Securities Lending by Mutual Funds

Transcription

1 Securities Lending by Mutual Funds Savina Rizova University of Chicago Booth School of Business Abstract Using hand-collected data for 2000 to 2008, I examine securities lending by U.S. equity mutual funds. First, I study the decision to start lending. It tends to be a family decision driven by economies of scale and past performance. Within participating families, smaller funds and funds with larger pricing impact are less likely to start lending. Second, I find that smaller and underperforming funds tend to earn more from securities lending. I also find families with superior investment performance learn to improve their value added from lending. Finally, fund entry in stock lending is associated with changes in fund turnover. University of Chicago Booth School of Business; srizova@chicagobooth.edu. I would like to thank Timothy Dore, Tobias Moskowitz, Marina Niessner, Lubos Pastor, and Jhe Yun for insightful discussions. 1

2 Securities lending plays an important role in financial markets by improving liquidity and price discovery. 1 The U.S. securities lending market is the largest in the world and is worth over 1 trillion USD. According to Data Explorers (a leading provider of securities lending data across all global market sectors), in 2008 securities lending generated 1.4 billion USD in gross income at U.S. mutual funds. As Figure 1 shows, the U.S. equity mutual funds, included in this study, obtained over 1 billion USD in net income from stock lending in Even though mutual funds are an important supplier in the market for securities lending, we know little about their participation in this market. [please insert Figure 1 here] The goal of this paper is to shed light on the behavior of mutual funds in the securities lending market by using hand-collected data on a comprehensive sample of U.S. equity mutual funds for 2000 to First, I investigate the determinants of fund entry into the securities lending market. The decision to enter tends to be a two-stage decision. At the first stage, the fund family decides whether to enter or not, and at the second stage the individual funds within the family decide whether to enter or not. Hence it is important to examine the determinants of both the family-level and the fund-level decisions. I first test whether economies of scale, pricing impact, past returns and turnover affect the family level decision. Economies of scale are likely to affect this decision if there are large fixed costs associated with launching a securities lending program (e.g., finding a lending agent, setting up the lending process, and implementing a monitoring system). Hence, in the presence of 1 For an overview of the securities lending market, see Geczy, Musto, and Reed (2002), D Avolio (2002) and Adams, Mansi, and Nishikawa (2011), among others. 2

3 large fixed setup costs, larger fund families are more likely to launch a lending program. Pricing impact fears could also drive the decision to enter. Families with larger holdings might fear that making their holdings available for shorting is likely to lower the prices of those holdings and make them more susceptible to speculative betting. Past returns could also be a factor in the decision process as poor past performance could push fund families to enter this market in an effort to add a few basis points to their portfolio returns. Finally, high turnover (due to manager-driven trading or to volatile fund flows) is likely to make funds less willing to enter the stock lending market as they anticipate frequent loan recalls. I find that the family decision to enter is driven by economies of scale and past performance. Worries about pricing impact do not appear to affect the family-level decision. The fund-level analysis, however, lends support to the pricing impact hypothesis. Within a participating family, funds with higher pricing impact are less likely to enter. More specifically, mid-cap and small-cap funds with larger assets under management are less prone to join the family lending program than mid-cap and small-cap funds with smaller assets. The fundlevel analysis also provides support for the economies of scale hypothesis. I find that larger funds, within a participating family, are more likely to enter this market. Interestingly, past performance does not appear to affect the fund-level decision. Taken together, the results suggest that the decision to enter is driven by economies of scale and past performance at the family level and by economies of scale and pricing impact at the fund level. The second part of the analysis examines the determinants of value added from securities lending - the ratio of income from securities lending to lagged fund assets. I document that there are strong year and asset class effects in value added from securities lending. Then I examine the drivers of value added in excess of year and asset class, which I refer to as abnormal value added or abnormal profitability. First, I test whether past performance af- 3

4 fects abnormal profitability from securities lending. On the one hand, the finding that poor past performance makes fund families more likely to start lending suggests that poor past performance could also make funds focus more on obtaining higher income from securities lending (by lending more stock and/or by lending stock more efficiently). On the other hand, stronger past performance could also be correlated with higher value added as stronger past performance could reflect better fund management. The data appear consistent with the hypothesis that poor past performance makes funds earn more from securities lending. I also test whether fund size affects negatively value added from securities lending. Worries about pricing impact are likely to make larger funds lend a smaller fraction of their assets, and this could lead to lower value added from securities lending. Indeed, I find that smaller funds tend to have higher abnormal value added from securities lending. I also test whether high fund turnover has a negative impact on profitability of stock lending. As already mentioned, frequent sales of holdings are likely to trigger frequent loan recalls and deteriorate the profitability of equity loans. In support of the view that high fund turnover has a negative impact on abnormal value added from securities lending, I find that index funds tend to earn more from securities lending, controlling for fund size, asset class, and year. Finally, I test whether fund family size affects positively value added from securities lending. Larger fund families might be able to dedicate more resources to the management of their securities lending program. In addition, larger families might have a larger bargaining power with lending agents. Interestingly, I do not find evidence that fund family size is related to abnormal value added from securities lending. The third part of my investigation examines whether funds learn from securities lending. Do funds improve their abnormal profitability from securities lending in the immediate years after fund entry? The data suggest that funds tend to learn how to improve their abnormal 4

5 profitability from securities lending, but the learning is concentrated in more successful fund families. Having explored fund entry, profitability, and learning in the securities lending market, I examine the relation between fund entry and changes in fund turnover in the same fiscal year, as well as in the subsequent year. For funds with low turnover, the entry in the securities lending market is associated with a rise in turnover, consistent with the hypothesis that the decision to enter is related to changes in fund management, strategy, and holdings. For funds with high turnover, however, entry in the equity lending market is related with lower current and future turnover, consistent with the hypothesis that funds are likely to lower turnover in order to avoid frequent recalls of stock loans. The paper has four main contributions to the financial literature. First, the findings in this paper improve our understanding of the behavior of mutual funds in the securities lending market. Although securities lending has been the focus of a few theoretical studies (e.g., Duffie, Garleanu, and Pedersen (2002), Duffie (1996), Krishnamurthy (2002)) and a growing number of empirical studies (e.g., Ofek and Richardson (2001), Mitchell, Pulvino, and Stafford (2001), D Avolio (2002), Geczy, Musto, and Reed (2002), Reed (2003), Cohen, Diether, and Malloy (2006), Saffi and Sigurdson (2010), Kaplan, Moskowitz, and Sensoy (2010)), there is almost no empirical research on the motivation and behavior of suppliers (and particularly mutual funds) in the securities lending market. The only other paper that analyzes securities lending from the perspective of mutual funds is Adams, Mansi, and Nishikawa (2011) who also collect data on securities lending by mutual funds, but they examine only index funds for 2003 to By examining a large number of active funds, in addition to index funds, I gain a much more comprehensive view of the behavior of U.S. equity mutual funds in the securities lending market. Note that in my sample, index funds 5

6 account for less than 30% of the aggregate securities lending income in Another important difference with Adams, Mansi, and Nishikawa (2011) is that they focus on the effect of lending agents and board structure on the return from securities lending, whereas I study why funds enter this market, what drives their abnormal value added from securities lending, whether funds learn from securities lending, and how funds change their trading behavior once they start lending stocks. The paper also contributes to previous research on the effects of mutual fund underperformance on mutual fund behavior. Lynch and Musto (2003) show that underperforming funds tend to change their investment strategies. The novel finding of this paper that poor past relative investment performance makes funds focus more on the securities lending market reveals a new mechanism through which funds attempt to offset poor performance. The paper also relates to the literature on learning by doing (e.g., Arrow (1962), Berk, Green, and Naik (2004), Seru, Shumway, and Stoffman (2010)) as it presents empirical evidence of learning by doing in the context of the opaque market of securities lending. Finally, the findings in this paper have important implications for the debate on whether mutual funds should make their holdings available for lending. The fact that in 2008, 70% of the larger U.S. equity mutual funds were lending stock suggests that the entry of any new funds into this market is unlikely to affect prices and volatilities of U.S. stocks and helps explain the findings of Kaplan, Moskowitz, and Sensoy (2010) that the recent entry of a relatively large supplier did not significantly impact stock prices and volatilities. Moreover, this paper suggests that entry in the securities lending market might benefit mutual fund shareholders not only through income from securities lending but also through inducing funds with extremely high turnover to reduce trading. The remainder of the paper is organized as follows. Section I presents the tested hypothe- 6

7 ses. Section II describes the sample selection process and the data. Section III presents the main empirical results. Section IV concludes. I Background Information and Testable Hypotheses This section provides a brief overview of the U.S. securities lending market and develops the testable hypotheses that guide the empirical analysis. The U.S. securities lending market is a relatively opaque market in which security owners lend their holdings, usually through a lending agent, to borrowers who typically short sell the securities in anticipation of future price drops. In a standard transaction with a U.S. stock, the borrower receives the security from the lender and posts cash collateral of 102% of the shares value. The lending agent invests the collateral. The spread between the interest earned on the collateral and the interest paid to the borrower of the stock (the rebate rate) is split between the lender and the lending agent. The two main risks for the lender are borrower s default and underperformance of the invested collateral. The risk of borrower s default is mitigated by the collateral which exceeds the borrowed value and is marked to market on a daily basis. In addition, the contract between the lender and its agent often stipulates that the lending agent will indemnify the lender in case of borrower s default. The lender can further mitigate the risk of default by monitoring its lending agent. As for collateral investment risk, the lender can mitigate this risk by limiting the set of eligible investments. The overview above implies that securities lending can be a relatively safe source of additional income for security holders such as mutual funds. Not all mutual funds, however, have entered the securities lending market yet. Hence, it is important to understand the determinants of the fund entry decision. 7

8 A Entry in the Securities Lending Market Economies of scale. Economies of scale are likely to affect the fund entry decision if there are large costs associated with launching a securities lending program (e.g., finding a lending agent, setting up the lending process, and implementing a monitoring system). Hence, in the presence of large setup costs, larger fund families are more likely to launch a lending program. Moreover, Saffi and Sigurdsson (2007) report that lending agents prefer to work with larger security lenders. Therefore, I test the following hypothesis: Hypothesis 1 (Economies of Scale and Fund Entry): Larger funds and fund families are more likely to enter the securities lending market. Pricing impact. Pricing impact could also drive the decision to enter as funds with larger holdings might fear that making their holdings available for shorting is likely to lower the prices of those holdings and make them more susceptible to speculative betting. Indeed, Kaplan, Moskowitz, and Sensoy (2010) mention that the fund family they worked with on a stock lending experiment had not lent out the stock it owned out of concern that doing so would lower the price of the stocks and increase their volatilities. Pricing impact worries are likely to matter more for funds holding large positions in smaller, less liquid stocks as trading in such stocks is more likely to have strong pricing impact. This suggests that: Hypothesis 2 (Pricing Impact and Fund Entry): Funds and fund families with large investments in small and mid-cap equities are less likely to enter the securities lending market. Past performance. Past returns could also be a factor in the decision process as poor past performance (especially relative to competitors) could push fund families to enter this market in an effort to add a few basis points to their portfolio returns. Previous research shows that underperforming funds do search for ways to improve performance (Lynch and Musto (2003)). Hence it is plausible that funds decide to launch a stock lending program in 8

9 an effort to achieve higher fund returns. Hypothesis 3 (Past Performance and Fund Entry): Funds and fund families with poor past performance are more likely to enter the securities lending market. Turnover. High fund turnover (due to manager-driven trading or to volatile fund flows) is likely to make funds less willing to enter the stock lending market as they anticipate frequent loan recalls and thus lower profitability from securities lending. This implies that: Hypothesis 4 (Turnover and Fund Entry): Index funds, funds with lower manager-driven turnover, and funds with less volatile fund flows are more likely to enter the securities lending market. B Value Added from Securities Lending Past performance. On the one hand, poor past performance could make funds focus more on obtaining higher income from securities lending by lending more stock and/or by lending stock more efficiently. On the other hand, stronger past performance could also be correlated with higher value added as stronger past performance could reflect better portfolio management. 2 Therefore, I suggest two alternative hypotheses: Hypothesis 5A (Poor Past Returns and Value Added): Funds with poor past performance are likely to obtain higher value added from securities lending. Hypothesis 5B (Strong Past Returns and Value Added): Funds with strong past performance are likely to have better portfolio management skill and hence to obtain higher value added from securities lending. Fund size. Fund size could affect negatively value added from stock lending because 2 Note that stronger past performance could be due not only to luck or to skill in picking stocks but also to skill in designing, managing, and trading portfolios efficiently. 9

10 worries about pricing impact are likely to make larger funds lend a smaller fraction of their assets and this could lead to lower value added. Hypothesis 6 (Fund Size and Value Added): Larger funds should obtain lower value added from securities lending. Fund family size. Larger fund families are likely to have stronger bargaining power with lending agents. Thus they are likely to negotiate better revenue split with lending agents. Moreover, larger fund families should be able to dedicate more resources to the management of their securities lending program. Therefore I expect fund family size to be positively related to value added from securities lending. Hypothesis 7 (Fund Family Size and Value Added): Larger fund families should obtain higher value added from securities lending. Turnover. As already mentioned, frequent sales of fund holdings are likely to trigger frequent equity loan recalls and deteriorate the profitability from equity loans. Therefore, funds that trade less often should deliver higher value added from securities lending. In addition, index funds are likely to focus more on securities lending because they have fewer sources of additional value added. Hypothesis 8 (Turnover and Value Added): Index funds, funds with lower manager-driven turnover, and funds with less volatile fund flows should earn more from securities lending. C Learning The securities lending income that the lender obtains is driven by four main factors: the size of the portfolio on loan, the rebate rate, the interest rate on the collateral, and the split between the lender and its lending agent. The lender has the ability to impact all four factors 10

11 in an effort to improve income from securities lending. First, the lender can make a larger fraction of its portfolio available for lending. By law, U.S. funds are not allowed to have more than one third of its assets on loan, but few funds ever reach that ceiling (Weinberg, 2011). Second, the lender can try to improve the loan fees and the collateral investment rates it receives by monitoring more closely its lending program and by seeking a more experienced and efficient lending agent. Finally, the lender could negotiate a more beneficial split of the revenue with the lending agent. There is little information about the actual splits between lenders and their lending agents, but the reported splits by U.S. equity mutual funds in my sample vary from 86% for the lender and 14% for the lending agent to 50% for the lender and 50% for the lending agent. The fact that lenders could affect their securities lending income suggests that mutual funds could learn how to improve their profitability from securities lending. Given the opaqueness of the securities lending market, the learning is likely to be gradual. Moreover, learning is likely to be concentrated in funds with larger assets under management as they have more resources to dedicate to this process and more to gain from an improvement in securities lending income. Learning is also likely to be stronger for older and more successful funds as both longer history and higher abnormal performance could signal better fund management skill. Therefore, I test the following hypothesis: Hypothesis 9 (Learning): Funds should improve their profitability from securities lending in the immediate years after fund entry in this market. Moreover, learning is likely to be more pronounced for larger, older, and more successful funds. 11

12 D Trading The decision to enter the securities lending market is likely to be related to changes in fund turnover. On one hand, the decision to enter is likely to be associated with changes in fund management, strategy, and holdings. Hence funds that enter the stock lending market could also experience a rise in turnover as they change strategy and rebalance their portfolios. On the other hand, since high fund turnover is likely to affect negatively profitability from securities lending (through frequent loan recalls), we can expect manager-driven trading to decline with fund entry into the stock lending market. Therefore, I examine two alternative hypotheses: Hypothesis 10A (Fund Entry and Strategy Change): Funds that enter the stock lending market are likely to experience a rise in manager-driven turnover. Hypothesis 10B (Fund Entry and High Past Turnover): Funds with high manager-driven turnover are likely to lower their turnover when entering the securities lending market. II Data The data for this study are from the CRSP Survivor-Bias-Free US Mutual Fund Database and from the Securities and Exchange Commission (SEC) EDGAR database. Mutual fund monthly returns, total net assets, age, turnover, index fund status, Lipper objective, and fund family history are from CRSP. 3 Securities lending status and income are hand-collected from the annual shareholder reports, forms N-30D (prior to 2003) and NCSR (from 2003 on), that registered investment advisors must file twice a year. These filings are available electronically in the SEC EDGAR database. Since securities lending is performed at the 3 Fund turnover reported by CRSP is the minimum of total fund purchases and sales scaled by fund assets. Hence I refer to it as manager-driven turnover. 12

13 fund level, I aggregate fund share classes into a single fund. More specifically, fund returns and turnover are asset-weighted averages of the returns and turnover of the individual share classes. Fund total net assets are the aggregate assets of all share classes. The other fund attributes are based on the oldest share class. To construct my sample, at the end of each year from 1999 to 2007, I identify all U.S. equity mutual funds in CRSP by using the Lipper class objective (codes EIEI, G, LCCE, LCGE, LCVE, MCCE, MCGE, MCVE, MLCE, MLGE, MLVE, SCCE, SCGE, SCVE, and SPSP). 4 Then I assign U.S. equity funds to eight asset classes (Marketwide, Large Cap, Mid Cap, Small Cap, Small Growth, Small Value, Large Growth, and Large Value) based on three-year correlations of their monthly returns with the returns on asset class indices (CRSP Value-Weighted Market Index, FF Large Cap, FF Mid Cap, FF Small Cap, FF Small Value, FF Small Growth, FF Large Value, and FF Large Growth). Each fund is assigned to the asset class whose index it is most correlated with. All indices are from Ken French s website. FF Large Cap is the value-weighted portfolio of common stocks with market capitalization in the top three size deciles of the NYSE. FF Mid Cap is the value-weighted portfolio of common stocks with market capitalization in fourth and fifth size deciles of the NYSE. FF Small Cap is the value-weighted portfolio of common stocks with market capitalization in the bottom five size deciles of the NYSE. FF Large Value is the value-weighted portfolio of stocks whose book-to-market ratio is in the bottom 30% of the NYSE and whose market capitalization is in the top 50% of the NYSE. FF Large Growth, Small Value, and Small Growth are defined similarly. I use past returns to assign asset classes because according to Brown and Goetzmann (1997), past returns are more appropriate for style classification 4 Evans (2004) reports that the CRSP Mutual Fund Database suffers from an incubation bias, the tendency of fund families to improve return histories by selecting to report a fund with superior performance from a group of incubated funds. Since I use fund returns in my analysis, I address the incubation bias by following Kacperczyk et al (2006) and excluding funds before their first offer date. 13

14 than self-reported styles or portfolio holdings. Within each asset class, I identify the largest 100 funds (with at least $10 million in assets under management). I then collect securities lending data for those funds for the fiscal year ending in the following calendar year. The information that I collect is whether the fund participated in securities lending during the fiscal year and the income it earned from securities lending. The final sample consists of 6,522 fund-year observations for 1,517 U.S. equity mutual funds, representing 411 fund families, over the period from 2000 to Table 1 presents summary statistics for the final sample. For each year from 2000 to 2008, the table reports the number of U.S. equity mutual funds for which data were collected, the fraction of funds that lend securities, the number of fund entries into and exits from the securities lending market observed in the sample, and the mean value added from securities lending (in basis points). Value added from securities lending is defined as the ratio of securities lending income for the fiscal year ending in month t divided by the average monthly total net assets for months t-1 to t-12. To shed light on the differences in securities lending participation and profitability across asset classes, the table also reports the summary statistics broken down by asset class. Panel A of Table 1 summarizes the overall sample. The annual sample sizes range from a low of 677 in 2007 to a high of 757 in The participation rate increases monotonically from 39% in 2000 to 69% in The fact that the participation rate almost doubles over the sample period implies that U.S. equity mutual funds aggressively entered the securities lending market during the sample period. Indeed, I observe 252 fund entries over the sample 5 The number of funds is lower than 800 in each year for two reasons. First, the assignment of asset class does not guarantee that there will be a hundred funds in each asset class. Second, securities lending data are missing when the fund is a fund of funds, when a fund merges or liquidates during the year, when reports are not available, and when it is impossible to determine whether the fund participated in securities lending or not. 14

15 period and only 67 exits. 6 The number of entries is highest in 2001 and 2002 and lowest in The average value added from securities lending is below 4 basis points for 2000 to In 2007 and 2008, with the collapse of the U.S. stock market and the surging demand for shorting securities, the value added from stock lending grows to 4.2 basis points in 2007 and 8.7 basis points in Therefore, participation, entry, and profitability from securities lending vary substantially through time. Panels B to I summarize the securities lending data by asset class and reveal great differences across asset classes. The participation rate is highest for Large Cap and Large Growth and lowest for Small Value. In contrast, profitability from securities lending is lowest for Large Cap, Large Growth, and Marketwide and highest for Small Cap, Small Growth, and Small Value. This finding is consistent with prior research as small-cap funds typically hold the smallest and least liquid stocks and these stocks tend to earn the highest loan fees, or equivalently the lowest rebate rates (D Avolio (2002), Geczy, Musto, and Reed (2002), Cohen, Diether, and Malloy (2006)). It is worth noting that during the financial crisis of 2007 and 2008 the income from securities lending was quite substantial for Small Cap and Small Value funds, adding more than 15 basis points, on average, to their returns in III A Main Results Entry in the Securities Lending Market In this subsection I analyze the determinants of a fund s decision to enter the stock lending market. The first step in this analysis is to find out whether the decision to enter tends to be 6 The low number of observed exits makes it impossible to conduct a statistically powerful analysis of the decision to exit. 15

16 a family-level decision or a fund-level decision. To answer this question, I calculate the fraction of family funds that enter jointly in a calendar year. 7 Figure 2 plots the distribution of this fraction in the sample. As Figure 2 shows, the distribution has a mode at 1. 42% of the family entries involve all family funds. Moreover, 70% of the family entries involve at least half of the family funds. Hence the decision to enter the securities lending market appears to be a family-level decision. The existence of dispersion in fund entry and participation within a fund family, however, suggests that we need to explore the fund-level decision as well. [please insert Figure 2 here] A.1 Family Entry I first examine the determinants of the family-level decision. To test the hypotheses that the family decision is driven by economies of scale, pricing impact considerations, past performance, and fund turnover, I consider two specifications. The first one is F amilyentry i,t = α + β 1 Log(Assets) i,t 1 + β 2 Log(F unds) i,t 1 + β 3 (SmallCapAssets/Assets) i,t 1 + β 4 (SmallCapAssets/Assets) i,t 1 Log(Assets) i,t 1 + β 5 (MidCapAssets/Assets) i,t 1 + β 6 MidCapAssets/Assets) i,t 1 Log(Assets) i,t 1 + β 7 Excess Ret i,t 1 + β 8 (Indexed Assets/Assets) i,t 1 + β 9 T urnover i,t 1 + β 10 F low V olatility i,t 1 + β 11 Y r t + e i,t where F amilyentry i,t is an indicator variable equal to one if fund family i entered the securities lending market in year t and equals zero otherwise. Log(Assets) t 1 is the log of the 7 Single funds are excluded from this calculation. 16

17 family total net assets for fiscal year end in t-1. Log(F unds) t 1 is the log of the number of family funds for fiscal year end in t-1. (SmallCap Assets/Assets) t 1 is the fraction of family total net assets that are in Small Cap, Small Growth, and Small Value funds at fiscal year end in t-1. (MidCap Assets/Assets) t 1 is the fraction of family total net assets that are in Mid Cap funds at fiscal year end in t-1. (Indexed Assets/Assets) t 1 is the fraction of family total net assets that are in index funds at fiscal year end in t-1. Excess Ret t 1 is the asset-weighted average of the excess returns of the family funds for the 12 months ending at fiscal year end in t-1. A fund excess return is defined as a fund return in excess of the median return of all U.S. equity funds in the same asset class over the same period. T urnover t 1 and F lowv olatility t 1 are asset-weighted averages of the turnover and flow volatility of the family funds for fiscal year t-1. Fund flow volatility is the standard deviation of monthly fund flows (relative to assets) for the 12 months ending at fiscal year end in t-1. 8 Y r t is a calendar year fixed effect. To estimate this model, I use OLS pooled cross-sectional regressions and cluster standard errors by fund family. The results from the estimation of the model are reported in Panel A of Table 2. In Columns 1 and 2, I test the economies of scale hypothesis. Column 3 adds variables that aim to capture pricing impact concerns. Column 4 includes the average lagged one-year excess return of the family in order to examine the impact of past performance on the decision to enter. Columns 5-7 incorporate three alternative measures of portfolio turnover. The specification in Column 1 tests whether fund family size, based on assets under management, affects the family decision to enter the securities lending market. The coefficient on lagged family assets is 0.02 with a t-statistic of Hence there is a positive and reliable relation between family entry and family size. In Column 2, I add an alternative measure 8 To avoid giving weight to extreme outliers, fund-level turnover and flow volatility are winsorized at the 99% level throughout the paper. 17

18 of fund family size, the log of the number of family funds. The loading on the number of family funds is 0.05 with a t-statistic of The magnitude of the coefficient implies that doubling the number of family funds raises the probability of family entry by 5 percentage points. Given that the sample probability is 8%, this is an important effect. Note that adding the number of family funds to the specification makes the slope on family assets drop in magnitude and lose significance. This suggests that there are large fixed costs associated with launching a securities lending program and that these costs are easier to swallow when divided among more funds. 9 To test whether pricing impact worries also affect the decision to enter, I include in Column 3 the fractions of family assets in small and mid cap funds and interact these fractions with the log of family assets. If pricing impact worries affect the family-level decision, the probability of entering should decline with the greater size of assets in small and mid cap stocks. Hence we should expect negative coefficients on the interaction terms. None of the newly added variables, however, is statistically significant. The regression results suggest that at the family level, pricing impact worries do not appear to impact the decision to enter. Next I examine whether past performance of family funds influences the family decision. The slope on the family average lagged excess return is negative, large, and statistically reliable. This finding implies that fund families with poor relative past performance are more likely to enter the securities lending market perhaps because they view securities lending as a way to improve overall fund returns. The last hypothesis I test is whether high holdings turnover is likely to make fund families 9 The strong relation between the number of family funds and the decision to enter could alternatively be due to the fact that families with a larger number of funds are likely to have a larger number of fund managers, which raises the probability that at least one manager advocates entry into the securities lending market. 18

19 less willing to enter the stock lending market. To measure holdings turnover, I use the fraction of family assets in index funds (Column 5), the family average of manager-driven fund turnover (Column 6), and the family average of fund flow volatility (Column 7). The results in Columns 5-7 show that only the coefficient on manager-driven fund turnover is marginally significant. The negative sign on that coefficient is consistent with the hypothesis that highly active families are less likely to enter because they expect lower income from securities lending due to more loan recalls. In summary, Panel A of Table II suggests economies of scale, past relative performance, and manager-driven fund turnover are important determinants of the family-level entry decision. To examine further the family-level decision, I rerun the analysis from Panel A replacing the indicator F amilyentry i,t as a dependent variable with F racf amentry i,t, which measures the fraction of family funds that enter the securities lending market in the first year of family participation. By using the fraction of family funds entering in the first year as the dependent variable, I can explore the intensity of the family entry decision. I estimate this model using tobit pooled cross-sectional regressions, clustering standard errors by fund family. The results from this estimation are reported in Panel B of Table 2. The results in Panel B are similar to those in Panel A. As in Panel A, in Panel B the number of family funds and past excess returns have positive and statistically reliable coefficients. Hence economies of scale and past performance affect not only the family decision to enter the securities lending market, but also the fraction of family funds through which this decision is implemented. [please insert Table II here] 19

20 A.2 Fund Entry Having examined the family-level decision to start lending stock, I now turn to investigating the fund-level decision. To test the hypotheses that the fund-level decision is driven by fund size, pricing impact worries, past investment returns, and holdings turnover, I run the following OLS specification: F undentry i,t = α + β 1 Log(Assets) i,t 1 + β 2 SmallCap i,t 1 + β 3 (SmallCap) i,t 1 Log(Assets) i,t 1 + β 4 MidCap i,t 1 + β 5 MidCap i,t 1 Log(Assets) i,t 1 + β 6 Excess Ret i,t 1 + β 7 IndexF und i,t 1 + β 8 T urnover i,t 1 + β 9 F low V olatility i,t 1 + β 10 Y r t + β 11 F amily i + e i,t where F undentry i,t is an indicator variable equal to one if fund i entered the securities lending market in year t. Log(Assets) t 1 is the log of the fund total net assets for fiscal year end in t-1. SmallCap t 1 is a dummy equal to one if the fund is in the Small Cap, Small Growth, or Small Value asset class at fiscal year end in t-1. MidCap t 1 is a dummy equal to one if the fund is a Mid Cap fund at fiscal year end in t-1. IndexF und t 1 is a dummy equal to one if the fund is an index fund. Excess Ret t 1 is the excess return of the fund for the 12 months ending at fiscal year end in t-1. T urnover t 1 and F lowv olatility t 1 are the manager-driven fund turnover and fund flow volatility, respectively, for fiscal year t-1. Y r t is a calendar year fixed effect. F amily i is a family fixed effect. I estimate this model using OLS pooled cross-sectional regressions, clustering standard errors by fund family. 10 In order to focus the analysis on the fund-level decision, I run the analysis using only funds from families with varying fund participation. 10 Clustering standard errors by fund yields similar results. 20

21 Table III presents the results from the estimation of this model. The specification in Column 1 tests whether fund size affects the fund decision to enter the securities lending market. The coefficient on lagged fund assets is 0.06 with a t-statistic of Hence there is a positive and statistically reliable relation between fund entry and fund size. 11 To test whether pricing impact worries also affect the decision to enter, I include in Column 2 indicators for whether the fund is a small-cap or mid-cap fund and interact these indicators with the log of fund assets. The loadings on SmallCap and MidCap are positive and statistically reliable suggesting that small-cap and mid-cap funds are more likely to enter probably because loan fees tend to be higher for small-cap and mid-cap stocks. The loadings on the interactions, however, are negative. The coefficient on the interaction with SmallCap is with a t- statistic of Similarly, the slope on the interaction with MidCap is with a t-statistic of These results suggest that small-cap and mid-cap funds are less likely to enter if they have larger assets under management. 12 Therefore, pricing impact fears play an important role in the fund decision to start lending stocks. The finding that pricing impact worries matter at the fund level but not at the family level implies that fund families do not consider pricing impact when they decide whether to enter or not but family funds do consider pricing impact when they pick which funds to include in the stock lending program. The differences between the determinants of the fund-level and family-level decisions continue in Columns 3-6. Column 3 reveals that the loading on past excess return is indistinguishable from zero. Hence, past performance does not appear to drive the fund-level decision. Columns 4-6 show that holdings turnover does not appear to affect the fund-level decision either. 11 Adams, Mansi, and Nishikawa (2011) find that participation in securities lending is positively related to fund size, using a sample of index funds. 12 More specifically, the estimates imply that if a small-cap fund moves from the 5th to the 25th percentile of small-cap fund assets, the effect of being small-cap on the likelihood of entering drops by 9 percentage points. Similarly, if a mid-cap fund moves from the 5th to the 25th percentile of mid-cap fund assets, the effect of being mid-cap on the probability of entering drops by 25 percentage points. 21

22 To summarize, the results in Table III suggest that the main determinants of the fundlevel decision to start lending stock are fund size and pricing impact worries. Within families with varying fund participation, funds are more likely to enter when they have larger assets but are less likely to enter when these larger assets are invested in smaller, less liquid stocks. [please insert Table III here] B Value Added from Securities Lending In the previous subsection, I analyzed the factors that affect the decision to enter the securities lending market. In this subsection, I examine the determinants of the value added from securities lending (the ratio of annual income from securities lending to lagged average fund assets). Table I revealed strong year and asset class effects in value added from securities lending. Hence in the analysis that follows I examine the determinants of a fund s value added in excess of the annual median for the fund s asset class. I refer to this measure as abnormal value added or abnormal profitability from securities lending. To test the hypotheses that the abnormal value added is driven by past performance, fund size, family size, and holdings turnover, I use the following specification: V alueadded i,t = α + β 1 ExcessRet i,t 1 + β 2 ExcessRet i,t 1 LowRet i,t 1 + β 3 LowRet i,t 1 + β 4 ExcessRet i,t + β 5 Log(Assets) i,t 1 + β 6 (Indexed Assets/Assets) i,t 1 + β 7 T urnover i,t 1 + β 8 F low V olatility i,t 1 + β 9 Log(F amassets) i,t 1 + β 10 Y r t + e i,t The dependent variable is a fund s value added from securities lending (in bps) for the fiscal year ending in calendar year t in excess of the median for all sample funds in that asset 22

23 class in that year. To test the two alternative hypotheses about the relation between past performance and future profitability from stock lending, I include as an explanatory variable ExcessRet i,t 1, the 12-month return ending at fiscal year end t-1 in excess of the median for all U.S. equity funds in that asset class in that year, and interact it with LowRet t 1, an indicator equal to 1 if ExcessRet i,t 1 is negative. The interaction term allows us to see if the relation between past fund performance and current value added from securities lending differs between funds that outperform the competition and funds that underperform the competition. Under the hypothesis that past abnormal fund performance captures fund management skill, the relation between past abnormal performance and current abnormal profitability from securities lending should be positive. Under the hypothesis that underperforming funds focus more on securities lending as a way to improve overall performance, the relation between past abnormal performance and current abnormal profitability should be negative (at least for underperforming funds). To account for the possibility that past fund performance affects subsequent profitability from securities lending simply because fund holdings exhibit momentum (e.g., fund holdings that have performed poorly in the near past are likely to attract short sellers and to continue to perform poorly), I add as a control the fund return for fiscal year t. To investigate the effects of fund and family size on abnormal value added, I also include lagged fund and family assets, Log(Assets) t 1 and Log(F amassets) t 1. Finally, to test whether holdings turnover has a negative impact on profitability from securities lending, I add an index-fund indicator, as well as manager-driven turnover and flow volatility for fiscal year t-1. I estimate the model using OLS pooled crosssectional regression with year fixed effects (to account for the time effect in fund returns). Standard errors are clustered both at the year and family level Clustering at the year and fund level yields similar results. 23

24 Table IV presents the results from regressing abnormal value from securities lending on lagged fund performance, fund size, family size, and holdings turnover. Column 1 shows that the coefficient on lagged excess fund return is statistically indistinguishable from zero. To see if this result is due to offsetting effects for outperformers and underperformers, in Column 2 I interact lagged excess return with an indicator for whether the lagged excess return is negative. The loading on the interaction is large, negative, and statistically reliable implying that the effect for underperformers is negative. 14 Hence I find that for an underperforming fund, the lower its return relative to the competition, the higher its subsequent abnormal value added from securities lending. This finding is consistent with the hypothesis that underperforming funds focus more on securities lending as a way to enhance returns. It is important to note that adding the interaction term makes the loading on lagged excess return larger although it is still not statistically different from zero. In column 3, I add the excess fund return for year t to see if the effect of past performance on current profitability from stock lending is due to stock return momentum. The slope on current excess return is large, negative, and statistically reliable. Not surprisingly, when a fund s stock holdings underperform, the fund tends to earn higher profitability from stock lending probably because of higher demand for short selling. Controlling for current excess return, however, does not impact the estimated slopes on lagged excess returns. 15 Therefore, the results do not provide evidence that past performance affects current value added from stock lending only through persistence in the returns of fund stock holdings. To test whether fund size affects negatively profitability from securities lending, I include in Column 4 the lagged fund total net assets. The coefficient on lagged fund assets is Indeed, the hypothesis β 1 + β 2 = 0 is rejected at the 1 percent significance level. 15 I have also repeated the analysis using raw fund returns, instead of excess fund returns, and obtained identical results. 24

25 with a t-statistic of -5.81, suggesting that larger funds tend to earn lower value added from securities lending probably because they tend to lend a smaller fraction of their portfolios. Indeed, Adams, Mansi, and Nishikawa (2011) report that larger index funds tend to lend a smaller percentage of their assets under management. In columns 5, 6, and 7 I test the hypothesis that holdings turnover has a negative impact on abnormal value added from securities lending. Column 5 adds an indicator for index fund as a measure of holdings turnover. The coefficient on index fund is positive and marginally significant. Hence, index funds tend to earn more from securities lending. This finding is consistent with the turnover hypothesis. Column 6 adds fund manager turnover as another measure of holdings turnover. The loading on fund manager turnover is tiny and indistinguishable from zero. Therefore, fund manager turnover does not appear to have any marginal explanatory power. Column 7 adds the third measure of holdings turnover, volatility of fund flows. The slope on flow volatility is negative and marginally significant suggesting that funds with more volatile flows tend to earn less from stock lending. This result is also consistent with the turnover hypothesis. Finally, I test the hypothesis that larger fund families have a positive effect on value added from securities lending. Column 8 adds lagged family assets to the model. The coefficient on lagged family size is close to zero and statistically unreliable. Hence family size does not appear to affect abnormal profitability from securities lending over and beyond the other considered factors. Examining the full specification in Column 8, we note that the positive coefficient on lagged excess return is now marginally significant, while the loading on the interaction term remains negative and statistically reliable. Hence, the results in Table IV provide strong evidence in favor of the hypothesis that underperformers tend to focus on securities lending 25

26 and offer some evidence in favor of the hypothesis that strong relative performance captures fund management skill. In addition, Table IV supports the hypotheses that fund size and holdings turnover have negative impact on abnormal value added from securities lending. [please insert Table IV here] C Learning In the previous subsections, I examined fund entry into the securities lending market and fund profitability from participation in this market. In this subsection, I investigate whether funds learn from securities lending. That is, do funds improve their abnormal profitability from securities lending in the immediate years after they start lending? To test this hypothesis, I run pooled cross-sectional OLS regressions of family-level abnormal value added from securities lending on years of participation. The dependent variable is the asset-weighted family average of abnormal value added from securities lending (in bps) for the fiscal year ending in calendar year t. I look at the family average because learning is likely to spill over across family funds given that family funds tend to enter jointly and to use common lending agents and lending processes. The explanatory variables are the years of family participation and the square of the participation years. The hypothesis that funds learn from securities lending implies that the loading on years of participation should be positive. If there are decreasing marginal returns to learning, the loading on the squared years of participation should be negative. Before presenting the results, it is worthwhile to point out that this specification cannot distinguish between learning about how much to lend and learning how to lend efficiently. Since I lack data on the size of the lending program of each fund, I cannot disentangle these 26

27 two mechanisms for learning. 16 For a mutual fund investor, however, what ultimately matters is the overall value added from securities lending. Hence it is important to find out whether funds learn how to improve the overall value added from securities lending. Table V presents the results from regressing family-level abnormal value added from securities lending on years of participation. Only fund families for which I observe entry into the securities lending market are included in this analysis. Column 1 and 2 report results for all considered fund families. Columns 3 and 4 report results separately for large and small families based on lagged family total net assets. Columns 5 and 6 report results for old and young families based on lagged family age. Columns 7 and 8 report results for families with high and low past excess returns based on family average three-year lagged excess fund returns. Finally, Columns 9 and 10 report results for families with high and low past alphas based on family average fund alphas estimated over the prior 3 years from the Fama-French model augmented with momentum. Standard errors in all columns are clustered by family. Columns 1 and 2 of Table V do not provide strong evidence of learning for the overall sample. The coefficient on years of participation is positive and the coefficient on the squared years is negative, which is consistent with decreasing returns to learning, but both coefficients are not statistically different from zero. The lack of a strong relation between abnormal value added from securities lending and years of family participation could be due to the fact that this relation is strong only for a subset of the families. To explore this possibility, I split the families into two groups by assets, by age, and by relative past performance based on the median values of those characteristics for the families included in Table V. Columns 3 and 4 show the results for large and small families, respectively, based on total net assets under 16 Adams, Mansi, and Nishikawa (2011) collect annual data on the size of the lending program for their sample of index funds, but since the percentage of portfolio on loan could change dramatically during the year, it is not clear how informative annual snapshots can be. 27

28 management. In Column 3, the coefficient on years of participation is 1.25 with a t-statistic of 2.99, while the coefficient on the squared term is with a t-statistic of Hence going from 1 to 2 years of participation increases the family average abnormal value added from securities lending by 1 basis point for larger families. The coefficients in Column 4 are smaller and statistically unreliable. The lack of statistical power for smaller families is likely due to the greater dispersion in profitability from securities lending across smaller families. Statistical tests for the difference in coefficients between large and small fund families cannot reject the null that the coefficients for large and small families are the same. 17 Hence I do not find strong evidence that larger families learn more from securities lending than smaller families. In Columns 5 and 6 I compare the relation between value added from stock lending and years of participation between old and young families. Once again the coefficients for old families are statistically reliable while the coefficients for younger families are statistically unreliable, but we cannot reject the hypothesis that the loadings for the two groups are the same. Thus we cannot conclude that older families tend to learn more from securities lending than younger families. Columns 7-10 show that these patterns change dramatically when I sort families on past abnormal performance. For example, for families with high 4-factor alphas (Column 9) the loading on years of participation is 2.11 with a t-statistic of 2.48, and the loading on the squared term is with a t-statistic of For families with low 4-factor alpha (Column 10), the estimated slopes are tiny and statistically indistinguishable from zero. Moreover, we can reject the hypothesis that the loadings for the two groups are equal at the 10% significance level. Hence Table V provides evidence that successful families (families with high 17 Using the number of family funds to classify families as large or small yields identical results. 28

29 abnormal past performance) tend to learn more from securities lending. This finding is consistent with the hypothesis that families with higher abnormal past performance are likely to have superior fund management skill, which helps them learn more from stock lending. To summarize, Table V shows that funds tend to learn from securities lending but this learning is concentrated in families with higher abnormal past performance, as measured by the family average three-year excess fund return or by the average 4-factor fund alpha, estimated over the prior three years. [please insert Table V here] D Turnover Having explored mutual fund entry, profitability, and learning in the securities lending market, I now turn to examine the relation between fund entry and fund turnover. On one hand, the decision to enter is likely to be associated with changes in fund management and strategy. Hence funds that enter the stock lending market could also experience a rise in turnover as they change strategy and rebalance their portfolios. On the other hand, since high fund turnover is likely to affect negatively profitability from securities lending (through frequent loan recalls), we can expect manager-driven trading to decline with fund entry into the stock lending market. The decline, however, should be concentrated in funds with high turnover as these funds are likely to benefit most from a decrease in holdings turnover. To test these two hypotheses, I use the following specification: T urnover i,t = α + β 1 F undentry i,t + β 2 F undentry i,t HighT urn i,t 1 + β 3 HighT urn i,t 1 + β 4 Log(Assets) i,t 1 + β 5 F i + e i,t 29

30 The dependent variable is fund i s turnover for the fiscal year ending in calendar year t in excess of the median for all U.S. equity funds in the same asset class over the latest fiscal year ending on or before fund i s fiscal year end. Using abnormal turnover allows me to control for year and asset class effects in turnover. To test the hypothesis that fund entry is associated with higher turnover, I include as a key explanatory variable F undentry i,t, an indicator for fund entry in fiscal year end t. To test the hypothesis that fund entry lowers turnover for funds with high turnover, I interact fund entry with HighT urn t 1, an indicator equal to 1 if the fund s turnover for fiscal year t-1 is above the median of all U.S. equity funds in the same asset class over the same period. The interaction term allows us to see if the relation between fund entry and turnover differs between funds with high and low lagged turnover. To account for the possibility that fund entry might affect turnover merely because as funds grow larger they are more likely to enter the securities lending market and to lower their turnover (due to pricing impact), I add as a control the lagged fund assets, Log(Assets) i,t 1. Only active funds are included in this analysis. I estimate the model using OLS pooled cross-sectional regressions with fund fixed effects in some of the specifications. The fund fixed effects allow me to use a fund before entry as a control for itself after entry. Standard errors are clustered at the fund level. Panel A of Table 6 presents the results from this estimation. Column 1 reports a univariate regression of fund turnover on fund entry in the securities lending market. The coefficient on fund entry is tiny and statistically indistinguishable from zero. In Column 2, I add the interaction of fund entry with an indicator for high past turnover. Now the coefficient on fund entry is positive and statistically reliable, implying that low-turnover funds tend to experience an increase of 10 percentage points in their turnover in the fiscal year when they start lending stocks. In contrast, the loading on the interaction term is with a t-statistic 30

31 of -2.62, suggesting that high-turnover funds tend to experience a decrease of 10 percentage points in their turnover when they start lending stock. Column 3 shows that controlling for the lagged fund assets does not have any impact on the relation between fund entry and turnover. In Columns 4-6 I repeat the analysis using fund fixed effects. Controlling for all fixed fund characteristics does not appear to have a great impact on the observed relation between fund entry and turnover. Since fund entry in the securities lending market can occur during any month in a fiscal year, entry in that market is likely to be related to changes in turnover not only in the same year, but also in the following year. Therefore in Panel B I examine the relation between fund entry and subsequent fund turnover. The results in Panel B are almost the same as in Panel A. For example, in Column 6, the coefficient on lagged entry is 0.12 with a t-statistic of 2.09 and the coefficient on the interaction term is with a t-statistic of Thus we find that fund entry predicts an increase in turnover for funds with low past turnover and a decrease in turnover for funds with high past turnover. It is important to note that these results could not be attributed merely to mean reversion in turnover because the coefficient on HighT urn t 1 is positive in all specifications and statistically reliable in four out of six implying that turnover is highly persistent. Overall, the results in Table 6 appear consistent both with the hypothesis that fund entry is associated with changes in fund strategy and holdings as well as with the hypothesis that high-turnover funds tend to lower their turnover when they start lending stock. [please insert Table VI here] 31

32 IV Conclusion This paper examines empirically the behavior of mutual funds in the securities lending market by using hand-collected data on a comprehensive sample of U.S. equity mutual funds for 2000 to I begin the analysis by investigating the determinants of fund entry into the securities lending market. The decision to enter tends to be a two-stage decision. At the first stage, the fund family decides whether to enter or not, and at the second stage the individual funds within the family decide whether to enter or not. I find that the family decision to enter is driven by economies of scale and past performance. Worries about pricing impact do not appear to affect the family-level decision. The fund-level analysis, however, lends support to the pricing impact hypothesis. Within a participating family, funds with higher pricing impact are less likely to enter. More specifically, mid-cap and small-cap funds with larger assets under management are less prone to join the family lending program than mid-cap and small-cap funds with smaller assets. The fund-level analysis also provides support for the economies of scale hypothesis. I find that larger funds, within a participating family, are more likely to enter. Interestingly, past performance does not appear to affect the fund-level decision. Taken together, the results suggest that the decision to enter is driven by economies of scale and past performance at the family level and by economies of scale and pricing impact at the fund level. I also examine the determinants of value added from securities lending. I first document that there are strong year and asset class effects in value added from securities lending. Then I examine the drivers of the value added in excess of year and asset class. The data appear consistent with the hypothesis that poor past performance makes funds earn more from securities lending. In addition, I find that smaller funds tend to have higher abnormal value added from securities lending. In support of the view that high fund turnover has 32

33 a negative impact on abnormal value added from securities lending, I also find that index funds tend to earn more from securities lending, controlling for fund size, asset class, and year. Interestingly, I do not find evidence that fund family size is related to abnormal value added from securities lending. The third part of my investigation examines whether funds learn from securities lending. The data suggest that funds tend to learn how to improve their abnormal profitability from securities lending, but the learning is concentrated in fund families with superior past investment performance. Having explored fund entry, profitability and learning in the securities lending market, I finish this study by examining the relation between fund entry and fund turnover. For funds with low turnover, the entry in the securities lending market is associated with a rise in turnover, consistent with the hypothesis that the decision to enter is related to changes in fund management, strategy, and holdings. For funds with high turnover, however, entry in the equity lending market is related with lower current and future turnover, consistent with the hypothesis that funds are likely to lower turnover in order to avoid frequent recalls of stock loans. The findings in the paper provide several venues for future research. First, it is important to investigate whether the documented time-series and cross-sectional patterns in fund entry, participation, and profitability from securities lending are also present in international equity funds as well as in bond funds. Second, it is interesting to examine further the link between fund entry in the stock lending market and changes in fund trading behavior by exploring whether index funds tend to incur higher tracking error when they are engaged in securities lending. Finally, it would be interesting to analyze theoretically and empirically the fund exit decision. 33

34 Table I: Summary Statistics for Securities Lending by U.S. Equity Mutual Funds, This table presents summary statistics of the sample of U.S. equity mutual funds for which data on securities lending participation and income were hand-collected from annual shareholder reports. At the end of each year, from 1999 to 2007, I assign U.S. equity mutual funds to one of eight asset classes (Large Cap, Large Growth, Large Value, Marketwide, Mid Cap, Small Cap, Small Growth and Small Value). Within each asset class, I sort funds on their net assets under management. I then collect securities lending data for the 100 largest funds in each asset class (with net assets under management of at least $10mln). N represents the number of observations, Mean Value Added is the mean of the ratio of net income from securities lending to average net assets over the fiscal year (in percent), Entries is the number of observed fund entries into the securities lending market, and Exits is the number of observed fund exits. Year N Participation Rate Mean Value Added (bps) Entries Exits Panel A: All Funds Panel B: Large Cap Funds Panel C: Large Growth Funds

35 Table I: Summary Statistics for Securities Lending by U.S. Equity Mutual Funds, (continued) Year N Participation Rate Mean Value Added (bps) Entries Exits Panel D: Large Value Funds Panel E: Marketwide Funds Panel F: Mid Cap Funds

36 Table I: Summary Statistics for Securities Lending by U.S. Equity Mutual Funds, (continued) Year N Participation Rate Mean Value Added (bps) Entries Exits Panel G: Small Cap Funds Panel H: Small Growth Funds Panel I: Small Value Funds

37 Table II: Family-Level Entry in the Securities Lending Market, This table presents pooled cross-sectional regressions of family entry in the securities lending market on lagged family characteristics. Panel A reports OLS regressions in which the dependent variable is an indicator for family entry during the fiscal year ending in calendar year t. Panel B reports tobit regressions of the fraction of family funds entering jointly in the first year of family participation. In both panels the explanatory variables are lagged family characteristics. Log(Assets) t 1 is the log of family total net assets and Log(F unds) t 1 is the log of the number of family funds for fiscal year end in t-1. (SmallCap Assets/Assets) t 1 is the fraction of family assets in Small Cap, Small Growth, and Small Value funds at fiscal year end in t-1. (MidCap Assets/Assets) t 1 and (Indexed Assets/Assets) t 1 are defined analogously. Excess Ret t 1 is the family average of fund returns (in excess of the median of all U.S. equity funds in the same asset class over the same period) for the 12 months ending at fiscal year end in t-1. T urnover t 1 and F lowv olatility t 1 are asset-weighted averages of the turnover and flow volatility of the funds in the family for fiscal year t-1. Fund flow volatility is the standard deviation of monthly fund flows (relative to assets) for the 12 months ending at fiscal year end t-1. All regressions include year fixed effects. Standard errors in all columns are clustered by family. T-statistics are reported below the coefficient estimates. Panel A: OLS Regressions of Family Entry on Lagged Family Characteristics Dependent Variable: Family Entry t (1) (2) (3) (4) (5) (6) (7) Log(Assets) t [3.04] [0.64] [0.52] [0.57] [0.51] [0.17] [0.22] Log(F unds) t [3.97] [3.80] [3.71] [3.26] [3.49] [3.42] (SmallCap Assets/Assets) t [0.59] [0.45] [0.31] [0.18] [-0.03] (SmallCap Assets/Assets) t 1 Log(Assets) t [-0.58] [-0.42] [-0.24] [-0.05] [0.12] (MidCap Assets/Assets) t [-0.80] [-0.83] [-0.75] [-0.86] [-0.96] (MidCap Assets/Assets) t 1 Log(Assets) t [0.62] [0.65] [0.59] [0.72] [0.82] Excess Ret t [-2.28] [-2.23] [-2.12] [-2.37] (Indexed Assets/Assets) t [1.19] [1.13] [1.02] T urnover t [-1.90] [-1.99] F low V olatility t [1.24] N Adj R Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Standard Errors Clustered by Family Yes Yes Yes Yes Yes Yes Yes 37

38 Table II: Family-Level Entry in the Securities Lending Market, (continued) Panel B: Tobit Regressions of Fraction of Family Entering on Lagged Family Characteristics Dependent Variable: Fraction of Family Entering t (1) (2) (3) (4) (5) (6) (7) Log(Assets) t [2.79] [0.05] [-0.44] [-0.40] [-0.43] [-0.81] [-0.76] Log(F unds) t [2.78] [2.76] [2.71] [2.57] [2.66] [2.61] (SmallCap Assets/Assets) t [-0.52] [-0.75] [-0.80] [-1.00] [-1.12] (SmallCap Assets/Assets) t 1 Log(Assets) t [0.47] [0.75] [0.82] [1.12] [1.24] (MidCap Assets/Assets) t [-0.97] [-1.04] [-0.96] [-1.04] [-1.13] (MidCap Assets/Assets) t 1 Log(Assets) t [0.86] [0.93] [0.86] [0.97] [1.07] Excess Ret t [-1.96] [-1.99] [-2.08] [-2.20] (Indexed Assets/Assets) t [1.13] [1.02] [0.88] T urnover t [-1.39] [-1.61] F low V olatility t [1.39] N Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Standard Errors Clustered by Family Yes Yes Yes Yes Yes Yes Yes 38

39 Table III: Fund-Level Entry in the Securities Lending Market, This table presents pooled cross-sectional OLS regressions of fund entry in the securities lending market. The dependent variable is an indicator for fund entry during the fiscal year ending in calendar year t conditional on varying participation within fund family in t. The explanatory variables are lagged fund characteristics. Log(Assets) t 1 is the log of the fund total net assets for fiscal year end in t-1. SmallCap t 1 is an indicator equal to one if the fund is in the Small Cap, Small Growth, or Small Value asset class at fiscal year end in t-1. MidCap t 1 is defined analogously. Excess Ret t 1 is the fund excess return for the 12 months ending at fiscal year end in t-1. A fund excess return is defined as a fund return in excess of the median return of all U.S. equity funds in the same asset class over the same period. T urnover t 1 is for fiscal year t-1. F lowv olatility t 1 is the standard deviation of monthly fund flows (scaled by assets) for the 12 months ending at fiscal year end t-1. All regressions include year and fund family fixed effects. Standard errors in all columns are clustered by family. T-statistics are reported below the coefficient estimates. OLS Regressions of Fund Entry Conditional on Varying Participation in Fund Family Dependent Variable: Fund Entry t (1) (2) (3) (4) (5) (6) Log(Assets) t [2.92] [3.30] [3.29] [3.35] [3.34] [3.34] SmallCap t [2.44] [2.47] [2.21] [2.22] [2.15] SmallCap t 1 Log(Assets) t [-2.76] [-2.79] [-2.65] [-2.66] [-2.61] MidCap t [3.83] [3.70] [3.87] [3.93] [3.86] MidCap t 1 Log(Assets) t [-4.36] [-4.20] [-4.40] [-4.45] [-4.37] ExcessRet t [0.44] [0.45] [0.48] [0.41] IndexF und [-0.86] [-0.88] [-0.87] T urnover t [-0.55] [-0.54] V olatility of F lows t [1.02] N Adj R Standard Errors Clustered by Family Yes Yes Yes Yes Yes Yes Year Fixed Effects Yes Yes Yes Yes Yes Yes Family Fixed Effects Yes Yes Yes Yes Yes Yes 39

40 Table IV: Abnormal Value Added from Securities Lending, This table presents pooled cross-sectional OLS regressions of abnormal value added from securities lending. The dependent variable is a fund s value added from securities lending (in bps) for fiscal year ending in calendar year t in excess of the median for all sample funds in the same category over the fiscal year ending in t. Value added from securities lending is the net income from securities lending for fiscal year ending in month m divided by the average fund net assets for months m-1 to months m-12. The explanatory variables are fund and family characteristics. Log(Assets) t 1 is the log of the fund total net assets for fiscal year end in t-1. Log(F amassets) t 1 is the log of the fund family total net assets for fiscal year end in t-1. T urnover t 1 is for fiscal year t-1. F lowv olatility t 1 is the standard deviation of monthly fund flows (relative to assets) for the 12 months ending at fiscal year end t-1. LowRet t 1 is an indicator equal to 1 if the fund return for the 12 months ending at fiscal year end t-1 is below the median return of all U.S. equity funds in the same asset class over the same period. ExcessRet t is the fund return for the 12 months ending at fiscal year end t in excess of the median return of all U.S. equity funds in the same asset class over the same period. All regressions include year fixed effects. Standard errors in all columns are clustered by year and family. T-statistics are reported below the coefficient estimates. Dependent Variable: Abnormal Value Added from Securities Lending t (1) (2) (3) (4) (5) (6) (7) (8) ExcessRet t [0.34] [1.43] [1.50] [1.27] [1.42] [1.60] [1.67] [1.71] ExcessRet t 1 LowRet t [-2.93] [-2.92] [-2.36] [-2.49] [-2.82] [-2.88] [-2.92] LowRet t [-1.11] [-1.13] [-1.14] [-0.87] [-0.80] [-0.86] [-0.87] ExcessRet t [-2.82] [-3.01] [-3.02] [-3.10] [-3.12] [-3.02] Log(Assets) t [-5.81] [-6.99] [-6.89] [-6.67] [-3.65] IndexF und [1.80] [1.78] [1.83] [1.91] T urnover t [0.10] [0.18] [0.27] F lowv olatility t [-1.74] [-1.94] Log(F amassets) t [-0.80] N 3,563 3,563 3,563 3,563 3,563 3,540 3,536 3,536 Adj R Year Fixed Effects Yes Yes Yes Yes Yes Yes Yes Yes SE clustered by Year and Family Yes Yes Yes Yes Yes Yes Yes Yes 40

41 Table V: Family-Level Learning in Securities Lending, This table presents pooled cross-sectional OLS regressions of family-level abnormal value added from securities lending. The dependent variable is the asset-weighted family average of abnormal value added from securities lending (in bps) for the fiscal year ending in calendar year t. A fund s abnormal value added from securities lending is value added in excess of the median for all sample funds in the same category for the fiscal year ending in t. Value added from securities lending is the net income from securities lending for fiscal year ending in month m divided by the average fund net assets for months m-1 to months m-12. The explanatory variables are the years of family participation and the square of the years. Columns 3 and 4 report results for large and small families based on lagged family assets. Columns 5 and 6 report results for old and young families based on lagged family age. Columns 7 and 8 report results for families with high and low past excess returns based on family average three-year lagged fund return in excess of the median return of all U.S. equity funds in the same asset class over the same period. Columns 9 and 10 report results for families with high and low past alphas based on family average fund alpha estimated over the prior three years from the Fama-French model augmented with momentum. Only fund families for which I observe entry into the securities lending market are included in this analysis. Column 2 includes family fixed effects. Standard errors in all columns are clustered by family. T-statistics are reported below the coefficient estimates. Dependent Variable: Family-Level Abnormal Value Added from Securities Lending All Families TNA Age Excess Ret 4-factor Alpha Large Small Old Young High Low High Low (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) P art Y rs [1.56] [1.54] [2.99] [0.18] [2.31] [0.72] [1.98] [0.28] [2.48] [-0.01] P art Y rs [-1.58] [-1.57] [-2.57] [0.20] [-1.95] [-0.65] [-2.01] [-0.32] [-2.72] [-0.07] N Adj R Family Fixed Effects No Yes No No No No No No No No SE Clustered By Family Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes 41

42 Table VI: Turnover and Entry in Securities Lending, This table reports pooled cross-sectional regressions of fund turnover. The dependent variable is the fund turnover for fiscal year end in calendar year t in excess of the median turnover for all U.S. equity funds in the same asset class. In Panel A, the key explanatory variables are Entry t, an indicator equal to one if the fund entered the securities lending market in fiscal year t, HighT urnover t 1, an indicator equal to 1 if the fund turnover for the prior fiscal year is above the median based on all U.S. equity funds in the same asset class over the same period, and the interaction between the two variables. Log(Assets) t 1 is the log of the fund net assets for the prior fiscal year end. In Panel B, the variable Entry is lagged one period. Only active funds are included in this analysis. Standard errors in all columns are clustered by fund. Fund Fixed Effects are included in columns 4-6. T-statistics are reported below the coefficient estimates. Panel A: Regressions of Fund Turnover on Contemporaneous Fund Entry (1) (2) (3) (4) (5) (6) Entry t [1.35] [2.15] [2.22] [1.28] [2.45] [2.76] Entry t HighT urn t [-2.62] [-2.58] [-2.11] [-2.34] HighT urn t [16.69] [16.15] [3.45] [2.63] Log(Assets) t [-1.54] [-3.33] N Adj R Fund Fixed Effects No No No Yes Yes Yes SE clustered by Fund Yes Yes Yes Yes Yes Yes Panel B: Regressions of Fund Turnover on Lagged Fund Entry (1) (2) (3) (4) (5) (6) Entry t [2.42] [3.05] [3.12] [0.20] [2.03] [2.09] Entry t 1 HighT urn t [-2.32] [-2.29] [-2.15] [-2.20] HighT urn t [14.95] [14.31] [1.16] [0.67] Log(Assets) t [-1.61] [-3.29] N Adj R Fund Fixed Effects No No No Yes Yes Yes SE clustered by Fund Yes Yes Yes Yes Yes Yes 42

43 Figure 1: Securities Lending Income of U.S. Equity Funds, This figure plots the aggregate net income (in millions of US dollars) from securities lending for a hand-collected sample of 1,517 U.S. equity mutual funds over the period from 2000 to