose the best possible outcome to interested parties by verifiable disclosures as an effective penalty would always annihilate reporting false evidence. Although venture capital funds base their valuation on realized prices, Chakraborty and Ewens (2015) find that their GPs delay write-offs which speaks against the hypothesis of manipulating NAVs. 3 Data and Sample Representativeness The data in my study were provided to me by one of the largest international LPs in the world on an anonymous and confidential basis. Although the data source is a large, global investor that invests in various private equity asset classes, I restrict my analysis to U.S. buyout funds to narrow the scope of investment focus with regard to the research question. The data set comprises 2,776 fund-investment pairs of 136 funds. While 19.2% (532) of these investments were publicly traded at some point 6
during their holding period, 7.7% (213) are fund investments in publicly traded stocks which remained publicly traded at least until their exit. These numbers correspond to 2,400 unique sample portfolio companies, of which 287 are publicly held and 202 are investments in publicly traded stocks. Being able to differentiate between investments held by several funds, I can identify 239 (127) privately held companies in the portfolios of at least two different funds (at least two different GPs). Of the 136 funds, 121 actually went into marketing to raise a new fund, whereas 15 did not even try to advertise a new fund. Eventually, 101 successfully raised a follow-on fund and 20 failed trying. Although my sample is smaller compared to the fraction of funds and portfolio companies in Thomson One for which I don t have information, my data allows me to analyze reported NAVs, distributions, investment costs and observe the GICS industry group in the 2,776 investments of the 136 funds. I consider funds with vintage years between 1996 and 2010. For this time period I am able to observe the potential date of a new fundraising with disclosed performance for all current fund investments, since my data on portfolio companies is as of 12/31/2013. This is a big advantage, as normally GPs transmit only quarterly net-of-fees fund level performance (see, e.g., Metrick and Yasuda 2010, Robinson and Sensoy 2013). In addition, I have access to the date of the focal funds reported performance in the due diligence package for the follow-on fundraising. Thus, in contrast to previous fundraising related studies, I am able to pin down the event quarter in which the GP actually promotes her current performance. With respect to this paper, a fundraising event is referred to precisely this quarter. Please see Table 1 Table 1 displays the overview of my sample data and compares descriptive statistics to the overall investment universe approximated by Thomson One in order to analyze the representativeness of my sample in more detail. 7
I find that my sample consists of more recent funds as well as statistically and economically larger funds than the average and median from Thompson One. Besides, funds raised by more established GPs are significantly more likely to be included in my sample. Partly, this is attributable to the fact that the large size of the Investor in question precluded them from investing in small funds. However, if compared to other recent studies with fund level NAV data, e.g., Barber and Yasuda (2015) and Brown et al. (2015), I find a smaller gap to the size of my analyzed funds (3,107 million USD vs. 1,532 million USD as in Barber and Yasuda (2015) or 1,324 million USD as in Brown et al. (2015) vs. 579 million USD as in Thomson One). A suitable comparison however is difficult, since their studies are based on commercial databases. Recent studies as in Metrick and Yasuda (2010) that also use data from a large independent LP report similar average fund sizes (1,238 million USD) as in, e.g., Barber and Yasuda (2015) and Brown et al. (2015). Considering that Metrick and Yasuda (2010) have data until 2006 and thus leave out the heydays of fundraising in the U.S. buyout market between 2006 and 2008, my average fund size seems to be comparable. In relation to Thomson One s data universe, my GPs are on average older and larger in terms of previous investment activity. To capture this, I divide the total capital invested in the GP s funds, whose first closing was in the ten calendar years prior to the year that this fund closed, by the total amount raised in all U.S. buyout funds in these years. My sample GPs appear more experienced as active investors instead of displaying zombie firms with on average 5 funds for a life time of 15 years versus an average of 2 funds in 11 years as in Thomson One. A data set of overall more experienced GPs might raise concerns that these GPs should have a low incentive to improve funds NAVs. However, a fraction of 26% of these funds did not raise a follow-on fund. This fraction is slightly lower compared to related studies (31% as in Chakraborty and Evens 2015, 44% as in Jenkinson et al. 2013). Brown et al. (2015) find that more experienced GPs are more likely to have a follow on fund, although the relationship is not monotonic. A typical example, discussed 8
in the media, is the failed fundraising attempt of the fourth U.S. buyout fund by J.W. Childs Associates in 2007. The private equity firm was in business for 12 years at that time and had around 3 billion USD under management. 4 Publicly Traded Investments Turning to my key identifier of the veracity of reported performance I distinguish between different types of publicly traded investments held by U.S. buyout funds. Based on previous research it is well known that buyout funds acquire a majority stake in publicly traded companies to take them private, mostly in the form of an LBO. It has also been documented that the GPs keep investments in their portfolio upon an IPO. Besides, a combination of both is possible, i.e., going-private and going-public transactions for the same portfolio company. Barely any evidence has been provided on buyout funds investments in publicly traded stocks. 2 4.1 Basic Summary Statistics on Publicly Traded Investments About 60% of the funds in my sample invested in publicly traded stocks -with a Schedule 13D filing- which were not taken private before their exit. Fractions remain similar for the subsamples of successful, unsuccessful and unattempted fundraisers. Please see Table 2 Table 2 reports descriptive statistics on the funds fractions of all types of publicly traded portfolio companies. Fractions of publicly held portfolio companies are based 2 Two studies on German data (Mietzner and Schweizer 2014, Petry 2015) as well as one study using Schedule 13D filings (Chen et al. 2014), report of these type of investments held by private equity firms. Chen et al. (2014) consider 77 investments with a Schedule 13D filing of a buyout fund, without any information of the remaining portfolio companies. 9
on the number as well as on the USD investment amount of the funds. On average -irrespective of the GPs quality- funds hold 20 companies in their portfolio during the sample fund life (SFL) and around 14 at a fundraising event (FRE). In line with Kaplan and Schoar (2005), there is a monotonic relationship between quality and fund size. The average invested capital of a fund for successful fundraisers is smaller compared to unsuccessful and even more so compared to unattempted fundraisers (2,9 billion USD vs. 2,4 billion USD vs. 2,2 billion USD). Taking a look at the publicly held portfolio fractions, investments in publicly traded stocks make up around 8% (11% in terms of invested capital) which is the same portion for going-public transactions. A lower portfolio fraction constitutes of going-private transactions (on average around 3% (5%)) as well as going-private followed by going-public transactions (on average below 1% (1%)). Averages remain similar across the subsamples of fundraisers, and point out the economic relevance of funds investments in publicly traded stocks -also around a fundraising event. Please see Table 3 Statistics in Table 3 show the distribution of funds publicly held investments over time. Since fundraising events differ between funds, the observations are on the fundinvestment pair level. Except for going-private followed by going-public transactions, publicly held investments generally span the entire sample period, with peaks in numbers around the late 90s / early 2000s and 2005 to 2008. This suggests that investment holding times are procyclical, which is in line with the finding by Robinson and Sensoy (2013). 4.2 Ownership, Trading Liquidity and Holding Time So far, I presented sample statistics for various types of publicly traded investments. While I find that GPs typically file a Schedule 13G upon an IPO of a portfolio company, 10
they file a Schedule 13D for publicly traded stock investments if they passed the 5% ownership threshold. I classify an investment as a publicly traded stock investment if I am able to observe a Schedule 13D filing by a sample fund and this investment is not taken private while held in the portfolio. 3 Although a Schedule 13D filing is also present for investments in publicly traded companies which are taken private in my sample, I excluded these from the group of investments in publicly traded stocks. 4 Thus, the four groups of publicly held portfolio companies are separated into two buckets. Investments in publicly traded stocks remain one group and the remaining publicly held investments as well as purely privately held portfolio companies presents the group of privately held investments. So far, I was able to present evidence that investments in publicly traded stocks represent a small but common portion in the portfolio of buyout funds. Yet, it is still unclear whether these investments are suitable to display observable returns to activism instead of pending attempts to take control of the company with low trading activity and short holding times. Please see Table 4 Statistics in Table 4 show, while stakes in these companies are larger compared to activist hedge funds, buyout funds remain minority shareholders. Only at the 95 th percentile in the sample of stock investments, the stake is 65.5% and slightly jumps in the majority control range. The median initial (maximum) percentage stake that a fund invests in a publicly traded stock is 14.15 (20.7). Information on the initial ownership in column 1 is captured from the associated Schedule 13D filing and data on the maximum ownership comes from amendments to the 13D filings (Schedule 13D/A). For the 17 3 17 investments in stocks were not accompanied by a Schedule 13D filing because the funds ownership levels in the company were below the 5% threshold. For these companies I used information based on news reports in Lexis/Nexis Academic. 4 Various literature has dealt with the question why companies are taken private. In line with this literature there should be differences between investments that were held publicly traded and those that were taken private, which I control for by forming separate groups. 11
non-schedule 13D events, the information is collected from news reports in Lexis/Nexis Academic. I record the highest stake by a filing party in the company. In order to observe transparent prices for these investments, stocks must be frequently traded. Columns 3 and 4 compare trading liquidity, measured by the Amihud (2002) illiquidity measure, which is defined as the yearly average of 1000 Return /(DollarTradingVolume). The measure is based on daily data in the fiscal year immediately preceding the initial 13D filing date. Liquidity is about the same for private equity funds stock investments and a set of industry/market value/book-to-market matched companies. Column 5 shows that median holding time is 3.75 years versus one year for hedge funds (Brav et al. 2008). The distribution of holding times is nearly identical between stock investments and privately held companies. It appears that buyout funds do not seek control in these public companies, given their minority stakes and same holding periods as in their private portfolio. Similar trading liquidity relative to the matched universe of firms suggests that they reflect transparent market prices. 4.3 Stock-price reaction to buyout fund investors Do returns of stock investments by buyout funds reflect performance in excess of the public market? This fundamental question asks whether fund s observable and unobservable performance can be compared, in the sense that stock should reflect abnormal returns consistent with findings for private investments (Cochrane 2005). If the market perceives a activism by a buyout fund, a run-up in performance is expected around the time the investment becomes publicly shared information. Figure 1 plots the average buy-and-hold return, in excess of the buy-and-hold return on the U.S. MSCI industry indices, in event time - 20 days before and 20 days after the Schedule 13D filing. Abnormal returns are plotted for the full sample as well as for the subsample of funds that successfully raised a new fund and those that failed. In the full sample, there is a run-up of about 15% between 12 days to 1 day prior to filing. The run-up is 12
about four times as large as for activist hedge funds -considering that the buyout fund s initial stake in the traded company is also about 3 times as large. The filing day sees a run-up of 4.8% with an increase up to a total of 23.2% in 20 days. Importantly, the runups occur proportionally in the same manner for successful fundraisers / unsuccessful fundraisers, with a jump on the filing day and the following day of 7.6% / 3.1% and a total run-up in 20 days post event of around 29.9% / 12.5%. One potential explanation for the high abnormal return around the filing date could be a temporary price impact caused by building up positions and buying pressure by other investors due to the investing buyout fund. However, Figure 1 shows no reversal of abnormal returns 20 days after the announcement. 4.4 Private Equity Factor exposures For a better understanding of the announcement return for activism, I test if abnormal returns are linked to systematic risk factors. I implement the calendar-time portfolio approach to regress portfolio excess returns based on stocks in a PE fund on the three Fama and French (1993) factors and the Carhart (1997) momentum factor, as in following equation: R p,t R f,t = α + β p (R m,t R f,t )) + s p SMB t + h p HML t + e p,t, (1) where R p,t is the equal or value-weighted return for calendar month t for the portfolio of stocks that had a PE activist event in the preceding or following month(s). For example, the portfolio with holding period +1,+12, consists of PE target stocks that have had an activst event in the preceding month and holds on to these companies through a year after there respective activism event. Please see Table 5 13
Table 5 shows positive factor loadings on SMB indicating that targeted public companies comove with small firms. The Beta loading is above one in all event windows in the equal weight specification and for the one year around the event in the value-weight specification. GPs seem to slightly increase the degree of leverage in smaller sample firms as the positive factor loadings on Beta peak in the year following the event in the equal weight specification. With both equal and value weighing, alpha is around 5% and significant in the year after the event [Event]. Alphas revert close to zero during the 12 months after the event and basically remain zero in the following two year. This refutes the possibility that te market overreacts for up two three years. Most importantly, the significant positive alphas following the announcement of activism show that that the four factor model does not capture the factors associated with PE investments and that private equity returns cannot be appropriately benchmarked with market portfolios. While a small portion of the abnormal announcement return can be explained by a slightly higher Beta risk in the smaller firms in my sample, the statistically significant alphas seem to reflect an important additional factor. This finidng is in line with Cochrane (2005) who suspects an additional factor to explain alphas in VC investments. It appears unlikely for private equity that the risk-return tradeoff is about traded factors but potentially about compensation structure in contracts in line with findings by Huether et al. (2015). 5 Raw (Un)observable Investment Performance After showing evidence of buyout fund activism in publicly traded stocks, I can now address the question of whether observable and unobservable investment returns differ. In particular, I compare performance, in terms of public market equivalents (PMEs), between investments in publicly traded stocks and privately held companies. The PME expresses the total return to a private equity investment in terms of its excess over 14
a publicly investable benchmark -U.S. MSCI industry equity returns. In contrast to the previously reported abnormal buy-and-hold returns, PMEs control for differences in investment and exit times across portfolio companies. By using this measure we can see in Table 6 that investments in publicly traded stocks on average have a PME of 1.33, which is only slightly above the abnormal buy-and-hold returns around the Schedule 13D filing. 5 While part of the abnormal buy-and-hold returns are not earned by the GP due to the volume weighted average price effect, holding returns of investments in publicly traded stocks are still mainly attributed to abnormal returns around the annoucement of activism. 6 The PMEs of GPs investments in publicly traded stocks are not significantly different from the average PME of 1.44 of privately held investments. Differences in average and median PMEs remain insignificant for the subsamples of successful, unsuccessful and unattempted fundraisers. This result suggests that, on average, GPs improve performance to the same extent in market priced as well as in private companies. Please see Table 6 In terms of scope for manipulation, indifferences could stem from realized investments which are observable and less prone to overstated numbers, thus Table 6 also differentiates between realized and unrealized investments. A differentiation between realized and unrealized returns is only possible with data on investment level. Returns based on fund-level data bear the risk that unrealized funds also pick up returns of realized investments. My data does not show any significant difference in returns between realized (unrealized) privately held and realized (unrealized) publicly traded investments. Unrealized investments are investment that are still in the fund s portfolio at the end of the sample period as in 12/31/2013. 5 As before in Table 5, I winsorize the investment PMEs at the 5% extremes to deal with outliers. 6 Holding-period returns for investments in stocks that are calculated on NAVs reported by the GP are in line with calculations based on quoted prices. This suggests that GPs use quoted prices for the valuation of their investments in publicly traded stocks. 15
There is a notable gap between returns of realized and unrealized privately held investments, whereas realized returns for investments in publicly traded stocks are only marginal higher. This can be explained by the immediate run-ups for stocks after the announcement, as shown in section 4.4. The higher average and median returns of realized versus unrealized publicly traded stocks are in line with the notion that the market slightly underestimates the value of ex post successful activism (e.g., Bond, Goldstein and Prescott 2007). An important take away from Table 6 is that unsuccessful fundraisers, which are presumably at the highest risk to manipulate, do not significantly differ in their PMEs of unrealized privately held companies and public stock investments (means: 1.224 vs. 1.165; medians: 0.869 vs. 0.704). While these numbers reveal no evidence of manipulation of quarterly NAVs, performance around a fundraising event might show a different picture. 6 Reported Returns around Fundraising If GPs signal strong performance in some sort, I expect to observe a peak in fund performance around a new fundraising event in line with previous studies. If these peaks are a result of timing fund-raising rather than window-dressing, I should see a post-event declining performance for the funds fraction in private companies as well as in stocks. The quarterly reported return (rt NAV ) of a fund / fund fraction / investment is the key variable for most of my analysis in this main section of the paper. Since the quarterly NAV of a fund can be expressed as: NAV f t = NAV f t 1 (1 + r NAV,f t ) + C f t D f t, (2) 16
where C f t and D f t denote the contributed and distributed capital in quarter t, the fund s quarterly reported return is written as: r NAV,f t = NAVf t ( NAV f t 1 + C f t D f t NAV f t 1 ). (3) If the enumerator in equation 2 is negative, it is equal to the definition of a NAV markdown as in Barber and Yasuda (2015). Since I am interested in the overall NAV management before and after fundraising, instead of merely isolating the effect of a markdown, I consider quarterly reported returns for my analysis. 6.1 Timing of Abnormal Performance Figure 2 presents average cumulative abnormal returns of funds, which are the cumulative excess returns over the S&P500, since fund inception (Panel A) and -12/+12 quarters around the fundraising event (Panel B) (i.e. the quarter for the reported current fund performance in the due diligence). These are generated by pooling all cumulative abnormal returns by fund age for funds with a next fund and those without (left graphs Panel A and B) and then plotting the average cumulative abnormal returns. The right hand side graphs split up the funds without a follow-on fund in unsuccessful and unattempted fundraisers. Based on the average current fund s life for a new fundraising in my sample, I define a hypothetical fundraising event for unattempted fundraisers as the thirteenth quarter of the fund s life. The plots in Panel A show a declining pattern after approximately 3 years into the fund life for funds without a follow-on fund. I find the same pattern in Panel B, keeping in mind that my sample funds on average go into fundraising every three years. Overall, the left hand side graphs of Panel A and Panel B confirm the finding previously shown by Brown et al. (2015). The graphs on the right hand side reveal that the spike in performance around a new fundraising event is driven by attempted fundraisers who ultimately fail to raise. This finding emphasizes 17
the conjecture that funds in the highest need of showing performance, have a spike in cumulative abnormal returns when they try to sell a new fund. Yet, on average, they are unsuccessful. Brown et al. (2015) interpret this result -although they do not differentiate between unsuccessful and unattempted fundraisers- as evidence that low-quality funds inflate NAVs which is seen through by the market. The declining cumulative abnormal returns after the fundraising event are also in line with documented markdowns of NAVs by Barber and Yasuda (2015) raising concerns of manipulation. To put this suspicion to the test, I break down the quarterly reported returns of the entire fund into fractions of quarterly observable, unobservable and realized returns. Figure 3 plots graphs for the funds fraction of privately held portfolio companies, further divided into unrealized and realized companies, as well as for the funds fraction of publicly traded investments. Since, manipulation is mostly expected for funds where the GP tried but failed to raise a follow-on fund, I also differentiate between successful, unsuccessful and unattempted fundraisers. In the top left graph of Figure 3 we observe increasing average cumulative abnormal returns across all four portfolio fractions of successful fundraisers. No sign of a discontinuity around the fundraising event is evident. The same pattern emerges from unattempted fundraisers with a flatter incline. In contrast to these two subsamples, the data shows differences in the pattern of reported performance across the returns for the fund fractions of unsuccessful GPs. Cumulative abnormal returns for privately held realized companies in event time t, displayed by the red line, show a rapid run-up immediately before the fundraising event and a steep decline in quarter t 0 +6. This finding suggests that GPs, that are virtually on the edge of making it, display conservatism in their reporting to surprise LPs with exits before fundraising. They also appear to delay write-offs in line with Chakraborty and Ewens (2015). Both, cumulative abnormal returns of fund fractions in publicly traded stocks and privately unrealized investments increase up to the fundraising event and remain steady afterwards. On the one hand 18
this finding refutes the suspicion of overstating NAVs and on the other hand shows that the markdowns after the attempted fundraising (Barber and Yasuda 2015, Brown et al. 2015, Jenkinson et al. 2013) are a result of delayed write-offs -in line with evidence by Chakraborty and Ewens (2015). Thus, there are spikes in cumulative reported returns of funds resulting from delayed write-offs and peaks remaining from observable and unobservable returns in unrealized portfolio companies. Figure 4 zooms in on the difference in cumulative abnormal returns for successful and unsuccessful fundraisers by the fungibility of the assets. The evolution of cumulative abnormal returns is visibly similar between privately held (left graph) and publicly traded investments (right graph), with an average decline after the fundraising event only for unsuccessful fundraisers. While cumulative abnormal returns on average increase by 55% in fund fractions of privately held companies within the 12 quarters before the fundraising event, they increase only by 36% in fund fractions of investments in publicly traded stocks. The smaller increase reflects the immediate price reaction of the announcement. Section 4.4 showed that stock investment returns by unsuccessful fundraisers see a lower run-up around the announcement with no subsequent reversal in prices. Thus, the question remains why cumulative abnormal returns of these fund fractions increase to the same extent, compared to successful fundraisers, but decline in the post event period. An answer to this question is provided by Figure 5, which illustrates the distribution of investments times according to whether the fund is successful in fundraising or not. The left graph of Figure 5 depicts the distribution of investment times as a function of fund age around fundraising in event time for privately held companies. There is very little difference in distribution of investment times. The investment times are approximately normally distributed with the median around two quarters before the event, meaning that funds have made half of their investments in privately held companies half a year before attempting to raise a new fund. 19
Unlike with privately held companies, there are substantial differences in the distribution of investment times with publicly traded stocks by fundraising type, as can be seen in the right hand side graph of Figure 5. For successful fundraisers the mean / median investment occurs again around half a year before the event, while for unsuccessful fundraisers the mean / median investment is made about two years before the fundraising event. These differences in means / medians are accompanied with substantial variations in distributions. This suggest that unsuccessful fundraisers try to mask their lower announcement returns with pulling stock investments into the pre-fundraising period. By doing so, they can profit from the run-up upon announcement but don t have to fear a reversal in returns, since the market has already priced their lower quality. As a result, pooling cumulative abnormal returns in these investments leads to an incline in performance until the fundraising period and a stagnation in abnormal performance thereafter. 6.2 Peer-Chasing and Fundraising Probability If GPs try to time strong current performance relative to peers around a new fundraising event, we should see an effect of GPs performance rank based on varifiable and unverifiable returns on the probability of fund-raising. To analyze the effect of current fund performance compared to peers on the probability of fundraising, I use a duration model in line with Barber and Yasuda (2015). Since GPs have contractual latitude when they try to raise a new fund before the current fund expires (afterwards they would lose their steady flow of income), with an increasing probability toward the middle and a declining toward the end of the fund s life, I use a Cox proportional hazard rate model. Because of the panel set-up, it is well suited to control for the temporal variation in the probability of raising a new fund. The failure event is the date a GP reports the current fund performance to potential LPs in the due diligence process in order to raise a new fund, which is the fundraising event (quarter). The current fund is at risk during its 20
entire lifetime. In order to avoid a bias in the results, I only consider funds that actually tried to raise a follow-on fund. I specify the hazard rate of a fundraising event for fund i at quarter t as: h(t x t ) = h 0 (t)exp(x T itβ), (4) where h 0 (t) is the baseline hazard function common to all funds in the sample, x it is a vector of covariates that includes time-varying and time-invariant fund characteristics of fund i and β is a vector of coefficients. For the first set of models I include the following 4 covariates: 3 dummy variables that are equal to one if a fund s return multiple rank in quarter t is in the top (second/third) quartile among its vintage year cohort funds and the log of fund size. I use these relative performance measures based on return multiples in comparisson with vintage year cohorts, since theses are performance measures LPs focus on in their assessment process (see also Barber and Yasuda 2015). For the second set of models, I add interaction terms between the performance quartile dummies and a dummy variable that takes the value of one if the GP of fund i failed to raise a next fund (TF). The interaction terms point out the effect whether the current fund s performance rank has a stronger effect on the probability of going into fundraising if the GP ultimately failed to raise a new fund. Please see Table 7 Column (1) of Table 7 includes all investments of fund i in quarter t. The hazard ratios of the top quartile and second quartile dummies are statistically significant and imply a 114.2% and 93.7% increase in the hazard of a fundraising event relative to a fund in the bottom quartile performance. Results in section 4.4 showed that activism for unsuccessful fundraisers is on average valued lower, suggesting a lower quality. Therefore, I expect their fundraising probability to be more sensitive to their current fund s performance rank than for successful fundraisers. The interaction terms with unsuccess- 21
ful fundraisers of the model in column (2) are statistically significant and substantially larger compared to the single effects of the performance quartiles. In the group of unsuccessful fundraisers, a fund that has a return multiple in the top quartile or second quartile has a hazard ratio of 8.33 (= 8.193 + 0.137) or 5.149 (= 5.013 + 0.137) times that of a fund in the bottom quartile. So far, I was basically able to show that my results are in line with Barber and Yasuda (2015). If GPs should have an incentive to manipulate NAVs, the returns for fund fractions in unrealized investments compared to its vintage year cohort funds should effect the probability of a fundraising event. I present results in columns (3) and (4). The hazard ratios for top quartile funds remain statistically significant but drop from 2.142 column (1) to 1.811 column (3) and for unsuccessful funds from 8.33 column (2) to 3.569 column (4). This shows that the probability of a fundraising event is partly explained by successful exits. The remaining effect of quarterly unrealized performance quartiles is quite similar to quartile ranks based on investments in publicly traded stocks. In this specific case I benchmark the return multiple from quarterly pooled stock investments with funds return multiple in all investments. Thus I am able to analyze funds outperformance in quarter t based on quicker returns to activism in public than in private investments and its effect on the funding event. Being in the top quartile implies a hazard ratio of 3.185 (column (5)), which is statistically significant. The top quartile effect of varifiable returns on the timing of fund-raising mainly stems from unsuccessful fundraisers. The hazard ratio for a top quartile unsuccesful fundraiser is 4.983 (= 4.404 + 0.579) versus 3.095 for top quartile successful fundraiser, both statistically significant. Overall, the results in Table 7 imply that current fund s quarterly performance rank -based on all investments, fraction in unrealized privately held investments / stock investments- has a has a positive and significant affect on the probability of a successor fundraising event. 22
6.3 Reported Returns Around Fundraising While scoring with true performance early on, appears as the important mechanism behind timing fund-raising with strong interim performance ranks, visual evidence of peaks in absolute returns presented in Figures 2-5 lacks a statistical test. In line with Figures 2-5, I will focus on absolute reported returns for the rest of my analysis instead of return multiple quartiles, as in section 6.2, which are less precise. Results of the multivariate analysis displayed in Tables 8-11 are not only based on test statistics but also control for further heterogeneity in the data. First, I fosus on potential biases between observable and unobservable returns in my panal data setting. I compute buy-and-hold cumulative abnormal returns (CARs) for each quarter in the +/- three-year fundraising event window. I use these CARs to analyse whether there are systematic biases in the CAR of the fund s fraction in unrealized privately held deals versus the CAR of the fund s fraction in publicly traded stock investments around a fundraising event. The CAR is calculated as the difference between compounded portfolio return and compounded market benchmark return based on the S&P500, which amounts to: CAR NAV t = t (1 + r s ) s=1 t (1 + rs m ), (5) where t and s index the period in event time, r s is the portfolio return in event period s, i.e., over the time interval from s 1 to s, and r m s is the market benchmark return in event period s. The event is defined as the period of one year proceding and following the date a fund advertises a new fund. I estimate the following regression model: s=1 CAR NAV,private it = δdfr it + βcar NAV,public it + γ it + u it, (6) 23
where dfr it denotes the fundraising event period dummy for quarters [-4,+4] and γ it displays fund quarter life time fixed effects. All models are estimated within the time period [-12,+12]. Please see Table 8 Table 8 presents the results. Changes of cumulative abnormal returns in the fund s fraction of unrealized privately held companies are significantly correlated with changes in the portfolio s fraction of stock investments. Controlling for the latter explains any bias in publicly unobservable private equity fund returns around fundraising. CAR NAV,public t Since cannot be stricly exogenous (CAR this quarter affects CAR after this quarter), I account for serial autocorrelation. Since estimates would be unbiased and consistent but inefficient, I use clustered standard errors that are robust to correlation between error terms of same fund and heteroskedasticity over time. To control for robustness, I also estimate different multiple-equation models and account for the fact that error terms are not i.i.d. and lead to a conditional error variance matrix. In a next step, I analyze changes over time comparing reported quarterly returns before and after the fundraising event. All following Tables report estimates for following regression model: r NAV it = α + x T itβ + γ T it1 + ɛ it, (7) where r NAV it is the quarterly reported return of either a fund, a fund fraction or investment, x it is a vector of covariates and γ it is a vector of fixed effects. The main explanatory variable is a dummy variable that takes the value of one in quarters +1 to +12, where 0 is the quarter for which performance of the current fund is reported in the due diligence package. I include calendar year fixed effects to control for heterogeneity in market conditions and industry fixed effects to control for variations across industries. Since all models are estimated in event time, I also include fund (investment) life fixed 24