How Smart are the Smart Guys? A Unique View from Hedge Fund Stock Holdings

Transcription

1 How Smart are the Smart Guys? A Unique View from Hedge Fund Stock Holdings BY JOHN M. GRIFFIN AND JIN XU * April 3, 2006 Preliminary * John Griffin is an Associate Professor at the University of Texas at Austin and can be reached at John.griffin@mccombs.utexas.edu or Ph# (512) Jin Xu is a Doctoral Candidate at the University of Texas at Austin. We wish to thank Zhiwu Chen, David Hsieh (the FMA discussant), Roger Ibbotson, Owen Lamont, Matt Spiegel, Will Goetzmann, Antti Petajisto, Geert Rouwenhorst, Laura Starks, Sheridan Titman, Christian Tiu, and brownbag participants at the Financial Management Meetings in Chicago, University of Lisboa in Portugal, University of Texas at Austin, and the Yale School of Management.

2 How Smart are the Smart Guys? A Unique View from Hedge Fund Stock Holdings Abstract: We provide what we believe to be the first comprehensive examination of hedge fund stock long-equity positions and the performance of these stock picks. Compared to mutual funds, hedge funds act as if they provide value. They have high turnover, hold stocks that take them further away from the market portfolio, and also prefer smaller opaque securities. In aggregate, hedge funds do not exhibit the strong momentum tilt documented for mutual funds but actually are overweight past loser stocks. Despite their active nature, we cannot reject the null that hedge funds in general are any better at long stock picking or timing sectors than mutual funds. However, hedge fund firms seem to have more differential ability in stock picking than mutual funds with the best hedge fund outperforming in stock picking. Overall, our study questions the ability of long-equity hedge funds to add value.

3 I. Introduction An underlying assumption by many on Wall Street is that the best and brightest managers migrate to the hedge fund industry. This assumption is clearly a contributor to the 20 fold increase in hedge fund assets under management since While hedge funds typically are known for investing in a variety of positions, roughly 1/3 of the industry simply invests in equities. This paper examines the long equity positions of hedge fund managers and asks whether these managers exhibit talent in stock picking and sector timing. We also examine differential ability within the hedge fund industry. In the process, we shed some light on how hedge funds seek to obtain profits in terms of the types of securities they hold and their trading intensity. Since hedge funds cater to sophisticated investors, they are afforded an extra level of opacity by the SEC and are not required to report their holdings semi-annually in SEC N- 30D filings like mutual funds. Hence, the evaluation of hedge funds has largely been left to measuring the effect of factor exposures to total returns. With the exception of an analysis of the gaming behavior of forty Australian mangers by Brown, Gallagher, Steenbeek, and Swan (2004), this lack of analysis of holdings data has led to almost no understanding of what type of stocks hedge funds typically hold. Examinations of hedge fund returns can be beneficial in that the total return for a fund is reported but the approach is also sensitive to the benchmarking factors and approach. This, along with different samples, has led to widely varying opinions regarding the performance and timing ability of hedge funds [Fung and Hsieh (1997), Ackermann, McEnally, and Ravenscraft (1999), Brown, Goetzmann, and Ibbotson (1999), Agarwal and Naik (2000), and Amin and Kat (2003)]. This paper backs out the stock holdings of 306 hedge fund firms from 1980 to present through a unique and labor intensive process. It should be noted that hedge fund 1

4 firms often contain the activity of several hedge funds. Our approach yields quarterly hedge fund activity in stocks during March 1980-September 2004 if we use our full data (Hedge A), or March 1986-September 2004 if we adopt a more conservative approach, where we only examine funds the year after they are available from one of our hedge fund data sources (Hedge B). This approach will shed light on a variety of different facets of hedge fund performance that have largely been ignored. To our knowledge no paper has used a holdings-based approach to examine large-scale issues of hedge fund holdings and performance. Brunnermeier and Nagel (2004) examine quarterly 13F holdings of hedge funds as we do, but their focus is on whether 53 hedge fund firms increased or decreased their positions and made money in high price-to-sales (P/S) stocks during the dot.com bubble from Our first question concerns issues of how active hedge funds are in their trading and their aggressiveness relative to the market. We find that although there is broad heterogeneity in hedge fund turnover, the median hedge fund has about twice the turnover of the median mutual fund. At the individual firm level, we find that hedge funds have holdings that are less correlated with the market than the average mutual fund. We next turn to examining what types of securities hedge funds hold. We do this first by examining average holdings in 25 portfolios formed according to size and BE/ME or size and momentum; and second, in a regression context with many firm characteristics. Relative to mutual funds, hedge funds are overweight in the bottom three size quintiles and underweight in the largest two size quintiles. Hedge funds tend to prefer small value stocks but strongly avoid the largest quintile of value stocks. The comparison based on momentum and size show that hedge funds are not heavy momentum players, and in fact, relative to CRSP weights and mutual funds, they have larger holdings in past loser stocks. 2

5 Our cross-sectional regressions reveal that unlike mutual funds, which have a strong and growing preference for stocks with analyst coverage, hedge funds in aggregate do not seem to prefer stocks with high analyst coverage over those with low analyst coverage. Hedge funds also have a much milder liquidity preference than mutual funds. They tend to hold much smaller and somewhat more volatile securities than mutual funds as well. We find that mutual funds throughout the mid 1990s to 2004 avoid high price-to-sales securities, but hedge funds exhibit no aversion to these stocks. Overall, our holdings analysis suggests that hedge funds, while more active traders, tend to prefer less traded, less analyzed, and smaller securities. We now turn to examining hedge fund performance. DGTW (1997) list many benefits to holdings based approaches. The first main advantage is that characteristic matching allows for benchmarks that explain more of the realized variance in returns than those based on factor loadings and should have more statistical power to detect abnormal performance than factor models. Second, the benchmark allows for a more comprehensive and accurate return decomposition into the components of stock selection ability [Characteristic Selectivity (CS), Characteristics Timing (CT), and Average Style (AS)]. However, for hedge funds this approach has even more advantages. Since hedge funds report total returns and not the returns separately for their equity-only investments, with total returns it is extremely difficult to answer whether hedge funds are able to earn excess returns in stocks. Given the complex nature of the assets held by hedge funds, linear factor models will have difficulties answering whether any perceived abnormal performance is due to model mis-specification, unique asset mix, or true ability. Fung and Hsieh (2001) find that trend-following strategies can fool standard linear-factor benchmarks and that even benchmarks designed to capture trend-following strategies may fail without adequate 3

6 knowledge of fund operations. Additionally, informationless strategies [Weismann (2002)] using derivatives such as writing options on low probability events can increase a fund s Sharpe ratio at the expense of increasing downside risk, and these strategies will not be detected using factor models. By focusing on the equity-only portion of investments, we exclude markets where most of the informationless investing occurs. An important limitation of our approach is that we only examine stock market performance through long positions. We see this cost as necessary given our focus on the ability of hedge funds in the stock markets and the paucity of evidence on this topic. We follow DGTW (1997) and classify the return space using 125 benchmark portfolios formed on size, BE/ME, and stock return momentum, due to the general agreement that these patterns have been consistently related to past realized stock returns. We find over the entire period that hedge funds stock picking added 2.15 percent per year in return value as compared to only 0.82 percent for mutual funds. Over the 1986 to 2004 period, hedge funds added a statistically insignificant 1.43 percent in stock selection ability within the benchmarks. Most of this performance is generated in 1999 and Even if the extra performance were significant, some simple extrapolation suggests that it is unlikely to cover their fees. We hope future research will investigate holdings performance in conjunction with hedge fund cost structure. A major selling point of hedge funds is that they are said to contain the ability to time stock picks across asset classes. At least in the long-equity stock universe, this assertion is unfounded. Over the entire period there is no evidence that hedge funds are better at rotating between sectors (with size, BE/ME, and momentum) portfolios than mutual funds. Hedge funds tend to hold better average style portfolios than mutual funds over this period as well (providing 0.92 percent in additional return). 4

7 We examine whether hedge funds that perform well one year outperform the next year. We find that they do not. However, we find that the best hedge fund stock pickers do seem to have talent at picking stocks despite finding that they are bad sector timers. The remainder of our paper is as follows. Section II explains the sample construction and displays some simple summary statistics. Section III examines the activeness of hedge funds relative to mutual funds. Section IV displays the preference of hedge funds for firm characteristics. Section V investigates whether mutual funds follow hedge funds. Section VI and VII examine the performance of hedge funds and their performance persistence followed by a conclusion in Section VIII. II. Data A. Data Collection Procedure Here we describe the particulars of the labor intensive collection procedure for compiling our hedge fund sample. Since 1978 all institutions with over $100 million under management are required to fill out 13F forms quarterly for all U.S. equity positions where they own greater than 10,000 shares or $200,000. Domestic and foreign hedge funds with over $100 million of 13F securities under management are not exempted from these requirements. 1 A limitation of the 13F data is that the shorting activity of hedge funds is not contained in these reports, so all of our results will be based on examining the long-side of the portfolio strategy. Our overall task is to identify hedge funds from several sources, find their management or holding company name, and match them up with the 13F holdings 1 The SEC website posts the commonly asked question of whether a foreign institutional investment manager must file the 13F form and says: Yes, if they: (1) use any means or instrumentality of United States interstate commerce in the course of their business; and (2) exercise investment discretion over $100 million or more in Section 13(f) securities. A similar answer is posted for a non-sec registered investment advisor. See 5

8 data. Finally, we check the management company to find their main line of business and exclude all funds whose primary business is not in the hedge fund industry, as we describe in detail below. We obtain hedge funds from six sources: AltVest, the MAR graveyard of dead funds, firms from Hoovers.com (premium access, information as of 1997), firms in Table 2-4 in Cottier (1997) that have management in excess of $500 million as of December 1995, hedge funds listed in annual Nelson's directory books from 1988 to 2002, and TASS from February 1977 to May 2000 containing both alive and dead funds. Getmansky (2004) claims that TASS is the industry s only unbiased and most comprehensive database. Although TASS includes the largest number of hedge funds, we find that other databases are equally important in obtaining our final sample. Our overall process of backing out hedge fund holdings from 13F filings is similar in spirit to the pioneering approach of Brunnermeier and Nagel (2004). However, we use a much more comprehensive list of hedge funds to start with and obtain a final sample which over the entire period contains nearly six times as many hedge fund firms as the 52 hedge fund firms analyzed in their sample. Many hedge funds have a holding company firm with a different name from the hedge fund. It is the holding company firm name that we must identify in the 13F database. An advantage of Nelson's directory is that it includes both hedge funds and their affiliated (or holding) firm names and the fraction of hedge fund business in their affiliated firms. We then use the holding company name (when they differ) to match the firms up to the 13F institutional holdings database which match funds in each of the above data sources. We manually check the matches and remove any mismatches. To avoid spurious matches, for firms where the matches are less than perfect, we obtain 6

9 additional information from the websites of particular funds like the location and total net asset to verify the fund. After matching the holding firm names we then start to examine whether the holding firm s major line of business is hedge funds in using one of four main ways. First, we look up the holding company names in Nelson s Directory and only include firms with hedge fund assets that constitute over 50 percent of a holding company s total assets. Second, we include firms whose major line of business is described as a hedge fund in Cottier (1997). Third, if firms are unavailable from Nelson s directory, we manually check the SEC ADV forms and (like Brunnermeir and Nagel, 2004) require that a company have over 50 percent of its investment listed as other pooled investment vehicles (e.g., hedge funds) or over 50 percent of its clients as high net worth individuals. In addition to this criterion we also require that the fund charge performance-based fees. Fourth, for funds whose names are found in 13f but are still unidentified in any of these ways we check their websites to see if their primary business is hedge funds. Only those funds with websites that claim they are hedge funds and only available to accredited investors are included in our sample. 2 Funds not identified in any of these four ways are not included in our sample. Finally, for all funds that are identified as hedge funds in one of these four ways we perform a further check by examining whether a mutual fund in the CDA/Spectrum database has a holding company of the same name as one of the hedge funds. To avoid any delisting bias, we keep hedge fund firms in our sample until they begin to offer mutual funds since hedge fund companies that 2 For a definition of Accredited Investors, please refer to SEC website: accred.htm. 7

10 perform well may later open mutual funds. The 13F database has some reporting issues which we are aware of and carefully address. 3 Mutual fund holdings are from Thomason Financial CDA/Spectrum S12 data. The funds that we include have the following self-declared investment objective: Aggressive Growth, Growth, Growth and Income, and Balanced--codes 2, 3, 4 and 7 respectively. Sector, bond, preferred, international, and any fund with an investment objective that is not oriented to general equity is excluded. The main difference between the mutual fund (from N-30D) and hedge fund holding (from 13F) databases are as follows: 1. Mutual fund holdings are at the fund level, whereas hedge fund holdings are at the holding company or firm level. 2. There is no threshold that we are aware of on mutual fund reporting size, whereas only hedge fund firms above $100 million must file Hedge fund filings are required to be reported quarterly, whereas mutual funds are only required to report semi-annually Mutual funds are required to report all of their long stock holdings, whereas 13F filings are only required on security positions that are greater than 10,000 shares or $200,000. These differences seem logical to lead to smaller mutual funds and smaller mutual fund positions being reported. When equal-weighting performance across funds or firms, the differences in capitalization between the groups should be important, but in value-weighted 3 These reporting problems and solutions to these problems are discussed recently in an appendix by Griffin, Harris, and Topaloglu (2004). 4 However, Wermers (1999) notes that some small mutual funds do not report. 5 Nevertheless, there are missing quarters of data on the 13F database and some mutual funds voluntarily report quarterly. 8

11 returns, the distinction between the two databases becomes less relevant. We explore empirical differences in the datasets below. B. Sample summary statistics Panel A of Table I identifies how many hedge funds are available in each of our data sources and how many of these funds end up in our final sample of hedge fund holdings. Many funds are available from multiple sources but we identify a fund according to its earliest reported source. The majority of our firms come from Nelson s directory, Altvest, TASS, and the MAR graveyard. It is important to note that many of our data sources end in 2000, and MAR and Nelson s directory end in Updated versions of these databases would likely add some newer hedge funds but these funds would only have limited track records. In all, we are able to obtain quarterly stock positions of over 306 fund firms. It is important to note that each of these fund firms may often have multiple hedge funds under management so the sample may be representative over 1,000 funds. However, we have no accurate method to judge the ultimate number of funds represented in our sample. Because examining portfolio performance is one of our objectives, we are particularly concerned with constructing a sample free of potential biases. Thus, we construct two samples. The first sample, named Hedge A, is larger and more comprehensive but may suffer from reporting biases while the second sample, named Hedge B, should be relatively free from potential biases. Since reporting of a hedge fund is voluntary, hedge funds may choose to report their performance to standard data sources such as TASS after they have had a period of good performance. Since we use these datasources to locate hedge funds, we are concerned about examining their holdings prior to the period in which their holdings appear in our first 9

12 datasource. 6 To control for this issue we also construct another sample, Hedge B, which only contains information on funds in the years after we obtain their name from one of the sources we list above. Thus, this database (Hedge B) should not suffer from self-selection bias. 7 Hedge B is more applicable to measuring performance whereas both samples can provide information about hedge fund holdings. Hedge A includes all hedge funds which are in our database both before and after the time where we locate their name. Annual summary statistics on both hedge fund samples are contained in Panel B of Table I. In 1980 our Hedge A sample contains 25 hedge fund firms but it grows slowly throughout the 1980s and 1990s to include 231 firms in 2000, before falling slightly to 191 firms in Panel B of Table I shows that Hedge B starts much later (because of the later start of hedge fund data sources). Although we only have one Hedge B firm tracked in 1984 and 1985, the number of firms grows rapidly and includes 37 hedge fund holding companies in 1992, 105 firms in 1995 and 193 firms in In addition, it is important to note that the shorter period is longer than that in many other important studies. 9 The number of funds and holdings of mutual funds from N-30D filings compiled by the CDA/Spectrum Mutual Fund Holdings database is also reported for comparison purposes. It is important to note that this is an unfair comparison in the sense that the hedge fund firms are more likely to handle many hedge funds within the same firm whereas each mutual fund is reported separately. Additionally, it is important to note that our hedge fund sample is only a partial sample of the industry as we have excluded hedge funds with holding 6 Although this good performance could come from a variety of positions in other markets, performance in the equity market is one factor that can drive positive performance. 7 The data could suffer from backfill bias to the extent that the original data reported by the vendors was filled back in time. To counteract this effect we have two databases of dead funds. 8 Since our TASS sample ends in 2000 there are likely many new funds in that are not included in our sample. The exclusion of these funds should not result in any systematic biases. 9 For example, Brown, Goetzmann, and Ibbotson (1999) use a sample of only seven years, whereas is 13 years. 10

13 companies having other primary businesses or also owning a mutual fund. Our sample shows a decline in the number of mutual funds from 3092 in 1997 to 2229 in We also examine the average and total number of stocks held in both hedge fund samples and in mutual funds. As seen in Panel B of Table I, the average number of stocks held by each Hedge A holding company is 120 in 1980, 131 in 1990, 169 in 2000, and 217 in The numbers for the Hedge B sample are slightly lower. The average mutual fund holds fewer firms, particularly earlier in the sample. These differences in the number of firms between mutual funds and hedge funds will have some relevance to the preciseness of calculations performed later in the paper. The total number of stocks held by Hedge A holding companies is 1,247 in 1980 which is 65.6 percent of the stocks with reported holdings by mutual funds. However, this number grows to 83.5 percent in 1990 and 85.6 percent in Hedge B holdings are far below those of Hedge A, but the holdings grow close by the mid 1990s. III. How active are hedge funds? We first examine how active traders of hedge funds are by first examining turnover and then we look at the relation between their hedge fund weights and market weights to judge the dispersion of their trading strategies. A. Turnover Hedge funds are not required to report their turnover like mutual funds. Thus, we must estimate turnover from quarterly stock positions. This measure is by nature an understatement of true turnover since intra-quarter trades will not be captured. Nevertheless, examining turnover in this manner can serve as a useful comparison between hedge funds and mutual funds. We compare the current holdings of a firm in stock X to its position at 11

14 the beginning of the quarter. So if the firm held 100,000 shares in stock X and it increased its positions to 1,000,000, then the change in holdings would be 900,000 shares times the stock price at the beginning of the period. We then sum this dollar value of trading across all stocks. This is done separately for sells as well, and to avoid sales due to flows of capital, the smaller of the buys and sells is picked and normalized by the firm s beginning of the period capitalization. 10 Figure 1 reports the distribution of turnover for both hedge funds and mutual funds. The distribution shows striking differences between the trading of mutual funds and hedge funds. Most hedge funds have turnover above that of most mutual funds. The median turnover using quarterly holdings for hedge funds of 1.02 is almost twice of the median mutual fund turnover (0.63). The distribution is also right skewed for hedge funds and similar results are obtained for both hedge fund samples. These findings reinforce the conventional wisdom that hedge funds are active traders. B. Weights relative to the market We now turn to estimating the correlations between hedge fund positions and market positions. Hedge funds may take positions in opposite directions that may lead to different conclusions from examining individual hedge fund firm holdings than their aggregate holdings. In every quarter for every hedge fund, we calculate the correlations of individual hedge fund firm weights and market weights. We then average these correlations across quarters for each fund and then compute annual averages across funds. As shown in Figure 2, during early years of the sample hedge funds (Hedge A) have a stronger relative correlation with market weights, but mutual fund correlations with the market grow in later 10 If the fund does not file for a particular quarter, then we carry over all the positions and no turnover occurs that quarter. This procedure follows the way that the CRSP mutual fund database reports turnover and is described in Chen, Jegadeesh and Wermers JFQA (2000, p349). Because CDA/Spectrum does not calculate turnover, we calculate turnover in this manner for both hedge funds and mutual funds for consistency. 12

15 years of the sample period. In 2004 the average correlation of the value-weighted market weights and those of the average mutual fund is whereas it is for hedge fund sample A and for sample B. In unreported results, we also calculate the correlation between the aggregate holdings of mutual funds and the market. This aggregate correlation is in 2004, as compared to for aggregate hedge fund holdings (Hedge A). We also compare weights using a deviation approach similar to Cremers and Petajisto (2006) by taking the difference between the weights in each stock ($ holding in each stock/$ total position in all stocks for a fund) and the weights of the CRSP market index in each stock. For each fund, the value-weighted sum of these deviations in each stock is then summed up across stocks and divided by two, since for every position over-weight there is an offsetting under-weight. Figure 3 presents the long-equity weight deviation distribution across funds where 0 indicates no deviation and 1 indicates a total deviation from either the CRSP value-weighted market or S&P 500 index. Panel A shows that hedge fund firms generally have larger weight deviations, except for the largest categories where some mutual funds deviate 100 percent from the index. These are likely funds with other objectives, such as a small cap with other benchmarks than the ones used here. To further examine these deviations, we decompose the total deviation into deviation across industries and that within industries. The cross-industry weights are the fraction of capital that the firm designates to sectors that is different from, say, the S&P 500. Hedge funds are generally much more aggressive in deviating from the industry, whereas mutual funds exhibit higher within-industry weight deviations, suggesting that they are more focused on localized stock-picking. In general, both through average weight correlations and observing the distribution of weight deviations we learn that hedge funds generally deviate 13

16 more from the value-weighted market indices than mutual funds and that these deviations usually come from taking larger cross-industry bets. IV. What type of stocks do hedge funds prefer? We examine whether hedge funds have preferences for stocks with certain characteristics. We do this in two main ways. First, we compare hedge fund weights relative to market weights on twenty-five size and book-to-market equity (BE/ME) portfolios and twenty-five size and momentum portfolios. We then compare these weights relative to mutual funds. Second, we perform cross-sectional regressions across firms of hedge fund holdings on a variety of characteristics including firm age, analyst coverage, beta, turnover, price-to-sales (P/S), standard deviation of returns, as well as size, BE/ME, and momentum. We also compare the preferences of hedge funds for these characteristics to the preferences of mutual funds. A. Hedge fund weights in size, BE/ME, and momentum portfolios Panel A and B of Figure 4 presents the positions of our Hedge B funds in stocks of a particular size and book-to-market equity quintile. We measure their holdings first relative to market weights (Panel A) and then relative to mutual funds (Panel B). The holdings are calculated on an annual basis and then summed across time and the weights in each portfolio are standardized by the market weights in each group so that percent weightings are comparable across portfolios. Interestingly, hedge funds (sample A) hold more in mediumsized stocks than a market-weighted investor and hence hedge funds are generally underweighted in the largest quintile of stocks relative to the market. The amount that the weights vary in the large size quintile varies from a slight overweight of 1 percent in the large growth portfolio to an underweighting of 18.3 percent in the large size value group 4 14

17 quintile. Hedge funds are also underweighted in stocks in the smallest size quintile (relative to market weights). The average preference for value/growth seems less clear. On average, hedge funds seem to have a preference for both extreme value and growth stocks in the medium size quintiles but not for small stocks. Preferences for these categories are likely to be highly time-varying, which we will examine in more detail in the next sub-section. Panel B compares hedge funds to mutual funds. Relative to mutual funds, hedge funds actually have a strong positive preference for stocks in the smallest three quintiles. The only exception is that hedge funds actually hold slightly more large growth stocks. Hedge funds also tend to show a preference for value stocks in the smaller quintiles although they are overweighted in growth stocks to a lesser degree in these quintiles as well. Panel C and D of Figure 4 examines hedge fund holdings in excess of market weights in securities sorted on size and momentum. Hedge funds prefer high momentum stocks in the medium size groups but not in the smallest size quintile. However, they actually have an even stronger preference for loser stocks in all size groups. Panel B shows that relative to mutual funds the preference of hedge funds for high momentum stocks is even more distinct. Past research has shown that mutual funds have a strong preference for stocks with high past stock returns (Grinblatt, Titman, and Wermers, 1995). Panel B shows that hedge funds have a weaker preference for winner stocks than do mutual funds. In addition, hedge funds seem to have a strong preference for low momentum stocks, particularly those with low BE/ME. The preference for low momentum stocks would seem to be a largely losing strategy since Hong, Lim, and Stein (2000) show that small loser stocks have more underperformance. However, it is important to note that they exclude stocks in the smallest NYSE size quintile where there is little evidence of momentum. Relative to mutual funds, 15

18 hedge funds exhibit a positive bias towards smaller stocks, past losers, and value stocks. The sorting approach has the advantage that it can allow for many non-linear patterns in ownership but has the disadvantage in that only several characteristics can be examined together. We now turn to an analysis of firm ownership on a broader set of firm characteristics. B. Regressions of stock ownership on firm characteristics We seek to understand the preference of hedge funds for holding firms with a variety of firm characteristics. Our approach to understanding the relation between ownership and firm characteristics is similar to Falkenstein (1996) for understanding the stock preference of mutual funds in 1991 and Each year, we estimate a cross-sectional Tobit regression of 1+hedge fund ownership for a particular stock on firm characteristics from the previous year. We use the yearly time-series to construct Fama/McBeth coefficients and standard errors. The firm characteristics we consider are the age of the stock measured by the number of months on the CRSP database, the number of IBES analysts, the monthly Dimson betas, the BE/ME, turnover, firm market equity, price-to-sales (P/S), the previous twelve month return (momentum), and a stock s standard deviation. Gompers and Metrick (2001) examine the preferences of institutions in general for firm characteristics and Bennett, Sias, and Starks (2003) examine in detail how these preferences have changed through time. Table II reports regressions both for mutual funds and hedge funds. Hedge funds seem to have a slight preference for younger securities, high beta securities, more liquid securities with higher turnover, smaller stocks, and high past returns. Consistent with Falkenstein s (1996) main results, mutual funds tend to have a strong preference for stocks 16

19 with more analysts, more liquidity, and avoid stocks with low volatility. They also tend to prefer stocks with high beta, low BE/ME, low P/S, and high past return. To further trace the dynamics of these securities over time, we plot the annual crosssectional regression coefficients in Figure 5. Mutual funds exhibit a large and growing preference for holding stocks with analyst coverage whereas hedge funds tend to not care about analyst coverage. Mutual funds seem to vary their preference for beta much more over time than do hedge funds (at least Hedge sample B), with a strong preference for beta in the mid 1990s but a negative preference in the 2000s. Mutual fund preference for BE/ME also tends to vary more across years although both mutual funds and hedge funds have tended to prefer low BE/ME securities since the mid 1990s. Mutual funds have a strong and growing preference for liquidity throughout the period, whereas the preference of hedge funds for liquidity is consistently lower, and in the case of Hedge A this preference has slightly decreased. Hedge funds had almost no preference for size in the 1990s, in contrast to mutual funds where there was a large increase in their preference for firm size. In the 1990s, mutual funds display an aversion to P/S that is not there for hedge funds. Mutual funds tend to have a stronger preference for momentum through most of the period, although this preference in both hedge funds and mutual funds seems to have disappeared in the last two years. Hedge funds do not seem to display an aversion to standard deviation through the entire period. In contrast, mutual funds avoid high standard deviation stocks until the last six years of the sample. Panel D compares the hedge fund sample to the mutual fund sample and further clarifies the average statistical comparisons between mutual funds and hedge funds. Relative to mutual funds, hedge funds prefer stocks with fewer analysts, less liquidity, smaller size, lower past return, and higher standard deviation. These findings are robust to both hedge 17

20 fund samples with the exception being that Hedge B, which is free from self-selection bias and starts later, also demonstrates a preference for high price-to-sales securities relative to mutual funds. Overall, our findings indicate that hedge funds prefer less opaque firms that may be more costly to trade and less analyzed. These results for stock selection are generally consistent with the findings of Fung and Hsieh (1997) that the factor exposures of hedge funds and mutual funds returns are dramatically different. Our findings indicate a potentially important and different role for hedge funds in generating stock pricing efficiency. V. Evaluation of fund performance A. Performance We use the methodology of Daniel, Grinblatt, Titman and Wermers (1997) to separate out the performance of hedge funds into their excess return relative to 125 size, industry adjusted BE/ME, and momentum benchmarks. 11 Characteristic selection (CT) measures the ability of managers to pick stocks within the size, BE/ME, and momentum benchmarks. The characteristic selectivity (CS) uses the change in the hedge fund weights from the previous year to measure the ability of funds to time selection within the benchmark portfolios. The average style (AS) uses the previous year s portfolio weights to measure the return due to a manager s propensity to hold stocks of a particular style. The combination of AS, CT, and CS equals a fund s average return. We closely follow DGTW (1997) and Wermers (2000) in calculating performance relative to the benchmarks. For hedge fund firms we calculate the performance measures on an annual basis for each firm and then perform equal-weighted averages across firms. For mutual funds we also calculate equal-weighted averages across firms and thus the performance of the investors in the 11 The industry adjusted book-to-market equity is consistent with Wermers (2004). We thank Russ Wermers for graciously providing the benchmark returns. 18

21 industry may differ because of the weighting. We examine performance measures for our Hedge sample B that only examine stocks in the year after they appear in the database. Because this database has fewer firms in early years of the sample, we begin reporting in 1986 when four hedge fund firms report. Our estimates quickly become more reliable in later years of the sample when more hedge fund firms are available. We compare the performance of our hedge funds to a CDA spectrum sample of actively managed mutual funds. Panel A of Table III reports year by year value-weighted performance measures and Panel B reports equal-weighted. The gross return indicates that there is substantial variation in the returns from hedge funds from both the market indices and mutual funds. We examine the causes of that return through the benchmark decomposition. We first report performance measures for hedge funds and mutual funds and then we compare differences in the performance measures. In Panel B, the CS measure for hedge funds is positive in 13 of 19 years for hedge funds but 15 of 19 years for mutual funds. The largest deviation in the characteristic selectivity measure for hedge funds is percent in 1999 when hedge funds outperformed mutual funds by 9.72 percent. This large outperformance may be due to the high weights of tech stocks or high P/S stocks as documented by Brunnermeier and Nagel (2004). Both hedge funds and mutual funds tend to generate negative characteristic timing in many years of the sample although in many years hedge funds have worse timing than mutual funds. In terms of average style it is not clear if mutual fund or hedge fund style dominates. Panel B also presents summary statistics for the sample from , from 1995 to 2004, and from We choose to divide the sample here because our sample (B) during later periods contains more hedge fund firms and thus performance statistics here are 19

22 much more reliable than those in early years of the sample. Overall, we find no evidence of either hedge funds (sample B) or mutual funds engaging in successful characteristic selectivity or timing in the earlier sub-period. However, both do seem to have positive average style over this period with hedge funds delivering slightly lower style performance than mutual funds. Over the period mutual funds deliver a characteristic selectivity return of 0.87 percent per year and hedge funds deliver 2.66 percent. In this period hedge funds outperform mutual funds by a statistically significant 1.79 percent per year in stock picking. Also, in terms of timing styles hedge funds have shown strong ability (1.27%) but the difference from that of mutual funds is positive but not significant. In terms of average style hedge funds pick an average style that delivers them an average return of 0.49 percent greater than mutual funds over the period. Over the whole period both hedge funds and mutual funds deliver positive and significant stock selection ability. Hedge funds deliver an additional 1.65 percent in stock selection returns but this difference is not statistically significant. Hedge funds tend to be slightly better timers and have a slightly better style but the difference is not significant. It is not clear whether we should rest our conclusions on the later period where there are more hedge funds available for judging their performance. If one does so then you are left with the impression that hedge funds may be better at picking stocks and have on average chosen a better average style. As discussed in DGTW (1997) it is not clear whether to give managers credit for picking the style or if they fell into that style. Even if one gives managers credit for picking the average style over the period, it would seem that hedge fund managers generated 2.75 [1.79 (CS) (CT) (AS)] in value 20

23 over mutual funds. Recall that much of the CS return (shown in Panel A) is generated in 1999 and 2000 and that only marginal value is created in other years. Since book-to-market benchmarks are inadequate for classifying tech stocks, we reexamine performance using ten price-to-sales deciles, similar to Brunnermeier and Nagel (2004). We specifically wish to address whether hedge funds performance during the late 1990 s was confined to high tech stocks (high price-to-sales). Table IV shows that this is largely the case out of the total 9.72% performance difference between hedge funds and mutual funds in 1999, 6.95% (1.40% % %) comes from the three highest priceto-sales deciles. B. Discussion The average outperformance of hedge funds over mutual funds is 1.48 percent (Table III) over the 1986 to 2004 period. We are not able to address whether this is enough to justify higher fees, since there is no reliable source on hedge fund leverage ratios available that we are aware of. 12 Nevertheless, it is clear that for a hedge fund to compensate for its higher fees relative to a mutual fund, it would need significant leverage or to make more profits on the short-side. However, shorting is on average a loosing proposition. It is also true that a large selling point of hedge funds is the value they appear to create by investing in various asset classes and timing capital between asset classes. While we can only examine long stock positions here, the lack of hedge funds to create any positive returns in timing characteristics (like picking growth, value or momentum at the right times) casts considerable doubt on the assertion that hedge funds can time markets. 12 We thank David Hsieh for informing us of the various problems with leverage ratios in TASS and other sources. 21

24 VI. Are some managers better than others? Although the average hedge fund is not much better than mutual funds, there may be more variation in ability for hedge funds, particularly given that they are less likely to index as compared to mutual fund managers. Consistent with hedge fund firms taking positions further away from the market index, Figure 6 shows that hedge fund firms have wider tails or more dispersed performance than mutual funds. Figure 6 also shows that there seems to exist a slightly larger group of hedge fund managers (in the right-hand tail) that outperform than is found in the mutual fund industry. We turn to examining differential ability first by looking at hot hands and then by ranking managers according to their entire history of performance. A. Performance Persistence We consider whether hedge funds have any hot hands and whether managers that perform well in one year also perform well in the next. We rank managers according to their ability to generate gross return and then examine the three components of the return as well as their gross return in the following year. Because of the limited number of hedge fund firms in earlier years of the sample we begin the ranking in 1991 and examine returns from Panel A of Table V examines the gross return and return components of five quintiles of past gross hedge fund or mutual fund return performance. Hedge funds with the highest past quintile of fund performance in the previous year tend to outperform those hedge funds with low past performance by 5.54 percent per year but this difference is not statistically significant. Further, this difference (5.54 percent) may be due to superior stock selection skill, but again there is no significant difference in the CS measure for past winner and loser hedge fund firms. 22

25 We also examine these measures for mutual funds. Using the benchmark approach, DGTW (1997) found that the previously identified patterns of hot hands were largely due to momentum and that mutual fund managers did not seem to exhibit hot hands. Over a much more recent period Table V confirms their findings. Mutual funds that performed well in the previous year continue to outperform by about 553 basis points, but this difference is not significant. None of the components of mutual fund performance are statistically different between past winners and losers either. We also perform tests with semi-annual rebalancing and although we find larger differences in hedge fund returns between winners and losers, there is no significant difference in returns. Using simple strategies it does not appear that past performing winner funds perform reliably better than loser funds. B. Performance using each firm s whole history Table VI examines performance for hedge fund firms based on their entire past picking of CS performance (Panel A) or stock picking and timing ability (Panel B). There is some evidence that good stock pickers, continue to outperform in terms of stock picking ability but they are bad market timers, resulting in insignificant total return differences. This finding has important implications for the way fund-of-fund investors pick funds. VII. Robustness Issues A. Turnover One potential problem with our findings is that many hedge funds may use stocks not primarily to make money but for hedging purposes for either options or bonds. Additionally, some hedge funds specialize in high frequency trading. To address these concerns we stratify hedge funds into two groups based on whether the estimated turnover in the previous year is less than or greater to one. Table VII presents value-weighted DGTW 23

26 performance numbers for both of these groups and shows that the aggregate performance of the low and high turnover groups are very similar. Only looking at low turnover funds does not alter our conclusions. B. Aggregate Returns Our results have shown some limited outperformance of hedge funds. One question is whether using this information in aggregate is valuable. Chen, Jegadeesh, and Wermers (2000) show that mutual funds purchases have statistically predictive power for future returns. Hedge funds are a much smaller proportion of ownership and take more disparate strategies than mutual funds, but we nevertheless examine a similar question of whether aggregate hedge fund ownership can predict future returns. In cross-sectional regressions of stock returns on the last quarter s return, we find statistically significant evidence of past hedge fund trading predicting future returns. However, this predictability is subsumed by other variables that have been known to have predictive power for stock returns like lagged mutual fund trades, a stock s past return, past breadth of mutual fund ownership, and past changes in institutional non-hedge fund ownership. VIII. Conclusion We provide the first comprehensive examination of the nature of hedge fund holdings in U.S. stocks and a holdings-based analysis of the equity performance of hedge funds. In comparison to mutual funds, hedge fund firms have much higher turnover and deviate more from market positions. Despite this preference to trade more frequently, however, hedge funds in comparison to mutual funds overweight stocks with fewer analysts, less turnover, smaller size, and more standard deviation. Hedge funds also tend to strongly buy small losers stocks and avoid winner stocks in comparison to mutual funds. These 24

27 patterns are inconsistent with the momentum preference found for mutual funds [Grinblatt, Titman, and Wermers (1995)]. We also measure stock return performance and performance persistence. The use of holdings to quantify performance is particularly important because unlike most mutual funds who hold mostly positions in stocks and bonds, hedge funds hold many disparate asset classes besides stocks and they do not typically report the performance of their stock-only portfolios. Thus, even if accurate measures of a fund s total performance are obtained, it is difficult for an investor to assess whether a hedge fund s abnormal performance is due to the nature of its activity in derivative markets and/or through superior skill in traditional equity market positions. We classify hedge fund performance by: a) the style of the fund, b) moving in and out of styles before they become popular, or c) picking winners within styles that simply perform better. We find some evidence that hedge funds have some talent at stocks picking (by 2.15 percent per year for value-weighted portfolios and 1.65 for equal-weighted) over the period. For value-weighted portfolios they are better stock pickers than mutual funds by a statistically significant 1.32 percent per year. However, the ability of hedge funds to pick styles is slightly worse, though not statistically different from mutual funds. Those hedge funds that have stock picking ability in the past continue to pick stocks better than other hedge funds, although they are poor market timers. Without the leverage positions of hedge funds (which few report), we cannot give precise statements about whether the far tail of equity hedge funds add enough value through their long positions to justify their higher fees compared to mutual funds. However, even for the far tail of stock picking, our evidence indicates that it would take significant leverage for these investors to earn abnormal returns 25