Flows, Performance, and Managerial Incentives in the Hedge Fund Industry Vikas Agarwal Georgia State University Naveen D. Daniel Georgia State University and Narayan Y. Naik London Business School JEL Classification: G10, G19 This version: September 1, 2003 Vikas Agarwal and Naveen D. Daniel are from Georgia State University, Robinson College of Business, 35, Broad Street, Suite 1221, Atlanta GA 30303, USA: e-mail: vagarwal@gsu.edu (Vikas) and nav@gsu.edu (Naveen) Tel: +1-404-651-2699 (Vikas) +1-404-651-2691 (Naveen) Fax: +1-404-651-2630. Vikas Agarwal is a Research Associate with EDHEC. Narayan Y. Naik is from London Business School, Sussex Place, Regent's Park, London NW1 4SA, United Kingdom: e-mail: nnaik@london.edu Tel: +44-207-262-5050, extension 3579 Fax: +44-207-724-3317. We would like to thank Nicole Boyson, Conrad Ciccotello, William Fung, Mila Getmansky, Paul Gompers, William Goetzmann, Jason Greene, Roy Henriksson, David Hsieh, Robert Kosowski, Lalitha Naveen, Sebastien Pouget, Stefan Reunzi, Krishna Ramaswamy, Ivo Welch, and participants at the Autumn seminar of INQUIRE Europe in Stockholm, All-Georgia conference, EFA 2003 meetings in Glasgow, FMA European 2003 meetings in Dublin, Georgia State University, London Business School and Wharton Hedge Fund conference for many helpful comments and constructive suggestions on an earlier version of this paper. This paper was adjudged the best paper on hedge funds at the European Finance Association (EFA) 2003 meetings in Glasgow. We are grateful for funding from INQUIRE Europe. We are grateful to Josh Rosenberg of Hedge Fund Research Inc., Chicago, TASS Investment Research Ltd., London and Zurich Capital Markets, Switzerland for providing us with the data on individual hedge funds and hedge fund indexes. We are thankful to Otgontsetseg Erhemjamts, Purnendu Nath, and Subhra Tripathy for excellent research assistance. We are responsible for all errors.
Flows, Performance, and Managerial Incentives in the Hedge Fund Industry Abstract Using a comprehensive database of individual hedge funds and funds of hedge funds, we investigate the determinants of money-flow and performance in the hedge fund industry. We have several important findings. First, good performers in a given year experience significantly larger money-flows in the subsequent year and this performance-flow relation is convex. Second, funds with persistently good (bad) performance attract larger (smaller) inflows compared to those that show no persistence. Third, we find that money-flows are positively associated with managerial incentives measured by the delta of the option-like incentive-fee contract offered to hedge fund managers. Fourth, when we examine the relation between flows and future performance, we find that larger hedge funds with greater inflows are associated with worse performance in the future, a result consistent with decreasing returns to scale in the hedge fund industry. Fifth, we find that funds with better managerial incentives (those with greater delta) are associated with better performance in the future. Finally, we find that unlike individual hedge funds, funds of hedge funds enjoy economies of scale. Overall, these results significantly improve our understanding of the complex interaction between money-flows, performance, and managerial incentives in the hedge fund industry. 2
Flows, Performance, and Managerial Incentives in the Hedge Fund Industry In recent years, the hedge fund industry has emerged as an alternative investment vehicle to the traditional mutual fund industry. It differs from the mutual fund industry in two important ways. First, hedge funds are much less regulated compared to mutual funds, with limited transparency and disclosure. Second, hedge funds charge performance-based incentive fees, which help align the interests of manager and investors. These differences have important implications for investment behavior of capital providers and the incentives of the hedge fund managers to deliver superior performance. Therefore, in this paper, we investigate the complex interaction between flows, performance, and managerial incentives in the hedge fund industry. Due to legal restrictions placed on advertising by hedge funds and limited disclosure, investors face significant search costs in selecting hedge funds. While we have a reasonable idea of the factors investors consider before placing their money in mutual funds (Ippolito, 1992; Chevalier and Ellison, 1997; Goetzmann and Peles, 1997; Sirri and Tufano, 1998), pension funds (Del Guercio and Tkac, 2002) and private equity funds (Kaplan and Schoar, 2003), we have limited understanding of the determinants of money-flows in the hedge fund industry. Given the institutional differences between hedge funds and traditional asset management vehicles, different factors may be influencing the investors fund selection process. This leads us to our first research question. What are the determinants of money-flows in the hedge fund industry? In particular, how do the money-flows relate to a fund s past performance (returns and persistence in returns), managerial incentives (call option-like incentive fee structure), and other fund characteristics (size, volatility, and age)? Active money management has long been considered as a puzzle to financial economists (Gruber, 1996). For example, it is not clear why investors devote resources to evaluate past 3
performance and chose funds on that basis, when future performance seems to be unrelated to past performance. To resolve this puzzle, Berk and Green (2002) present a model of active portfolio management where such behavior arises as a rational response on the part of the investors. They argue that investors propensity to chase past performance drives the lack of performance persistence found in the empirical literature. An important assumption underlying their model is that the managers face decreasing returns to scale and therefore, money-flows into better performing funds lower their abnormal returns in future. This leads us to our second research question. What is the relation between money-flows and fund s future performance (returns and the likelihood of persistence in its future returns)? Furthermore, in case of hedge funds, managerial incentives in the form of option-like incentive-fee contract play an important role in motivating the managers to perform better. Therefore, we also examine the relation between managerial incentives (proxied by the delta of the option-like incentive-fee contract) and future performance. Finally, we examine funds of hedge funds (henceforth FOFs), a rapidly growing segment of the hedge fund industry. These intermediaries invest in other hedge funds. In order to search, select, monitor, and track the performance of individual hedge funds, they charge a fixed management fee and a performance-based incentive fee. Thus, an investor in FOFs encounters an additional layer of agency. However, FOFs offer certain benefits including diversification, due diligence, professional management, and potential access to star managers of hedge funds that may be closed to new investors. This leads us to our third and final research question. What are the determinants of money-flows and performance in FOFs and whether they are different from those in case of individual hedge funds? We conduct our investigation by merging three leading hedge fund databases: Hedge Fund Research (HFR), TASS, and Zurich Capital Markets/Managed Accounts Reports 4
(ZCM/MAR). Our database contains return histories for 1776 live and 1655 dead individual hedge funds and 352 live and 187 dead FOFs. Using the merged database, we make four important contributions to the hedge fund literature. First, we conduct a comprehensive study of the determinants of money-flows into the hedge fund industry. Goetzmann, Ingersoll, and Ross (2003) (henceforth GIR) is the only paper to study this issue by conducting a univariate analysis of money-flows and past returns. We believe that in addition to past returns, investors consider various other factors such as fund size, age, volatility, and more importantly, managerial incentives in the form of degree of moneyness of the manager s option-like incentive-fee contract. For example, an investor may be reluctant to invest his money in a small, young, and highly volatile fund, which is significantly below its high-water mark. 1 In addition to these factors, investors may also be paying attention to consistency in performance of a fund. Arguably, investors prefer to place their money in funds with persistently above-median returns. Therefore, we also examine the relation between flows and persistence in past performance, a previously unstudied phenomenon. Second, we make an important contribution to the extant literature by explicitly measuring the managerial incentives through the delta of the manager s option-like incentive-fee contract. Performance-based compensation is an important tool to motivate the manager to perform well. 2 However, the incentive fee as a percentage of profits, per se, does not take into account how far the fund is relative to its high-water mark. For example, Ackermann, McEnally, and Ravenscraft (1999, pp. 860) acknowledge the limitations of using only incentive fees as a metric for managerial incentives as follows: The problem is that the relationship between high- 1 Fund size and age may proxy for managerial skill and hence, investors confidence in a particular fund. 2 Prior literature on hedge funds (Ackermann, McEnally, and Ravenscraft, 1999; Brown, Goetzmann, and Ibbotson, 1999; Liang, 1999; Edwards and Caglayan, 2001) has examined the relation between incentive fees and performance 5
water marks, incentive fees, and volatility is complicated. The relationship should depend on where the fund is relative to its high-water mark. This is further complicated by the fact that new investors may have different high-water marks than original investors. We specifically address this issue and determine the managerial incentives by measuring the delta of the manager s incentive-fee option. For this purpose, we compute a fund s yearly money-flows since the start of a fund s return history. We then estimate fund s delta after applying the hurdle rate and highwater mark provisions to yearly money-flows from both new and original investors. We measure delta as the dollar increase in the incentive-fee-based compensation of the manager for an increase of one percent in the fund s return. Clearly, delta depends on the degree of moneyness of the options granted to the manager by annual inflows. The larger is the value of delta, the greater is the incentive to deliver superior returns. This is the first attempt in the hedge fund literature to empirically quantify the level of incentives, in dollar terms, offered by the performance-based-compensation contracts. Third, we examine the relation between current flows and future performance and address the issue of decreasing returns to scale in the hedge fund industry. This is an important issue that underpins Berk and Green s (2002) model. In addition, as argued above, highpowered incentives should motivate the manager to perform better. Towards that end, we also examine the relation between manager s option-delta and future performance, an issue hitherto unexplored in the hedge fund literature. Finally, we make an important contribution to the literature by analyzing the determinants of money-flows and performance for FOFs. Due to the double-fee arrangement, Brown, Goetzmann, and Liang (2002) argue that FOFs face different incentives compared to and, in general, has found a positive relation between the two. See Elton, Gruber and Blake (2003) for the performance of mutual funds that charge incentive fees. 6
individual hedge funds. Therefore, we also examine how money-flows and performance relate to managerial incentives. We have six main findings. First, we find money-flows chase returns and the performance-flow relation is convex. This finding is consistent with that of Chevalier and Ellison (1997), Goetzmann and Peles (1997), and Sirri and Tufano (1998) in the mutual fund industry. Our finding, however, differs from that of GIR (2003), who find that the best performers experience outflows. They argue that this may be due to good performers being reluctant to accepting new money. It is important to note that presence of such managers in our sample biases against finding evidence of money chasing good performance. We believe that the differences in regulatory and market environments during the two sample periods may be responsible for the differences in GIR s and our results. For example, the National Securities Markets Improvement Act of 1996 removed the restriction on the maximum of 99 qualified investors. Further, during our sample period, data on individual hedge funds and indexes started becoming widely available, which made it easier for investors to compare the performance a hedge fund with its peer group. To assess the impact of these changes, we re-run GIR s univariate specification using only offshore funds from our composite database during 1989-1995. We find that instead of outflows, the top performers experience flows that are indistinguishable from zero. This suggests that the nature of the hedge fund industry has indeed changed over time, and in recent years it has started resembling the mutual fund industry. Our second finding pertains to the relation between flows and persistence in performance. Following Brown, Goetzmann, and Ibbotson s (1999) approach, we label a fund as a winner (loser) if its return in a given year is above (below) median return of peer-group funds. We classify a fund as persistent winner (loser) if it is a winner (loser) for two successive years. We 7
find that money-flows are significantly higher (lower) for funds that are persistent winners (losers). Our third finding concerns the relation between money-flows and managerial incentives. We find a positive relation between flows and delta, which confirms that investors prefer to place their money in funds where the manager has greater incentives to perform better. Since delta incorporates information about the entire history of returns and money-flows, this finding suggests that investors take into account historical returns as well as money-flows. Our fourth finding pertains to the relation between current money-flows and future performance. We find that flows into larger funds are associated with lower returns in the future. Further, they are also associated with lower probability of those funds showing good persistence in the future. This finding is consistent with the presence of decreasing returns to scale in the hedge fund industry. It is also consistent with the findings of Chen et al (2002) for the mutual fund industry. Fifth, we find future performance to be positively related to managerial incentives. This suggests that investors prefer funds where the manager has higher incentives to perform well in the future, where good performance is reflected in higher returns and higher probability of exhibiting good persistence. Our final result relates to FOFs. When we analyze this intermediated sector of the hedge fund industry, we find that FOFs with good performance (higher returns or good persistence) attract greater money-flows. While this finding is similar to that for individual hedge funds, there are two important differences. First, we find no significant relation between managerial incentives, flows, and future performance. This suggests that investors preferences and funds performance in response to managerial incentives are different for FOFs. Second, we find that larger flows into bigger FOFs are positively related to future returns and to the probability of 8
showing good persistence. This suggests that FOFs may be enjoying economies of scale. It is also consistent with the notion that FOFs with substantial assets under management may have greater bargaining power and enjoy better access to well-performing managers that may otherwise be closed for new money. The rest of the paper is organized as follows. Section I describes the data. Sections II and III examine the relation between new money-flows, managerial incentives, and past performance. Section IV and V investigate the relation between money-flows, managerial incentives, and future performance. Section VI offers concluding remarks with suggestions for future research. I. Data A. Data Description In this paper, we construct a comprehensive hedge fund database that is a union of three large databases namely HFR, TASS, and ZCM/MAR. 3 This enables us to resolve occasional discrepancies among different databases as well as create a sample that is more representative of the entire hedge fund industry. We consider both individual hedge funds as well as FOFs for the purpose of our study. Our sample period extends from January 1994 to December 2000. We focus on this period for three reasons. First, the number of funds prior to our sample period is relatively few. Second, the databases do not extensively cover dead funds before 1994. 4 Finally, publicly disseminated data on hedge fund indexes became available only since 1994, 3 In the past, researchers have used one or more of the three major hedge fund databases. For example, Fung and Hsieh (1997) use TASS database, Ackermann, McEnally, and Ravenscraft (1999) use HFR and ZCM/MAR databases, Agarwal and Naik (2000, 2003) and Liang (2000) uses HFR and TASS databases, while Brown, Goetzmann, and Park (2001) use data from Offshore Funds Directory and TASS. 4 It is important to note that the word dead is misleading and missing-in-action may be a more appropriate term as they include funds that are liquidated, merged/restructured, and funds that stopped reporting returns to the database vendors but may have continued operations. However, in order to be consistent with previous research, we 9
which enabled investors to assess the relative performance of a fund more easily. Therefore, we conduct our analysis using 1994-2000 data. Table I (Panels A and B) provides the breakdown of funds from different data vendors. After merging the three databases, we find that there are 1776 (352) live hedge funds (FOFs) and 1655 (187) dead hedge funds (FOFs). 5 Figure 1 reports the overlap between the three databases via Venn diagrams. The Venn diagrams in Figure 1 show that there is an overlap of 30% (41%) in the number of hedge funds (FOFs) across the three databases, with HFR having the largest coverage (54%). These Venn diagrams demonstrate that there are a large number of hedge funds that are unique to different databases and thus, merging them helps in capturing a more representative sample of the hedge fund industry. Even though hedge funds market themselves as absolute performers, investors arguably evaluate the performance of a hedge fund relative to its peers. Unfortunately, there is no universally acceptable way of classifying hedge funds into different styles. Academic research (Fung and Hsieh, 1997; Brown and Goetzmann, 2003) shows that there are five to eight distinct style factors in hedge fund returns. Following these insights, we classify the reported hedge fund strategies into four broad categories: Directional, Relative Value, Security Selection, and Multi- Process Traders. Appendix A describes the mapping between the data vendors classification and our classification. Table I Panel C reports the distribution of hedge funds across the four broad strategies. Table II reports the means and medians of monthly net-of-fee returns, volatility of returns, age, management and incentive fees for individual hedge funds and FOFs. We find that, in continue to call them dead funds. 5 We exclude managed futures, natural resources, mutual funds, and other hedge funds since these categories are not usually considered as typical hedge funds. We also exclude long-only, Regulation D and funds with missing strategy information. 10
general, FOFs have significantly lower returns and volatility, higher age and management fee but lower incentive fee compared to individual hedge funds. Lower returns of FOFs can be attributed, in part, to lesser biases compared to hedge funds (Fung and Hsieh, 2000) while the lower volatility of FOFs can be explained by the diversified nature of their portfolios. B. Computation of Money-Flows We first compute dollar flows for fund i during month m as follows ( 1 ) Dollar Flow = AUM AUM + R (1) im, im, im, 1 im, where, AUM im, and AUM im, 1are the size for fund i at the end of month m and month m-1 and Rim, is the return for fund i during month m. 6 We aggregate the monthly flows during the year t to estimate annual flows (Annual Dollar Flow i,t ). As in Chevalier and Ellison (1997) and Sirri and Tufano (1998), we scale annual dollar flows by beginning-of-year assets under management to capture the change in size due to net money-flows. Flow i,t = AnnualDollarFlow it, (2) AUM it, 1 In Table III, we report the trend in assets under management (AUM) and money-flows in the hedge fund industry. As can be seen from Table III, the total assets under management for hedge funds have grown five-fold from $40 billion to $201 billion from December 1993 to December 2000. Over the same period, the total assets under management for FOFs have also grown from $9 billion to $25 billion. We observe that, in general, more than 25% of funds experience 6 This formula assumes that the fund flows occur at the end of the month. For the sake of robustness, we also compute money-flows assuming that they occur at the beginning of the month and find very similar results. When AUM data is not available at a monthly frequency, we compute flows for coarser intervals. 11
outflows. Further, the mean flows are systematically higher than the median flows suggesting that some funds experienced significant growth in the assets under their management. C. Computation of Managerial Incentives One of the distinguishing differences between mutual funds and hedge funds is the manager s compensation contract. Unlike most mutual funds, hedge funds charge incentive fees as a fixed percentage of the profits earned. Typically, incentive-fee contracts include hurdle rate and high-water mark provisions. Hurdle rate (usually a rate of return on cash or LIBOR) is the minimum return that the manager needs to achieve before claiming any incentive fees. Highwater mark requires the manager to make up for the previous losses before he can earn incentive fees. Hence, incentive-fee contracts are essentially a call-option on the assets under management with the exercise price depending on the hurdle rate and high-water mark provisions. An incentive fee contract amounts to the investor having written a fraction of a call option on the assets under management. For example, if incentive fee equals 20 percent, then it is equivalent to the investor having written 0.2 of a call option on the money invested. When money gets invested in a fund at different points in time, each investment is associated with its own high-water mark. In addition, when a hurdle rate is specified, the manager must exceed it before he can claim an incentive fee. Therefore, in general, incentive-fee contracts endow the manager with a portfolio of call options. The value of the each of the call options depends on the time when the money came in, its high-water mark and whether the return exceeds the hurdle rate or not. 12
The option-like compensation provides the fund manager with incentives to deliver superior returns. 7 We proxy these incentives by the delta of the portfolio of options, which equals the dollar change in the incentive fee for a one percent change in the fund s return. The greater the delta, larger is the incentive to deliver superior returns. We describe the procedure of computing delta in Appendix B. By definition, delta depends on the degree of moneyness of the portfolio of options generated by money-flows over time, which in turn, depends on whether a fund has had a negative return in a year or in previous years. Clearly, if the fund has had negative returns during a year, the fund will be below its high-water mark. However, a fund with a positive return in a given year could still be below its high-water mark if its cumulative negative returns in prior years exceed the positive return in the current year. We report the summary statistics of funds with negative returns and those below high-water mark in Panel B of Table III. In case of individual hedge funds, we find that the percentage of funds with negative returns has increased from 8.4% to 33.1% from December 1993 to December 2000. The mean (median) losses also increased from 11.2% (7.9%) to 19.9% (15.0%) over this period. Consistent with our discussion above, a larger proportion of funds are below their high-water mark (50.2% and 51.3% in December 1993 and December 2000 respectively). The shortfall below high-water mark for these funds will however be smaller than the magnitude of loss in that year because it includes funds that have positive returns during that year and are below high-water mark due to poor return history. As expected, the mean (median) shortfall below high-water mark varies from 2.9% (0.3%) in December 1993 to 8.3% (1.2%) in December 2000. The results for FOFs are qualitatively similar to those for individual funds. 7 It is important to note that option-like payoffs also influence risk-taking incentives of hedge fund managers. Agarwal, Daniel, and Naik (2003) examine these and find that fund volatility is positively related to both implicit 13
Finally, we report the summary statistics of delta in Panel C of Table III. 8 Over the sample period, we find that mean (median) delta across all hedge funds has increased from $130,000 ($30,000) in December 1993 to $240,000 ($50,000) in December 2000. In contrast, across all FOFs, the mean (median) delta has decreased from $170,000 ($20,000) to $80,000 ($10,000) over the same period. 9 II. Determinants of Money-Flows In this section, we investigate the determinants of money-flows into and out of individual hedge funds and FOFs. It may be that investors follow a top-down approach where they first choose the broad strategies in which to invest, and then decide in which funds to invest. Therefore, as a first step, we explore the performance-flow relation at a strategy level. For this purpose, we sum up annual dollar flows across all funds within a strategy and obtain aggregate strategy-level flows. We plot in Figure 2 the annual dollar flows against prior-year s weighted average returns of the funds following different strategies. We use both equally-weighted and value-weighted (weighted by AUM) returns at the strategy level. Figure 2 suggests that, in general, money-flows chase recent performance for all four strategies as well as for all the funds considered together. We investigate this finding further at an individual fund level using our composite database. When investors infer a manager s ability through past returns, one would expect to find that funds that have performed especially well in the recent past would attract larger money-flows. and explicit risk-taking incentives. 8 We measure delta in millions of dollars. However, it is simple to convert our dollar delta into the standard Black and Scholes (1973) call option delta by dividing our dollar delta by (0.01*incentive fee*investors assets). 9 Like hedge fund managers, top executives of corporations receive option-like payoffs, which create similar managerial incentives. It is interesting to compare the level of managerial incentives in these two industries. Coles, Daniel, and Naveen (2003) report the mean (median) delta of executive stock options for the top 1500 firms in S&P 14
For poorly performing funds, one may not find significant outflows due to impediments unique to the hedge fund industry such as lockup period, notice period, and redemption period. There are two additional factors that may also affect the performance-flow relation. First, investors may place less weight on the recent performance as some of the hedge fund managers could show a good previous track record. Second, some of the well-performing hedge funds may be closed to new investment. For these funds, we would find no money inflows despite their good performance. This will lower the likelihood of finding that money-flows chase good performance. Our empirical analysis captures the net effect of all these factors. To examine the performance-flow relation, we need to categorize performance as good or bad. While it would be useful to estimate risk-adjusted performance, this is a perilous task given the non-normality in hedge funds returns and option-like dynamic trading strategies adopted by hedge funds (Fung and Hsieh, 1997, 2001; Goetzmann et al, 2002; Agarwal and Naik, 2003). Therefore, we use returns relative to one s peer group as a measure of performance. In order to compare our findings for hedge funds, with those of Sirri and Tufano (1998) for mutual funds, we follow an identical empirical procedure. We classify each fund-year observation into one of five quintiles based on performance. We find that the average inflow into funds in the top (first) quintile is 63% compared to an average outflow of 3% for funds in the bottom (fifth) quintile. We find a similar performance-flow relation for FOFs as well. In particular, the top quintile performers attract inflow of 25% compared to outflow of 15% for the bottom quintile performers. 10 during 1992-2000 to be $584,000 ($196,000). See Murphy (1999) and Core, Guay, and Larcker (2003) for a survey of literature on executive compensation. 10 For hedge funds, the t-test shows that the difference in the flows between all the adjacent quintiles (e.g., top and second, second and third, and so on) is significantly different (p-value=0.000). For FOFs, the t-test shows that only the difference in the flows between top and second, and fourth and bottom quintile is significantly different. 15
A. Performance-flow relation for individual Hedge Funds Arguably, investors consider factors in addition to performance, such as size, volatility, age, managerial incentives, and past flows before making their investment decisions. Therefore, we examine the relation between flows and performance using a multivariate regression that controls for other potential determinants of money-flows. We specify the following regression: j ( ) ( ) ( ) ( ) 5 j it, = β0 + β1 it, 1 + β2 it, 1 + β3 it, 1+ β4 σit, 1 + β5 it, 1 j= 1 Flow Qrank Delta Size I AgeTop 3 s 6 I( AgeBottomit, 1) 7 ( MFeeit, ) 8 ( Flowit, 1) 9 ( Rit, ) 10 IStrategy ( is, ) it, s= 1 + β + β + β + β + β + ε where, Flowit, and Flowit, 1are the money-flows in fund i in years t and t-1, Qrank j it, 1is the fractional rank of fund i in quintile j for year t-1, Deltait, 1is the natural logarithm of delta of the managers incentive-fee-contract for fund i as of end of year t-1, 11 Sizeit, 1is the size of the fund measured as the natural logarithm of the AUM for fund i at time t-1, σ it, 1is the standard deviation of the monthly returns of fund i during year t-1, ( it, 1) (3) I AgeTop is an indicator variable that takes the value 1 if the fund age is in the top one-third of funds at the end of year t-1, ( it, 1) I AgeBottom is an indicator variable that takes the value 1 if the fund age is in the bottom one-third of funds at the end of year t-1, MFeeit, is the management fees charged by fund i in year t, it, R is the return of fund i in year t, ( is, ) 1 if fund i belongs to strategy s, and εit, is the error term. I Strategy are strategy dummies that take the value 11 We measure delta in millions of dollars and take its natural logarithm to mitigate the outlier effect. 16
We construct the fractional rank quintiles, Qrank j it, 1 as per Sirri and Tufano (1998). First, each fund is given a fractional rank, Frank, from 0 through 1 based on returns in year t-1. This fractional rank represents its percentile performance relative to other funds in the same period. We estimate the coefficients on fractional ranks using piecewise linear regression framework over five quintiles. Towards that end, we define 1 Qrank, the bottom quintile rank, to equal Min (0.2, Frank), 2 1 Qrank = Min (0.2, Frank- Qrank ) and so forth up to the highest performance 5 quintile, Qrank, i.e., the top quintile. The slope coefficients on these piecewise decompositions of fractional ranks represent the slope of the performance-flow relation over their range of sensitivity. Sirri and Tufano (1998) argue that the use of Fama and MacBeth (1973) procedure produces more conservative estimates of the significance levels of their coefficient estimates compared to those from pooled regression. Hence, we follow Fama and MacBeth (1973) procedure. For robustness, we repeat our analysis using pooled regressions and obtain similar results. We report the results of regression based on Fama-MacBeth (1973) procedure in Table IV. 12 The results for individual hedge funds in Model 1 show that only the coefficients on the top and the third quintile are significantly positive at the 1% level. This suggests that the wellperforming funds attract significantly higher flows compared to the poorly performing ones. The coefficient on the top quintile (coeff=1.757) implies that if a fund moves 10 percentile in the highest performance group, it will attract incremental inflows of 18%. The magnitude of the top quintile coefficient is similar to that documented by Sirri and Tufano (1998) for mutual funds. Also, the R-square of our model (about 13%) is of the same order of magnitude as Sirri and 12 To minimize the influence of outliers, we winsorize top 1% of the annual returns, standard deviation, assets under management, and net flows. 17
Tufano (1998). Our R-square is considerably higher than R-square (about 4%) obtained from univariate analysis in GIR (2003), confirming the importance of controlling for fund-specific factors in addition to fund returns. To examine if there is convexity in flow-performance relation, we conduct a Chow test on a pairwise basis on adjacent performance quintiles. In case of individual hedge funds, we find that only the third and fourth quintile coefficients are significantly different from each other (pvalue=0.08). This suggests that in case of individual hedge funds, the kink in the performanceflow relation occurs at the 40 th percentile as compared to the 80 th percentile kink documented by Sirri and Tufano (1998) in case of mutual funds. 13 Chevalier and Ellison (1997) document that the money-flows of older funds are less sensitive to recent returns. Thus, we examine if the convexity of performance-flow relation varies across age for individual hedge funds. For this purpose, we interact the terms for the five performance quintiles and those for the two age dummies and re-estimate the regression in equation (3) after including these interaction terms (not reported for the sake of brevity). Chow test results indicate kinks at the 40 th, 60 th, and 80 th percentile for younger funds, and none for the middle-age and older funds. This suggests that similar to mutual fund industry, flows into the younger funds appear to be more sensitive to recent performance. We investigate the possibility that investors might consider coarser performance groups by combining the middle three quintiles into one group as in Sirri and Tufano (1998). 14 The results in Model 2 of Table IV show that the coefficient of the top quintile (coeff=1.474) and the middle three quintiles pooled together (coeff=0.743) are significantly positive at the 5% and 1% level 13 Interestingly, Kaplan and Schoar (2003) find a concave performance-flow relationship in private equity funds. 14 Here, the middle quintile rank is given by Min (0.6, Frank Qrank 1 ). 18
respectively, while the bottom quintile coefficient is indistinguishable from zero. These results are once again consistent with those from the five-quintile specification in Model 1. A.1 Relation between Flows and Managerial Incentives As discussed earlier, performance-related fee provides strong incentives to the manager to perform better. Managers whose funds are near or at their high-water marks face better incentives compared to those substantially below their high-water marks. We capture these incentives through delta. It is important to note that delta incorporates information about the entire history of returns and money-flows. Greater the delta, greater are the managerial incentives. Recognizing this effect, investors may prefer to invest in funds with greater delta. Therefore, we expect to find a positive relation between flows and delta. The results for individual hedge funds (in Models 1 and 2) show that the slope coefficient for delta is positive and significant (coeff.=0.011). The impact of delta is economically significant. An increase in the delta from 25 th percentile ($10,000) to 75 th percentile ($131,000) is associated with an increase in annual flows by 2.8% compared to the median flow of 11.7%. These findings suggest that investors recognize the importance of managerial incentives and accordingly invest in funds where managers have better incentives, namely, funds with higher delta. 15 A.2 Relation between Flows and other factors Next, we examine the relation between money-flows and other fund characteristics such as size and age. From Table IV, we observe that the slope coefficient on fund size is significantly 15 For robustness, we examine if there is non-linear relation between flows and delta by including square of delta in equation (3) and find the slope coefficient to be positive but not significant. 19
negative in both specifications, implying that smaller funds get higher flows. We also find that the coefficient of younger funds dummy is significantly positive, implying that younger funds experience higher flows than the middle-aged funds. Prior literature on mutual funds (Gruber, 1996) has used lagged flows in addition to the fund characteristics discussed above. Lagged flow is an important determinant of flows into hedge funds for a number of reasons. First, lagged flows can proxy for non-performance variables such as reputation and marketing efforts that can influence investors decisions. Second, investors may want to stick to their prior decision and continue to invest in the same fund, behavior deemed as status-quo bias. Finally, uninformed investors may herd with others and select funds that have recently experienced larger flows. Therefore, we include lagged flow in equation (3) and find that it is positively related to flows. Previous researchers (Chevalier and Ellison, 1997) have also studied the response of money-flows in mutual funds to contemporaneous returns. Hence, we also include returns R t at time t in equation (3) and find that flows are positively related to contemporaneous returns. One may argue that lagged flows and contemporaneous returns may not be in the information set of investors. Hence, for robustness, we repeat our analysis by excluding these variables from equation (3) and find that our results remain unchanged. Recent research in mutual funds suggest that there are significant spillover effects, i.e., star or well-performing funds within a fund family can lead to an increase in the flows to other funds in that family (Khorana and Servaes, 2000; Massa, 2000; Ivkovich, 2001, Nanda, Wang, and Zheng, 2002). We examine if there are similar spillover effects in the hedge fund industry. Unlike mutual funds, multiple-fund families are rare in hedge funds. The average number of funds per family across our sample period is 1.3 (minimum of 1 fund and a maximum of 11 funds per family) with 80% of the funds not belonging to a family. In order to examine spillover 20
effects, we include performance of other funds in a family as an additional explanatory variable in equation (3). In unreported results, we find the slope coefficient on this variable to be positive but not significant. A.3 Comparison of results with Goetzmann, Ingersoll and Ross (2003) Although our findings are consistent with those of Chevalier and Ellison (1997) and Sirri and Tufano (1998) for mutual funds, they differ from those of GIR (2003). We believe that the differences in the regulatory or market environments during the periods covered in the two studies are responsible for the differences in the findings. For example, during our sample period, the restriction on the maximum number of qualified investors was relaxed. Further, data on individual hedge funds as well as a range of hedge fund indexes started becoming widely available. These changes made it easier for investors to obtain information about hedge funds and to compare their performance with peer group. To examine if the differences in the results are indeed attributable to the changing nature of the industry, we repeat our analysis, as in GIR, using only offshore funds from our sample during 1989-1995. Instead of GIR s finding of outflows for top performers, we find that the top performers experience flows that are indistinguishable from zero (see Appendix C). 16 This suggests that the nature of the hedge fund industry has changed over time, and in recent years it has started resembling the mutual fund industry. B. Robustness checks 16 Please note that the two samples will still not be identical as GIR use Offshore Funds directory while we use offshore funds included in the our composite database created by merging HFR, TASS, and ZCM/MAR databases. 21
For our performance-flow regression in equation (3), we need data on annual flows. However, if a fund disappears from our database during a year, it will not have annual flows in the year of its disappearance. Unlike mutual funds, where liquidation due to poor performance is the only reason for fund s disappearance, hedge funds may disappear for other reasons as well. These reasons include delisting by the data vendor, non-reporting by fund, change in fund names, and mergers (Fung and Hsieh, 2000). Goetzmann and Peles (1997) assign a flow of -100 percent in the last year for those funds that disappear due to poor performance. Following their insights, we examine the robustness of our results by assigning a flow of -100 percent in the last year for funds that disappear due to liquidation. We continue to find no relation between flows and returns for poorly performing funds. This confirms that our results are not sensitive to exclusion of last year s flows for liquidated funds. This is consistent with the findings of Sirri and Tufano (1998) for mutual funds. Next, we investigate whether the investors consider longer-term performance for making investment decisions. For this purpose, we analyze the fund flows based on the fund s two-year performance at time t-1. In particular, we form five fractional rank variables based on recent biannual performance. For the sake of brevity, we do not report the results here. We find that none of the performance quintiles come out significant. However, delta continues to be positively related to flows. 17 At first sight, lack of significant relation between bi-annual performance quintiles and flows may suggest that investors care more about recent performance. However, it is important to note that delta incorporates the fund s performance history and investors response (in terms of money-flows) to this history. Hence, our finding of positive and significant 17 In addition, size, lagged flows, and contemporaneous returns exhibit significant relation with flows as in Models 1 and 2 in Table IV. 22
relation between flows and delta suggests that investors do take into account the long-term performance of the funds. C. Performance-flow relation for Funds of Hedge Funds Since FOFs are an intermediated sector of hedge fund industry with an additional layer of fees and agency issues, in this subsection, we separately investigate the performance-flow relation for FOFs. For this purpose, we re-estimate regression in equation (3) for FOFs and report our results in Table IV. Results for Model 1 show that the slope coefficient for top quintile is positive and significant at the 5% level (coeff.=2.028). This implies that if a FOF moves 10 percentile in the highest performance group, it will attract incremental inflows of 20%. This increase is of similar order of magnitude to that observed in case of individual hedge funds (18%). Further, we investigate the possibility of investors considering coarser performance groups by combining the middle three quintiles into one group. The results in Model 2 of Table IV show that only the coefficient of the top quintile (coeff=1.850) is significantly positive at the 5% level. These results suggest that the well-performing FOFs attract significantly higher flows than the poorly performing ones in this intermediated sector of the hedge fund industry as well. We conduct Chow test on the pairwise differences in the slope coefficients for adjacent performance rank quintiles. Our results indicate that there are no kink points based on Model 1. However, when we pool the middle three quintiles (Model 2), we obtain a kink at the 80 th percentile. Thus, our results using coarser performance groups suggest that there seems to be some evidence of convex performance-flow relation for FOFs. 23
Similar to individual hedge funds, we find the slope coefficient for delta and lagged flows to be positive, but not significant. Furthermore, like in case of hedge funds, we find flows are negatively related to size and positively related to contemporaneous returns. Overall, the findings in this section confirm that money-flows chase recent performance both for individual hedge funds and FOFs. Arguably, in addition to recent performance, investors consider consistently good performance before investing their money into hedge funds and FOFs, an issue we examine in the next section. III. Relation between flows and persistence in prior performance A stronger signal of manager s ability is persistence in performance and not simply one year of superior performance. Investors therefore should direct more flows to managers who show persistence. Therefore, in this section, we examine the relation between money-flows and persistence in prior performance. To capture persistence, we follow Brown, Goetzmann, and Ibbotson (1999) and Agarwal and Naik (2000) and define a fund to be a winner (loser) in year t, if its returns are greater (less) than the return of the median fund in its peer group in that year. The indicator variable I(Persistent Winner i,t-1 ) equals 1 if fund i is a winner in years t-1 and t-2, and equals 0 otherwise. Similarly, the indicator variable I(Persistent Loser i,t-1 ) equals 1 if fund i is a loser in years, t-1 and t-2, and equals 0 otherwise. We investigate the persistence-flow relation by estimating the following regression: ( ) ( ) ( ) ( ) ( ) ( ) ( ) Flow = γ + γ I PersistentWinner + γ I PersistentLoser + γ Delta it, 0 1 it, 1 2 it, 1 3 it, 1 + γ Size + γ σ + γ I AgeTop + γ I AgeBottom + γ MFee 4 it, 1 5 it, 1 6 it, 1 7 it, 1 8 it, 3 s 9 ( Flowit, 1) 10 ( Rit, ) 11 IStrategy ( is, ) eit, s= 1 + γ + γ + γ + (4) 24
We report the results from regression in equation (4) in Table V. Since, we have both Persistent Winner and Persistent Loser indicator variables in the regression, the excluded category of funds is the one that shows reversals, i.e., those who win in one year and lose in the other. We find that the coefficient of Persistent Winner is significantly positive (coeff=0.205) for individual hedge funds implying that the flow is 20.5% higher for funds which exhibit persistently good returns compared to those that exhibit reversals. Similarly, the coefficient of Persistent Loser is significantly negative (coeff=-0.137) for individual hedge funds. This implies that consistently poorly performing funds experience 13.7% lower flows compared to the funds exhibiting reversals. As before, slope coefficient on delta is significantly positive (coeff.=0.012) suggesting that funds with better managerial incentives attract higher flows. An increase in delta from 25 th to 75 th percentile is associated with 3.1% increase in annual flows. Finally, the slope coefficients of control variables are similar to those in Table IV. In particular, we find that larger funds and higher-volatility funds experience lower flows. We repeat this analysis with FOFs and also report our results in Table V. Like hedge funds, we find that FOFs that are persistent winners experience significantly higher flows compared to those that exhibit reversals. However, unlike hedge funds, we do not find that persistent losers experience lower flows. We also do not find a significant relation with delta. Overall, these findings suggest that investors direct more flows towards persistent winners and less towards persistent losers. Having seen that the flows are positively related to past performance and persistence in past performance, we next examine the relation between flows and future performance as well as persistence in future performance. IV. Relation between flows, incentives, and future performance 25
New money-flows into a fund may hinder its future performance due to decreasing returns to scale. 18 In this section, we shed light on this issue by examining the relation between money-flows into a fund in a given year and its future performance. A. Flows, incentives, and future returns OLS specification We examine the flow-future performance relation by regressing a fund s annual returns on prior year s flows, managerial incentives, size, and other control variables. In particular, we estimate the following regression: ( ) ( Deltait ) ( it ) I( AgeTopit ) I( AgeBottomit ) R = λ + λ Flow + λ Size + λ Size Flow it, 0 1 it, 1 2 it, 1 3 it, 1 it, 1 + λ + λ σ + λ + λ 4, 1 5, 1 6, 1 7, 1 3 s 8 ( MFeeit, ) 9 IStrategy ( is, ) it, s= 1 + λ + λ + ξ (5) If there are decreasing returns to scale in the hedge fund industry, one would expect this effect to be especially prominent for larger funds experiencing higher inflows. In order to capture this, we include interaction of size and flows Sizeit, 1 Flowit, 1 in our regression. As before, we also include controls for age, volatility, and management fee. Since higher delta implies higher incentives to perform better, we expect to see a positive relation between delta and future performance. Table VI reports our findings from the regression in equation (5). The OLS results (Model 1) in Table VI (Panel A) show that the coefficient on size is significantly negative suggesting that larger funds are associated with lower returns in the following year. Further, we find coefficient of interaction term between size and flow to be significantly negative. As this is an interaction between two variables, we examine the effect of the two individually as well as jointly. An increase in the size from 25 th to 75 th percentile is 18 Chen et al (2002) find that fund size erodes performance in the mutual fund industry. They attribute it to reasons 26