A First Glimpse into the Short Side of Hedge Funds*

Transcription

1 A First Glimpse into the Short Side of Hedge Funds* This Draft: March 2016 Abstract We develop a novel algorithm to identify both the opening and covering of hedge fund short sales by combining trading and holding data of institutional investors. Our algorithm successfully identifies short sales, as evidenced by its capturing a dramatic dip in short sales during the short sale ban in September Using this identification of hedge fund short sales, we show that hedge funds possess skills in both opening and covering of short trades. Their short positions are highly profitable and generate an abnormal return of up to 14 basis points per day for trades covered within a week. This profitability of short sales is at least partly due to informed trading, as hedge fund short sales predict negative earnings surprises. The performance of hedge fund short positions is also persistent. These findings reconcile contrasting results in the literature that hedge funds generate positive abnormal returns but that their long positions are not profitable. Keywords: Hedge funds; short sale profitability

2 1. Introduction It is well documented that short sellers are informed traders and are able to generate superior performance. 1 Although as yet there is no direct evidence regarding the identity of these informed short sellers, the consensus among both academics and practitioners is that they include institutional investors, including hedge funds. Since hedge funds trading is largely secretive except for their long side positions disclosed in the 13F forms, 2 the literature lacks direct evidence on the short side of hedge funds. There is only conjecture regarding the profitability of hedge funds short positions. This paper provides direct evidence regarding the profitability of hedge fund short positions. We identify the opening and covering of short sales by hedge funds and also other institutional investors by developing a novel algorithm that takes advantage of data on both the detailed transactions and holdings of institutional investors. The short sale data that we construct using the algorithm provide unique advantages as compared to the data used in previous research. In particular, we are able to identify both the opening and covering of short and long side trades at the level of individual investment managers. Exploiting this feature of our data on short sales, we examine the profitability of hedge fund short trades, including whether hedge funds have skills in both short opening and covering. This contrasts with existing studies that are unable to identify short sales at the manager level but rather examine whether short sales (by unidentified investors) and changes in aggregate short interest predict future stock returns. We identify the short selling activities of hedge funds by combining two different datasets consisting of trades and holdings of institutional investors. The former database is 1 See, for example, Asquith and Meulbroek (1996), Desai et al (2002), Asquith, Pathak, and Ritter (2005), Diether, Lee, and Werner (2009), and Boehmer, Jones, and Zhang (2008), among many others. 2 Brunnermeier and Nagel (2004) and Griffin and Xu (2009) examine the long side holdings of hedge funds by employing the Form 13F data.

3 provided by ANcerno, a trade execution cost consulting firm, and reports all transactions by a set of institutional investors. 3 These data include the names of fund management companies and both their long and short transactions. We combine these data with institutional holdings obtained from 13F forms, which report only the long positions of fund management companies. The idea of our algorithm is to combine quarterly snapshot of long positions from the 13F data with the ANcerno data on all trades to determine the long or short position of each investment manager at the time of each trade. Since the SEC 13F reporting rule does not allow netting out long and short positions of the same stocks in different accounts by the same management firm, we are able to identify institutional trades as short sales using the ANcerno database if the stocks for those trades do not appear in the long position of the same managers 13F filings. This algorithm can also identify three other types of trades: buys of stocks in short positions (short buys) and buys and sells of stocks in long positions (long buys and long sells). We then categorize investment management firms into hedge funds and non-hedge funds, using the investment managers ADV filings with the SEC. Through this algorithm, we identify short sales, buys that cover short sales, long buys, and sales that close long positions, for 53 hedge fund management companies and 141 nonhedge fund management companies from January 1999 through September We refer to these four types of trades as short sells, short buys, long buys, and long sells, respectively. The data include a total of 2,967,918 daily hedge fund net trades, of which 213,337 are trades that either open or close short positions. As a validity check on whether the algorithm correctly identifies short sales, we examine identified short sales in September and October of 2008, during which period the short sale ban was implemented on a set of financial firms by the SEC. 3 For example, Anand, Irvine, Puckett, and Ventakaraman (2012), Puckett and Yan (2011), and Jame (2014), use this database for investigating institutional investors trading ability.

4 We find that our algorithm identifies short sales successfully, since we show there is a dramatic decrease in short sales by hedge funds in financial companies on the date of the ban and a similar dramatic increase in short sales on the release of the ban, which is also reported in Boehmer, Jones, and Zhang (2013). Using our data on short sales, we first provide evidence about the profitability of short sales by hedge funds in comparison to other institutional investors, or non-hedge funds. To this end, we exploit the unique feature of our data and examine the profitability of the closed positions of hedge fund short sale. We find that hedge fund short sales in our sample are profitable, especially for short horizons. Specifically, hedge fund short trades that are covered within five trading days on average earn an abnormal return based on characteristic-matched portfolios (Daniel, Grinblatt, Titman, and Wermers, 1997) of 14 basis points per day, which translates to greater than 35% per year. In comparison, non-hedge funds do not exhibit significant profitability from short trades. Their short sales covered within five trading days generate on average a loss of 10.8 basis points per day. Longer-term non-hedge short sales covered within one to three months are only weakly profitable, providing an abnormal return of only 2.6 basis point per day. We obtain similar results by examining the abnormal returns of alphas of calendar-time portfolio return regressions using Carhart (1997) four-factor model of Carhart (1997). Having establishd that hedge fund short trades are profitable, we next test whether the profitability in hedge fund short trades is due to informed trading. Specifically, we examine whether increased short selling activities by hedge funds have predictive power for future negative earnings surprises. For this purpose, we construct short intensity measures for each stock using the ratio of total dollar amounts shorted by all hedge fund managers to the total

5 dollar amounts of all trades by all managers. We find that greater short selling intensity by hedge funds predicts more negative earnings surprises. This predictability of negative earnings surprises is concentrated in short horizons, say, within five days, which is consistent with our earlier results that hedge fund short trades are particularly profitable for short horizons. Also, hedge fund short trades that are open during earnings announcement and covered within five trades tend to be profitable. In contrast, short selling by non-hedge funds does not show any correlation with firm s earnings surprises. Lastly, we investigate whether hedge funds are persistently skilled short sellers. Our data set provides a unique advantage over previous databases to examine short selling skills, since it tracks short trades at the individual manager level, allowing us to track the persistence of hedge fund short positions. In particular, we examine whether past successful short sellers who are profitable on their short positions also profit from their short positions in the following periods. We find evidence suggesting that hedge funds possess short selling skills. The portfolio of hedge funds in the highest quintile of profitability on their past short positions generates negative returns in subsequent quarters, implying that these funds make consistent profits in their short positions. In sum, our results indicate that hedge funds are skilled not only in opening but also in covering short positions, as evidenced by our profitability analysis. Hedge fund short trades are profitable especially for short-horizon trades that are covered less than a week. In addition, we find this profitability is consistent with informed trading by hedge funds, as they can also predict negative earnings surprises. Through a thorough examination of hedge fund profitability on their short trades, our paper makes an important contribution to the literature. As noted, previous studies, e.g., Griffin

6 and Xu (2009), find that hedge funds long equity holdings do not generate significant positive abnormal performance. It is thus of particular interest whether hedge funds have superior performance with their short positions. Our paper adds to this literature by documenting that hedge funds are profitable on their short portfolios and thus complement the existing results in the literature. Also, although there are existing studies that investigate what kinds of short sales are informed based on account types (e.g., Boehmer, Jones, and Zhang 2008), and while Kelly and Tetlock (2014) examine short selling by retail investors, it has not been studied yet how particular types of institutional investors, e.g. hedge funds, perform with short sales. To our knowledge, we provide the first piece of evidence regarding the performance of hedge fund short sales. Furthermore, we make a contribution to the literature by documenting that hedge fund trades are profitable and thus they have skills in both opening and covering short trades, unlike most previous studies that focus on short openings. 2. Data and Variable Construction Our analysis of hedge fund short sales requires identification of institutional investors and their holdings. We provide the description of the main data for institutional investor transactions and also describe the matching process between institutional transaction data with the holdings data in the Form 13F filings. We then explain our algorithm on hedge fund short sale transactions Institutional Trading Data from ANcerno We use a dataset of transaction-level institutional trading data from ANcerno. The dataset, over a rolling four-quarter period ending 3Q-2010, consists of 133 billion traded shares

7 with market equity of $3.4 trillion in over 8,390 stocks and ADRs. It covers quite a representative set of transactions for institutional investors in the database. For example, Agarwal et al. (2014), Anand et al. (2012), Bethel et al. (2009), Chemmanur et al.(2009), Goldstein et al.(2009), Green and Jame (2011), Green et al. (2014), Hu (2009), and Hu et al. (2014) also employ the database. It covers January 1999 to September 2011 provides institutional manager name, detailed information on institutional stock transactions including CUSIP and ticker identifiers, transaction dates, the numbers of shares transacted, transaction types (whether transactions are buys or sells), and transaction prices Matching Transaction Data with Hedge Fund Holdings Data We match institutional managers in ANcerno with those in the Thompson 13F database by comparing both the manager names and their quarterly holdings changes. We use the methodology in Chemmanur, He and Hu (2009) to match ANcerno institutional managers to Thompson 13F institutions by comparing cumulative quarterly ANcerno institutional managers trades to institutions quarterly changes in Thompson 13F reports. We only retain matches where the ANcerno institutional manager names exactly match the 13F management company names. By matching on both names and cumulative quarterly trades we ensure that the managers in the ANcerno data and the Thompson 13F reports are the same and that the ANcerno data captures close to all the trades made by the managers during the quarter. We identify management companies as hedge funds based on their SEC Form ADV filings, following the methodology in Jame (2015). We classify a management company (asset manager, or fund company) as a hedge fund if the following two conditions hold. First, item 5D

8 of Form ADV lists more than half of the management company s investors are high net worth individuals or pooled investment vehicles and second, item 5E of Form ADV lists that the management company charges a performance fee. We are able to match 194 ANcerno investors to 13F management companies. We identify 53 of the management companies as hedge funds and 141 as non-hedge funds Short Sales Classification Algorithm After matching the ANcerno investor data to Thompson 13F data, we then compare cumulative net daily ANcerno positions against Thompson 13F positions to identify short sales using the below algorithm. Our main idea is to exploit the difference between the Form 13F filing data and ANcerno transaction data: the Form 13F mainly contains the long side of institutional holdings, while ANcerno reports both long and short transactions. Thus, if we find sell transactions on a stock from ANcerno when there are no corresponding long side holdings, we identify those transactions as short sales. We detail the algorithm below. Let I(t, A, X) represent the inventory of ANcerno institutional manager A in stock X at end of trading day t. Thus, negative values of I(t, A, X) mean A has an open short position in stock X. We initialized I(t, A, X) by setting the variable equal to the number of shares listed in 13F filings where t is the report date of the 13F filing. For example, if on 13F report date q, A states in its 13F that it holds 2,000 shares of X then we set I(q, A, X) = 2,000. The ANcerno data is then used to assign values to I(t, A, X) on trade dates between 13F reports. For example, using the ANcerno data, if at t= q+5, A buys a total of 200 shares of X, then I(q+5, A, X) = 2,200 (i.e. 2, ). If at t = q+6, A sells a total of 3,000 shares of X, then I(q+6, A, X) = -800 (i.e.

9 2, ,000). In a similar manner we use ANcerno trades on dates prior to a 13F report to calculate the value of I(t, A, X). 4 Investor A traded shares of X on dates between 13Fs that report holdings in X: Let Q(1), Q(2),,Q(N) be report dates where A has a 13F report for stock X. Consider the case where A s 13Fs state holdings in stock X for two quarters: Q(n) and Q(n+1). We do not require Q(n) and Q(n+1) to be consecutive end of quarter dates, but we do require Q(n) < Q(n+1). Let I1(t, A, X) represent the inventory of A in stock X at time t with the initial inventory level being determined by the number of shares in stock X at quarter Q(n) 13F. Let I2(t, A, X) represent the inventory of A in stock X at time t with the starting point being the number of shares in stock X at quarter Q(n+1). I1(t, A, X) is updated with every ANcerno listed trade by A in stock X, as t increments from Q(n) to Q(n+1). I2(t, A, X) is updated with every ANcerno trade by A in stock X, as t decrements from Q(n+1) to Q(n). 5 If for the same t, I1(t, A, X) and I2(t, A, X) have different values, we use the following weighting algorithm to determine I(t, A, X). 6 Let T = Q(n+1) Q(n). For example, say T = 90 for consecutive quarters and T= 180 if there is one quarter missing a 13F report. We calculate I(t, A, X) as I(t, A, X) = w * I1(t, A, X) + (1-w) * I2(t, A, X) (1) 4 Adjustments are made for stock splits when mapping trades to portfolio positions. 5 E.g. consider the scenario where institutional manager A reports 13Fs on June 30, 1999 and September 30, 1999 where the manager lists inventories in stock X of 3,000 and 5,000 shares respectively. The ANcerno data reports A buys 1,000 shares of X on July 30, 1999 and a further 995 shares on August 30, The algorithm assigns Q(1) = June 30, 1999 and Q(2) = September 30, Furthermore, I1(June 30, 1999, A,X) = I2(June 30, 1999, A,X) =3,000 and I1(Sep 30, 1999, A,X) = I2(Sep 30, 1999, A,X) = 5,000. Incrementing trades from = June 30, 1999 results in I1(July 30, 1999, A,X) = 4,000 and I1(August 30, 1999, A,X) = 4,995. Decrementing trades from = September 30, 1999 results in I2(August 30, 1999, A,X) = 5,000 and I2(July 30, 1999, A,X) = 4, E.g. in footnote 5 above, we have I1(July 30, 1999, A,X) = 4,000 while I2(July 30, 1999, A,X) = 4,005; and I1(August 30, 1999, A,X) = 4,995 while I2(August 30, 1999, A,X) = 5,000. In this example differences between I1(t, A, X) and I2(t, A, X) occur because the cumulative change in 13F holdings is 2,000 shares and the cumulative ANcerno trades over the period is 1,995 shares.

10 where w = ( Q(n+1) t ) / T. For example, suppose Q(n) and Q(n+1) are 180 days apart, which means T = 180. Further suppose t is ten days after Q(n), then w = ( ) / 180 = 170 / 180. Investor A traded shares of X prior to A s first 13F that reports holdings in X: Let Q(1) be the first end of quarter date that institutional manager A 13F reports holdings in stock X. For all ANcerno trades that occur at time t < Q(1), I(t, A, X) is calculated in a manner similar to I2(t, A, X). Investor A traded shares of X post A s last 13F that reports holdings in X: Let Q(N) be the last end of quarter date that A s 13F reports holdings in X. For all ANcerno trades that occur at time t > Q(N), I(t, A, X) is calculated in a manner similar to I1(t, A, X). Investor A traded shares of X and has never filed a 13F with holdings in X: If A never reports a 13F with holdings in stock X, but ANcerno data show A trades in stock X, we calculate the relative holdings, RI(t, A, X), of A in stock X by assuming the initial relative holdings of A in stock X over the date range of the ANcerno dataset is 0. 7 That is, if the first trading observation in the ANcerno data has A buying a total of 200 shares of X at time t=t, then RI(t, A, X) = 200. Then if at t=t+6, A sells a total of 50 shares, we then calculate RI(t+6, A, X) = 150 (i.e ). We treat RI(t, A, X) as having the same distribution as the true inventory, I(t, A, X), for which we do not know the initial shares held. After calculating RI(t, A, X), we shift down the values of RI(t, A, X), to approximate I(t, A, X). This approach allows us to be conservative in that we underestimate short positions. In other words, most likely our estimate for I(t, A, X), will be greater than the true I(t, A, X). Specifically, we shift RI(t, A, X), so that 7 13F filings only report long positions in stocks.

11 max(i(t, A, X)) for the entire trading period is 0. That is, I(t, A, X) = R I(t, A, X) max (RI(t, A, X)). In the example above, this results in I(t, A, X) = 0 and I(t+6, A, X) = -50. The above algorithm calculates the holdings of A in stock X for scenarios where 1) A has never filed a 13F that reports holdings in X, 2) ANcerno data has A transacting shares of X prior to A s first 13F that reports holdings in X, 3) ANcerno data has A transacting shares of X on dates between 13Fs that report holdings in X, and 4) ANcerno data has A transacting shares of X post A s last 13F that reports holdings in X Identifying Short Sales, Short Buys, Long Sales and Long Buys Using the estimates of I(t, A, X) we are able to identify short positions and short sales. A has a short position (or a negative position) in stock X when I(t, A, X) <0. A short sale occurs when I(t, A, X) is initially non-positive and the net number of daily shares bought in X is negative. Using a similar approach we are able to identify daily transactions as short buys, long sells and long buys. A short buy occurs if I(t, A, X) is initially negative, the net number of shares bought in X is positive and the new value of I(t, A, X) remains negative. A long buy occurs if I(t, A, X) is initially non-negative, and the net number of shares bought in X is positive. A long sell occurs if I(t, A, X) is initially positive, the net number of shares bought in X is negative, and the new value of I(t, A, X) remains positive. This classification scheme allows for short buys and long buys to occur on the same date, as well as long sells and short sells. Once we have the number of shares in short sales, we calculate the dollar volume short sold by multiplying the number of shares in the short sale by the closing day price. Summary statistics of the trade classifications are presented in Table 1. The ANcerno data has 10,368,220 net daily stock-manager trades. Long buys are the majority of the daily

12 trades with a sample size of 4,744,899; followed by long sells (4,707,011), short sells (547,667), and short buys (461,565). Short sells consist of 5.3% of the sample trades. For hedge fund only transactions, of the 2,588,401 net daily stock-hedge fund trades; 1,145,713 are long buys, 1,164,148 are long sells, 168,813 are short sells and 137,270 are short buys. Short sells consist of 6.5% of the sample hedge fund trades. 3. Short Sales During the Financial Crisis: A Reality Check of the Algorithm 3.1. Hedge Fund Short Sales during the Banned Short Selling of Financial Stocks. To verify the validity of our trade classification algorithm, we compare our identified short sales to a period in the market when short sales were banned. Graphs of hedge fund short sales and long sales during the ban are presented in Figure 1. During this period, our algorithm, if correct, should identify short selling by hedge funds close to zero. In an effort to protect the integrity and quality of the securities market and strengthen investor confidence, the Securities and Exchange Commission banned short selling in financial companies on September 19 th, This short selling banned remained in effect until short selling in financial sector companies were allowed to resume on October 9 th, Prior to these regulations, on September 17 th, 2008, the SEC banned naked short selling in all stocks. 9 On September 18th, just prior to the SEC ban on short selling in financial companies, from our algorithm, we calculate hedge funds had an average of just over $25 million in short sales proceeds. Following the ban on short sales, the average dollar value of shares sold short 8 See 9 See

13 drops steeply to close to $0. 10 The average dollar value of financial shares short sold remains close to $0 until the ban is lifted and the average dollar volume short sold by hedge funds rises dramatically to slightly over $40 million on October 9 th, While the ban was intended to halt the slide in the price of financial stocks, in actuality the ban resulted in a short run bump to prices, but by the end of the ban, the financial sector was down nearly 26%. 11 Our algorithm calculates a steep decline short sales on the implementation of the ban and an even greater increase in short sales at the end of the ban. Our algorithm produces results that are consistent with financial events. We also illustrate the dollar-value of long selling by hedge funds in financial stocks. Following the short sale ban, and the temporary increase in the price of financial stocks, hedge funds reduce their long selling in financial stocks. However, unlike the drastic reduction in short sales, the average dollar volume of long sales on September 19 th is $275 million i.e. long selling does not fall anywhere close to $0 on 09/19/08. During the period of the short sale ban, the long selling by hedge fund increases as financial stock prices decline. Average long sales by hedge funds reach $163 million on September 29 th, When the ban on short sales in financial stocks is lifted (on 10/09/08), the proceeds from long sales drop significantly from $157 million the previous day to approximately $25 million as hedge funds resume short selling. The differences in short selling and long selling trading activity by hedge funds during the short sale ban are proof that our algorithm correctly identifies short sales and long sales Hedge Fund Short Sales and the Great Recession 10 The reason the graph does not show short selling falling to $0 during the 09/19/08 to 10/08/08 period is SEC banned short sales in 799 financial companies. We examine short selling in all financial companies based on SIC code. 11 See

14 We further examine the hedge fund short sales during the 2008 financial crisis. Figure 2 illustrates the selling activity of hedge funds from 01/01/2007 to 06/30/2009. We map our identified short selling activity against events during the financial crisis. The timeline of events during the Great Recession is quoted from the Federal Reserve Bank of St. Louis. 12 Short selling by hedge funds reaches its lowest average dollar value on September, 2007 and its highest value on October, October, 2008 was the peak of the Great Recession. There is an increase in short selling by hedge funds in 2007, just prior to Bear Sterns June 2007 disclosure of losses from its subprime mortgages investments. 13 The overall trend of short selling by hedge funds during the Great Recession is an upward trend. Following the peak of the Great Recession and the subsequent market recovery (November 2008 and onwards), there is a sharp decline in hedge fund short selling. While there is an overall trend in hedge fund short selling (increasing as the market declines, decreasing as the market improves), the intermittent movements in short selling can be attributed to actions by the Securities Exchange Commission (SEC). For example, in July, 2008 the SEC issues an emergency order temporarily prohibiting naked short selling in the securities of Fannie Mae, Freddie Mac, and primary dealers at commercial and investment banks and we observe a decline in hedge fund short selling in July, Overall, our calculated short selling by hedge funds is consistent with what one will expect during the financial crisis. In contrast, our calculated long selling by hedge funds (Figure 2b) does not demonstrate similar patterns during the Great Recession. Hedge fund long selling appears to decline over period of the Great Recession as opposed to illustrating an upward time trend. These differences 12 See 13 On June 2007, Bear Stearns needed to bail out its High-Grade Structured Credit Strategies Enhanced Leverage Fund.

15 between hedge fund short selling and long selling show that we are able to distinguish short sales from long sales. The fact that our calculated short sales mirror behavior that is expected during the Great Recession adds credence to our algorithm correctly identifying short sales. 4. The Profitability of Hedge Fund Short Sales There is an extensive literature on the predictability of short sales. Most these studies, however, rely on aggregate measures of short trades and do not have access to account-level data, and do not examine actual profitability of short sales. The advantage of our data over previous databases is that we can figure out both when investors open short trades and also close them and thus can back out the profitability of short sales, while previous studies only examine the ability of opening short trades Calculating Days to Cover Even if hedge fund managers short sale has predictability for negative future stock returns, it is still not enough to say it is profitable. If they close their short position at a poor timing, a manager may end up with mediocre profit from short sales. To capture how a manager incorporates the predictability into profitability, we measure trading days taken for a shorted stock to be covered, or days to cover (DTC). To capture managers profitability of managers short trades, we base our DTC calculation from closed short trades. Specifically, for each short buy transactions in our data, we identify the most recent trading day short sell trades that correspond to the short buy transactions. Suppose a manager has a short position of 100 shares in a stock and buys back ten shares (i.e., short buys) on January 12 th, 2005 and the short position becomes 90 shares. We then track back

16 from January 12 th to find the most recent short sales that resulted in the short position of 100 shares. If there are multiple such short sales on different dates, we assign all these dates as separate opening dates. For example, if the short position goes from 90 to 95 on Jan 5 th, 2005 and 95 to 100 on Jan 6 th, 2005 and thus the most recent short sales are two-five transactions on those two dates, we assign five and four trade days as DTC to the short sales on Jan 5 th and 6 th, respectively. In this sense, our DTC calculation is based on the last-in-first-out (LIFO) rule. We use the LIFO rule also due to our data restriction. Our data starts from January of 1999 and if a stock s short position is opened earlier than January of 1999, we cannot identify the starting date of the short position. Rather than arbitrarily assuming a starting date of a stocks short position, looking at the lastly opened position compared to neighboring closing transaction allows consistent measure of DTC across the whole shorted stocks. For the further analysis based on DTC measure, we only include short sell observations with computable DTC. There are stocks that are shorted but never covered throughout our sample period. These observations take around 20% of our total short sell sample. There are two possible explanations for it. The stocks may be actually not covered during our sample period but covered only after end of our sample period (September 30th, 2011). For the first case, if the last short sales of the stocks appear before September of 2010 and there is no following covering trade of the short position, DTC is going to be longer than 1 year. Since our focus of analysis is in the profitability of short sell trading that is covered shorter than 1 year, excluding the observations does not seriously compromise our conclusion. Table 2 provides the summary statistics of DTC. Panel A compares average length of short side DTC between hedge fund managers and non-hedge fund managers. On average, hedge fund managers DTC is shorter than the non-hedge fund managers by 2.84 days. Panel B shows

17 the distribution of long side DTC between hedge fund managers and non-hedge fund managers. Hedge funds take more days to cover their long sides. Figure 3 shows the comparison of short sell DTC distributions between hedge funds managers and non-hedge fund managers. The distribution of hedge funds DTC is more densely distributed over shorter period than the distribution of non-hedge funds DTC. In contrast, nonhedge funds DTC shows higher frequency in the longer periods. Hedge funds DTC is featured with high spike in the shorter DTC range while non-hedge funds DTC has relatively smoother distribution. The average gap between short sell DTC of hedge funds and that of non-hedge fund, which is suggested in panel 2A, comes from these differences in the distribution Are Hedge Fund Short Sales Profitable? We form five value-weighted portfolios of shorted stocks based on DTC ( DTC<=5, 5<DTC<=21, 21<DTC<=63, 63<DTC<=252, and DTC<=252) to examine the profitability of hedge fund short sales. Specifically, we sort stocks that are short sold each day into one of the five DTC buckets and keep them in the portfolio during their days to close. Portfolio weights are based on dollar amounts shorted. We form the DTC portfolios for both hedge funds and nonhedge funds. Panel A of Table 3 reports DGTW-adjusted returns on the four DTC portfolios for hedge funds (HF) and non-hedge funds (NHF). Hedge funds have strong profitability for short positions covered less than five trading days. For example, the average DGTW-adjusted return on the portfolio is negative 14 bps per day with a t-statistic of -2.36, which indicates that the short sale portfolio is highly profitable. In contrast, non-hedge funds exhibit negative profitability; the portfolio generates positive 11 bps per day, which is statistically significant at 5% level. We find

18 evidence that non-hedge fund short-sale portfolios are profitable for longer horizons (between one and three months) but the economic magnitude is not significant (only three bps per day). Also, although hedge funds short-term short sales have strong profitability, we find no evidence that their long-term short positions are profitable in our sample. In sum, the results in Panel A shows strong profitability for hedge fund short sales only in short-term positions. Table 3 Panel B reports abnormal returns (alphas) on the DTC portfolios estimated from the Carhart four-factor model (Carhart 1997). We find results that are consistent with those from Panel A. Hedge fund portfolios exhibit negative alphas only for short-term positions. The estimated alpha from the four-factor model is negative 14 bps daily with a t-statistic of In contrast, non-hedge fund short sale portfolio in the short horizon (within one week) shows a positive alpha (0.13% daily), indicating that non-hedge funds tend to lose money on shorthorizon short sale positions. In the longer horizon, one to three months, however, non-hedge funds become profitable albeit weakly, as can be seen from an alpha of negative three basis point with a t-statistic of As a robustness check, we estimate alphas using Dimson s sum beta (Dimson, 1979) and we find largely the similar results. In unreported results, we estimate alphas for equal-weighted DTC portfolios and obtain qualitatively similar results. The results in Panels A and B show that hedge funds short positions are profitable in the short horizon. We ask how this profitability on the short side compares with profitability from the long side. This is an interesting question, given that previous studies, e.g., Griffin and Xu (2009), report that hedge funds do not particularly perform well on their long positions. We form DTC portfolios using stocks in the long positions, similar to the short sale portfolios. Panels C and D report average DGTW-adjusted returns and alphas from the four-factor models, respectively, for hedge fund and non-hedge fund long positions. The results indicate that

19 hedge funds do not perform well on their long side positions. Most abnormal returns tend to be close to zero or even negative, if any. Interestingly, non-hedge funds performance on the long side of their positions is particularly subpar. For example, their portfolios generate negative daily 23 bps in Panel D for DTC<=5. In summary, we find that hedge funds are profitable on their short positions especially for short-horizon trades. They earn approximately 14 bps per day on their short positions, which translates to almost 35% annum. In contrast, we find no significant positive performance for nonhedge funds on short positions Do Hedge Fund Short Sales Predict Earnings Surprise? The results in the previous section show that hedge funds short positions are profitable. A natural question that arises is, what is the source of the predictability? Previous studies document that institutional short sales are informed (e.g., Boehmer, Jones, and Zhang 2008) and common conjecture is that hedge funds comprise the majority of these informed institutions. In this section, we examine whether hedge fund short sales can predict negative earnings surprise announcement. In particular, we run the following regression: EEEE ii,ττ = αα + ββ SSSSSS ii,tt + CCCCCCCCCCCCCCCC ii,tt + εε ii,tt (2) Earnings surprises (EEEE ii,ττ ) for firm i quarter τ are calculated following DellaVigna and Pollet (2009). The main explanatory variable is short sale intensity, (SSSSSS ii,tt ), which is defined as jj SS SS ii,jj,tt jj (SSSS ii,jj,tt + SSSS ii,jj,tt + LLLL ii,jj,tt + LLLL ii,jj,tt ) where SSSS ii,jj,tt, SSSS ii,jj,tt, LLLL ii,jj,tt, and LLLL ii,jj,tt are the dollar volumes of short sells, short buys, long sells, and long buys by manager j at time t respectively. Short sales intensity captures the degree or intensity of short selling in a stock on a given trading day by all managers. The measure captures the overall managers view on a given

20 stock. The greater the short selling by managers in the stock, the higher the Short Sales Intensity. To examine differential predictability across time horizons, we use average short intensities during one week before earnings announcement (τ -5, τ -1) and during the period between one and four weeks before earnings announcement (τ-21, τ-6). As control variables, we follow Fama and French (2006) and include a dummy variable indicating negative previous earnings (NEGE), accruals per share (ACC),percent changes in total assets (AG), a dummy variable indicating zero dividends (DD), dividends per share (DPS), and log book-to-market (BE/ME). The regressions are pooled and standard errors are clustered by date and firm. Table 4 reports the results for earnings surprise prediction. We find strong evidence that hedge fund short sales can predict negative earnings surprises especially in the short horizon. For example, the coefficient on the short intensity of hedge funds during one week before earnings announcement highly statistically significant at the 1% level. The economic magnitude is also sizable. A one-standard-deviation increase in short intensity (0.419) corresponds to 4.1% of a one-standard-deviation (1.70%) decrease in earnings surprises (4.1%=0.17*0.419/1.703). For longer horizon between one to four weeks before earnings announcement, we find much weaker predictability of earnings announcement using hedge fund short intensities. Combined with the results reported in Table 3, these results in Table 4 indicate that the short-horizon profitability of hedge funds short position suggest that hedge funds are informed short sellers. We also find that non-hedge funds are not likely informed traders who can predict negative earnings surprises as the coefficient estimates are not statistically significant for non-hedge fund results in Panel B. In the last four columns of Table 4, we reports the regression of earnings surprises on long sell intensities to examine whether unwinding of long positions has similar predicting

21 power for future earnings announcement. We find the coefficient estimates tend to be negative, but in none of the cases considered are they statistically significant at the conventional levels. Thus, when institutional investors in our sample maintain long positions in a stock, future earnings are not particularly lower. We further examine whether hedge funds make profits through short sales during earnings announcement periods. Table 5 provides abnormal returns on hedge funds short positions that are open during earnings announcements and closed after. We form two portfolios based on days to close: one for DTC less than or equal to five (DTC<=5) and the other for DTC greater than five and less than or equal to twenty one (5<DTC<=21). For the hedge fund portfolio with DTC less than or equal to five, we find the average abnormal return is % per day, slightly greater than the abnormal return on the portfolio with DTC less than equal to five days that is reported in Table 3A (-0.141%). Although statistically insignificant due to the small size of the earnings announcement sample, this result suggest that hedge funds make profits by being able to predict negative earnings surprises. The implicit flip side result, though, is that hedge funds are also profitable during non-earning seasons, indicating that information for future earnings surprises is not the only source of their profitability. Overall, our results in Tables 4 and 5 show that our sample hedge fund short sales predict future negative earnings surprise and suggest that these hedge funds are informed short sellers Are Hedge Fund Short Sale Profitability Persistent? In this section, we test how persistent hedge fund short sell profitability is by examining whether hedge funds are skilled short sellers. Following Daniel, Grinblatt, Titman, and Wermers (1997), we form quintile portfolios of individual hedge fund managers short positions. During

22 portfolio formation quarter, we rank investment managers based on the past year s performance on short positions. We sorted hedge funds and non-hedge funds separately for comparison. After portfolio formation, we trace equal weighted portfolio of these managers short positions for the next four quarters. Table 6 provides the persistence results of short sell performance for hedge funds and non-hedge funds, separately. All reported returns are DGTW adjusted average daily portfolio returns for subsequent quarters (Q1 to Q4). The first two columns (t=0) report returns at the sorting period. For the subsequent quarters after portfolio formation, performance of top ranked managers in both of institution drops down. Top hedge fund managers profitability in terms of past short sale performance persists until the next three quarters. The top hedge funds managers keep performing better than top non-hedge funds managers. Overall, the results provided in Table 6 suggest that hedge funds have skills in short selling. 5. Conclusion Using databases for holdings and trades of institutional investors, we provide an algorithm to identify short sells, short buys, long sells, and long buys of hedge fund managers. The main idea of the algorithm is to exploit the fact that the ANcerno database contains both long and short trades, while the 13F filing only reports the long side of institutional holdings. This algorithm successfully identifies short sales by hedge funds, by capturing the 2008 short sale ban episodes. We obtain novel empirical results using these data. First, we find that hedge funds, compared with other institutional investors, tend to make profits for short-horizon trades that are

23 covered within one week. Second, we document that hedge fund short sales have predictive power for future abnormal returns and earnings announcement. Third, hedge fund short positions exhibit persistent performance. In sum, we provide the first evidence in the literature that hedge funds are skilled short sellers.

24 References Ackermann, C., R. McEnally, and D. Ravenscraft, 1999, The performance of hedge funds: Risk, return, and incentives, Journal of Finance 54, Agarwal, V., and N. Y. Naik, 2004, Risks and Portfolio Decisions Involving Hedge Funds, Review of Financial Studies, 17, Agarwal, V., N. D. Daniel, and N. Y. Naik, 2007, Why is Santa so Kind to Hedge Funds? The December Return Puzzle! Working Paper, Georgia State University. Agarwal, V., N. M. Boyson, and N. Y. Naik, 2009, Hedge Funds for Retail Investors? An Examination of Hedged Mutual Funds. Journal of Financial and Quantitative Analysis, 44, pp Agarwal, V., W. Jiang, Y. Tang, and B. Yang, 2013, Uncovering hedge fund skill from the portfolio holdings they hide, Journal of Finance 68, Agarwal, V., G. D. Gay, and L. Ling, 2014, Window Dressing in Mutual Funds, Review of Financial Studies 27, Akbas, F., E. Boehmer, B. Erturk, and S. Sorescu, 2013, Short interest, returns, and fundamentals, Unpublished working paper, University of Kansas. Amihud, Y., 2002, Illiquidity and stock returns: Cross-section and time-series effects, Journal of Financial Markets 5, Anand, A., P. Irvine, A. Puckett, and K. Venkataraman, 2012, Performance of institutional trading desks: An analysis of persistence in trading costs, Review of Financial Studies, 25, Aragon, G. and P. Strahan, 2012, Hedge funds as liquidity providers: Evidence from the Lehman bankruptcy, Journal of Financial Economics 103, Asness, C., R. Krail, and J. Liew, 2001, Do Hedge Funds Hedge?: Be Cautious in Analyzing Monthly Returns, Journal of Portfolio Management, 28, Asquith, P., and L. Meulbroek, 1996, An empirical investigation of short interest, Working paper, Harvard University Asquith, P., P. A. Pathak, and J. R. Ritter, 2005, Short interest, institutional ownership, and stock returns, Journal of Financial Economics, 78, Bethel, J., G. Hu, and Q. Wang, 2009, The market for shareholder-voting rights around mergers and acquisitions: evidence from institutional daily trading and voting. Journal of Corporate Finance, 15,

25 Boehme, R. D., B. R. Danielsen, S. M. Sorescu, 2006, Short sale constraints, differences of opinion, and overvaluation, Journal of Financial and Quantitative Analysis, 41, Boehmer, E., C. M. Jones, and X. Zhang, 2008, Which shorts are informed? Journal of Finance 63, Boehmer, E., C. M. Jones, and X. Zhang, 2013, Shackling Short Sellers: The 2008 Shorting Ban Review of Financial Studies 26, Brunnermeier M. K. and L. Pedersen, 2009, Market liquidity and funding liquidity, Review of Financial Studies 22, Brunnermeier, M. K. and S. Nagel, 2004, Hedge Funds and the Technology Bubble, Journal of Finance, 59, Campbell, J., S. Grossman, and J. Wang, 1993, Trading volume and serial correlation in stock returns, Quarterly Journal of Economics 108, Carhart, M., 1997, On persistence in mutual fund performance, Journal of Finance 52, Chemmanur, T. J., H. Shan, and G. Hu, 2009, The role of institutional investors in seasoned equity offerings, Journal of Financial Economics, 94, Chen, H., N. Jegadeesh, and R. Wermers, 2000, The Value of Active Mutual Fund Management: An Examination of the Stockholdings and Trades of Fund Managers, Journal of Financial and Quantitative Analysis, 35, Chen, Y., and B. Liang, 2007, Do Market Timing Hedge Funds Time the Market?, Journal of Financial and Quantitative Analysis, 42, Cohen, L., K. B. Diether, and C. J. Malloy, 2007, Supply and demand shifts in the shorting market, Journal of Finance 62, Cremers, K. J. Martijn, and A. Petajisto, 2009, How Active is Your Fund Manager? A New Measure that Predicts Performance, Review of Financial Studies 22, Daniel, K., M. Grinblatt, S. Titman, and R. Wermers, 1997, Measuring mutual fund performance with characteristic-based benchmarks, Journal of Finance 52, Danielsen, B. R. and S. M. Sorescu, 2001, Why do option introductions depress stock prices? Journal of Financial and Quantitative Analysis 36, DellaVigna, S. D. and J. M. Pollet, 2009, Investor Inattention and Friday Earnings Announcements, Journal of Finance 64, Desai, H, K. Ramesh, S. R. Thiagarajan, and B. V. Balachandran, 2002, An investigation of the informational role of short interest in the Nasdaq market, Journal of Finance 57,

26 Diamond, D. W., and R. E. Verrecchia, 1987, Constraints on short-selling and asset price adjustment to private information, Journal of Financial Economics 18, Diether, K. B., C. J. Malloy, and A. Scherbina, 2002, Differences of opinion and the cross section of stock returns, Journal of Finance 57, Diether, K. B., K. Lee, and I. M. Werner, 2009, Short-sale strategies and return predictability, Review of Financial Studies 22, Dimson, E., 1979, Risk measurement when shares are subject to infrequent trading, Journal of Financial Economics, 7, Engelberg, J. E., A. V. Reed, and M. C. Ringgenberg, 2012, How are shorts informed? Short sellers, news, and information processing, Journal of Financial Economics 105, Fama, E. F., and J. D. MacBeth, 1973, Risk, return, and equilibrium: empirical tests. Journal of Political Economy 81, Fama, E. F., and K. R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, Fama, E. F., and K. R. French, 2006, The Value Premium and the CAPM. Journal of Finance, 61: Fung, W., and D. A. Hsieh, 1997, Empirical Characteristics of Dynamic Trading Strategies: The Case of Hedge Funds, Review of Financial Studies, 10, Fung, W., and D. A. Hsieh, 2000, Performance Characteristics of Hedge Funds and Commodity Funds: Natural vs. Spurious Biases, Journal of Financial and Quantitative Analysis, 35, Fung, W., and D. A. Hsieh, 2001, The Risk in Hedge Fund Strategies: Theory and Evidence from Trend Followers, Review of Financial Studies, 14, Fung, W., and D. A. Hsieh, 2006, Hedge Funds: An Industry in Its Adolescence, Economic Review, 91, Fung, W., D. A. Hsieh, N. Y. Naik, and T. Ramadorai, 2008, Hedge Funds: Performance, Risk, and Capital Formation, Journal of Finance, 63, Green, C. T. and R. Jame, 2011, Strategic trading by index funds and liquidity provision around S&P 500 index additions, Journal of Financial Economics, 14, Green, C. T., R. Jame, S. Markov, and M. Subasi, 2014, Broker-hosted investor conferences, Journal of Accounting and Economics, 58,

27 Goldstein, M., P. Irvine, E. Kandel, and Z. Wiener, 2009, Brokerage commissions and institutional trading patterns, Review of Financial Studies, 22, Griffin, J. and J. Xu, 2009, How smart are the smart guys? A unique view from hedge fund stock holdings, Review of Financial Studies, 22, Hu, G., 2009, Measures of implicit trading costs and buy sell asymmetry, Journal of Financial Markets, 12, Hu, G., R. D. McLean, J. Pontiff, and Q. Wang, 2014, The year-end trading activities of institutional investors: Evidence from daily trades, Review of Financial Studies, 27, Jame, R., 2015, Liquidity Provision and the Cross-Section of Hedge Fund Returns, Working Paper Jegadeesh, N., and S. Titman, 1993, Returns to buying winners and selling losers: implications for stock market efficiency, Journal of Finance 48, Kelley, E. K. and P. C. Tetlock, 2013b, Retail Short Selling and Stock Prices, Working Paper. Kosowski, R., A. Timmermann, R. Wermers, and H. White, 2006, Can mutual fund "stars" really pick stocks? New evidence from a bootstrap analysis, Journal of Finance 61, Puckett, A. and X. S. Yan, 2011, The interim trading skill of institutional investors, Journal of Finance 66, Rendleman, R., C. Jones and H. Latane, 1982, Empirical Anomolies Based On Unexpected Earnings and the Importance of Risk Adjustments, Journal of Financial Economics 10, Rinne, K. and M. Suominen, 2010, Short-term reversals, returns to liquidity provision and the costs of immediacy, Working paper. Sadka, R., 2010, Liquidity risk and the cross-section of hedge-fund returns, Journal of Financial Economics 98, Wermers, R., 2004, Is money really 'smart'? New evidence on the relation between mutual fund flows, manager behavior, and performance persistence, Working paper.