Institutional Money Manager Mutual Funds *

Transcription

1 Institutional Money Manager Mutual Funds * William Beggs September 1, 2017 Abstract Using Form ADV data, I document the extent to which investment advisers to mutual funds manage accounts and assets for other clientele. I find that the risk-adjusted performance of an investment adviser s mutual funds is positively associated with the sophistication of its client base. Additionally, I affirm several empirical predictions of the Gârleanu and Pedersen (2017) model which links the efficiency of the market for asset managers to that for assets. This includes showing that the well-documented relationship between the activeness of mutual funds (as measured by active share and R 2 ) and future risk-adjusted returns is concentrated in funds managed by investment advisers with the highest percentages of sophisticated, institutional clientele. Overall, results suggest that sophisticated investor search costs incurred to find skilled investment advisers allow for the identification of actively managed mutual funds likely to outperform. JEL classification: G11, G23 Keywords: mutual funds, institutional investors, investment advisers, Form ADV * I thank Richard Sias, Jonathan Reuter, Scott Cederburg, David C. Brown, Brian Melzer and seminar participants at the University of Arizona for helpful suggestions. I also thank Kathleen Kahle for data assistance. All errors and omissions are my own. Eller College of Management, University of Arizona, Tucson, AZ, address: beggs@ .arizona.edu. Phone:

2 1. Introduction Investment advisers catering to sophisticated, institutional clientele have long operated investment strategies in mutual funds side-by-side with their institutional clients. For example, LSV Asset Management (LSV) opened its first mutual fund in 1999, the LSV Value Equity Fund, five years after its founding in The opening of this fund brought the firm s expertise in value investing, previously available only to institutional clients, to retail investors. As mutual fund assets have benefitted from growth in defined contribution plans over the recent past, it is no surprise that this trend has continued [e.g., Sialm, Starks, and Zhang (2015)]. Since 1999, LSV has opened five additional mutual funds. Furthermore, many other notable investment advisers to sophisticated clientele, such as AQR Capital Management, have either launched or increased their footprint in the mutual fund industry. This paper examines how the presence of sophisticated, institutional clientele is associated with the performance of an investment adviser s mutual funds. Theoretically, the relationship is ambiguous. First, as pointed out in the Gârleanu and Pedersen (2017) model, large and sophisticated clients benefit from searching for manager skill because search costs are low relative to capital. As a result, mutual funds managed by advisers with a large institutional client base would be expected to outperform. On the other hand, recent work suggests that investment advisers may cross-subsidize by siphoning performance from less important, less sophisticated mutual fund clients to institutional clients [e.g., Ben-Rephael and Israelsen (2017); Del Guercio, Genc and Tran (2017)]. Ultimately, the impact of side-by-side institutional clients presence on mutual fund performance is an empirical question. 3 According to their website, LSV managed $94 billion in value equity portfolios for 350 clients as of December 31, Their average client size is approximately $269 million. Since LSV only manages six mutual funds, this suggests the average client has a significant asset base (i.e. resources) to engage in enhanced manager research activities. 2

3 A positive relationship between the presence of institutional clientele and mutual fund performance is also consistent with a causal governance explanation [e.g., Evans and Fahlenbrach (2012) and Del Guercio and Reuter (2014)]. 4 Specifically, increases in the presence of institutional clients for the adviser could cause the adviser s investment personnel to exert more managerial effort because sophisticated clients are more sensitive to performance and thus more likely to terminate the adviser due to poor performance [Del Guercio and Tkac (2002); Goyal and Wahal (2008)]. I do not discern between selection (i.e., search costs and due diligence) and governance. These explanations are complementary and simply two sides of the same coin, selection and termination, because the governance mechanism at work is manager termination. For example, when a pension plan hires a consultant to provide recommendations on managers for their plan, it is likely the same personnel in the research department of the investment consultant who initiate recommendations to hire and fire managers [Jenkinson, Jones and Martinez (2016)]. This suggests that the overall skill of a client s manager research function will be highly correlated with both of these explanations. 5 This study has three key contributions to the literature. First, I show that the risk-adjusted performance of an investment adviser s mutual funds is positively associated with the sophistication of its client base. I find that this relationship is economically significant and amounts to nearly a one basis point increase in annual outperformance (4-factor alpha) for every percentage point increase in sophisticated, institutional clientele. Consistent with the notion that these 4 Del Guercio and Reuter (2014) examine the performance and flow-performance sensitivity of retail direct-sold versus retail broker-sold mutual funds and find that direct sold shareholders are more sensitive to performance and that directsold funds tend to outperform broker sold funds by 9.6 basis points per month. I build upon and extend the work of Evans and Fahlenbrach (2012) by using regulatory filings to identify institutional clientele by type and presence at the investment adviser level. 5 Furthermore, to the extent that funds exhibit time-varying skill due to changes in operational factors or other unobservables, it is plausible that sophisticated clientele may be able time purchases and sales of adviser strategies in advance of within fund differences in performance. 3

4 advisory firms are more skilled and/or informed, I find that a significant portion of this outperformance can be attributed to higher return gaps [Kacperczyk, Sialm, and Zheng (2008)] and higher active share [Cremers and Petajisto (2009)]. Next, I test several empirical predictions of the Gârleanu and Pedersen (2017) model. First, I examine Gârleanu and Pedersen s prediction that large institutional investors are more likely to outperform because search costs are low relative to capital. Consistent with this prediction, baseline results are strongest for adviser firms with significant sums of assets managed in vehicles other than mutual funds. Second, I examine the empirical prediction that the efficiency of markets for asset managers is linked to that of underlying assets. Consistent with this prediction, I show that baseline results are stronger for strategies focusing on less efficient assets (i.e., small and mid cap funds). Finally, if sophisticated, institutional investors incur search costs to find more highly skilled and informed managers as assumed, I posit that investment advisers with more sophisticated client bases must be better managers of funds which take on greater active risk. As a result, I present evidence that the positive relationship between fund activeness, as measured by both active share and R 2, and future risk-adjusted returns is primarily concentrated in mutual funds managed by advisers with high percentages of institutional clientele [Cremers and Petajisto (2009); Amihud and Goyenko (2013)]. Third, using annual filings of each mutual fund adviser s Form ADV, I introduce a measure of sophisticated, institutional clientele which takes into account varying levels and changes in levels over time. This improves on prior literature which has proxied for institutional clientele using binary variables based on the existence of separate account composites in commercial databases and/or institutional mutual funds. Importantly, Form ADV is a mandatory regulatory filing for all investment advisers to mutual funds and separate account client data are provided by 4

5 client type (e.g., individuals, pension plans, charitable foundations, etc.) and are based on actual client presence. 6 Form ADV data further allow for me to document the extent to which mutual fund advisers serve other clients. I show that the average mutual fund is managed by an investment adviser whose client base consists of only 43% mutual funds. This indicates that, on average, the majority of mutual fund advisers clientele consists of other types of clients. Other types of clients also tend to make up a significant portion of mutual fund advisers assets under management (AUM). The average mutual fund is managed by an investment adviser who derives 37% of its AUM from clients other than its mutual funds. Finally, I find flow-performance sensitivities for funds managed by advisers with higher levels of institutional clientele are consistent with the baseline results. Furthermore, I show that the baseline results are robust to a number of alternative explanations. These include various operational arrangements of the investment adviser which could present conflicts of interest or differences in incentives shown to impact returns. Finally, I conduct a battery of robustness tests which include further validating my Form ADV measure of sophisticated, institutional clientele using another dataset. Specifically, I show that the measure is positively associated with informed trading for SEC registered investment advisers who file quarterly 13F holdings. The remainder of the paper is structured as follows: Section 2 reviews related literature, Section 3 presents the data, Section 4 examines the relationship between sophisticated, institutional clientele and mutual fund performance, Section 5 tests several empirical predictions of the 6 Literature which examines the side-by-side management of mutual funds and hedge funds provides an example of how different conclusions can be reached based on identification via commercial databases versus regulatory filings. Nohel, Wang, and Zheng (2010) use commercial hedge fund databases to identify 112 cases where the same fund portfolio manager simultaneously manages mutual funds and hedge funds and find that mutual funds managed by these portfolio managers outperform. Del Guercio, Genc, and Tran (2016) use mutual fund N-SAR filings to identify portfolio managers who simultaneously manage hedge funds and mutual funds and find that mutual funds managed by these portfolio managers underperform. 5

6 Gârleanu and Pedersen (2017) model, Section 6 discusses supplemental analyses and robustness tests, and Section 7 concludes. 2. Background Gârleanu and Pedersen (2017) consider an imperfect market for asset management due to search frictions. 7 Their model is consistent with the empirical evidence of Sirri and Tufano (1998), Jain and Wu (2000), Hortaçsu and Syverson (2004) and Choi et al. (2010). Gârleanu and Pedersen suggest that large and sophisticated clients benefit from searching for informed active managers since their search costs are low relative to capital. Hence, managers with larger and more sophisticated investors are expected to outperform. Gârleanu and Pedersen s model proposes that the relative sophistication of an investment adviser s clientele is positively associated with the performance of the mutual funds they manage due to sophisticated investors superior search and selection efforts. This paper offers tests of their model using Form ADV data on adviser clientele and the returns realized in advisers mutual funds. Another strand of literature examining side-by-side management of mutual funds and other clientele has focused on conflicts of interest which can arise. If institutional investors are preferred clientele, then it is conceivable that investment advisers may siphon performance to their accounts at the expense of mutual fund investors through favorable trade allocations, i.e. cross-subsidization. 8 For example, Del Guercio, Genc, and Tran (2017) examine this issue with regard to side-by-side management of mutual funds and other clients who have incentive fee structures. They find that mutual funds whose portfolio managers also manage hedge funds 7 In contrast, Berk and Green (2004) consider the implications of fully efficient asset management markets. 8 See Gaspar, Massa and Matos (2006) and Bhattacharya, Lee, and Pool (2013) for cross-subsidization behavior in the mutual fund industry. Chaudhuri, Ivkovic, and Trzcinka (2017) examine cross-subsidization in the institutional asset manager industry. 6

7 significantly underperform their peers, consistent with the conflicts of interest hypothesis. In another related study, Ben-Rephael and Israelsen (2017) use ANcerno transaction-level data to identify bunched trades where an advisory firm trades the same stock in the same direction on the same day on behalf of many different clients. They find that some advisory firms assign the same average transaction price to each client account, while others assign different prices. At firms that allow this discretion, they find supportive evidence that certain clients systematically receive better prices than others. Furthermore, they show that mutual fund clients are less likely to be favored relative to other client types. In contrast to the findings of Del Guercio, Genc and Tran (2017) and Ben-Rephael and Israelsen (2017), the results in this paper suggest that any conflicts of interest arising from side-by-side management of sophisticated, institutional clients and mutual funds are dominated by the positive implications of the search costs incurred by sophisticated clientele. This study is also related to literature that examines the role of investment advisers in mutual fund management, which has been relatively underexplored. For example, Deli (2002) empirically investigates the investment advisory contracts of mutual funds. Two recent and highly related papers use hand gathered samples of Form ADV data. Chen et al. (2013) identifies mutual fund families that outsource portfolio management to unaffiliated investment advisory firms using Form ADV (subadvisory relationships) and Casavecchia and Tiwari (2016) use Form ADV data to examine how the cross-trading practices of investment advisers impact mutual fund performance. There exists two closely related prior studies which have examined how the presence of large institutional shareholders/clients impacts mutual fund performance using the existence of institutional funds or separate account products to proxy for their presence at the individual fund level [i.e., James and Karceski (2006); Evans and Fahlenbrach (2012)]. 9 My paper differs from 9 James and Karceski (2006) first address this question by focusing on institutional mutual funds. They use Morningstar s classification system to identify institutional mutual funds, which are those with initial investment 7

8 these studies as I measure the levels of institutional clientele at the investment adviser level as well as changes in levels over time. My data and approach allow for several direct tests of the Gârleanu and Pedersen model s empirical predictions. Furthermore, I focus exclusively on sophisticated clientele outside of the adviser s managed mutual funds. This approach is consistent with data provided in Gerakos, Linnainmaa, and Morse (2017) and by the Investment Company Institute (ICI) which indicate relatively little institutional investment in mutual funds compared to other vehicles such as separate accounts and commingled funds. 10 Moreover, my data on institutional clientele originate from required regulatory filings and include the broader population of both retail and institutional US mutual funds, as registered investment advisers which provide portfolio management services to investment companies registered under the Investment Company Act of 1940 are required to file Form ADV on at least an annual basis. In this filing, advisers are required to disclose information on the types of clients for which they manage assets. These data provide measurement on the institutional client presence for each investment adviser on an annual basis and a sample free from selection biases. Lastly, my study is not subject to the unknown variation in the presence of individual investors in institutional requirements of at least $100,000 or funds that designate themselves as institutional. They designate big institutional funds as those with initial investment requirements over $500,000 and small institutional funds as those with lower minimum investments. Overall, they find that institutional funds with larger minimum investment requirements significantly outperform other institutional funds, and this difference in returns cannot be attributed to the difference in fees. Evans and Fahlenbrach (2012) also use the Morningstar database and examine the performance of 463 retail mutual funds which also have an institutional mutual fund twin or institutional separate account twin. Consistent with the reduction of agency problems, they find that retail funds with an institutional twin outperform other retail funds by 1.5% per year. After the institutional twin is created they find evidence that expenses decrease while managerial effort increases. 10 Gerakos, Linnainmaa, and Morse (2016) cite P&I surveys 2012 data which shows that out of $48 trillion in delegated institutional assets only $5 trillion was managed in mutual funds. The ICI s 2016 factbook offers a similar picture of institutional investment in mutual funds. According to their figures, $15.6 trillion was invested in mutual funds as of December 31, Of this, $13.5 trillion or 87% consisted of individual investors and $2.1 trillion consisted of institutional investors. 8

9 mutual funds and separate account products, since Form ADV data allow for direct identification of separate account clients by type Data and Variables 3.1 Form ADV Data Mutual fund investment adviser clientele data come from each adviser s Form ADV filed annually with the SEC. Form ADV data was gathered via FOIA request from the SEC and includes the complete sample of ADV filings between 2001 (the beginning of electronic filing) and March Section 203A of the Investment Advisers Act of 1940 requires that all investment advisers to mutual funds register with the SEC and submit an annual amendment to their Form ADV within 90 days of their fiscal year end, providing a natural panel dataset free of selection biases. My analyses focus on data from ADV Item 5D. Item 5D presents a breakdown of client types for each investment adviser. Designated client types include: individuals, high net worth individuals, banking or thrift institutions, investment companies, business development companies, pooled investment vehicles (other than investment companies), pension and profit 11 Institutional investor levels may vary widely across institutional mutual fund and separate account products. The ICI s 2013 Trends in the Expenses and Fees in Mutual Funds shows that gross sales of institutional no-load mutual fund shares has grown nearly monotonically from $289 billion in 2004 to over $1.3 billion in The ICI attributes this growth to two primary sources, both of which consist of individual end clients: 1) the transitioning of retail financial advisers to fee-based business structures where there is no longer a need for load and 12b-1 fees, and 2) defined contribution plans (e.g., 401(k) plans) and other retirement accounts. Sialm et al. (2015) shows that defined contribution (DC) assets are less sticky than non-dc assets suggesting an added amount of sophistication due to plan sponsor and/or consultant actions. However, the sophistication of these assets has not been compared to other delegated institutional assets which do not include individual end clients (e.g. defined benefit pension plans, foundations, endowments, etc.). Thus, it remains unclear whether DC assets can be thought of as a sophisticated equivalent to other institutional investors assets. Additionally, prior literature offers a highly negative view of financial advisers ability to select mutual funds [e.g. Linnainmaa, Melzer, Previtero (2016); Bergstresser, Chalmers, and Tufano (2009)]. Moreover, separate account strategies in the Morningstar database have widely varying account minimums which could attract varying types of clientele. Elton et al. (2014) indicate that account minimums for separate account composites in the Morningstar database range from $100,000 up to $25 million, suggesting that certain separate account products may be geared toward less sophisticated clientele. 12 There has been limited research in the finance literature conducted using Form ADV filings because previous researchers used samples of hand gathered data. Advisers are also required to update their Form ADV if material changes to their business occur during the course of the year. 9

10 sharing plans (but not the plan participants), charitable organizations, corporations or other businesses not listed above, state or municipal government entities, other investment advisers, insurance companies and other entities. For each investor type, advisers are required to select one of the following percentage band categories: none, up to 10%, 11-25%, 26-50%, 51-75%, 76-99%, and 100%. Client type data by percentage of clients are available for all electronically filed Form ADVs since the beginning of the sample. 13 Appendix A provides a sample Item 5D from a Form ADV filing. I measure the sophistication of an adviser s client base by drawing on prior literature to identify sophisticated, institutional equity investors likely to incur manager search costs. I compute an investment adviser s percentage of institutional clients using band midpoints for each client type on Form ADV. For example, if individual investors are listed as 51-75% of clients, a value of 63%, the midpoint is assigned. Once midpoints are obtained, the percentage of institutional clients (%Inst_Clients) for investment adviser j in year t is calculated as % Inst _ Clients Funds. (1) j, t % Publicj, t % Corporatej, t % Charity j, t % Private_ j, t Where %Publicj,t is the percentage of state or municipal government entities which make up investment adviser j s client base in year t, %Corporatej,t is the percentage of corporations which make up investment adviser j s client base in year t, %Charityj,t is the percentage of charitable foundations which make up investment adviser j s client base in year t; and %Private_Fundsj,t is 13 AUM data by client type are also available beginning in I choose to focus on percentage of clients as opposed to percentage of client assets because data are available over a longer a time period and are presented in more precise percentage bands, giving my tests more power. 10

11 the percentage of pooled investment vehicles (other than investment companies) which make up investment adviser j s client base in year t. I include %Public and %Corporate in %Inst_Clients because Goyal and Wahal (2008) indicate 50% of corporate and 82% of public pension plans employ investment consultants to engage in manager research activities. 14 Significantly large public and corporate pension plans may further employ internal investment staff to conduct manager research activities [Jenkinson, Jones, and Martinez (2016)]. Charitable foundations are included for similar reasons. Charitable foundations and non-profit entities tend to employ internal investment staff or hire investment consultants to search for asset managers [Jenkinson, Jones, and Martinez (2016)]. Additionally, I include private funds in the measure of sophisticated, institutional clientele. Private funds managed by mutual fund advisers may consist of long-only commingled funds, hedge funds, or private equity funds. To assess the sophistication of private funds as clients, I consider underlying shareholders who make buy and sell decisions in these funds. Since private funds may not make a public offering of their shares, they are generally limited to accredited investors or qualified purchasers in order to comply with Sections 3(c)(7) or 3(c)(1) of the Investment Company Act of This requires investors in private funds to have significant net worth, income or investable assets, i.e. institutional investors or high net worth individuals, suggesting that underlying investors in private funds generally have resources to incur search costs to find informed managers. For a further discussion on the inclusion and exclusion of various client 14 Item 5D allows for double counting in client categories if a client falls into more than one category. As an example, a public pension plan such as CALPERS would be counted both as a municipal or state government entity and a pension plan. A positive correlation (0.43) between the sum of %Public j,t and %Corporate j,t and the percentage of an adviser s clients which are pension funds indicates substantial overlap in these categories. This is not surprising as corporate and public pension plans (e.g., General Motors, CALPERS, etc.) are among the largest and most prevalent institutional asset owners. 11

12 types from %Inst_Clients, see Appendix B. 15 In Section 6, I conduct robustness checks on my decisions to exclude or include various groups of clientele and find that the baseline results are not sensitive to any single decision. Table 1 presents summary statistics on mutual fund investment adviser clientele. The %Inst_Clients measure shows that the mean percentage of institutional clientele for the average mutual fund-month observation in the sample is 18.7% while the median percentage of institutional clientele is 15.0%. The standard deviation, 10 th and 90 th percentile values for %Inst_Clients demonstrate its significant variation. While some advisers primarily manage mutual funds, some manage a significant percentage of institutional accounts. The average fund-month observation is managed by an adviser whose client base consists of 43.3% mutual funds, indicating that many advisers focus heavily on serving mutual fund clientele. Specific institutional clientele 15 In this measure, pension plans are not included as there exists double counting between pension plans and the %Corporate j,t and %Public j,t categories which could introduce further measurement error in the variable. In Section 6 tests, I replace the sum of %Public j,t and %Corporate j,t with the percentage of the firm s clientele consisting of pension plans in this measure and baseline results are statistically and economically similar. I do not include banks, individuals, high net worth individuals, investment companies, insurance companies, other investment advisers, business development companies and other categories in my measure of sophisticated institutional clientele. Insurance companies, other investment advisers and business development companies are excluded due to insufficient data, as these categories were added more recently in Individuals and high net worth individuals are excluded by definition (i.e., they are not institutions). Admittedly, there exists a subset of highly sophisticated ultra-high net worth family office investors. As a robustness test in Section 6, I add high net worth individuals to my measure of sophisticated, institutional clients and baseline results remain statistically significant. Commercial banks or thrifts are not considered primarily due to asset class constraints. For example, according Office of the Comptroller of the Currency s Comptroller s handbook, national banks are generally prohibited from investing in stocks though there are certain exceptions. Trust departments of banks may hire third party equity managers for their high net worth clients, but in this case, the end client (i.e., asset owner) would very likely be counted in the investment adviser s Form ADV as a high net worth individual. However, because banks meet the definition of institutional clientele, I also add them to my measure of institutional clientele as a robustness test in Section 6 and baseline results are statistically and economically similar. Finally, investment companies (i.e., open and closed-end mutual funds and ETFs) are not considered sophisticated clientele. This is because mutual fund shareholders are primarily individuals. See the ICI s 2016 factbook. According to their figures, $15.6 trillion was invested in mutual funds as of December 31, Of this, $13.5 trillion or 87% consisted of individual investors and $2.1 trillion consisted of institutional investors. Furthermore, prior literature has documented the relative lack of sophistication exhibited by mutual fund investors. For example, Sirri and Tufano (1998) and Del Guercio and Tkac (2002) find that mutual fund investors use raw return performance to evaluate funds and flock disproportionately to recent winners but do not withdraw assets from recent losers. 12

13 categories with the greatest means are pension plans (9.2%), corporations (6.0%), private funds (4.8%), charities (4.6%) and public clients (3.4%). [Insert Table 1 here] 3.2 Mutual Fund Data I manually match Form ADV data with the CRSP Survivor-Bias-Free US Mutual Fund Database on investment adviser name. The baseline sample covers 2003 to CRSP provides information on fund returns, total net assets (TNA), investment objectives and other fund characteristics. Fund characteristics and returns are aggregated across share classes on an assetweighted basis using the CRSP class group variable (crsp_cl_grp). I use the oldest available share class to compute fund age. CRSP returns are net after fees, expenses and brokerage commissions but before any front-end or back-end loads. I convert all net fund returns to excess returns by subtracting the corresponding risk-free rate. 17 Holdings data are from Thomson Reuters and are merged with CRSP using the MFLINKS table. Annual data on active share for the full sample are obtained from Martijn Cremers website [see Cremers and Pareek (2016)]. Monthly time series data on the market, size, value, and momentum factors are from Kenneth French s website. The analyses focus on diversified actively managed domestic equity funds. I drop ETFs and variable annuities from the sample. Index funds are removed from the sample using the CRSP index fund flag and manual checks on fund name. Using the CRSP style code, I retain only domestic equity funds with specified objectives of large cap, mid cap, small cap, micro cap, 16 The sample begins in 2003 because lagged clientele data are used and many advisers did not begin filing Form ADV electronically until I obtain data on the monthly risk-free rate from Kenneth French s website. See I thank Kenneth French for making these data available. 13

14 growth, growth and income and equity income. Following Evans (2010), I address potential incubation bias concerns by requiring that funds have a minimum age of at least three years prior to entering the sample and a minimum TNA value of $5 million as of the previous month-end. I also require funds to have between 80% and 105% of their portfolio invested in common stocks. For funds in which holdings data are available, I require funds to hold at least 10 stocks. This yields a baseline sample which includes 199,186 fund-month observations from 2,270 mutual funds, 677 fund families, and 734 investment advisers. I compute a baseline fund performance measure which is a one-month forward-looking 4- factor alpha which follows an estimation period of 24 months [Amihud and Goyenko (2013)]. In the estimation period, I regress monthly fund excess returns (over the one-month T-bill rate) on the Fama-French and Carhart (FFC) factor returns. Using these factor loadings and the factor realizations for time t, a benchmark return is calculated as follows: Benchmark MKT MKT SMB SMB HML HML MOM MOM ri, t t 1, t 24rt t 1, t 24rt t 1. t 24rt t 1, t 24rt (2) The difference between the monthly fund excess return at time t and the monthly benchmark return gives the one-month forward-looking 4-factor alpha. Similarly, I calculate a one-month forward looking CAPM alpha using only the market factor. I also examine known predictors of mutual fund performance from prior literature including the return gap [Kacperczyk, Sialm, and Zheng (2008)] and active share [Cremers and Petajisto (2009)]. Control variables generally follow Del Guercio and Reuter (2014) and include lagged variables which have been shown to affect fund performance. These include: the natural log of fund total net assets (Log(TNA)), log of fund total net assets squared (log(tna) 2 ), log of fund family total net 14

15 assets (Log(Family TNA)), turnover (Turnover), expense ratio (Expense), fund age (Age), the sum of fund flows over the previous 12 months (Sumflows12) and the standard deviation of fund flows over the previous 12 months (Stdflows12). To account for a traditional proxy of institutional investors in mutual funds, I control for the percentage of a fund s assets which are derived from institutional share classes as identified using the institutional fund flag in CRSP (Inst_ratio). In further analyses, I consider the adviser s AUM which are not managed in mutual funds. This measure is obtained from the total AUM reported on the firm s ADV Item 5F minus the sum of the total net assets of the adviser s mutual funds in CRSP for the corresponding time period. All continuous variables (including ADV clientele variables) are winsorized at 1 and 99 to account for the possibility of outliers. Panel A of Table 2 presents summary statistics on dependent and control variables. The average actively managed mutual fund charges an annual expense ratio of 1.20% and underperforms by 1.12% per year on an after-fee risk-adjusted basis which is consistent with Fama and French (2010). The average fund-month observation is managed by an adviser that has 37% of its assets owned by non-mutual fund clients. [Insert Table 2 here] Panel B of Table 2 partitions fund observations in the sample at the lagged median values of %Inst_Clients for each month and examines the differences in means of dependent and control variables. The results indicate that funds managed by advisers with above median levels of institutional clientele generate a significantly greater one-month 4-factor alpha than funds managed by advisers with below median levels of institutional clientele before controlling for any 15

16 other observables. The difference in means suggests that a fund with above median %Inst_Clients earns approximately 37 basis points higher annual 4-factor alpha. Differences in means for the control variables show that funds managed by advisers with a greater percentage of institutional clients are significantly smaller ($471 million) and belong to families with significantly lower TNA ($71.2 billion) on average. These funds also have significantly lower turnover (2.5%), annual expenses (0.09%), and a greater percentage of assets are derived from institutional share classes (10.3%). In Section 6, I conduct propensity score adjusted analyses as a robustness check and show that the baseline results are not driven by systematic differences in observables between mutual funds managed by advisers with high and low percentages of institutional clientele. 4. Do Institutional Mutual Fund Advisers Earn Higher Risk-adjusted Returns? I examine differences in risk-adjusted performance within a regression framework that controls for observables. Table 3 presents panel OLS regressions which regress measures of mutual fund performance on lagged measures of investment adviser institutional clientele and controls. Each regression includes investment objective-by-month fixed effects so that each performance measure is relative to other actively managed funds with the same investment style, operating in the same month. The main variable of interest is the %Inst_Clients measure of sophisticated, institutional clientele for each fund s investment adviser which is standardized. Standard errors are clustered on both investment adviser and month. [Insert Table 3 here] 16

17 Columns (1), (2) and (3) of Table 3 show that the percentage of institutional clientele for an investment adviser is positively associated with higher risk-adjusted, after fee returns realized by its actively managed equity mutual funds in the next month. The coefficient estimates for %Inst_Clients in all three columns are statistically significant at the 1% level. Column (1) regresses 4-factor alpha on %Inst_Clients without control variables. The coefficient estimate indicates that a one standard deviation (19%) increase in the percentage of institutional clientele for the investment adviser is associated with a higher annual 4-factor alpha of 19 basis points. 18 Once controls are introduced into the model in columns (2) and (3), the results show that a one standard deviation increase in the percentage of institutional clientele for the investment adviser is associated with higher annual CAPM and 4-factor alphas of 18 basis points and 16 basis points respectively in the managed mutual funds. The estimates are economically significant and amount to approximately a one basis point increase in annual outperformance for every additional percentage point increase in institutional clientele for the fund s investment adviser. Column (4) of Table 3, focuses on the return gap measure of Kacperczyk, Sialm and Zheng (2008), which is the difference between fund i s actual gross return and the gross return implied by the fund s lagged reported holdings. This measure captures unobservables such as the value added by a skilled manager or favorable IPO allocations, or the value destroyed by poor trade executions or agency costs. I find that a significant portion of the difference in returns between funds managed by advisers with varying levels of institutional clientele is due to differences in return gaps. The results in column (4) for the return gap measure are consistent with the idea that advisers with more sophisticated client bases are better skilled and/or informed. 18 This is calculated by multiplying the standardized monthly coefficient estimate by 12 to obtain an annual estimate. 17

18 Column (5) of Table 3 examines the active share measure of Cremers and Petajisto (2009), which has been shown to predict mutual fund performance. Active share is the fraction of fund i s assets that would need to be traded to obtain a portfolio that mirrors fund i s benchmark. Observations in column (5) for active share are lower due to data availability being at an annual frequency. Controlling for observables, I find evidence that funds managed by advisers with higher percentages of institutional clientele are actively managed to a greater extent. The estimates in column (5) indicate that an adviser s percentage of institutional clientele has a significant and positive association with the active share of its mutual funds. A one standard deviation increase in institutional clientele for the adviser is associated with approximately a one percentage point increase in the active share of its mutual funds. Consistent with Gârleanu and Pedersen (2017), the taking of greater active risk relative to a fund s benchmark also suggests advisers with more sophisticated client bases are likely better informed. Taken together, the baseline multivariate results present evidence that mutual funds managed by investment advisers with more sophisticated client bases earn higher risk-adjusted returns. Much of the difference in 4-factor alphas can be explained by the difference in return gaps as well as the difference in active share of the funds, suggesting that these advisers exhibit superior interim trading skills and/or are better informed. The results are consistent with Gârleanu and Pedersen (2017) who predict that managers with more sophisticated clients should outperform. 5. Testing Empirical Predictions of the Gârleanu and Pedersen Model 5.1 Sophisticated, Institutional Investor Capital Levels Analyses up to this point consider the percentage of institutional clients for each investment adviser. It is possible for an adviser to mutual funds to have many smaller institutional clients and 18

19 a few large mutual funds which drive the adviser s revenues and decision to invest in information collection. In this case, the adviser s funds may not be expected to outperform because investors may not have sufficient capital levels to incur search costs. Gârleanu and Pedersen (2017) suggest that larger institutional clients would be more likely to incur search costs because they have resources to employ manager research personnel or investment consultants. Hence, I predict the positive relationship between %Inst_Clients and fund risk-adjusted performance is primarily driven by advisers which derive a significant percentage of their AUM from large institutional clients outside of their mutual funds. 19 Table 4 presents the results of OLS regressions of one-month ahead 4-factor alpha on %Inst_Clients and the percentage of non-mutual fund AUM for the investment adviser. For the percentage of non-mutual fund AUM managed by the adviser, I create a dummy variable equal to one if fund i is managed by investment adviser j manages an above median percentage of their AUM for clients other than mutual funds in month t-1 and zero otherwise (%Non-MF AUM dummy). Controls, fixed effects and clustered standard error specifications are identical to the baseline specifications. Interestingly, the coefficient estimate for %Non-MF AUM dummy is negative and significant in column (3) when controlling for %Inst_Clients. This result is plausible because non-mutual fund AUM may be owned by less sophisticated separate account clients, such as individuals. In fact, the coefficient on %Non-MF AUM dummy in column (4) represents the 19 The key assumption is that significant sums of assets not managed in mutual funds are likely derived from large institutional clients. To the extent that an investment adviser s non-mutual fund assets are attained in significant quantities from individuals directly, this would likely bias against finding results. Industry data sources indicate that the vast majority of institutional assets are not invested in mutual funds, but instead are invested in separate accounts and private commingled funds. Gerakos, Linnainmaa, and Morse (2017) cite P&I surveys 2012 data which shows that out of $48 trillion in delegated institutional assets only $5 trillion was managed in mutual funds. The ICI s 2016 factbook offers a similar picture of institutional investment in mutual funds. According to their figures, $15.6 trillion was invested in mutual funds as of December 31, Of this, $13.5 trillion or 87% consisted of individual investors and $2.1 trillion consisted of institutional investors. 19

20 estimate for %Non-MF AUM dummy when %Inst_Clients is equal to zero, suggesting the negative relationship is driven by the presence of outside assets owned by less sophisticated clientele. 20 [Insert Table 4 here] Column (4) of Table 4 presents the results of one-month ahead 4-factor regressed on %Inst_Clients, %Non-MF AUM dummy and the interaction term between the two variables. The coefficient on the interaction term indicates that the relationship between %Inst_Clients and 4- factor alpha is significantly stronger when the adviser derives a larger percentage of its assets from clients other than mutual funds. For these funds, the added coefficients from %Inst_Clients and the interaction term imply a one standard deviation increase in the adviser s percentage of sophisticated, institutional clientele is associated with a 31 basis point increase in annual 4-factor alpha realized in its mutual funds. The results indicate that the positive relationship between an adviser s percentage of sophisticated clientele and the performance of its mutual funds is only significant when institutional clients make up a substantial fraction of the investment adviser s total assets. This is consistent with Gârleanu and Pedersen s prediction that larger institutional clients have greater resources to incur search costs. 5.2 Asset Market Efficiency and Asset Manager Market Efficiency Empirical predictions by Gârleanu and Pedersen (2017) further link the efficiency of the market of underlying assets to those for asset managers. One of their key results suggests that the 20 In untabulated tests, I regress one-month ahead 4-factor alpha on the percentage of the investment adviser s separate account clients which are individuals using the baseline specification. Consistent with this result, I find the coefficient estimate is negative but not statistically significant. 20

21 outperformance of investors which incur search costs is larger in less efficient markets. I test this prediction in Table 5 by restricting the sample to small and mid cap funds and repeating the analyses performed in Table 3. To the extent that pricing of mid and small cap stocks is less efficient than large cap stocks, the returns to investing in active management should be higher among mid cap and small cap funds according to the model. [Insert Table 5 here] In Table 5 where the sample is restricted to small and mid cap funds, the coefficient estimates indicate that differences in percentages of institutional clientele are associated with greater differences in risk-adjusted performance relative to the entire sample. For small and mid cap funds, a one standard deviation (20%) increase in the percentage of institutional clientele for the adviser is associated 23 basis points higher annual CAPM and 4-factor alpha. Furthermore, the coefficient estimates for return gap and active share in columns (4) and (5) are also larger in magnitude and exhibit greater statistical significance. In summary, the results in Table 5 confirm the empirical prediction of Gârleanu and Pedersen (2017) that investors who incur search costs outperform to a greater extent in less efficient markets. 5.3 Returns to Highly Active Funds and the Presence of Sophisticated Clientele Results in Section 4 provide evidence that the presence of institutional clients for an investment adviser is positively associated with the degree to which their mutual funds are actively managed. However, if advisers with higher percentages of sophisticated clients are truly better informed than other advisers, then it would be expected that this outperformance would manifest to the greatest 21

22 extent in funds which take greater active bets relative to indices or known risk factors. Based on predictions from Gârleanu and Pedersen (2017), I expect that the positive relationship between a fund s activeness and future risk-adjusted returns to be most significant for funds which are managed by advisers with higher percentages of sophisticated, institutional clientele. The subsequent analyses examine both returns-based and holdings-based measures of fund activeness: the R 2 measure of Amihud and Goyenko (2013) and the active share measure of Cremers and Petajisto (2009). Table 6 presents results of OLS regressions of one-month ahead 4-factor alpha on the adviser s lagged percentage of institutional clientele and lagged active share or R Both active share and R 2 are standardized in each sample. Controls, fixed effects and clustered standard error specifications are identical to baseline specifications. I create two dummy variables from the %Inst_Clients measure of clientele. The first, in Panel A, is equal to one if fund i managed by investment adviser j has an above median percentage of institutional clientele in month t-1. The second, in Panel B, is equal to one if fund i managed by investment adviser j has a percentage of institutional clientele which is in the top quartile of the sample in month t-1. This is done to explore the relationship between fund activeness and alpha for funds managed by advisers with both high and low percentages of institutional clientele. The use of different dummy variables in Panels A and B allows for varying cutoff levels to designate a high percentage of institutional clientele for a given investment adviser. Columns (1) and (3) show that the adviser s percentage of institutional clientele, both as an above median and top quartile dummy variable, is still positive and significantly associated with 4-factor alpha. Consistent with prior literature, the coefficient on 21 R 2 is obtained from the regression of excess returns on FFC factors over the prior 24 months per Amihud and Goyenko (2013). Active share is the reported active share from Cremers website in the prior calendar year. 22

23 active share is positive and statistically significant. The coefficient on R 2 is negative but not significant in the baseline sample. 22 [Insert Table 6 here] Columns (2) and (4) of Table 6 examine the interaction between fund activeness and the %Inst_Clients dummy. The results in Panel A show that the significant positive (negative) relationship between fund activeness (R 2 ) and future risk-adjusted performance is only present in the subset of mutual funds managed by advisers with high percentages of sophisticated, institutional clients. Panel B indicates that the results are even more pronounced amongst advisers in the top quartile of the sample based on %Inst_Clients. The table shows that the well-documented relationships between both active share and R 2 and future risk-adjusted performance appear to be concentrated within mutual funds managed by investment advisers with higher percentages of sophisticated clientele. The difference in results between Panel A and Panel B suggests that as the adviser s percentage of institutional clientele increases, there are incremental returns to highly active funds. The findings presented in this section suggest that advisers with higher percentages of institutional clientele are better managers of highly active funds than advisers with lower percentages of institutional clientele. This is consistent with the notion that sophisticated institutional clients hire managers who are better skilled or more informed. The results also offer 22 This result differs from the findings of Amihud and Goyenko (2013). I am able to obtain statistically significant results after correcting for differences in sample selection by excluding funds with lower than $15 million in TNA from the sample and restricting the time period to , the period which overlaps with their study. I also find that statistical significance for R 2 is highly sensitive to the decision to cluster standard errors on month. 23