Share Restrictions and Investor Flows in the Hedge Fund Industry*

Transcription

1 Share Restrictions and Investor Flows in the Hedge Fund Industry* Bill Ding, Mila Getmansky, Bing Liang, and Russ Wermers This Draft: November 29 Abstract This paper studies the effect of share restrictions on the flow-performance relation of individual hedge funds. As such, we reconcile previous research that shows conflicting results for this relation without explicitly considering restrictions. Specifically, we find that hedge funds exhibit a convex flowperformance relation in the absence of share restrictions (similar to mutual funds), but exhibit a concave relation in the presence of restrictions our evidence is consistent with both a direct effect of the binding restrictions and an indirect effect that is due to investors endogenizing expected future binding restrictions when investing their money. Further, we find that live funds exhibit a concave flow-performance relation due to stricter flow restrictions than defunct funds, which display a convex relation. Finally, we find that money is smart, that is, fund flows predict future hedge fund performance; however, this smart money effect is eliminated among funds with greater share restrictions. Key words: hedge fund flows, share restrictions, asset illiquidity, live/defunct funds, smart money effect JEL classification: G23, G11 *We thank Nick Bollen, Ben Branch, Stephen Brown, Kalok Chan, William Goetzmann, Wei Jiang, Nikunj Kapadia, Hossein Kazemi, Jose Martinez, Lubos Pastor, Thomas Schneeweis, Hany Shawky, Christian Tiu, Heather Tookes, Xuemin (Sterling) Yan, Xiaoyan Zhang, and seminar participants at the 29 Oxford-Man Institute Hedge Fund Conference, the 28 China International Conference in Finance, the 28 Asian FA/NFA meetings, the 27 Western Finance Association meetings, the 27 European Finance Association meetings, the 26 European Financial Management Association meetings, Babson College, University of Venice, SUNY University at Buffalo, European Central Bank (ECB-CFS conference), Sveriges Riksbank conference, CISDM Hedge Fund Conference, and the University of Massachusetts, Amherst for helpful comments and suggestions. Financial support from The Institute for Quantitative Research in Finance (the Q-Group) is gratefully acknowledged. We thank Chris Schwarz for excellent research assistance. All remaining errors are our own. Bill Ding is at Department of Finance, School of Business, State University of New York at Albany, 14 Washington Avenue, Albany, NY 12222, , bding@uamail.albany.edu; Mila Getmansky is at Isenberg School of Management, University of Massachusetts, 121 Presidents Drive, Amherst, MA 13, , msherman@som.umass.edu; Bing Liang is at Isenberg School of Management, University of Massachusetts, 121 Presidents Drive, Amherst, MA 13, , bliang@som.umass.edu; Russ Wermers is at Robert H. Smith School of Business, University of Maryland at College Park, College Park, MD , , wermers@umd.edu. Electronic copy available at:

2 Over $1.4 trillion is now managed by hedge funds, with substantial additional sums flowing into this industry each year. 1 For example, total inflows during the second quarter of 29 were $1 billion. However, investors often withdraw money from hedge funds following (or in anticipation of) performance losses; during the third quarter of 28, a record $31 billion was withdrawn. In addition, substantial sums are moved between different hedge funds by investors each year. Consequently, managers of individual hedge funds are often faced with large inflows or outflows over the short-term, necessitating a great deal of trading activity which, in turn, has the potential to destabilize markets. For instance, hedge funds have been cited as one likely source of the market turmoil of 28, with total investor redemptions reaching nearly $4 billion. 2 As a first step in understanding the impact of hedge funds on markets, it is important to study the behavior of investor flows to individual hedge funds, and how these flows react to changing fund performance. To date, much of the research on hedge funds has focused on manager skills and risk-return tradeoffs; less attention has been directed toward the money flows of investors and how these flows might impact hedge fund trading. In the mutual fund industry, Sirri and Tufano (1998) and Chevalier and Ellison (1997) study the behavior of money flows and document a convex flow-performance relation. Further research documents deviations from this relation under certain conditions. Specifically, Chen, Goldstein, and Jiang (29) study bank runs in mutual fund flows due to the presence of asset illiquidity, which can explain a concave relation between mutual fund flows and past returns, as opposed to the usual convex relation. Huang, Wei, and Yan (27) study the impact of participation costs, such as the cost of searching for superior funds, on the sensitivity of fund flows to past performance. They find that mutual funds with higher participation costs have lower flow sensitivity in the medium performance region and higher flow sensitivity in the high performance region, compared to their 1 This estimate is reported by Hedge Fund Research for the second quarter of Hedge Fund Liquidation, New York Law Journal, March 2, Electronic copy available at:

3 lower participation-cost peers. Berk and Tonks (27) study fund flows of the worst performing mutual funds and find that the fund flow-return sensitivity in this region depends on investors' (heterogeneous) willingness to move their capital. Although the behavior of mutual fund flows is well-documented, hedge funds have proven much more elusive. Beyond the difficulty in obtaining complete data that is free of reporting biases, hedge funds exhibit many complex and unique features that impact flows, relative to mutual funds. For example, common hedge fund characteristics include statutory restrictions on the number of investors, high minimum investments, lock-up periods, forced redemptions, closure to new investments, capacity constraints, asset illiquidity, required delay periods for subscriptions and redemptions, and the ability to return capital to investors at the discretion of the funds. The impact of such restrictions on the behavior of flows, while likely important, has largely been ignored by past hedge fund research. Simply put, it is not clear how the presence of share restrictions might affect the reaction of investors to the perceived management quality of hedge funds, as represented by their past performance. Accordingly, this paper examines the structure of restrictions of individual hedge funds, and the influence of these restrictions on the flow-performance relation of the funds. While some previous studies have analyzed the flow-performance relation without specifically considering these flow restrictions, they have found conflicting results. For example, Agarwal, Daniel, and Naik (24) find a convex flow-performance relation for individual hedge funds, which is similar to that documented for individual mutual funds. In contrast, Goetzmann, Ingersoll, and Ross (23) find a concave flow-performance relation, while Baquero and Verbeek (29) find a linear relation. These conflicting results leave a muddled view of the true response of investors to hedge fund performance. We argue that the existence of restrictions on fund flows explains the disparity in the results of these studies. We replicate and reconcile these previous results by finding that different 2 Electronic copy available at:

4 levels of flow restrictions among the samples and time-periods used by these papers explain differences in the fund flow-performance relationship. Further, we are the first to document that hedge fund investors anticipate these restrictions, which further modifies the flow-performance relation. To be specific, we consider how lockup periods, redemption periods, advance notice periods, subscription periods, capacity restrictions, closure to new investment, certain onshore fund statutory restrictions (such as a limit on the allowable number of investors), and implicit restrictions driven by illiquidity in the assets held by hedge funds impact the flow-performance relation in various portions of the performance spectrum (i.e., low, middle, and high past-performing funds). Moreover, we consider the effect on the flow-performance relation of investor anticipation (endogenizing) of future binding restrictions when they make their investment decisions. Our empirical investigation of the impact of restrictions on the flow-performance relation of hedge funds brings several new insights. First, in the absence of significant restrictions, we find that the flow-performance relation for individual hedge funds is convex, similar to that observed for mutual funds. Second, in the presence of significant restrictions, we find a concave flowperformance relation, due both to the direct binding effect of the restrictions and to the indirect effect of investors endogenizing these restrictions. Specifically, we find that the indirect effect dominates in the low-performance region, due to investor aversion to becoming locked into an underperforming or liquidating fund and losing the option to redeem shares while we find that the direct effect dominates in the high-performance region, where investors are less averse to being locked out of a outperforming fund, since other high-performing funds may be available, making the option to stay invested less valuable. In addition, fund managers are much more likely to close wellperforming funds to new investors when the fund reaches its designed capacity, making it much 3

5 more difficult for investors to strategically place money into them. 3 Combined, these results give a concave flow-performance curve in the presence of share restrictions. Surprisingly, we find that more volatile funds actually impose fewer restrictions, even though this puts them more at-risk for failure. We interpret this finding as evidence that investors understand the costs that restrictions impose on them (consistent with the option argument, since restrictions take options away), and that they know these costs are higher when returns are more volatile. More volatile funds tend to be smaller and newer, which are funds that are competing more vigorously for inflows to attain an economic scale, and they tend to impose fewer flow restrictions. Next, we separately measure the flow-performance relation for funds in the Live and Defunct databases. Live funds exhibit a concave flow-performance relation, while defunct funds exhibit a convex relation. We show that the differences in these flow-performance relations are largely due to the Live database being populated with funds that have more restrictions than hedge funds in the Defunct database. 4 When we combine the live and defunct funds, we observe a linear flow-performance relation due to the combination of both highly restricted and less restricted funds. This is consistent with a recent study by Baquero and Verbeek (29), who find a linear flow-performance relation for all hedge funds. Therefore, our results demonstrate that it is important to consider the impact of share restrictions on the flow-performance relation and to distinguish live from defunct funds. For 3 For example John Meriwether at Long Term Capital Management unexpectedly returned about $2.7 billion of the fund s capital back to investors because investment opportunities were not large and attractive enough (The Washington Post, September 27, 1998). 4 In a signaling equilibrium, high quality managers will impose stricter share restrictions, while low quality managers cannot afford to mimic them. If low-quality managers set up strict outflow restrictions such as a long lock-up period, they will lock investors up in the first round, but as investors learn about the true low quality of the fund, they will tend to disinvest in the second round. As a result, we would observe a separation of restrictions based on fund quality, where low quality funds allow investors to keep their options which puts these funds more at risk to outflows. As a result, high quality funds which tend to impose stronger restrictions (i.e., funds in the Live database) more often survive, while lower quality funds which tend to impose fewer restrictions (i.e., funds in the Defunct database) more often tend to become defunct. 4

6 example, in predicting hedge fund failure, one should incorporate share restriction variables in the hazard rate model. Finally, in addition to the above relation between current fund flows and past performance, we examine the relation between current flows and future performance. That is, we examine whether investor flows are smart in predicting future hedge fund performance, and whether restrictions impede any smart money effect. In related papers, Zheng (1999) and Gruber (1996) find evidence that flows into mutual funds earn positive risk-adjusted returns. Among hedge funds, Agarwal, Daniel, and Naik (24) find that annual returns are negatively related to flows in the prior year, while Baquero and Verbeek (29) find that the flows can deliver outperformance, but for only one quarter. By conditioning on flow restrictions, we find stronger evidence that investor flows predict performance, i.e., that money is smart. In particular, funds that experience net inflows outperform those that experience net outflows in low-restriction funds. Specifically, a zero-cost, equally-weighted portfolio that is long funds with positive prior-quarter flows and short funds with negative prior-quarter flows generates a statistically significant alpha of 2.8% per month for funds with fewer restrictions. However, this smart-money alpha is insignificant among hedge funds with greater share restrictions. Thus, we show that conditioning on share restrictions is also important for studying the smart-money effect among hedge funds. Overall, our results indicate that the flow-performance relation and the smart-money effect for hedge funds are quite different from that of mutual funds due to the many restrictive features of hedge fund markets. Flow restrictions lead to a concave flow-performance relation, which contrasts strongly with the convex relation found in the mutual fund literature, and is consistent with the ability of investors to anticipate and endogenize restrictions in their investment choices. Money is smart only when the absence of restrictions allow it to flow. As such, our paper reconciles prior 5

7 findings on the flow-performance relation among hedge funds, and provides new insights on the behavior of capital among individual hedge funds. We believe that our findings at the individual hedge fund level help to provide a foundation for studying the behavior of aggregate capital among different hedge fund styles in order to examine such macro issues as the role of hedge funds in financial market contagion. Our paper is organized as follows. Section I develops hypotheses to be tested in the data. Section II describes the data. Methodology is developed in Section III. We report empirical results in Section IV, and conclude in Section V. I. Hypotheses Past research (e.g., Agarwal, Daniel, and Naik (24), Goetzmann, Ingersoll and Ross (23), and Baquero and Verbeek (29)) documents a strong, positive correlation between investor flows and the past performance of hedge funds, indicating that investors infer fund manager talent at least partly from prior performance. While a positive relation also exists in the mutual fund industry (e.g., Sirri and Tufano (1998)), hedge funds may be different because of the numerous restrictions on flows that are imposed. 5 We would expect these hedge fund flow restrictions to potentially change the flow-performance relation relative to mutual funds, which have fewer flow restrictions. That is, the presence of share restrictions and asset illiquidity may severely limit the ability of investors to capitalize on the information about hedge fund quality that is represented by prior performance. For instance, most hedge funds implement lockup provisions, subscription, redemption, and advance notice periods that delay or otherwise limit the responsiveness of fund flows to performance. Further restrictions are imposed by U.S.-based (onshore) funds, which are allowed to 5 Mutual funds also impose some restrictions, such as loads, short-term trading fees, or fund closures. However, the average mutual fund is much less restrictive than the average hedge fund. 6

8 cater only to accredited investors having a minimum of $1 million in net worth. 6 Moreover, onshore funds are not allowed to advertise to the general public, and differences in legal structures caused by tax provisions and the limit of 499 investors for each onshore fund also lead to more severe share restrictions, relative to mutual funds. 7 All of these restrictions may affect the flowperformance relation of hedge funds. Hedge fund managers argue that they impose restrictions on flows for at least a couple reasons. First, hedge funds engage in strategies that (1) may involve significant losses before they produce profits, and (2) invest in relatively illiquid and complex securities over long time horizons. In either case, outflow restrictions help to prevent a forced premature liquidation of fund assets. In addition to the above-mentioned standard restrictions, managers may post discretionary withdraw restrictions such as a gate on the amount of money requested for redemption or a suspension on withdrawals. 8 Second, hedge fund strategies are fundamentally different from the long-only portfolio strategies implemented by mutual funds (Goetzmann, Ingersoll and Ross (23), Liang (1999), and Fung and Hsieh (1997)). For instance, hedge funds often engage in arbitrage opportunities. By their very nature, arbitrage opportunities are not infinitely exploitable even the most successful hedge funds eventually reach an optimal level of assets and close to new investments. Flows may also be affected in a secondary way from restrictions that are imposed by the funds investors may endogenize investment restrictions when they react to past performance. For instance, an investor may be reluctant to invest in a fund with few liquid assets and volatile returns, since that investor may be restricted from withdrawing money at a later date when the fund is at 6 See 7 Onshore funds are usually organized as limited partnerships, with a limit of 1 investors for 3(c)(1) funds, and 499 for 3(c)(7) funds, while offshore funds are usually organized as open-ended investment companies with no restriction on the number of investors. These onshore partnerships are usually more illiquid than open-ended mutual funds, and have higher share restrictions (Liang and Park, 27). 8 See, for example, Blue Mountain Freezes $3.1 Billion Credit Hedge Fund (Bloomberg, November 3, 28). 7

9 risk of liquidating at fire-sale prices. In essence, the ability to quickly withdraw or invest money is a real option for investors and (as with financial options) is more valuable when the underlying assets are more volatile. Investors who understand costs associated with restrictions can anticipate the potential future binding effect of share restrictions by decreasing their inflows to poorly performing funds and decreasing their outflows from well-performing funds to avoid a potential adverse outcome because their options are restricted. 9 This endogenizing of restrictions may further impact the flow-performance relation, relative to mutual funds. In fact, there are two recent papers that model redemptions as either long options for investors or short options for managers. Ang and Bollen (29) model investor decisions about withdrawals as real options, and show that a two-year lockup with a three-month notice period can cost approximately 1% of the initial investment for risk-averse investors. Dai and Sundaresan (29) study the relations among hedge funds, prime brokers, and investors by defining redemption options from a manager s point of view as short options with investors, and funding options as short options with prime brokers. 1 In general, flows to different hedge funds may exhibit quite dissimilar sensitivities to past performance, due to the large array of restrictions that inhibit flows. The literature on hedge fund flows finds widely differing evidence on the flow-performance relation. 11 While these papers attribute their conflicting results to the use of different hedge fund databases and time periods, it is quite possible that restrictions on flows may be responsible for the differing results. To illustrate, consider funds that impose outflow restrictions, such as redemption or lockup periods. The direct effect of such restrictions would be to flatten the flow-performance relation, since investors would 9 For poorly performing illiquid funds, investors would like to strategically react to outflow restrictions by augmenting their withdrawals, similar to Chen, Goldstein, and Jiang (29). However, they are precluded from doing this due to share restrictions. Similarly, investors would like to strategically overinvest in a fund with inflow restrictions, but are precluded from doing so. 1 These papers only consider options to redeem, and not options to invest. 11 For example, Agarwal, Daniel, and Naik (24) find a convex flow-performance relation for individual hedge funds, which is similar to that documented for individual mutual funds by Sirri and Tufano (1998) and Chevalier and Ellison (1997). By contrast, Goetzmann, Ingersoll and Ross (23) find a concave relationship in the flow-performance relation, while Baquero and Verbeek (29) find a linear fund flow-performance relation. 8

10 be prevented from freely selling their fund shares in poorly performing funds. Now, suppose that investors endogenize these outflow restrictions when deciding on whether to buy shares in a fund. 12 In this case, investors may refrain from buying shares of a fund that has especially poor performance, since the disinvest option, the option to exit such a fund (in the event of continued poor performance) is especially valuable. 13 Compared to a poorly performing fund with no outflow restrictions, in which investors may invest a small amount (in case the poor performance was simply due to bad luck), a poorly performing fund with outflow restrictions will have even lower inflows thus, steepening the flow-performance relation. In the high-performance region, the direct effect of restrictions (i.e., subscription period, capacity constraint, decision to close, and onshore provisions), will serve to reduce the flowperformance relation, similar to the direct effect of outflow restrictions among poorly performing funds. Investors are also able to endogenize the inflow restrictions. If investors face binding restrictions on inflows, they will delay selling shares, as the remain invested option, the option to stay in the fund is valuable (in the event that manager ability is good). This option is lost once they redeem. 14 Thus, if investors endogenize potential binding restrictions by their reluctance to disinvest, the flow-performance sensitivity will be increased. 15 We develop several hypotheses about the effect of restrictions on the flow-performance relation to guide our empirical tests. To be specific, we consider the following hypotheses about the 12 Different investors face differing menus of binding restrictions on inflows and outflows, depending on the fund and the investor. For instance, some investors may be beyond the lockup period, while others may not, making the lockup period an outflow restriction only for the latter type who must be concerned with the actions of the former. 13 In this low-performance region the disinvest option is more valuable than the remain invested option, since manager ability is likely to be low. 14 In this high-performance region, the remain invested option is more valuable than the disinvest option, since manager ability is likely to be high. 15 An interesting example of the high value of this option occurred in mutual fund markets. The Schroeder Ultra Fund, the top-performing fund of the late 199 s, closed to new investors during 1998, but remained open to flows from existing shareholders. A market developed for shares, with new investors paying large amounts for a single share from other investors, in order to gain access to the inflow option. 9

11 relation between flow and performance due to restrictions, noting that they are not mutually exclusive in the sense that the two hypotheses may work together. Hypothesis 1A (Direct Effect of Share Restrictions): Significant share restrictions lead to a decrease in outflows from poorly performing funds, and a decrease in inflows to wellperforming funds due to the binding nature of these restrictions. This results in a flatter flow-performance relation in both the low- and high-performance regions of the curve. Hypothesis 1B (Indirect Effect of Share Restrictions): Treating the decisions to invest or disinvest as real options and knowing the value of these options, investors anticipate the potential future binding effect of share restrictions by decreasing their inflows into poorly performing funds and decreasing their outflows from well-performing funds because disinvest and remain invested options are valuable. Endogenizing the presence of restrictions results in a steeper flow-performance relation in the low- and high-performance regions of the curve. In Figure 1, we illustrate the opposing effects of Hypotheses 1A and 1B. The net effect of restrictions on the flow-performance relation, therefore, will depend on the relative importance of the direct and indirect effects. In the low-performance region, we would expect a very strong endogenous effect of restrictions due to the high option value of disinvesting. The disinvest option value increases dramatically as performance worsens. Investors will take this into account and thus will endogenize share restrictions. 16 < Insert Figure 1 > On the other hand, in the high performance region, the direct effect may be stronger and the endogenous effect might be weaker. The direct effect may be stronger in this region due to unexpected additional restrictions not stipulated in investment contracts and which are thus difficult to anticipate unexpected fund closures (to new investment) and the return of capital to investors 16 Moreover, if a hedge fund manager decides to liquidate a fund, this unexpected effect augments the sensitivity of fund flows to past performance in the low-performance region. 1

12 by managers. 17,18 These direct restrictions have no counterparts in low-performing funds. The indirect (endogenous) effect may also be weaker in the high-performance region because the main option value lies in the remain invested option (since managers in this region are likely to be skilled). This option value does not increase dramatically as performance improves; since being locked out of a fund merely means that the investor will need to take an alternative choice on another fund, perhaps with a similar performance record. Since the value of the remain invested option is relatively low, investors tend not to highly endogenize share restrictions in the highperformance region. In conclusion, investors wish to avoid being locked into a bad fund much more than being locked out of a good fund, making the endogenous effect much stronger in the low-performance region than in the high-performance region. In other words, the option value of withdrawing from a poorly-performing fund is much higher than the option value of remaining invested in a well-performing fund, especially for a risk-averse investor. Based on the above arguments, we conjecture the following about the flow-performance relation: Prediction 1 (Concave Flow-Performance Relation in the Presence of Share Restrictions): In the low-performance region, the indirect effect prevails, since the option value to disinvest increases dramatically as performance worsens. In the high-performance region, the option value to remain invested does not increase dramatically as performance improves, since investors have other choices of managers. In addition, investors find it difficult to endogenize the unexpected actions of hedge fund managers to close the fund to new money or return money, leading to a dominant direct effect. This leads to a concave flowperformance relationship in the presence of restrictions. 17 Investors might also endogenize the possibility of fund closures or capital returns. However, these events are much more difficult to forecast, since they are not clearly outlined in investment contracts. Thus, investors are less likely to be successful in anticipating the binding effect of such inflow restrictions. 18 The only restriction which is predetermined in investment contracts and is relevant in the high-performance region is a subscription period. However, we anticipate and show in the paper that the effect of endogenizing this restriction is clearly dominated by other inflow restrictions. 11

13 In the absence of flow restrictions, we would expect a flow-performance relation similar to that of mutual funds: Prediction 2 (Convex Flow-Performance Relation in the Absence of Share Restrictions): In the absence of share restrictions, the best performing funds command disproportionate flows, leading to a convex flow-performance relation, similar to mutual funds with daily liquidity. Our next hypothesis explores potential differences in the fund flow-performance relation between the Live and Defunct databases. Funds in the Live database are those that are available for investment; those in the Defunct database are mostly not (due either to fund liquidation or selfwithdrawal from the Live database because they already obtained sufficient funds from investors). Consistent with prior literature, we argue that the main reason for a fund to be in the Defunct database is poor performance (see Fung and Hsieh (2), Liang (2)). We also conjecture that there exists a separate signaling equilibrium for high-quality and low-quality managers: the former impose stricter share restrictions to curb redemptions and harvest the illiquidity premium, while the latter cannot afford to mimic the former. 19 In addition, high quality managers may set the correct level of share restriction parameters, such as more frequent subscription and less frequent redemption periods to effectively attract inflows and deter outflows. As a result, these restrictions distinguish high-quality from low-quality managers, ex ante, hence, live from defunct funds, ex-post. We conjecture that the flow-performance relation for live funds is concave because of stricter share restrictions as well as better performing funds closing to new investment (or returning capital to investors) due to capacity constraints. Specifically, better performing funds in the Live 19 Usually, these restriction variables are set at the inception of the funds and they hardly change over time. 12

14 database might decide to close after facing diminishing returns to scale (Berk and Green, 24) and an expected decrease in incentive fees due to capacity constraints. 2 Conversely, the flow-performance relation for defunct funds is convex, since these funds are less restricted. In addition, exceptionally poorly performing funds go through liquidation and may not have much money left for investors to redeem. Meanwhile, exceptionally well-performing funds experience extreme inflows and often withdraw from the data vendor (since these funds no longer need the publicity, and may prefer to reduce their public disclosure). 21 Hypothesis 2 (Live vs. Defunct Funds): Live funds exhibit a concave fund flowperformance relation due to stricter share restrictions and better performing fund closures to investment. Conversely, defunct funds exhibit a convex fund flow-performance relation due to fewer share restrictions, and to the presence of extremely well-performing funds that attract substantial new investments, and poorly-performing funds that are liquidated. Our final hypothesis analyzes the effect of inflow and outflow restrictions faced by hedge fund investors on the smart money effect, that is, the ability of hedge fund investors to move money into future winners and out of future losers. If money is smart, meaning that current fund flows can predict future fund returns, then the direct effect of restrictions on both inflows and outflows is to reduce the ability of flows to predict following-period hedge fund returns, since investors cannot fully respond to their superior information. For example, outflow restrictions such as lockup periods or redemption periods may delay the ability of investors to act when they gather information that indicates that fund performance will worsen. Hypothesis 3 (Effect of Share Restrictions on Smart Money): Restrictions on hedge funds will result in a reduced smart money effect. Specifically, restrictions on investing in and 2 The relative break-down between management and incentive fees depends on the age of the fund and total assets under management. Christoffersen and Rouah (27) argue that when CTAs are small and young, they take higher risk and heavily rely on incentive fees, while they take less risk and rely on management fees when they are large and old. 21 Funds in the Defunct database are not all liquidated. Liang and Park (28) develop a rule for indentifying truly liquidated funds by examining the past 6-month returns and past 12-month assets of funds in the defunct database. 13

15 withdrawing from hedge funds will make inflows and outflows more sticky which leads to reduced performance of the flows. II. Data A. TASS Database We use the TASS hedge fund database for our empirical analysis. 22 As of the third quarter of 25, this database tracks about $8 billion held by global single-manager hedge funds, excluding funds of funds, side pockets, and managed accounts. 23 There are other hedge fund databases, such as Morningstar Altvest, CISDM/MAR, and Hedge Fund Research (HFR). However, academic studies (e.g., Liang (2)) indicate that the TASS database is probably the most comprehensive database covering hedge funds. The TASS database consists of monthly returns, assets under management, and other fundspecific information. TASS also classifies hedge funds into 11 strategies: Convertible Arbitrage, Dedicated Short Bias, Emerging Markets, Equity Market Neutral, Event Driven, Fixed Income Arbitrage, Global Macro, Long/Short Equity Hedge, Managed Futures, Multi-Strategy, and Fund of Funds. The database is divided into two parts: Live and Defunct funds (hedge funds that are in the Live database are active as of September 25). Once a hedge fund is liquidated, restructured, merged with other hedge funds, or decides to stop reporting, the fund return history is transferred into the Defunct database from the Live database. The Defunct database became available in 1994, thus, funds that were dropped from the Live database before 1994 are not included in TASS. Therefore, survivorship bias is not a 22 The TASS database is distributed by Lipper, Inc. For further information about TASS, see 23 Not all hedge funds report to TASS. Also, managed accounts are not in TASS which hold a significant fraction of total hedge fund assets. Side-pockets are not included either, which usually track hard-to-value or illiquid assets. 14

16 concern after However, the database is subject to backfilling bias; specifically, when a fund decides to be included in the database, TASS adds the fund to the Live database and backfills all available prior performance data for the fund. 25 As of September 25, the combined database of both live and defunct hedge funds contained 6,97 funds having at least one monthly return observation 3,821 funds in the Live database and 2,276 in the Defunct database. The majority of the 6,97 funds report returns, net of management and incentive fees, on a monthly basis. 26 TASS converts all foreign-currency denominated returns to US-dollar returns using the appropriate exchange rates. TASS also reports assets in local currency and reports returns using local currency NAVs. We eliminated 54 funds that reported only gross-of-fee returns, leaving 6,43 funds (3,797 in the Live and 2,246 in the Defunct database). As mentioned above, the Defunct database starts in 1994; therefore, to avoid survivorship bias, we start our sample in As a result, we are left with 6,17 funds (3,792 in the Live and 2,225 in the Defunct databases). Furthermore, we omitted return observations associated with an incubation stage or a liquidation/lack of reporting stage, where a fund did not have any assets under management reported in the beginning or the end of the fund time series, respectively. Specifically, the beginning and/or the end of the fund time series were deleted for these funds. If a fund had one month of missing assets under management, we linearly interpolated the missing observation using adjacent assets under management. If assets under management observations were reported as being 24 For studies attempting to quantify the degree and impact of survivorship bias, see Brown, Goetzmann, and Park (21), Liang (2), Fung and Hsieh (2), Brown, Goetzmann, and Ibbotson (1999), Carpenter and Lynch (1999), Fung and Hsieh (1997), and Brown, Goetzmann, Ibbotson, and Ross (1992). 25 Also, due to reporting delays, some Defunct funds can be incorrectly listed in the Live database. TASS adopted a policy of transferring a fund from the Live to the Defunct database if its managers have not heard from the hedge fund or were not able to contact the hedge fund manager over a 3 month period. 26 TASS defines returns as the change in net asset value during the month (assuming the reinvestment of any distributions on the reinvestment date used by the fund) divided by the net asset value at the beginning of the month, net of management fees, incentive fees, and other fund expenses. Therefore, these reported returns should approximate the returns realized by investors. 15

17 constant for more than 4 months, we eliminated those observations. We also eliminated hedge funds having fewer than 12 months of observations. After these filters, it is feasible that a fund had two discontinuous time-series intervals with reported assets under management. In this case, we would report the largest continuous interval. Finally, we have 4,594 funds left after all filtering (75% of the initial fund sample). Out of this sample, 3,555 are individual hedge funds and 1,39 are funds of funds. B. Hedge Fund Share Restrictions The following three sets of share restrictions are considered in the analysis: 1. Outflow restrictions: lockup period, redemption period, advance notice period, and total redemption period. Lockup period is the initial amount of time investors are required to keep their money in the fund before they can redeem shares. Investors cannot access their money during this time period. Once the lockup period is over, an investor can withdraw money during the next redemption period. Redemption period is the time period an investor in a hedge fund must wait before withdrawing money. Advance notice period is the time period of advance notice that investors are required to give to hedge fund managers in advance of the redemption period. Total redemption period is defined as the sum of redemption and advance notice periods. 2. Inflow restrictions: subscription period, capacity constrained, and open to investment. Subscription period is a time delay between investing in a fund and actually purchasing fund shares. Funds are capacity constrained if they belong to emerging markets, fixed income arbitrage, event driven, and convertible arbitrage strategies following Getmansky (25). A fund manager can also decide whether a fund is open or closed to new investment. 3. Domicile: A fund is onshore domiciled if funds report United States as their domicile country. Onshore domicile can have implications for both inflows and outflows. Liang and Park 16

18 (27) show that onshore funds have both longer subscription and redemption periods than their offshore counterparts. All these share restrictions are clearly specified in an investment contract between a hedge fund and an investor. We also find that they are fairly constant across the life-cycle of a fund. III. Methodology A. Measuring the Flow-Performance Relation The fund flow-performance function is estimated using a piecewise linear relationship between current fund flows and past returns. We apply a modification of the methodology used by Sirri and Tufano (1998) in their study of the mutual fund flow-performance relation. First, a fractional rank, Frank i,t, is calculated for each fund, from to 1, based on returns during year t in each hedge fund category. Then, Trank 1 i,t, the bottom tercile rank, Trank 2 i,t, the middle tercile rank, and Trank 3 i,t, the top tercile rank are calculated as follows: Trank 1 i,t=min(1/3, Frank i,t) Trank 2 i,t=min(1/3, Frank i,t - Trank 1 i,t) (1) Trank 3 i,t=min(1/3, Frank i,t - Trank 1 i,t- Trank 2 i,t) We measure flows into each fund i as the percentage change of net assets of the fund (in local currency) between the beginning and the end of year t, net of investment returns, assuming flows are invested at the end of the period: Assets Assets (1 r i, t i, t 1 i. t Flow i, t. (2) Assetsi, t 1 ) The top one percent of flows are winsorized to prevent outliers from affecting our analysis. The following regression is specified to understand the determinants of fund flows in the presence of restrictions: 17

19 Flow Trank Trank Trank SD Assets it, i 1 it, 1 2 it, 1 3 it, 1 4 it, 1 5 it, 1 Live OpentoPublic HighWaterMark Leverage 6 i 7 i 8 i 9 i ManagementFee IncentiveFee LockupPeriod 1 i 11 i 12 i 13TotalRedemptionPeriodi 14SubscriptionPeriodi 15 StyleEffecti, t i, t (3) where SD i,t-1 is the standard deviation of monthly returns during year t-1; Assets i,t-1 is the natural logarithm of hedge fund dollar assets at the end of year t-1; Live i equals 1 if a fund is in the Live database, and if it is in the Defunct database; OpentoPublic i equals 1 if a fund is open to public, and otherwise; HighWaterMark i equals 1 if a high water market provision is present, and otherwise; Leverage i equals 1 if a fund uses leverage, and otherwise; ManagementFee i is the management fee (measured as a percentage of assets under management); IncentiveFee i is the incentive fee (measured as a percentage of a fund s upside above a specific threshold); LockupPeriod i is the lockup period (measured in months); TotalRedemptionPeriod i is the total redemption period, which is the sum of redemption and advance notice periods (measured in days); SubscriptionPeriod i is the subscription period (measured in days), and StyleEffect i,t is the average flow for the category to which fund i belongs during year t. To further study the impact of share restrictions on the flow-performance relation in Equation (3), we interact restriction parameters (subscription and redemption periods, advance notice period, lockup, onshore vs. offshore, asset illiquidity, and capacity restrictions) with performance tercile ranks (low, middle, and high). 27 In analyzing the flow-performance relation, we estimate (each year) a piecewise linear regression between current fund flows and past returns, as shown in Equation (3). Standard errors are adjusted for autocorrelation and heteroskedasticity using Newey-West (1987). 27 The restriction dummy variables are defined as zero for low restriction (parameter value less than or equal to the median) and one for high restriction (parameter value above the median). 18

20 B. Measuring Asset Illiquidity and Return Smoothing To quantify the impact of asset illiquidity and smoothing on hedge fund returns, we follow Getmansky, Lo, and Makarov (24) by asserting that a fund s true economic return in month t is given by R t, which represents the sum total of all the relevant information that would determine the equilibrium value of the fund s securities in a frictionless market. The authors assume that true economic returns are not observed. Instead, t, and let: Rt is a reported and observed monthly return in period R t [,1], j,1,2 j 1 R 1 t R 2 1 t 1 R 2 t 2 (4) which is a weighted average of the fund s true monthly returns, R t, over the most recent 3 months, including the current month., 1 and 2 are estimated using a maximum likelihood procedure. is an asset illiquidity and smoothing measure. If for a specific hedge fund is close to 1, then most of the real contemporaneous return is currently reflected in the observed data, thus, the hedge fund exhibits more liquidity and a lack of smoothing. However, a smaller signifies that a hedge fund is illiquid and is more likely to exhibit smoothing. Following Getmansky, Lo, and Makarov (24), we impose a 5-year filter in order to obtain reliable, 1, and 2 estimates. 28 C. Measuring the Performance of Flows C.1. Grinblatt-Titman Measure First adopted in Grinblatt and Titman (1993), the GT measure is a performance measure that does not require the knowledge of the benchmark for the evaluated investment portfolio. This 28 The filtering procedure can be done for 95 funds in our sample. 19

21 measure is especially appropriate for hedge funds, since they tend to follow absolute versus relative benchmark strategies, implement dynamic trading strategies, and invest in various asset classes with difficult to define benchmarks. In applying the GT measure for the performance of hedge fund flows, we assume that investors as a whole rebalance their hedge fund portfolios on a quarterly basis, where the portfolio weight for a fund is determined by its assets: 29 N t 1 i 1 GT ( w w ) R, (5) where w i,t and w i,t-1 are the weights of fund i measured by assets (relative to total hedge fund assets) at the end of quarters t and t-1, respectively; R i,t+1 is the raw return of hedge fund i for quarter t+1; and N represents the number of hedge funds during quarter t. If hedge fund investors are smart in allocating their capital across hedge funds, we expect to see a significantly positive average GT measure over the sample period. i, t i, t 1 i, t 1 C.2. Index Model to Measure the Performance of Flows As an alternative approach, we use an index model to measure the performance of investor flows. Considering the complexity in strategies that might be pursued by hedge funds, we use a multi-index model. 3 Included as independent variables are the Russell 3 index return, difference between the Russell 1 index return and the Russell 2 index return (LMS), difference between the Russell 1 value index return and the Russell 1 growth index return (VMG), the momentum factor downloaded from Ken French s web site (UMD), the Lehman Aggregate Bond index return, yield spread between Moody s BAA- and AAA-rated bonds, yield spread between the 1-year Treasury note and the 6-month LIBOR, return on the S&P5 at-the-money call option, A similar approach is used by Zheng (1999) for measuring the performance of mutual fund flows. 3 Similar multi-index models are used in Fung and Hsieh (24), Agarwal and Naik (24), and Liang (1999). 31 We thank Vikas Agarwal and Narayan Naik for kindly providing the option-based factor data, an earlier version of which was used in Agarwal and Naik (24). 2

22 the MSCI emerging market stock index return, the MSCI emerging market bond index return, the 6- month LIBOR, the Federal Reserve dollar index return, the gold index return, oil price change, and change in the volatility index (VIX). The intercept of the model measures the abnormal performance of a fund after all potential (known) risk factors are accounted for. The beta coefficients of the independent variables capture the exposure of the hedge fund returns to the market indexes. IV. Results A. Descriptive Statistics Table I provides an overview of the returns for each of the 11 styles from 1994 through 24. The Long/Short equity hedge style represents 4% (the largest number) of hedge funds in the database. Panels A, B, and C display statistics for all, live, and defunct funds, respectively. In each panel, statistics are provided for the equally-weighted portfolio of hedge funds within each category, as well as for an asset-weighted portfolio. 32 For example, Panel A shows that different categories of funds exhibit quite different returns: the long/short equity category earns the highest mean return of 1.6% per month, in contrast to -.16% for the dedicated short bias category. Not surprisingly, emerging markets, dedicated short bias, and managed futures have higher average standard deviations than other funds, consistent with the fact that these funds engage in riskier trading strategies through derivatives, short selling, and emerging market investments. Jarque-Bera (JB) tests also reject that emerging markets, event driven, and fixed income arbitrage fund returns are normally distributed. < Insert Table I > 32 AWR is the asset-weighted return using the last available assets under management for a hedge fund as an asset weight. 21

23 Several hedge fund categories display very large higher moments. For example, fixed income arbitrage funds display a negative skewness of -.88 and an excess kurtosis of 7.14, consistent with these hedge funds implementing strategies that are more speculative and highly levered, and, therefore, capturing non-linearities in prices around such events. The average first-order serial correlation coefficient, which is a proxy for illiquidity, is 1%. Consistent with Getmansky, Lo, and Makarov (24), hedge fund categories that hold and trade illiquid assets have the highest first order serial correlation coefficients: (convertible arbitrage (32%), emerging markets (16%), event driven (19%), and fixed income arbitrage (15%)). Multi-strategy and fund of funds also have high levels of autocorrelation (14% and 23%, respectively). Further insights are apparent when examining live and defunct fund returns in Panels B and C. Specifically, an equally-weighted portfolio of live funds has a much higher mean return (1.1%) than that of defunct funds (.53%). This is consistent with Liang (2), who shows that poor performance is the main reason for hedge fund attrition from the live dataset. Live funds are also less risky (with a mean standard deviation of 3.51%) than defunct funds (5.42%). Consistent with Brown, Goetzmann, Ibbotson, and Ross (1992), defunct funds take riskier bets, and, hence, are more likely to disappear because of the relation between extreme poor performance and fund attrition. The average autocorrelation coefficient is twice as high for live funds (13%) compared to defunct funds (7%). This is consistent with the finding that funds earn an excess return due to illiquidity exposure (Aragon, 27) and that these funds are more likely to survive. 33 < Insert Table II > Table II reports summary statistics for equally-weighted hedge fund quarterly flows for each category from 1994 through 24. Quarterly flows are reported for All, Live, and Defunct databases. Investor flows into each fund are defined as the percentage change of net assets of the 33 This is consistent with our finding to be shown shortly, that live funds have more share restrictions and invest in more illiquid assets. 22