Internet Appendix for. Liquidity Provision and the Cross-Section of Hedge Fund Returns. Russell Jame

Transcription

1 Internet Appendix for Liquidity Provision and the Cross-Section of Hedge Fund Returns Russell Jame This document contains supplementary material for the paper titled: Liquidity Provision and the Cross-Section of Hedge Fund Returns. It consists of five sections. Section IA.1 outlines the identification of hedge fund management companies. Section IA.2 compares the sample of ANcerno hedge funds to hedge funds that report to TASS. Section IA.3 compares ANcerno hedge funds to 13F-filing hedge funds. Section IA.4 compares transaction-level measures of liquidity provision to measures of liquidity provision inferred from changes in quarterly holding or estimated from time-series regressions of returns on a liquidityprovision factor. Lastly, Section IA.5 reports robustness tests. IA.1. Identifying Hedge Fund Management Companies I use Form ADV to identify hedge fund managers within the ANcerno sample. I find Form ADV for 534 of 653 managers in the ANcerno sample. 1 Following Brunnermeier and Nagel (2004) and Griffin and Xu (2009), I classify a manager as a hedge fund if more than half of its investors are categorized as high net worth individuals or pooled investment vehicles in item 5.D. In addition, I require that the manager charge a performance-based fee (item 5.E). However, this approach incorrectly includes some funds with no hedge fund operations. I thus visit each company s website and eliminate any firms that do not report any hedge funds on their website. This filter eliminates private equity firms (e.g., New Harbor Capital), real estate firms (e.g., ERE Rosen), and investment advisors who have high net worth investors but do not offer hedge fund products (e.g., Denver Investment Advisors). I also exclude large banks (e.g., Bank of America). After these filters, the sample includes 55 hedge fund management companies. 1 Beginning in March 2012, the Dodd-Frank Act required that nearly all investment advisors, including hedge funds, file Form ADV. In addition, a 2004 SEC investment advisor rule required all hedge funds to file Form ADV for a short period in Thus, I obtain Form ADV for nearly all hedge fund families that had operations in 2006, or from 2012 onwards, plus any funds that voluntarily filed Form ADV. Form ADVs can be downloaded from the SEC website: IA.1

2 An additional concern is that Form ADV fails to capture many hedge funds. I examine the Form ADV of Institutional Investor s Top 100 Hedge Funds and find that the Form ADV approach correctly classifies 78 of the 100 hedge funds. 2 Of the 22 remaining funds, the majority list pensions and profit sharing plans as part of their investor base (see, e.g., Bridgewater Associates). To capture additional hedge funds that do not meet the Form ADV criteria, I examine a list of roughly F-filing hedge funds provided by Morningstar. According to Morningstar, the list is self-reported by the money management company, and typically reflects whether the management company is predominantly a hedge fund manager. 3 I find that 27 management companies appear on the Morningstar list of hedge funds, but fail to meet the Form ADV hedge fund criteria. Of the 27 hedge funds, 24 charged performance-based fees and had over 50% of their investors as high net worth individuals, pooled investment vehicles, or pensions and profit sharing plans. For the other three funds, Form ADV was unavailable, but inspection of the company s website indicated significant hedge fund operations. Based on this information, I classify all 27 companies as hedge funds. IA.2. Comparison of ANcerno and TASS Hedge Funds In this section, I compare TASS hedge funds that appear in ANcerno to TASS hedge funds that do not appear in ANcerno. I begin by identifying 10,374 live and graveyard hedge funds, excluding funds of funds that report to the TASS database between 1999 and I exclude funds of funds since such funds would not appear in ANcerno. For each fund-year, I compute the following variables: Assets, Relative Return, Management Fee, Incentive Fee, Lockup Period, Restrictions, Leverage, Derivatives, Pct. Equity, Asset Illiquidity, Liquidity Risk, and β RLP. Definitions of all the variables are provided in the Appendix. To make the TASS data 2 The list of top 100 hedge funds is here: Ranking.html. 3 Unfortunately, the Morningstar list also fails to capture some hedge funds. For example, the Morningstar list identifies 55 of Institutional Investors top 100 hedge fund companies. However, using a combination of Form ADV and Morningstar correctly identifies 83 of the top 100 hedge fund companies. IA.2

3 congruent with the ANcerno data, I aggregate across all funds that belong to the same management company. I report average values of each characteristic by equally-weighting across all funds in the management company. 4 Since my main analysis excludes funds that belong to management companies that do not offer any equity-oriented funds (i.e., long/short equity or equity market neutral), I also exclude TASS companies that offer no equity funds. This results in a sample of 11,178 management company-year observations and 1,900 unique management companies. I manually search for ANcerno hedge funds within TASS. The intersection of ANcerno and TASS yields a sample of 24 management companies for which the TASS and ANcerno reporting periods overlap, resulting in a total of 139 management company-year observations. 5 I compare the sample of ANcerno hedge funds reporting to TASS to TASS hedge funds that do not report to ANcerno. I report the average across all management-companies year. Note, this approach is different from Table 5 which weighs each management company by the number of funds (i.e., client-manager pairs) that appear in ANcerno. Since most TASS hedge funds do not report to ANcerno, weighting by the number of ANcerno clients is no longer possible. As a result, the estimates are not directly comparable to Table 5. Panel A of Table IA.1 presents the results. Along most dimensions, TASS hedge funds that appear in ANcerno do not significantly differ from TASS hedge funds that do not appear in ANcerno. For example, the two groups do not differ significantly with respect to average fund size, performance (Relative Return), incentive fees, leverage, derivative usage, Pct. Equity, and β RLP. The insignificant difference of Pct. Equity is sensitive to limiting the sample to management companies with Pct. Equity >0. Before imposing this 4 I report equal-weighted averages rather than asset-under-management (AUM) weighted averages because many funds are missing information on AUM. However, results are similar using AUM-weighted averages. 5 The sample is slightly different from Table 5 for two reasons. First, unlike Table 5, I do not require that a fund appear in ANcerno for consecutive years. Second, I now require that the management company appear in both ANcerno and TASS during the same year. In Table 5, the fund characteristics considered (e.g., management fee) reflect a single snapshot in time, and thus matching on time is not critical. However, I now test for differences between ANcerno and non-ancerno hedge funds with respect to returns, which do vary over time. IA.3

4 filter, ANcerno funds have an average Pct. Equity = 0.62, which is significantly larger than the 0.34 average Pct. Equity of non-ancerno hedge funds. There are, however, a few significant differences. First, ANcerno hedge funds charge significantly lower management fees (1.17% vs. 1.29%). In a competitive equilibrium, one might expect net-of-fee alphas to be zero for all funds (Berk and Green, 2004). As the equity trading returns (ETR) measure is gross of management fees, this may bias the sample towards less skilled funds. However, in Table 5, I find no evidence that management fees are correlated with a fund s tendency to provide liquidity. Further, in untabulated analysis, I find that management fees are not significantly correlated with ETR1 or EHR. ANcerno hedge funds also have greater asset illiquidity. Table 6 and 7 shows that funds with greater asset illiquidity have higher ETR. This points to the possibility that my estimates of ETR1 are biased upwards. However, my primary emphasis is on contrasting the ETR1 of liquidity-supplying and liquiditydemanding hedge funds. I find no evidence that asset illiquidity is correlated with the magnitude of a fund s liquidity provision. Further, in Specification 5 of Tables 6 and 7, I show that my main conclusions are very similar after directly controlling for asset illiquidity. When using ANcerno data, I treat a manager-client pair (e.g., Falcon Point Capital trading on behalf of CalPERS) as a fund. This differs from the typical definition of a fund in commercial databases, which treats a specific fund-product (e.g., Falcon Point Long/Short) as the unit of measurement. In Panel B of Table IA.1, I explore the relationship between the number of ANcerno funds and the size of the management company in TASS. Specifically, each-year I sort ANcerno managers into four groups based on the number of ANcerno clients for which the manager is trading on behalf of in a given year. For each group, I report the average number of funds offered by the management company in TASS and the total assets managed by the management company. Both measures exclude funds of funds and management companies with Pct. Equity = 0. The results confirm that ANcerno managers who have more clients tend to be larger management companies. For example, ANcerno managers with one ANcerno client (i.e., one fund) offer, IA.4

5 on average, 2.88 funds and the average management company size is $278 million. In contrast, ANcerno managers with more than five ANcerno clients offer 3.75 funds and manage assets of roughly $752 million. IA.3. Comparison of ANcerno and 13F-Filing Hedge Funds I next compare 13F-filing hedge funds that appear in ANcerno to 13F-filing hedge funds that do not appear in ANcerno. I begin by collecting a list of all institutional investors that filed form 13F, and thus report quarterly equity holdings, over the period from 1999 to I then use the hedge fund classification procedure, as described in Section IA.1, to classify 13F-filing institutions into hedge funds and other institutions. The final sample consists of 1,168 hedge funds and 25,714 hedge-fund quarters. I manually search for ANcerno hedge funds in the 13F data. The intersection of ANcerno and the 13F data yields a sample of 61 hedge funds for which the reporting periods overlap, resulting in a total of 1,406 hedge fundquarter observations. For each hedge fund-quarter, I use the holdings data to compute the following fund-level variables: Total Net Assets (TNA), Stocks Held, Holding Return, Size, BM, Amihud Illiquidity, Idiosyncratic Volatility (IVOL) and Mom21. Mom21 is computed relative to the first day of the quarter. All stock characteristics (e.g., Size, BM, etc.) are measured as percentile rankings. Additional details of the variable construction are described in the Appendix. Panel A of Table IA.2 compares the characteristics of ANcerno hedge funds to other 13F-filing hedge funds that do not appear in ANcerno. I report the average values across all fund-quarters, and t- statistics are based on standard errors clustered by management company and quarter. Within a management company, all the stock characteristics are based on the principal-weighted average of the characteristic. ANcerno hedge funds are significantly larger than other hedge funds as measured by both TNA and Stocks Held. This finding is consistent with Puckett and Yan (2011), who also find that ANcerno IA.5

6 institutions are substantially larger than non-ancerno institutions. 6 Given the structure of ANcerno data, it is not surprising that the sample is tilted towards larger funds. Most ANcerno hedge funds enter the sample because they manage money on behalf of a plan sponsor that subscribes to ANcerno. Larger hedge funds manage money for more plan sponsors, which increases the likelihood that they appear in the ANcerno sample. On average, hedge funds in ANcerno have one-quarter ahead DGTW-adjusted returns on their holdings that are similar to those of other hedge funds in the 13F universe (0.08% vs. 0.19%). There are some differences in the characteristics of the stock held by ANcerno and Non-ANcerno hedge funds. Specifically, ANcerno hedge funds tend to hold larger stocks, more liquid stocks, less volatile stocks, and stocks with stronger past one-month returns. However many of these differences relate to the finding that ANcerno hedge funds are larger than non-ancerno funds. After controlling for fund size, the only characteristic that remains significantly different across the two samples is Mom21. I also examine whether the trading of ANcerno hedge funds is similar to that of the universe of 13F-filing hedge funds. Trading is computed as changes in quarterly holdings. For each fund, I estimate the fund s Portfolio Turnover and the principal-weighted DGTW-adjusted returns on stocks bought less the principal-weighted DGTW-adjusted returns on stocks sold over the subsequent quarter (Trading Return). I also examine principal-weighted stock characteristics of the portfolio of stocks bought less the portfolio of stocks sold. As in Panel A of Table IA.2, I examine the following characteristics: Size, BM, Amihud Illiquidity, IVOL, and Mom21. Panel B of Table IA.2 presents the results. The Trading Return of ANcerno hedge funds is not significantly different from other hedge funds. (0.31% vs. 0.55%), There is also no evidence that the net trading (i.e., buys sells) of ANcerno hedge funds differs from non-ancerno hedge funds along any dimension. The only significant difference in trading style is that ANcerno hedge funds have lower turnover 6 At the time of Puckett and Yan s (2011) study, the ANcerno data were anonymous; however the authors were able to obtain a list of the names of 68 institutions from ANcerno. Their analysis does not distinguish hedge funds from other institutions. IA.6

7 than other hedge funds (82% vs. 132%). This suggests that the long holdings periods I document in ANcerno may be overstated relative to the average hedge fund. However, even the average turnover of 132% for non-ancerno hedge funds implies a holding period of over 9 months. IA.4. Inferring Liquidity Provision from Quarterly Holdings or Returns A central contribution of the paper is to use transaction-level data to measure a fund s tendency to demand or supply liquidity. A natural question is whether one could develop similar measures using other available data sources such as quarterly holdings or monthly reported returns. To investigate this issue, I construct two alternative measures of liquidity provision. First, I construct a measure based on changes in quarterly holdings from ANcerno data. Specifically, for each fund and each stock, I aggregate all trades within a quarter and calculate the cumulative net trading as of the quarter end. This net trading measure corresponds to net changes in holdings from the previous quarter, and thus reflects a measure that is similar to quarterly trading inferred from13f data. 7 Following the literature that measures trading based on changes in quarterly holdings (see e.g., Chen, Jegadeesh, and Wermers, 2000), I assume that all trading occurred on the last day of the quarter. I then compute the Mom1&5 associated with each trade, which I denote Mom1&5 Q. For each fund-year, I compute the principal-weighted average Mom1&5 Q of stocks purchased less the principal-weighted average Mom1&5 Q of stocks sold, resulting in a fund-level measure of Mom1&5 Q. I also construct a second measure of a fund s tendency to provide liquidity based on returns. As ANcerno does not provide monthly returns, I estimate a fund s monthly return by averaging the daily estimates of EHR in a given month. I then estimate a fund s tendency to provide liquidity based on the beta of the monthly return with a liquidity-provision factor. I use the return to liquidity provision (RLP) factor constructed in Jylhä, Rinne, and Suominen (2014), as this reflects an existing approach used to estimate a 7 This measure will still differ somewhat from changes in quarterly holdings of the 13F data, since ANcerno captures some equity trading not included in the 13F data, such as short-sales and confidential filings. IA.7

8 fund s tendency to provide liquidity based on the realized returns of a fund. 8 For each fund year, I estimate a time-series regression of monthly EHR on the monthly returns to the RLP factor, resulting in an annual estimate of β* RLP for each fund. Since β* RLP should be negatively correlated with momentum trading, for ease of interpretation, I multiply β* RLP by negative one. Panel A of Table IA.3 reports the correlation between transaction-level measures of a fund s tendency to provide liquidity (i.e., Mom1&5, Mom1, Mom5, Mom21, and Shortfall) and the two nontransaction-based measures (Mom1&5 Q and β* RLP ). All measures are standardized to have mean 0 and variance 1 each year, and are winsorized at plus or minus 3 standard deviations. Table IA.3 indicates that all of the transaction-based measures of liquidity-provision are strongly correlated, with correlations ranging from 0.72 to However, the non-transaction measures are much more weakly correlated. For example, the correlation between Mom1&5 Q and the various transaction-based measures ranged from 0.20 to 0.27, and the correlation between β* RLP and the transaction measures ranges from 0.02 to If liquidity-provision reflects a dedicated trading strategy of the fund, then the measure should be highly persistent across years. However, if a fund s tendency to provide liquidity is measured with error, then persistence will be considerably reduced. In Panel B, I examine the persistence of the liquidityprovision measures by estimating panel regression of a fund s liquidity-provision in year t+1 (as measured by the various liquidity-provision proxies in Panel A), on the fund s liquidity provision in year t (using the same liquidity-provision proxy). As there is only one independent variable and both the dependent and independent variables are standarized, the coefficients can be interpreted as the correlation in the liquidityprovision measure for a fund across adjacent years. I cluster standard errors by management company and year. The results indicate that the transaction-level measures are highly persistent, with correlations ranging from 0.65 to The non-transaction measures are much less persistent with correlations of 0.17 and I thank Petri Jylhä, Kalle Rinne, and Matti Suominen for providing data on the liquidity provision factor. More details on the construction of the factor can be found in Suominen and Rinne (2011) and Jylhä, Rinne, and Suominen (2014). IA.8

9 The persistence tests suggest the non-transaction measures of liquidity provision are estimated with considerable error and are unlikely to be particularly useful for identifying funds with greater trading skill. To more directly examine whether the non-transaction measures of liquidity provision can identify funds with superior skill, I replicate Specification 3 of Tables 6 and 7, but replace Mom1&5 with Mom1&5 Q or β* RLP. Table IA.4 reports the results. The first five rows of Table IA.4 report the results using the transactionlevel measures. Row 6 indicates that Mom1&5 Q is not significantly related to ETR1 or EHR, and the coefficient on Mom1&5 Q has the wrong sign for the EHR regression. Similarly, Row 7 indicates that β RLP is not significantly related to ETR1 or EHR. In sum, Mom1&5 Q and β* RLP are only weakly correlated with transaction-level measures of liquidity provision, are not a persistent fund characteristic, and cannot identify funds with superior equity trading returns. These findings highlight the importance of using transaction data to investigate the relationship between trading style and trading skill. IA.5. Liquidity Provision and Equity Trading Skill: Robustness Tests In this section, I examine the robustness of the findings that hedge funds that follow liquiditysupplying strategies earn significantly larger ETS1 (Table 6) and EHS (Table 7). I now only consider Specification 3 from Tables 6 and 7 and only report the coefficient on the main variable of interest (i.e. Mom1&5 or an alternative liquidity-provision proxy). For reference, the first row of Table IA.5 reports the baseline estimates from Specification 3 of Tables 6 and 7. In Rows 2 through 5, I examine whether the results are robust to alternative proxies for liquidity provision. In particular, I replace Mom1&5, with Mom1, Mom5, Mom21, or Shortfall, as defined in Table 5. These alternative measures yield similar conclusions. Specifically, when the dependent variable is ETR1, the coefficient on the alternative liquidity provision variables range from to -0.46, and all are significant at a 5% level. When the dependent variable is EHS, the coefficients range from to -0.17, and all are significant at a 5% level, except Shortfall which is significant at a 10% level. In Row 6, I report the results net of trading commissions. Trading commissions are subtracted from the day 0 return. For example, if a fund paid total commissions of $100 (as reported in ANcerno), on a IA.9

10 $10,000 transaction, then 1% would be subtracted from the day 0 return for that transaction. Row 6 indicates that incorporating trading commissions does not influence the coefficient on Mom1&5. 9 In Rows 7, I compute ETR using five-factor alphas rather than DGTW-adjusted returns. Specifically, I compute ETR by estimating the following time-series regression: ETR n * f,t = α f + β f, jr jt, + ε f, t. j= 1 (IA.1) ETR * f,t is computed similarly to ETR f,t, however it is computed using raw returns rather than DGTW-adjusted returns. Further, I aggregate ETR1 * f,t to a monthly measure by averaging across all daily estimates in a given month. I aggregate to a monthly frequency since the returns on some of the factor portfolios are available only at a monthly frequency. As ETR * f,t already reflects the returns of a long-short portfolio, I do not subtract the risk-free rate from ETR * f,t. R j,t captures the return on the Fama-French (1993) three-factors, plus the Carhart (1997) momentum factor, and the Pastor-Stambaugh (2003) traded liquidity factor. I estimate Equation IA.1 for each fund and compute the fund s monthly alpha as the sum of the intercept plus the monthly residual. The five-factor alpha from this regression can be interpreted as the excess return relative to a portfolio of equivalent riskiness (with respect to the above five-factors). The results, reported in row 7, indicate that replacing DGTW-adjusted returns with five-factor alphas yield slightly stronger results. Row 8 confirms that the results are similar after excluding the small sample of hedge funds that directly subscribe to ANcerno (i.e., money managers), and Row 9 shows that the results are robust to including hedge fund management companies that are less likely to have an all-equity focus. In Row 10, I estimate the results using a panel regression. Following Petersen (2009), standard errors are clustered by both management company and time. The panel estimates are slightly smaller than the Fama-Macbeth estimates, but the coefficients remain statistically significant at a 5% level. 9 While trading commissions does not influence measures of relative skill, they do reduce the absolute skill. Incorporating trading commissions reduces ETR1 (EHS) by roughly 0.30% (0.03%) per month. IA.10

11 In Tables 6 and 7, if a fund drops out of the sample, I simply exclude the fund from the analysis. However, if funds that drop out of the sample (hereafter: non-surviving funds) have different ETR than funds that remain in the sample (hereafter: surviving funds), then estimates of ETR can be biased. Following ter Horst, Nijman, and Verbeek (2001) and Baquero, ter Horst, and Verbeek (2005), I attempt to correct for this look-ahead bias by modelling the probability of hedge fund survival. This approach allows for surviving and non-surviving funds to have different expected returns, and instead assumes that the return of a fund is independent of survival after conditioning upon determinants of a fund s survival. I model a fund s survival using the following logistic regression: Survive f,t =A t +β 1 ETR1 f,t-1 + β 2 EHS f,t-1 + β 3 Mom1&5 f,t-1+ β 4 Holdings Size f,t-1+ β 5 Age f,t-1 +e f,t. (IA.2) Survive f,t is a dummy variable equal to one if fund f appears in the sample period during year t. A t allows for different intercepts for each year of the sample (i.e., year fixed effects). Holdings Size and Age are in natural logs. In untabulated analysis, I find that Age and Holdings Size have significant positive coefficients. Thus, funds that have appeared in ANcerno for a longer period of time or larger funds are more likely to remain in the ANcerno sample. The results from Equation IA.2 provide estimates of the conditional probability of survival. I reweight each observation by the unconditional probability of survival, scaled by the fund s conditional probability of survival. Intuitively, funds that have a lower conditional probability of survival (e.g., funds that just recently joined ANcerno or smaller funds) are less likely to appear in ANcerno in the post-ranking period. To correct for their underrepresentation, the returns on these funds are given greater weight. Row 11 indicates that the results are virtually identical after correcting for look-ahead bias. IA.11

12 Appendix: Description of the Control Variables Fund Characteristics obtained from TASS Data: Assets: assets under management. Relative Return: the net-of-fee realized return on a fund less the average net-of-fee realized returns for all other funds in the same investment style (as defined in TASS). Management Fee: the management fee charged by the fund. Incentive Fee: the incentive fee charged by the fund. Lockup Period: the length of the lockup period, reported in months. Restrictions: the sum of the notice period and the redemption period. Leverage: a dummy variable equal to one if the fund uses any leverage. Derivatives: a dummy variable equal to one if the fund has any exposure to derivatives. Pct. Equity: The fraction of funds within the money management company that have investment objectives (as reported by the primary category variable in TASS) of either equity market neutral or long-short equity. Asset Illiquidity: the first-order serial correlation of a fund s returns. Liquidity Risk: the fund s beta with respect to the liquidity risk factor of Sadka (2006). β RLP : the fund s beta with respect to the returns to liquidity provision (RLP) factor as constructed in Jylhä, Rinne, and Suominen (2014). Fund Characteristics obtained from 13F Data: Characteristics of the Management Company Total Net Assets (TNA): The total value (shares held * price per share) of long-only equity holdings, computed quarterly. Stocks Held: the number of long-only equity positions reported by the management company per quarter. Holding Return: the one-quarter ahead DGTW-adjusted return of the equity holdings, as reported by the management company. Portfolio Turnover: The turnover of the fund, computed as: min(buy it, Sale it )/Holdings it-1, where Buy it (Sale it ) is the total value of stocks bought (sold) by fund i in quarter t and Holdings it-1 is the total equity holdings of fund i in quarter t-1. IA.12

13 Trading Return: the one-quarter ahead DGTW-adjusted return of the equity buy trades (i.e., changes in quarterly holdings) less the one-quarter ahead DGTW-adjusted return of the equity sell trades. Fund Characteristics obtained from 13F Data: Characteristics of a Fund s Holdings and Trades Note: All stock characteristics are reported as a percentile ranking (ranging from 1 to 100) based on the cross-sectional distribution of all stocks in the sample. At the fund-level, stock characteristics are computed from the principal-weighted average of holdings or trades. Size: market capitalization (share price * total shares outstanding) at the end of the year prior to the year of the trade. BM: book-to-market ratio computed as the book value of equity for the fiscal year ending before the most recent June 30 th divided by market capitalization on December 31 st of the same fiscal year. Estimated for the fiscal year prior to the year of the trade. Amihud Illiquidity: The Amihud (2002) measure computed using all daily data available for the year prior to the year of the trade. Idiosyncratic Volatility (IVOL): the square root of the mean squared residual from an annual regression of a firm s daily returns on market (value-weighted CRSP index) returns. Computed in the year prior to the year of the trade. Mom21: the return on the stock in the 21 trading days prior to the beginning of the quarter. Liquidity Provision Measures obtained from ANcerno data Note: All trade-level liquidity-provision characteristics are aggregated to a fund characteristic. The fundlevel measure is computed as principal-weighted average characteristics (e.g. Mom1) of stocks purchased less the principal-weighted average characteristic (e.g. Mom1) of stocks sold. Mom1: the market-adjusted return on the stock on the trading day prior to the day of the trade. Mom5: the market-adjusted return on the stock in the five trading days prior to the day of the trade. Mom1&5: the average of Mom1 and Mom5. Mom21: the market-adjusted return on the stock in the 21 trading days prior to the day of the trade. Shortfall: the principal-weighted implementation shortfall of a fund measured as: P 1t P 0t D t P 0t where P 1t measures the value-weighted execution price of ticket t, P 0t is the price at the time when the broker receives the ticket, and D t is an indicator variable that equals one for a buy ticket and minus one for a sell ticket. Mom1&5 Q : A measure of liquidity provision inferred from changes in quarterly holdings. For each fund and stock, I aggregate all trades within a quarter and calculate the cumulative net trading as of the quarter IA.13

14 end. I assume that all trading occurred on the last day of the quarter. Using this quarterly trading, I then compute the principal-weighted average Mom1&5 of purchases less the principal-weighted average Mom1&5 of sales. Β* RLP : A measure of liquidity provision inferred from returns. I estimate a fund s monthly return by averaging the daily estimates of EHR in a given month. For each fund-year, I estimate a time-series regression of monthly EHR on the returns to liquidity provision (RLP) factor as constructed in Jylhä, Rinne, and Suominen (2014). Β* RLP is the coefficient estimate from this time-series regression. I multiply Β* RLP by negative one. Other Fund Characteristics obtained from ANcerno Data: ETR1 (One-Month Equity trading returns): the DGTW-adjusted returns on the fund s long holdings less the DGTW-adjusted returns on the funds short holdings, where both long and short holdings are estimated based on a fund s trading over the prior 21 trading days (including the current trading day). A more detailed definition is provided in Section 3.2. o t(etr1): ETR1 scaled by the standard error of ETR1. EHR (Equity holding returns): the DGTW-adjusted returns on the fund s long holdings less the DGTWadjusted returns on the fund s short holdings, where both long and short holdings are estimated based on all of a fund s equity trading since entering ANcerno (including the current trading day). A more detailed definition is provided in Section 3.2. o t(ehr): EHR scaled by the standard error of EHR. Holdings Size1: the total value of a fund s long holdings and short holdings where both long and short holdings are estimated based on a fund s trading over the prior 21 trading days (including holdings established on the current trading day). Holdings Size: the total value of a fund s long holdings and short holdings where both long and short holdings are estimated based on all of a fund s historical trading since entering ANcerno (including the current trading day). This measure is computed for all fund-days in which the fund has been in the ANcerno sample for at least one year. Volume: the average quarterly trading volume of a fund. Actual/Implied: the ratio of actual quarterly trading volume to implied quarterly trading volume. Actual trading reflects the aggregate quarterly trading of a fund. Implied quarterly trading volume is computed as the absolute net dollar volume, buys sells, for a fund-stock-quarter, aggregated across all stocks traded by the fund over the quarter. o For example, if a fund purchased $50,000 of Microsoft and $100,000 of Apple in January 2008 and sold $20,000 of Microsoft in February 2008, the fund s total trading volume in quarter 1 of 2008 would be $170,000, while its implied trading volume would be $130,000. Actual/Implied would thus be 1.31 ($170,000/$130,000). IA.14

15 Com/Share: the dollar volume paid in commissions scaled by total share volume traded (reported in cents). Plan Sponsor: a dummy variable equal to one if the hedge fund enters the ANcerno sample because it manages money on behalf of a Plan Sponsor client that has hired ANcerno. This value is set to zero for hedge funds that enter the sample because they directly hire ANcerno (i.e., money manager clients). Money Manager: a dummy variable equal to one if the hedge fund directly subscribes to ANcerno sample. This value is set to zero for hedge funds that enter the sample because they manager money on behalf of a plan sponsor who subscribes to ANcerno (i.e., Plan Sponsors). IA.15

16 References: Amihud, Yakov, 2002, Illiquidity and stock returns: Cross-section and time-series effects, Journal of Financial Markets 5, Baquero, Guillermo, Jenke ter Horst, and Marno Verbeek, 2005, Survival, look-ahead bias, and persistence in hedge fund performance, Journal of Financial & Quantitative Analysis 40, Berk, Jonathan and Richard Green, 2004, Mutual fund flows and performance in rational markets, Journal of Political Economy 112, Brunnermeier Markus, and Stefan Nagel, 2004, Hedge funds and the technology bubble, Journal of Finance 59, Carhart, Mark, 1997, On persistence in mutual fund performance, Journal of Finance 52, Chen, Hsiu-Lang, Narasimhan Jegadeesh, and Russ 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, Fama, Eugene and Kenneth French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, Griffin, John and Jin Xu, 2009, How smart are the smart guys? A unique view from hedge fund stock holdings, Review of Financial Studies 22, Horst, Jenke ter, Theo Nijman, and Marno Verbeek, 2001, Eliminating look-ahead bias in evaluating persistence in mutual fund performance, Journal of Empirical Finance 8, Jylhä, Petri, Kalle Rinne, and Matti Suominen, 2014, Do hedge funds supply or demand liquidity? Review of Finance 18, Pastor, Lubos, and Robert Stambaugh, 2003, Liquidity risk and expected stock returns, Journal of Political Economy 111, Puckett, Andy and Xuemin (Sterling) Yan, 2011, The interim trading skill of institutional investors, Journal of Finance 66, Suominen, Matti and Kalle Rinne, 2011, A structural model of short-term reversals, Working paper. IA.16

17 Table IA.1: Comparison of ANcerno Hedge Funds and TASS Hedge Funds Panel A reports the average value of different fund characteristics for TASS hedge funds that report to ANcerno (N = 139 management company-years) and TASS hedge funds that do not report to ANcerno (N = 11,039 management company-years). Definitions for all fund characteristics are available in the Appendix. For both ANcerno and Non- ANcerno management companies, I report the equally-weighted average across all management-company years. Dif. is the difference between ANcerno and Non-ANcerno hedge funds. T(Dif) tests whether the difference is significantly different from zero. Statistical significance is based on standard errors clustered by management company and year. Panel B sorts ANcerno funds based on their number of clients in a given year, as reported in ANcerno. For each group, I report the number of fund-year observations, the average number of clients, the average number of funds offered by the management company in TASS, and total assets managed by the management company in TASS. Panel A: Average Characteristics of ANcerno and Non-ANcerno Hedge Funds ANcerno Other Dif. t(dif.) Log (Assets) (0.37) Relative Return (-0.56) Management Fee (-2.70) Incentive Fee (-0.31) Lockup Period (1.42) Restrictions (1.59) Leverage (0.07) Derivatives (-0.21) Pct. Equity (-0.77) Asset Illiquidity (3.39) Liquidity Risk (1.73) β RLP (-0.89) Panel B: Sorts by Clients per Management Company Clients per Management Company >5 Fund-Year Observations Average Clients/Manager TASS Funds Management Company Assets ($ mil) IA.17

18 Table IA.2: Comparison of ANcerno Hedge Funds and 13F Hedge Funds This table compares 13F-filing hedge funds that report to ANcerno to 13F-filing hedge funds that do not report to ANcerno. The sample includes 1,168 hedge fund management companies (hereafter hedge funds) and 25,714 hedge fund quarters. For each hedge fund-quarter, I examine whether the hedge fund reports to ANcerno during that quarter. The sample includes 1,406 ANcerno hedge fund-quarters. Panel A compares the average fund characteristics of the stocks held by ANcerno and Non-ANcerno hedge funds. Within a hedge fund-quarter, stock characteristics are based on the principal-weighted averages of the characteristic. The table presents averages across all fund-quarters. In the last column, I also test whether the averages for ANcerno and Non-ANcerno hedge funds are significantly different. T-statistics are based on standard errors clustered by management company and quarter. Panel B compares the trading, estimated from changes in quarterly holdings, of ANcerno and Non-ANcerno hedge funds. For each fund, I estimate the fund s Portfolio Turnover and the principal-weighted DGTW-adjusted returns on stocks bought less returns on stocks sold over the subsequent quarter (Trading Return). I also examine principalweighted characteristics of the portfolio of stocks bought less the portfolio of stocks sold. All stock characteristics are reported as percentile rankings. Additional information regarding variable construction is presented in the Appendix. Panel A: Holdings ANcerno Other Dif. t(dif.) Log (TNA) (9.04) Stocks Held (4.41) Holding Return (Quarterly) (-0.65) Size (2.76) BM (-1.24) Amihud Illiquidity (-1.93) IVOL (-2.30) Mom (3.18) Panel B: Trading ANcerno Other Dif. t(dif.) Portfolio Turnover (-7.57) Net Trading (Buys - Sells) Trading Return (Quarterly) (-1.03) Size (-1.15) BM (1.68) Amihud Illiquidity (0.20) IVOL (0.97) Mom (-0.45) IA.18

19 Table IA.3: A Comparison of Transaction and Non-Transaction Measures of Liquidity Provision This table reports the correlations of various liquidity-provision measures. I consider 5 liquidity-provision measures that are constructed directly from transaction data: Mom1&5, Mom1, Mom5, Mom21, and Shortfall. I also consider two measures of liquidity provision that could be estimated without transaction data: Mom1&5 Q and B RLP. The construction of each measure is described in the Appendix. All measures are standardized to have mean 0 and variance 1 each year, and are winsorized at plus or minus 3 standard deviations. Panel A reports the correlation in the measures across fund years. The sample consists of 1,298 fundyear observations. Panel B estimates panel regression of a fund s liquidity-provision measure in year t+1, on the liquidity provision measure in year t. The sample consists of 978 fund-years where data on liquidity-provision is available for a given fund over adjacent years. T-statistics, based on standard errors clustered by management company and year, are reported in parentheses. Panel A: Correlations of Liquidity Provision Measures Mom1&5 Mom1 Mom5 Mom21 Shortfall Mom1&5 Q B RLP Mom1& Mom Mom Mom Shortfall Mom1&5 Q B RLP 1 Panel B: Persistence in Liquidity Provision Measures X = Mom1&5 Mom1 Mom5 Mom21 Shortfall Mom1&5 Q B RLP ρ (X t, X t+1 ) (14.22) (9.69) (13.97) (15.83) (17.23) (2.53) (0.11) IA.19

20 Table IA.4: Liquidity Provision and ETR - Transaction versus Non-Transaction Based Measures of Liquidity Provision This table examines the relationship between liquidity provision and subsequent ETR1 (as in Table 6) or EHR (as in Table 7) using various liquidity-provision proxies. Row 1 presents the baseline results as reported in Specification 3 of Tables 6 and 7. Rows 2-5 consider four alternative liquidity-provision measures constructed from transaction data: Mom1, Mom5, Mom21, and Shortfall. Rows 6 and 7 consider two liquidity-provision proxies that could be estimated without transaction data: Mom1&5 Q and B RLP. Detailed definitions of all the liquidity-provision proxies are available in the Appendix. Obs. reports the average number of funds that are in the portfolio across all trading days in the sample. T-statistics, based on standard errors computed from the time-series standard deviation, are reported in parentheses. ETR1 (Table 6 Robustness) EHR (Table 7 Robustness) Row Specification Obs. Coeff. t-stat Obs. Coeff. t-stat 1 Mom1& (-2.73) (-2.39) Alternative Transaction-Based Liquidity Provision Proxies 2 Mom (-2.50) (-2.04) 3 Mom (-2.81) (-2.50) 4 Mom (-2.92) (-2.34) 5 Shortfall (-2.77) (-1.63) Non Transaction-Based Liquidity Provision Proxies 6 Mom1&5 Q (-0.42) (1.63) 7 B RLP (-1.14) (-0.73) IA.20

21 Table IA.5: Liquidity Provision and ETR Robustness This table presents robustness tests for the finding that a fund s tendency to supply liquidity is negatively related to subsequent ETR1 (Table 6) or EHR (Table 7). Row 1 presents the baseline results as reported in Specification 3 of Tables 6 and 7. Rows 2-11 examine the robustness of the baseline results to alternative specifications. In Rows 2-5, I replace Mom1&5 with liquidity provision proxies based on the stock s return over the prior one, five, or 21 trading days, (Mom1, Mom5, and Mom21) or a measure of price impact (Shortfall). Row 6 reports DGTW-adjusted returns net of trading commissions, and Row 7 reports the fivefactor alpha that includes the Carhart (1997) four-factors plus the Pastor-Stambaugh (2003) tradable liquidity-risk factor. In Row 8, I exclude hedge funds that enter the ANcerno sample as money managers, and in Row 9, I include hedge fund management companies that are less likely to pure equity-oriented funds. In Row 10, I estimate results using a panel regression with standard errors clustered by both management company and day. In Row 11, I use a weighting procedure designed to eliminate the potential bias due to fund attrition (i.e., the look-ahead bias). Obs. reports the average number of funds that are in the portfolio across all trading days in the sample, or in Row 10, the number of fund-days in the panel regression. T-statistics are reported in parentheses. ETR1 (Table 6 Robustness) EHR (Table 7 Robustness) Row Specification Obs. Coeff. t-stat Obs. Coeff. t-stat 1 Baseline Results (-2.73) (-2.39) Alternative Liquidity Provision Proxies 2 Mom (-2.50) (-2.04) 3 Mom (-2.81) (-2.50) 4 Mom (-2.92) (-2.34) 5 Shortfall (-2.77) (-1.63) Alternative Benchmark 6 DGTW-Returns less Trading Commissions (-2.64) (-2.38) 7 Five-Factor Alpha (-3.65) (-2.78) Alternative Samples 8 Exclude Money Managers (-2.81) (-2.39) 9 Include Funds with Pct. Equity = (-2.92) (-2.48) Alternative Methodologies 10 Panel Regression 197, (-2.27) 209, (-2.42) 11 Correction for "Look-Ahead" Bias (-2.70) (-2.41) IA.21