## Abstract

Hedge funds significantly reduced their equity holdings during the recent financial crisis. In 2008:Q3––Q4, hedge funds sold about 29% of their aggregate portfolio. Redemptions and margin calls were the primary drivers of selloffs. Consistent with forced deleveraging, the selloffs took place in volatile and liquid stocks. In comparison, redemptions and stock sales for mutual funds were not as severe. We show that hedge fund investors withdraw capital three times as intensely as mutual fund investors do in response to poor returns. We relate this stronger sensitivity to losses to share liquidity restrictions and institutional ownership in hedge funds.

Hedge funds are the investor class that most closely resembles textbook arbitrageurs: They engage in sophisticated trading strategies, use leverage, and take short positions. Despite these degrees of freedom, hedge funds depend on outside financing, which may curtail their ability to exploit profit opportunities (Shleifer and Vishny 1997; Gromb and Vayanos 2002; Vayanos 2004; Brunnermeier and Pedersen 2009; also see Gromb and Vayanos 2010 for a survey). Limits on hedge funds' arbitrage potential are likely to be more severe during market crises. At such times, in response to initial losses, capital providers, investors, and lenders may withdraw their funds and force the hedge funds to liquidate their positions prematurely. This behavior can deteriorate liquidity in the market and cause further losses. Some empirical evidence reveals that during the recent financial crisis, the returns from providing liquidity increased (Nagel 2011) and the stocks traded by hedge funds that used Lehman Brothers as prime broker became less liquid (Aragon and Strahan 2011), which indirectly suggests that hedge funds withdrew from the market.1

In this article, we provide direct evidence on hedge fund trading in the U.S. stock market during the financial crisis of 2007–2009. We document the change in hedge funds' long stock positions, investigate the economic determinants of their trades, and analyze the difference in their behavior relative to mutual funds. Our study relies on a new dataset that originates from matching the institutional ownership of U.S. stocks from 13F filings to a proprietary list of hedge funds. These data are then manually matched to the Lipper TASS Hedge Fund Database and a dataset of Form ADV filings to compile information on hedge fund characteristics, performance, and ownership structure.

The main message of the article is that hedge funds exited the U.S. stock market en masse as the financial crisis evolved, primarily in response to the tightening of funding by investors and lenders.2 Although hedge funds have provisions in place to limit redemptions, our results suggest that this was potentially a magnifying factor in causing hedge fund investors' withdrawals relative to mutual funds. Investors, fearing that the hedge fund would further constrain their ability to pull their money (e.g., by raising the gates), reacted promptly to the first signs of deteriorating performance. Our results support the theories of limits to arbitrage, which propose that arbitrageurs cannot operate in an unconstrained fashion due to their reliance on outside financing (e.g., Shleifer and Vishny 1997; Gromb and Vayanos 2002, Gromb and Vayanos 2010; Vayanos 2004; Brunnermeier and Pedersen 2009).

The stylized fact that we document and explore is the sharp decline in hedge funds' stock ownership during the recent financial crisis. Figure 1 plots the fraction of U.S. market capitalization held by the hedge funds in our dataset. The shaded areas denote the quarters around two events of special market stress: the Quant Meltdown (2007:Q3) and Lehman Brothers' bankruptcy (2008:Q3). The figure shows significant declines in hedge fund holdings around these events. More specifically, we find that hedge funds reduced their equity holdings by about 6% in each of the third and fourth quarters of 2007, and by about 15% in each of the third and fourth quarters of 2008, on average.3 We also document an aggregate decline in short interest4 in these four quarters (for which the short-selling ban on financial stocks in September and October 2008 is also responsible). We show that the decline in long holdings and short interest did not cancel each other out, as there was only a small overlap between stocks that were sold by hedge funds and those that were bought to cover short positions. Rather, they both serve as evidence of the large deleveraging process that took place in the hedge fund sector. Complementary to our findings, Ang, Gorovyy, and van Inwegen (2010) show that hedge fund leverage decreased substantially during the crisis.

Figure 1

Time series of hedge funds' equity holdings (% of total market capitalization)

The figure plots the fraction of U.S. stock market capitalization held by the hedge funds in our sample. The shaded areas denote the quarters around the Quant Meltdown (2007:Q3) and Lehman Brothers' bankruptcy (2008:Q3). The series ranges from 2004Q1 to 2009Q4.

Figure 1

Time series of hedge funds' equity holdings (% of total market capitalization)

The figure plots the fraction of U.S. stock market capitalization held by the hedge funds in our sample. The shaded areas denote the quarters around the Quant Meltdown (2007:Q3) and Lehman Brothers' bankruptcy (2008:Q3). The series ranges from 2004Q1 to 2009Q4.

After establishing that hedge funds withdrew significantly from the stock market during the financial crisis, the goal of the article is to understand the economic forces behind the withdrawal. Guidance for our analysis comes from theories suggesting that limits-to-arbitrage can emerge at times of market stress (e.g., Shleifer and Vishny 1997; Gromb and Vayanos 2002; Vayanos 2004; Brunnermeier and Pedersen 2009). These forces can manifest themselves as investors' redemptions, margin calls, and the risk limits that are in place to preempt future capital calls. In addition, the occurrence of liquidity dry-ups simultaneously across asset types (Chordia, Sarkar, and Subrahmanyam 2005; Goyenko 2006; Goyenko and Ukhov 2009; Baele, Bekaert, and Inghelbrecht 2010) obliges constrained arbitrageurs to close some potentially profitable positions in order to undertake other trades with greater expected risk-adjusted returns.

We provide direct evidence of the primary role redemptions played in causing the selloffs. Redemptions account for roughly 50% of the average decline in hedge fund equity holdings during the selloff quarters (that is, they explain 6% of the 12% average decline in equity holdings). Because we can directly measure redemptions, this finding is the soundest evidence of the motives behind equity sales during the crisis. Moreover, we use hedge fund average leverage as a proxy for lender pressure and show that this channel alone accounts for about 42% of selloffs. Overall, redemptions and leverage explain about 80% of the decline in average hedge fund equity holdings (that is, 9.5% of the 12% average decline in equity holdings). We view this set of findings as consistent with the limits-of-arbitrage literature cited above.

To bring further evidence to light about the role played by financial constraints, we study the characteristics of the stocks that hedge funds traded during the crisis. We report that hedge funds were more likely to close positions in high- rather than low-volatility stocks. Symmetrically, short interest decreased more strongly for high-volatility stocks. These results reveal a potential risk management motive. As the limits-to-arbitrage literature predicts, a reduction in exposure to high-volatility assets can derive from margin calls (Brunnermeier and Pedersen 2009) or from internal risk management practices, such as Value at Risk (VaR) models (Vayanos 2004; Brunnermeier and Pedersen 2009). Interestingly, hedge funds were more likely to sell liquid stocks during the crisis,5 consistent with Scholes's (2000) observation that, during a crisis, investors unwind their portfolios by selling the most liquid securities first. Moreover, hedge funds unwound their value and momentum strategies, consistent with a forced deviation from their standard strategies.

To give perspective to our findings about hedge funds, we compare their behavior with that of another important class of institutional investors, mutual funds. We show that mutual funds' equity portfolios did not significantly decrease during the financial crisis and that their redemptions were not nearly as severe as they were for hedge funds. This evidence calls for an investigation of the economic mechanisms that could make hedge funds more vulnerable to external funding than are other institutional investors. Unlike hedge funds, mutual funds do not use leverage, have no restrictions on investors' liquidity, and in general cater to a less sophisticated clientele.

This observation suggests that investors in these two asset classes may react differently to negative performance, especially if they fear that hedge funds may lock up their money in case of further losses. Prior studies find a convex flow-performance relation for mutual funds (e.g., Chevalier and Ellison 1997; Sirri and Tufano 1998) and document that individual investors do not immediately liquidate their investments in mutual funds after initial losses (Calvet, Campbell, and Sodini 2009). In contrast, the flow-performance relation for hedge funds seems to be concave (in accordance with Li, Zhang, and Zhao 2011), especially in the presence of share restrictions (corresponding to the results of Ding et al. 2009). We combine these two separate strands of the literature and, in a pooled analysis of the flow-performance relation, document that, following poor past performance, hedge fund investors withdraw almost three times more capital as do mutual fund investors. We corroborate the evidence in Ding et al. (2009) by showing that a large part of the difference is due to hedge funds with illiquid shares relative to mutual funds. This result is consistent with the idea that the potential for hedge fund shares to become even more illiquid at times of crisis generates a preemptive response on the part of investors once poor performance is observed. We take this evidence as one key element in defining the different responses of mutual and hedge fund investors to the crisis. Hence, this element has the potential to explain at least part of the difference in the portfolio liquidation behavior of these two groups of institutional investors.

Finally, we examine the hypothesis that the observed difference in trading behavior also originates from mutual and hedge funds' different clienteles. Institutional investors in hedge funds are likely to be more sophisticated than individual investors in mutual funds. Calvet, Campbell, and Sodini (2009) provide evidence that investor sophistication magnifies the speed of reaction to news. Moreover, institutional investors have risk management controls in place to preempt violations of capital requirements. Also, managers employed by institutional investors have career concerns, as their compensation depends on the performance of the funds they select. Overall, these mechanisms are likely to make institutional investors more reactive to bad news than are individual investors. We test these conjectures by exploiting the heterogeneity in the client base of hedge funds (drawn from the ADV form database). We document that hedge funds with a higher share of institutional investors experienced stronger redemptions during the crisis and also sold more equity. This evidence establishes the prevalence of an institutional client base in hedge funds as another likely channel of the large redemptions hedge funds suffered and their consequent selloffs during the financial crisis.

Our results provide perspective on the findings of other recent research. Boyson, Stahel, and Stulz (2010) show that hedge funds display contagion across asset classes. Our results suggest that redemptions by investors and credit constraints are potentially important factors that could generate systematic contagion effects across asset types that are traded by hedge funds. Khandani and Lo (2011) hypothesize that the unprecedented losses of a number of long-short hedge funds in the summer of 2007 were the result of forced deleveraging. We show that this deleveraging actually occurred and that it was related to tightening financial constraints. By studying the correlation of hedge fund returns with the stock market, Cao et al. (2009) and Billio, Getmansky, and Pelizzon (2010) conclude that hedge funds are able to time the market and avoid liquidity dry-ups. Our evidence suggests that much of this “timing” behavior is the result of hedge fund capital evaporating during crises. Sadka (2010) shows that hedge fund returns contain a premium related to aggregate liquidity risk. Our evidence can explain this premium in terms of the financial constraints that prevent hedge funds from providing liquidity in times of crisis. Our finding that redemptions are a major constraint to hedge funds' ability to capture the illiquidity premium resonates with previous results showing that hedge funds' performance is affected by the amount of investor capital available to them (Hombert and Thesmar 2009; Teo 2011). Since the first draft of our article, Boyson, Helwege, and Jindra (2010) have analyzed similar data, but with a smaller set of hedge funds. They found that during the financial crisis, hedge funds sold more equity holdings than needed to merely face redemptions. Our work suggests that in addition to redemptions, a large part of the selloffs can be explained by hedge funds responding to lenders' pressure to deleverage.6

The article proceeds as follows. Section 1 describes the data sources we use. Section 2 explores the aggregate behavior of hedge funds during the crisis. Section 3 explores the financial constraints channel for the stock selloffs. Section 4 presents the comparative analysis of the flow-performance sensitivity relative to mutual funds and explores the channels for the observed difference in sensitivity. Section 5 concludes.

## Data

### Data sources and sample construction

Our study combines several datasets related to hedge funds, mutual funds, stocks, and institutional ownership.

#### Hedge fund holdings data

The main dataset used in the study combines a list of hedge funds (by Thomson-Reuters), mandatory institutional quarterly portfolio holdings reports (13F), and information about hedge fund characteristics and performance (Lipper TASS Hedge Fund Database). The 13F mandatory institutional reports are filed with the Securities and Exchange Commission (SEC) on a calendar-quarterly basis and are compiled by Thomson-Reuters (formerly known as the 13F CDA Spectrum 34 database).7”Form 13F requires all institutions with investment discretion of over $100 million at the end of the year to report their long holdings (mainly publicly traded equity, convertible bonds, and options) in the next year.8Therefore, all hedge funds with assets in such qualified securities in excess of$100 million are required to report their holdings in 13F filings. 13F reporting is done at the consolidated management company level.9

We then match the list of 13F institutions in Thomson-Reuters with a proprietary list of 13F hedge fund managing firms and other institutional filers provided by Thomson-Reuters. Relative to the self-reported industry lists commonly used to identify hedge funds, the Thomson-Reuters list is more comprehensive, as it classifies all 13F filers.10 Moreover, the Thomson-Reuters hedge fund list identifies hedge funds at the disaggregated advisor level, not at the 13F report consolidated level.11,12 The 13F data available to us range from 1989:Q3 to 2009:Q4. Before applying the filters described below, the number of hedge funds in the Thomson-Reuters list varies from a few dozen in the early years to over 1,000 at the 2007 peak. With some caveats that we mention below, an additional advantage of 13F filings is that they are not affected by the selection and survivorship bias that occurs when relying on TASS and other self-reported databases for hedge fund identification (Agarwal, Fos, and Jiang 2010).

Data in the 13F filings have a number of known limitations. First, small institutions that fall below the reporting threshold ($100 million in U.S. equity) at the end of the year are not in the sample the following year. Second, institutions are not required to report positions that do not make the threshold of$200,000 and 10,000 shares. Third, short equity positions are not reported. Fourth, the holding reports as collected by Thomson-Reuters are aggregated at the management company level. Nevertheless, as mentioned above, the Thomson-Reuters classification allows us to separately identify the advisors within the management company. Fifth, we only observe end-of-quarter snapshots on hedge fund holdings. Despite these limitations, it must be stressed that our data are not plagued by survivorship bias as they also contain the filings of defunct hedge fund companies.

Because many financial advisors manage hedge-fund-like operations alongside other investment management services, we need to apply a number of filters to the data to ensure that the hedge fund business is the main line of operation for the institutions in our sample. Therefore, we drop institutions that have advisors with a majority of non-hedge-fund business, even though such institutions have hedge funds that are managed in-house and included with their holdings in the parent management company's 13F report. Thomson-Reuters's hedge fund list also provides the classification of non-hedge-fund entities that file under the same 13F entity. We use this list to screen out all companies with other reported non-hedge-fund advisors that file their 13F holdings along with their hedge funds. Additionally, we manually verify that large investment banks and prime brokers that might have internal hedge fund business are excluded from our list (e.g., Goldman Sachs Group, JP Morgan Chase & Co., American International Group Inc.). As a further filter, we double-check the hedge fund classification by Thomson-Reuters against a list of ADV filings by investment advisors since 2006, when available.13 We match those filings by advisor name to our 13F data. Then, following Brunnermeier and Nagel (2004) and Griffin and Xu (2009), we keep only the institutions with more than half of their clients classified as “High Net Worth Individuals” or “Other Pooled Investment Vehicles (e.g., Hedge Funds)” in Item 5.D (Information About Your Advisory Business) of Form ADV. Therefore, we believe that our final list of hedge funds contains only institutions with the majority of their assets and reported holdings in the hedge fund business, which we label “pure-play” hedge funds. Our final sample covers 79.8% of the number of 13F institutions that have any hedge fund business, which makes 25.3% of their aggregate equity portfolio. The institutions that are excluded from our sample provide a variety of other asset management and trading services, such as wealth management advisory services and brokerage services.

We augment our data with hedge fund characteristics and monthly returns from the Thomson-Reuters's Lipper-TASS database (drawn in July 2010).14 We use both the “Graveyard” and “Live” databases. We use hedge fund company names in TASS and map them to the advisor company name that appears in 13F filings. The Lipper-TASS database provides hedge fund characteristics (such as investment style and average leverage) and monthly return information at the strategy level. We aggregate the TASS data at the management company level on a quarterly frequency and match it to the 13F dataset using the consolidated management company name.15 To avoid potential data errors, especially arising from the fact that not all funds report assets under management to TASS, we exclude management companies for which the ratio of 13F assets to assets under management from TASS exceeds ten. This filter drops about 8% of the observations. Further, we exclude hedge funds with less than $1 million in total assets under management (0.6% of the observations), in order to ensure that our results are not driven by hedge funds with insignificant holdings. As argued in the introduction, we focus on the years surrounding the recent financial crisis; our sample starts in the first quarter of 2004. The sample-end coincides with the end of the 13F data availability (2009:Q4). Finally, for the fund-level regressions, we winsorize fund flows and changes in hedge fund equity holdings at the 5th and 95th percentiles within each quarter, as the distributions of these variables have fat tails. Panel A of Table 1 provides annual statistics for our sample of hedge funds. The first three columns show a rapid increase in the number of hedge funds and assets under management (AUM) up to 2007. The subsequent decline in the number of matched TASS funds and AUM is consistent with the recent patterns of hedge fund liquidations at the end of 2008 and in the first three quarters of 2009. The slow increase in the number of 13F funds in 2008–2009 (Column 1) is due to smaller new funds that do not report to TASS. According to Hedge Fund Research Inc., the total assets managed by hedge funds had, by 2009, decreased by around 19% due to the market crisis and the record-setting hedge fund closures in 2008 and 2009.16 This pattern is strongly reflected in Figure 1, which plots hedge fund equity holdings over time as a fraction of the total market capitalization. Panel A in Table 1 also provides summary statistics on quarterly portfolio turnover. As in Wermers (2000), Brunnermeier and Nagel (2004), and the CRSP mutual fund database, portfolio turnover is defined as the minimum of the absolute values of buys and sells during a quarter $$q$$ divided by the total holdings at the end of quarter $q−1$ , where buys and sells are measured with end-of-quarter $q−1$ prices. This definition of turnover captures trading unrelated to inflows or outflows. Because it is computed from quarterly snapshots, it is understated, but it nevertheless provides an important assessment of the relevance of quarterly holdings data. The average quarterly turnover in the sample is 39.4%. The magnitude of the turnover in our data is comparable to that found by Brunnermeier and Nagel (2004), and is higher than the 18.2% (quarterly) turnover for mutual funds in 1994 found by Wermers (2000) and the 14.2% quarterly turnover for the mutual funds in our sample. Despite the high turnover, a substantial part of the portfolio holdings survives on the quarterly horizon. As argued by Brunnermeier and Nagel, this finding legitimates the use of quarterly snapshots to capture the low-frequency component of hedge fund trading. Table 1 Summary statistics  Panel A: Hedge-fund level, by year Number of Mgrs Equity portfolio ($m, TASS match) Equity portfolio ($m, whole sample) Number of Stocks per manager Quarterly portfolio turnover Year 13F TASS match Total AUM in TASS ($bn) Mean Mean Median St. dev. Mean Median St. dev. Mean Median St. dev. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) 2004 436 104 93 466 754 254 1,810 104 47 197 0.45 0.40 0.32 2005 530 124 112 597 851 279 1,996 105 45 215 0.42 0.38 0.28 2006 606 133 147 747 901 259 2,286 106 41 235 0.42 0.38 0.29 2007 693 136 189 910 1011 286 2,762 102 38 228 0.41 0.37 0.29 2008 696 114 149 610 667 164 1,872 80 29 203 0.33 0.29 0.25 2009 621 98 147 521 611 139 1,605 81 29 200 0.47 0.40 0.41
 Aggregate Variables Description HF holdings over market cap (%) Total stock market hedge fund holdings in $, scaled by the total market capitalization. Data source: 13F, CRSP, Thomson-Reuters. Δ HF Holdings (%, share of equity holdings) Quarterly change in total hedge fund holdings at the previous quarter’s prices. For each stock, we total the changes in the number of shares owned by hedge funds and multiply by last-quarter prices. We aggregate across all stocks and scale by the value of the total hedge fund holdings in the previous quarter. Data source: 13F, CRSP, Thomson-Reuters. Δ HF Holdings (%, share of market cap) Quarterly change in total hedge fund holdings at the previous quarter’s prices. For each stock, we total the changes in the number of shares owned by hedge funds and multiply by last-quarter prices. We aggregate across all stocks and scale by the total market capitalization in the previous quarter. Data source: 13F, CRSP, Thomson-Reuters. MF holdings over market cap (%) Total stock market mutual fund holdings in$, scaled by the total market capitalization. Data source: CRSP Mutual Funds, CRSP. Δ MF Holdings (%, share of market cap) Quarterly change in total mutual fund holdings at the previous quarter’s prices. For each stock, we total the changes in the number of shares owned by mutual funds and multiply by last-quarter prices. We aggregate across all stocks and scale by the total market capitalization in the previous quarter. Data source: CRSP Mutual Funds, CRSP. Other institutional holdings over market cap (%) Total stock market holdings by other institutions in $, scaled by the total market capitalization. “Other institutions” are defined as institutions that report the 13F, except for hedge funds and mutual funds. Data source: 13F, CRSP. Δ Other institutional holdings (%, share of market cap) Quarterly change in the total holdings of other institutions at the previous quarter’s prices. For each stock, we total the changes in the number of shares owned by other institutions and multiply by last-quarter prices. We aggregate across all stocks and scale by the total market capitalization in the previous quarter. “Other institutions” are defined as institutions that report the 13F, except for hedge funds and mutual funds. Data source: 13F, CRSP. Non-institutional holdingsover market cap (%) Total stock market holdings by non-institutional investors in$, scaled by the total market capitalization. For each stock, “Non-institutional investor ownership” is defined as one minus the fraction held by 13F institutions. Data source: 13F, CRSP Mutual Funds, CRSP, Thomson-Reuters. Δ Non-institutional holdings (%, share of market cap) Quarterly change in total non-institutional investor holdings at the previous quarter’s prices. Calculated as the sum of the changes in the number of shares owned by non-institutional investors, multiplied by last-quarter prices, and scaled by the last-quarter total stock market capitalization. For each stock, “Non-institutional investor ownership” is defined as one minus the fraction held by 13F institutions. Data source: 13F, CRSP Mutual Funds, CRSP, Thomson-Reuters. Short interest ratio (SIR) (%) Total short interest in $, scaled by the total market capitalization. The total short interest is obtained by aggregating stock-level short interest. Data source: Compustat, CRSP, Exchanges. Δ Short interest ratio (Δ SIR) (%, share of short interest) Quarterly change in the total short interest at the previous quarter’s prices. Calculated as the sum of the changes in stock-level short interest across stocks, multiplied by the previous quarter’s prices, and scaled by the total short interest in the previous quarter. Data source: Compustat, CRSP, Exchanges. Δ Short interest ratio (Δ SIR) (%, share of total shares outstanding) Quarterly change in the total short interest, at the previous quarter’s prices. Calculated as the average of the changes in short interest across stocks, as a percentage of their total shares outstanding. Data source: Compustat, CRSP, Exchanges. Stock-level Data Total hedge fund holdings (%) The total number of shares owned by hedge funds, scaled by the total shares outstanding. Data source: TASS. Δ Total hedge fund holdings (%) Quarterly change in the stock’s hedge fund holdings. Calculated as the change in the number of shares owned by hedge funds over the quarter, scaled by the total shares outstanding. Data source: TASS. Short interest ratio (SIR) (%) Short interest (the sum of shares shorted) scaled by the total shares outstanding. Data source: Compustat, Exchanges. Δ Short interest ratio(SIR) (%) Quarterly change in the stock’s short interest. Calculated as the change in the number of shares shorted over the quarter, scaled by the total shares outstanding. Data source: Compustat, Exchanges. Volatility Previous 24-month return volatility. Data source: CRSP. High volatility indicator Indicates whether the stock has an above-median volatility within the month. Amihud ratio Stock liquidity is measured by the average ratio of the absolute value of daily returns to daily volume in the quarter (Amihud 2002). Data source: CRSP. High Amihud ratio indicator Indicates whether the stock has an above-median Amihud ratio within the month. Size ($ million) Market capitalization in $m. Data source: CRSP. High size indicator Indicates whether the stock has an above-median size within the month. Book-to-market Book value of assets (from the most recent 10Q filing) divided by the market value of equity at quarter end. Data source: Compustat, CRSP. High book-to-market ratio indicator Indicates whether the stock has an above-median book-to-market within the month. Past 6m ret Cumulative past six-month returns. Data source: CRSP. High past 6m ret indicator Indicates whether the stock has an above-median past 6-month return within the month. Hedge-fund-level Data Selloff quarter Indicator variable for 2007:Q3–Q4, 2008:Q3–Q4. Pre-crisis Indicator variable for 2004:Q1–2007:Q2. Crisis Indicator variable for 2007:Q3–2009:Q1. Post-crisis Indicator variable for 2009:Q2–2009:Q4. Δ HF Holdings (%, share of equity holdings) The value of shares added to a hedge fund’s portfolio multiplied by the previous quarter’s prices minus the value of shares sold from the portfolio multiplied by the previous quarter’s prices, scaled by the total value of the equity portfolio in the previous quarter. Data source: CRSP, 13F, Thomson-Reuters. Fund flows (%, share of AUM) Quarterly change in assets under management less the total returns over the quarter divided by assets under management in the previous quarter. Data source: TASS. Hedge Fund total return (%) Total return to investors (as reported). Data source: TASS. Equity portfolio return (%) Quarterly returns based on the quarterly change in the hedge fund long equity holdings from 13F, assuming that trades occur at quarter-end prices. Data source: CRSP, 13F, Thomson-Reuters. Assets under management (log(AUM)) Logged assets under management (AUM) as reported in TASS. Data source: TASS. Average (Avg) leverage Average leverage, as reported in TASS in August 2007. Data source: TASS. Multi-asset strategy dummy A dummy for whether the hedge fund has more than 50% of its AUM in one of the following strategies: convertible arbitrage, emerging markets, fixed income arbitrage, fund of funds, global macro, managed futures, or multi-strategy. Data source: TASS. Lockup period indicator An indicator of whether the fund has a lockup period: a period following an investment during which investors are not allowed to redeem their investment (in months). Data source: TASS. Redemption period > 90 days indicator An indicator of whether the sum of redemption notice and redemption frequency exceeds 90 days. Data source: TASS. Frank 3-month performance ranking between 0 and 1. Sorting could be across all hedge funds on a particular date, or within a style-date. Data source: TASS. TRank1 The minimum between 1/3 and FRank. Data source: TASS. TRank2 The minimum between 1/3 and FRank – TRank1. Data source: TASS. TRank3 The minimum between 1/3 and FRank – TRank1 – TRank2. Data source: TASS. Hedge Fund indicator An indicator of whether the entity is a hedge fund (as opposed to a mutual fund). Hedge Fund with constraints indicator Indicates whether the hedge fund has liquidity restrictions due to a lockup period or because their redemption notice period is longer than 30 days. Data source: TASS. Institutional ownership Institutional ownership calculated using self-reported data on ADV filings. Investors are considered institutional if they are not individuals or high-net-worth individuals. Data source: ADV. Mutual Funds Mutual funds’ quarterly returns Quarterly returns. Data source: CRSP Mutual Fund Database. Mutual fund flows The amount of investor funds that entered/exited mutual funds in a particular quarter. Calculated as the different in assets under management on quarter t minus the assets under management on quarter t – 1 times (1 + r), where r is the return of the mutual fund. Data source: CRSP Mutual Fund Database. Mutual fund trades The aggregate net change in mutual funds’ holdings. Calculated as the change in the total number of shares multiplied by the last quarter’s prices. Data source: CRSP, 13F. Frank 3-month performance ranking between 0 and 1. Sorting could be across all mutual funds on a particular date. Data source: CRSP Mutual Fund Database. TRank1 The minimum between 1/3 and FRank. Data source: CRSP Mutual Fund Database. TRank2 The minimum between 1/3 and FRank – TRank1. Data source: CRSP Mutual Fund Database. TRank3 The minimum between 1/3 and FRank – TRank1 – TRank2. Data source: CRSP Mutual Fund Database. Appendix Table A1. The determinants of hedge fund trades: Predicted flows  Dependent variable: Δ HF equity portfolio (%) Investor redemptions Lender Pressure All financial constraints (1) (2) (3) (4) (5) Selloff quarter -11.078*** -5.238 -12.243*** -6.637 -1.406 (-3.386) (-1.248) (-4.606) (-1.387) (-0.212) × Predicted flows (q+1) 0.789** 1.040** (2.284) (2.670) × Avg. leverage -6.471* -5.403* (-1.863) (-1.940) Predicted flows (q+1) 0.698*** 0.437 (3.382) (1.502) Avg. leverage 3.792** 3.752** (2.509) (2.487) Constant 10.002*** 10.407*** 10.622*** 7.401*** 7.747*** (5.810) (6.220) (6.640) (3.542) (3.747) Observations 1838 1838 1180 1180 1180 Adj R2 0.011 0.028 0.013 0.018 0.033  Dependent variable: Δ HF equity portfolio (%) Investor redemptions Lender Pressure All financial constraints (1) (2) (3) (4) (5) Selloff quarter -11.078*** -5.238 -12.243*** -6.637 -1.406 (-3.386) (-1.248) (-4.606) (-1.387) (-0.212) × Predicted flows (q+1) 0.789** 1.040** (2.284) (2.670) × Avg. leverage -6.471* -5.403* (-1.863) (-1.940) Predicted flows (q+1) 0.698*** 0.437 (3.382) (1.502) Avg. leverage 3.792** 3.752** (2.509) (2.487) Constant 10.002*** 10.407*** 10.622*** 7.401*** 7.747*** (5.810) (6.220) (6.640) (3.542) (3.747) Observations 1838 1838 1180 1180 1180 Adj R2 0.011 0.028 0.013 0.018 0.033 The table reports results from the OLS fund-level regressions in which the dependent variable is hedge fund trades as a fraction of the hedge fund equity portfolio, evaluated at prior-quarter prices. The explanatory variables include the selloff quarter dummy and the level and interactions of predicted fund flows (as of quarter q + 1), and average leverage. Selloff quarters are 2007:Q3–Q4 and 2008:Q3–Q4. Predicted flows are the fitted values from a regression of flows in quarter q + 1 on to total hedge fund returns in quarters q, q – 1, and q – 2. Standard errors are clustered at the calendar quarter level. t-statistics are reported in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. The sample period is 2004:Q1 to 2009:Q4. ## References Agarwal V Daniel N Naik NY Flows, Performance, and Managerial Incentives in Hedge Funds 2004 Working Paper, Georgia State University and London Business School ——— Role of Managerial Incentives and Discretion in Hedge Fund Performance Journal of Finance , 2009 , vol. 64 (pg. 2221 - 56 ) Agarwal V Fos V Jiang W Inferring Reporting Biases in Hedge Fund Databases from Hedge Fund Equity Holdings 2010 Working Paper, Columbia University Agarwal V Naik NJ Risk and Portfolio Decisions Involving Hedge Funds Review of Financial Studies , 2004 , vol. 17 (pg. 63 - 98 ) Amihud Y Illiquidity and Stock Returns: Cross-section and Time-series Effects Journal of Financial Markets , 2002 , vol. 5 (pg. 31 - 56 ) Anand A Irvine P Puckett A Venkataraman K Market Crashes and Institutional Trading 2010 Working Paper, Syracuse University Ang A Gorovyy S van Inwegen GB Hedge Fund Leverage 2010 Working Paper, Columbia University Aragon GO Strahan P Forthcoming Hedge Funds as Liquidity Providers: Evidence from the Lehman Bankruptcy. Journal of Financial Economics 2011 Baele L Bekaert G Inghelbrecht K The Determinants of Stock and Bond Return Comovements Review of Financial Studies , 2010 , vol. 23 (pg. 2374 - 428 ) Baquero G Verbeek M A Portrait of Hedge Fund Investors: Flows, Performance, and Smart Money 2009 Working Paper, ESMT European School of Management and Technology and Erasmus University Billio M Getmansky M Pelizzon L Crisis and Hedge Fund Risk 2010 Working Paper, University of Venice Billio M Getmansky M Lo AW Pelizzon L Econometric Measures of Systemic Risk in the Finance and Insurance Sectors Working Paper, Massachusetts Institute of Technology , 2010 Boehmer E Jones CM Which Shorts Are Informed? Journal of Finance , 2008 , vol. 63 (pg. 491 - 527 ) Boyson NM Helwege J Jindra J Crises, Liquidity Shocks, and Fire Sales at Financial Institutions 2010 Working Paper, Northeastern University Boyson NM Stahel CW Stulz RM Hedge Fund Contagion and Liquidity Journal of Finance , 2010 , vol. 65 (pg. 1789 - 816 ) Brown DB Carlin BI Lobo MS Optimal Portfolio Liquidation with Distress Risk Management Science , 2010 , vol. 56 (pg. 1997 - 2014 ) Brown GW Green J Hand JRM Are Hedge Funds Systemically Important? 2010 Working Paper, University of North Carolina at Chapel Hill Brunnermeier MK Nagel S Hedge Funds and the Technology Bubble Journal of Finance , 2004 , vol. 59 (pg. 2013 - 40 ) Brunnermeier MK Pedersen LH Market Liquidity and Funding Liquidity Review of Financial Studies , 2009 , vol. 22 (pg. 2201 - 38 ) Calvet LE Campbell JY Sodini P Fight or Flight? Portfolio Rebalancing by Individual Investors Quarterly Journal of Economics , 2009 , vol. 124 (pg. 301 - 48 ) Campbell JY Ramadorai T Schwartz A Caught on Tape: Institutional Trading, Stock Returns, and Earnings Announcements Journal of Financial Economics , 2009 , vol. 92 (pg. 66 - 91 ) Cao C Chen Y Liang B Lo AW Can Hedge Funds Time Market Liquidity? 2009 Working Paper, Pennsylvania State University Cella C Ellul A Giannetti M Investors' Horizons and the Amplification of Market Shocks 2011 Working Paper, Stockholm School of Economics and Indiana University Chevalier J Ellison G Risk-taking by Mutual Funds as a Response to Incentives Journal of Political Economy , 1997 , vol. 105 (pg. 1167 - 200 ) Chordia T Sarkar A Subrahmanyam A An Empirical Analysis of Stock and Bond Market Liquidity Review of Financial Studies , 2005 , vol. 18 (pg. 85 - 129 ) Coval J Stafford E Asset Fire Sales (and Purchases) in Equity Markets Journal of Financial Economics , 2007 , vol. 86 (pg. 479 - 512 ) Ding B Getmansky M Liang B Wermers R Share Restrictions and Investor Flows in the Hedge Fund Industry 2009 Working Paper, State University of New York at Albany Fama EF French KR The Cross-section of Expected Stock Returns Journal of Finance , 1992 , vol. 47 (pg. 427 - 65 ) ——— Common Risk Factors in the Returns on Stocks and Bonds Journal of Financial Economics , 1993 , vol. 33 (pg. 3 - 56 ) Fung W Hsieh DA Empirical Characteristics of Dynamic Trading Strategies: The Case of Hedge Funds Review of Financial Studies , 1997 , vol. 10 (pg. 275 - 302 ) ——— Hedge Fund Benchmarks: A Risk-based Approach Financial Analyst Journal , 2004 , vol. 60 (pg. 65 - 80 ) Getmansky M Lo AW Makarov I An Econometric Model of Serial Correlation and Illiquidity in Hedge Fund Returns Journal of Financial Economics , 2004 , vol. 74 (pg. 529 - 610 ) Goetzmann WN Ingersoll JE Ross SA High-water Marks and Hedge Fund Management Contracts Journal of Finance , 2003 , vol. 58 (pg. 1685 - 718 ) Sachs Goldman Hedge Fund Monitor 2010Q2 Portfolio Strategy Report , 2010 Goyenko R Stock and Bond Pricing with Liquidity Risk 2006 Working Paper, McGill University Goyenko R Ukhov A Stock and Bond Market Liquidity: A Long-run Empirical Analysis Journal of Financial and Quantitative Analysis , 2009 , vol. 44 (pg. 189 - 212 ) Griffin J Xu J How Smart Are the Smart Guys? Unique View from Hedge Fund Stock Holdings Review of Financial Studies , 2009 , vol. 22 (pg. 2531 - 70 ) Gromb D Vayanos D Equilibrium and Welfare in Markets with Financially Constrained Arbitrageurs Journal of Financial Economics , 2002 , vol. 66 (pg. 361 - 407 ) ———. Limits of Arbitrage: The State of the Theory Annual Review of Financial Economics , 2010 , vol. 2 (pg. 251 - 75 ) Hau H Lai S The Role of Equity Funds in the Financial Crisis Propagation. Working Paper, INSEAD 2011 He Z Khang IG Krishnamurthy A Balance Sheet Adjustments in the 2008 Crisis 2010 Working Paper, University of Chicago Hombert J Thesmar D Limits of Limits of Arbitrage: Theory and Evidence. Working Paper, HEC 2009 Jegadeesh N Titman S Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency Journal of Finance , 1993 , vol. 48 (pg. 65 - 91 ) Jotikasthira C Lundblad C Ramadorai T Asset Fire Sales and Purchases and the International Transmission of Financial Shocks. Working Paper, Said Business School 2009 Khandani AE Lo AW . What Happened to the Quants in August 2007? Evidence from Factors and Transactions Data Journal of Financial Markets , 2011 , vol. 14 (pg. 1 - 46 ) Lamont OA Stein JC Aggregate Short Interest and Market Valuations American Economic Review , 2004 , vol. 94 (pg. 29 - 32 ) Li H Zhang X Zhao R Investing in Talents: Manager Characteristics and Hedge Fund Performances Journal of Financial and Quantitative Analysis , 2011 , vol. 46 (pg. 59 - 82 ) Lemke TP Lins GT Disclosure of Equity Holdings by Institutional Investment Managers: An Analysis of Section 13(f) of the Securities Exchange Act of 1934 Business Lawyer , 1987 , vol. 43 (pg. 93 - 119 ) Manconi A Massa M Yasuda A The Behavior of Intoxicated Investors: The Role of Institutional Investors in Propagating the Crisis of 2007–2008 2010 Working Paper, University of California at Davis Metrick A Gorton GB Haircuts 2010 Working Paper, Yale University Mitchell M Pedersen L Pulvino T Slow-moving Capital American Economic Review P&P , 2007 , vol. 97 (pg. 215 - 20 ) Mitchell M Pulvino T Arbitrage Crashes and the Speed of Capital. Working Paper, CNH Partners 2011 Nagel S Evaporating Liquidity 2011 Working Paper, Stanford University Sadka R Liquidity Risk and the Cross-section of Hedge-fund Returns Journal of Financial Economics , 2010 , vol. 98 (pg. 54 - 71 ) Scholes M Crisis and Risk Management American Economic Review , 2000 , vol. 90 (pg. 17 - 21 ) Shleifer A Vishny RW The Limits of Arbitrage Journal of Finance , 1997 , vol. 52 (pg. 35 - 55 ) Sias R Starks L Titman S Changes in Institutional Ownership and Stock Returns: Assessment and Methodology Journal of Business , 2006 , vol. 79 (pg. 2869 - 910 ) Sirri ER Tufano P Costly Search and Mutual Fund Flows Journal of Finance , 1998 , vol. 53 (pg. 1589 - 622 ) Teo M The Liquidity Risk of Liquid Hedge Funds Journal of Financial Economics , 2011 , vol. 100 (pg. 24 - 44 ) Vayanos D Flight to Quality, Flight to Liquidity, and the Pricing of Risk Working Paper , 2004 London School of Economics Wermers R Mutual Fund Performance: An Empirical Decomposition into Stock-picking Talent, Style, Transactions Costs, and Expenses Journal of Finance , 2000 , vol. 55 (pg. 1655 - 95 ) ——— Money Fund Runs 2010 Working Paper, University of Maryland 1 Focusing on merger arbitrage following the 1987 crash and on convertible arbitrage in 2005, Mitchell, Pedersen, and Pulvino (2007) provide evidence that redemptions forced hedge funds to turn from liquidity providers into liquidity demanders. Mitchell and Pulvino (2011) focus on a set of relative value strategies during the recent crisis and argue that the disappearance of long-term financing caused arbitrageurs to withdraw liquidity from these markets, generating further price divergence. 2 He, Khang, and Krishnamuruthy (2010) report that the aggregate value of assets owned by hedge funds declined by$800bn. While their work portrays the balance sheets of institutions during the crisis, our study, in contrast, focuses on the active selling of U.S. stocks by hedge funds.
3
These figures translate to an exit of 0.2%, on average, of the total market capitalization in each of the third and fourth quarters of 2007, and to 0.4%, on average, of the total market capitalization in each of the third and fourth quarters of 2008.
4
Because hedge fund short equity positions are not disclosed, we rely on the conjecture that most short selling is performed by hedge funds. Goldman Sachs (2010) estimates that as of March 2010, 85% of all equity short positions going through their brokerage house is performed by hedge funds.
5
Anand et al. make similar observations about trading during the crisis for the universe of institutions.
6
Also since our article, Brown, Green, and Hand (2010) have argued that fire sales were not a widespread phenomenon during the crisis because many funds did not experience negative alphas. In our view, the absence of negative alphas is not by itself evidence of a lack of fire sales, especially if the distress condition is reflected in the risk factors themselves. Rather, one can interpret the fact that hedge funds' inability to capture underpriced securities (testified to by the lack of positive alphas) as consistent with our finding of severe financial constraints for the hedge fund sector in a depressed equity market.
7
According to Lemke and Lins (1987), Congress justified the adoption of Section 13F of the Securities Exchange Act in 1975 because, among other reasons, it facilitates consideration of the influence and impact of institutional managers on market liquidity: “Among the uses for this information that were suggested for the SEC were to analyze the effects of institutional holdings and trading in equity securities upon the securities markets, the potential consequences of these activities on a national market system, block trading, and market liquidity…
8
Specifically with regard to equity, this provision concerns all long positions greater than 10,000 shares or $200,000 over which the manager exercises sole or shared investment discretion. The official list of Section 13F securities can be found on the following SEC webpage: http://www.sec.gov/divisions/investment/13Flists.htm. More general information about the requirements of Form 13F pursuant to Section 13F of the Securities Exchange Act of 1934 can be found at http://www.sec.gov/divisions/investment/13Ffaq.htm. 9 13F filings have been used intensively for research concerning the role of institutional investors in financial markets. Sias, Starks, and Titman (2006) study the sources of correlation between institutional trades and returns. Brunnermeier and Nagel (2004) explore the behavior of hedge funds during the Internet bubble. Campbell, Ramadorai, and Schwartz (2009) combine 13F filings with intraday data to explore the behavior of institutional investors around earnings announcements. 10 This comprehensiveness depends on Thomson's long-lasting and deep involvement with institutional filings. The SEC has long contracted the collection of various institutional data out to Thomson-Reuters, dating back to when those reports were paper filings or microfiche in the public reference room. References to Thomson-Reuters (or the companies that it acquired, such as CDA/Spectrum, which was formerly known as Disclosure Inc. and Bechtel) can be found at: 1. http://www.sec.gov/rules/final/33-8224.htm (search for “Thomson”). 2. SEC Annual Reports, 1982, http://www.sec.gov/about/annual_report/1982.pdf (page 37 or 59 of the PDF file). 3. http://www.sec.gov/rules/final/33-7432.txt (search for “contractor”). 4. http://www.sec.gov/about/annual_report/1989.pdf (search for “contractor”). 11 For example, for Blackstone Group holdings in 13F data, Thomson-Reuters provides a classification of each of the advisors within Blackstone that reported their holdings in the same filing. There are three advisor entities within Blackstone Group L.P. that report their holdings in the same consolidated Blackstone Group report. Among the three advisors included, GSO Capital Partners and Blackstone Kailix Advisors are classified by Thomson-Reuters as Hedge Funds (which an ADV form confirms), while Blackstone Capital Partners V L.P. is classified as an Investment Advisor. See the “List of Other Included Managers” section in the September 30, 2009, Blackstone 13F reports filed on November 16, 2009: http://www.sec.gov/Archives/edgar/data/1393818/000119312509235951/0001193125-09-235951.txt. 12 As a shortcut, from now on we will refer to the observational unit in our dataset as a “hedge fund.” It should be clear, however, that 13F provides asset holdings at the management firm level, or at the advisor entity level, when a management firm and its advisors are different entities. Each firm/advisor reports consolidated holdings for all the funds it has under management. 13 All current advisor ADV filings are available on the SEC's investment advisor public disclosure website: http://www.adviserinfo.sec.gov. The ADV filings were mandatory for all hedge funds only for a short time in 2006. After that point, they were filed on a voluntary basis. 14 While we use a recent TASS data feed (July 2010) in our analysis, we use an older version (August 2007) to identify firms (because it includes hedge fund names). 15 We use strategy assets under management as weights in aggregating fund characteristics and total reported returns. 16 See BusinessWeek's article “Hedge Your Bets Like the Big Boys” by Tara Kalwarski, in the December 28, 2009, issue. 17 We first adjust shares held for splits and distributions. We then use the quarterly holding snapshots to derive the trades and make sure that we are filtering out changes in holdings that originate from changes in the universe of 13F filers. For this reason, we require hedge funds to appear in two consecutive quarters. When a hedge fund does not report (because it is below the$100 million assets-under-discretion cutoff), we eliminate the observation (as opposed to reporting a large drop in holdings). More details about the sample construction and trade derivation are available as a WRDS research application with the SAS code: “Institutional Trades, Flows, and Turnover Ratios using Thomson-Reuters 13F data,” http://wrds.wharton.upenn.edu/.
18
Notice that while Figure 1 does not show a drop in stock market participation in the fourth quarter of 2007, the decrease in the hedge fund portfolio is evident from Table 2. The difference in the two statistics results from the fact that the ratio in Figure 1 is affected by the change in market prices in both the numerator and denominator. To filter out this effect, one needs to rely on Table 2, which focuses on trades evaluated at prior-period prices.
19
Coval and Stafford (2007) and Hau and Lai (2011) find that fire sales affect the prices of other securities held by the same entities.
20
This method of imputing non-institutional investors' holdings provides an upper bound. The reason is that 13F filings do not include institutions that do not reach the \$100 million threshold. However, given the small size of the excluded institutions, we believe the approximation error to be modest.
21
2 * (–5.982% + 4.476%) = –3.012%.
22
1 – (–2.653% / –12.118%) = 78.135%.
23
Amihud (2002) computes a stock-level illiquidity measure as the average of the absolute value of daily returns over the daily dollar volume.
24
The number of shares is adjusted for stock splits.
25
Mutual funds' performance is compared to the universe of equity mutual funds in our database in the same quarter. For hedge funds, we offer two benchmark groups: either the universe of hedge funds in the same quarter or hedge funds of the same investment style in the same quarter. Table 7 presents results for both groups.
26
To see this result, add the coefficients on TRank1 and on the interactions with the indicators for hedge funds and constrained hedge funds.
28
Investor categories include individuals, high-net-worth individuals, banks, mutual funds, pension funds, pooled investment vehicles, endowments, corporations, the government, and other. Ownership fractions are broken into categories (e.g., up to 10%, between 10% and 20%, etc.). For calculating institutional ownership, we compute the midpoint for each relevant category and take the average.