Whom do you trust? Investor-advisor relationships and mutual fund flows

Transcription

1 Whom do you trust? Investor-advisor relationships and mutual fund flows Leonard Kostovetsky 1 Simon School, University of Rochester leonard.kostovetsky@simon.rochester.edu Abstract I measure the value that investors place on trust and relationships in asset management by examining mutual fund flows around announced changes in the ownership of fund management companies. I find a decline in flows of around 7% of fund assets in the year following the announcement date, starting after announcement and accelerating after the closing date of the ownership change. A decomposition into inflows and outflows shows that the overall decrease in flows is entirely driven by increasing outflows with no change in inflows. Retail investors and investors in funds with higher expense ratios are most responsive to ownership changes, providing new evidence that such investors place a significant value on trust and are more likely to respond to a relationship disruption by withdrawing their assets. Alternative explanations such as changes in distribution network, reactions to expected fund closure, expected or past manager changes, or poor expected returns do not seem to explain the results. 1 The author can be reached at the following address: Simon Graduate School of Business, University of Rochester, Rochester, NY Phone: (585) , leonard.kostovetsky@simon.rochester.edu

2 1. Introduction An important unanswered question in the field of delegated asset management is how much importance investors place on who is managing their money. Asset management companies spend more than a billion dollars each year on advertising (Gallagher, Kaniel, and Starks, 2006), much of it trying to persuade investors that their firm will provide them with trustworthy and dependable financial advice (Mullainathan, Schwartzstein, and Shleifer, 2008). Gennaioli, Shleifer, and Vishny (2012) propose that the well-documented empirical finding that average active mutual fund alphas are negative (e.g., Jensen, 1968) is due to a trust premium, which allows asset management firms to charge investors additional fees if there is a trusting relationship between them. They write that trust can be established through personal relationships, familiarity, persuasive advertising, connections to friends and colleagues, communication, and schmoozing, all of which are likely to be disrupted by an exogenous change in firm management. In this paper, I measure the value of trustworthy relationships between investors and asset management firms by examining mutual fund flows around management company ownership changes. My main finding is that mutual fund flows turn negative in response to announced changes in the parent company or ownership of a fund s advisor. A reduction in flows begins after the announcement date and is initially about 3% of assets (on an annualized basis), and then accelerates after the closing date to total approximately 7% of assets over the twelve months following the announcement date. The results are robust to controlling for fund characteristics such as the past five years of returns, age, fund and family size, and style, as well as parent company characteristics (for public parent companies) such as the parent s market capitalization and past year s stock returns. 1

3 An alternative empirical strategy would be to look at flows around individual manager changes. The main problem with that strategy is that manager changes are highly correlated with past fund performance (Chevalier and Ellison, 1999) and fund flows are also extremely sensitive to past performance (Sirri and Tufano, 1998), making it difficult to disentangle performancedriven outflows from outflows due to manager changes. In addition, there is a reverse causality problem if managers can anticipate future flows and voluntarily depart the fund when they expect fund outflows, and therefore reductions in assets under management and their own compensation. My empirical strategy begins with an examination of 185 events (covering 843 funds) from 1995 through 2011, where there is a change in the ownership of the fund s management company (mergers and acquisitions involving the management company itself or its parent). One example of such a merger occurred in 2001 when Deutsche Bank announced its purchase of Zurich Scudder, the manager of the Scudder Funds, from Zurich Financial. In order to control for parent company characteristics, I next restrict the sample to the 78 events (covering 391 funds) involving ownership changes of U.S. public parent companies. While management company ownership changes are less likely to be driven by a particular fund s performance than manager changes, it is still possible that they are related to the entire management company s past investment performance. In order to rule out such endogeneity concerns, I perform a more rigorous test, restricting the event space to the 70 events (covering 295 funds) in which the public parent company undergoing an ownership change derives a small share of revenues (<10%) from its mutual fund operations, and find similar results. Next, I test a number of different explanations for my main results. One explanation is that a group of investors attach significant value to their relationship with the fund s management 2

4 company (e.g., due to advertising or past experience), and move their savings elsewhere when the ownership change disrupts this relationship. However, new investors might also avoid investing in the fund while its management is transitioning from one owner to another owner. I decompose fund flows into inflows (purchases of shares by investors) and outflows (sales of shares by investors), and find that while there is little change in inflows around the announcement date, there is a large increase in outflows that leads to the reduction in total fund flows. The importance of trust and relationships should also be more important for less sophisticated investors who don t have the skills or resources to monitor the fund s management. I test the trust hypothesis by separately looking at the effect of ownership changes on retail class flows and institutional class flows to examine whether investor sophistication is an important factor. I find that outflows are driven by retail class investors, and that investors in institutional classes do not react adversely to changes in ownership. This result might also explain why flows only slowly react to announcement changes. Limited attention is well documented among retail investors (Barber and Odean, 2008), which is why retail investors slowly find out about the ownership change. After the closing date, the news of the ownership change is more likely to filter through to retail investors, as it appears in the fund prospectus and other disclosure documents. In a similar vein, I separate my sample of funds into high-expense funds and low-expense funds, and then examine the effect of ownership changes on the flows of each group. I find the decline in flows from pre-announcement to post-announcement is anywhere from 25% to 100% bigger for high-expense funds relative to those with low expense ratios. This supports the thesis that a component of the expense ratio is a trust premium, since investors in funds with higher 3

5 expense ratios are more likely to withdraw their money when their prior relationship is broken due to a change in ownership. Next, I examine whether characteristics of the acquiror or the purpose behind the acquisition affect the fund flows after the announcement date. Interestingly, I find that investors react in a more negative way when a bank acquires their fund s management company than when the acquiror is an insurance or securities firm. However, neither the purpose of the merger, nor the past stock performance or past fund performance (either of the whole family or just funds in the same style as the target fund) of the acquiring firm affects post-announcement fund flows. This result might indicate that investors don t believe the past performance of the acquiring firm will necessarily carry through to their fund, and/or that the reason for their sale is the disruption of their relationship with the prior organization that had been managing their fund. I then test a number of alternative explanations for the paper s main results. Changes in ownership of management companies may also coincide with changes in distribution channels. Del Guercio, Reuter, and Tkac (2010) document the various distribution channels used by mutual fund families, and Bergstresser, Chalmers, and Tufano (2009) provide evidence on the importance of brokers in portfolio decisions made by retail investors. A change in distribution channel might lead brokers to counsel their clients to pull out money from the fund, an effect that would have nothing to do with a disruption of trust between investors and fund management. I test this hypothesis by controlling for the main distribution channel used by the fund, and find that my results are robust to these controls. I then drop any funds that underwent both an ownership change as well as a change in the primary type of distribution channel, leaving just funds whose distribution channel remained the same after the announcement of the ownership 4

6 change. Even among this subset of funds, there are economically and statistically significant outflows after the announced ownership change. The dot-com bubble presents some concerns as well. There is a significant clustering of M&A events around the dot-com bubble since merger activity usually peaks during market booms and the financial industry was undergoing consolidation at the time after the repeal of the Glass-Steagall Act. I test whether my results are coming from this clustering by dropping all fund-month observations from 1999 and I find that my main results are robust to exclusion of the period around the dot-com bubble. Another possible explanation is that advisor ownership changes are associated with an increase in manager turnover and fund closures as the new owners tweak the array of offered funds and the managers of those funds. Investors might be reacting to expected or realized manager changes or announced fund closures by withdrawing money from the fund. I test this hypothesis by including dummy variables that indicate whether the fund will close in the next six months and whether the manager will change in the next six months or has changed in the prior six months. My results are robust to inclusion of these controls. Another possibility is that the decline in asset flows after an announced change in ownership is a rational reaction to expectations of lower returns. For instance, during the period of transition, the management firm might not be putting in maximum effort in fund management, and investors might temporarily be leaving the fund to avoid this period of lower expected returns. I test whether performance is affected by ownership changes, and find no evidence that mutual funds underperform in the year following an announced ownership change. This paper builds on the growing literature focusing on the importance of trust, familiarity, and loyalty in investment. For instance, Guiso, Sapienza, and Zingales (2008) 5

7 highlight the importance of trust in stock market participation. French and Poterba (1991), Coval and Moskowitz (1999), and Huberman (2001) all provide evidence on the importance of familiarity and geographic proximity for investment decisions. Cohen (2009) highlights the effect of loyalty by studying employee decisions to invest in their company s stock. My findings complement this literature by highlighting and measuring the role that trust and familiarity play in investors choices of asset managers, through the use of exogenous breaks in the adviserinvestor relationship. This paper also contributes to prior research on the role of mutual fund parent companies. Sialm and Tham (2011) find positive spillover effects from the performance of the parent company s stock to the ability of the mutual fund to attract investors. A number of papers including Ferris and Yan (2009) and Adams, Mansi, and Nishikawa (2012) highlight the importance of agency issues at advisory firms. Massa and Rehman (2008) show that information flows from bank parent companies to affiliated mutual funds, allowing these mutual funds to outperform on stock investments in companies that have borrowed from (and therefore provided private information to) the bank. My paper also uses mergers and acquisitions of financial institutions as exogenous identification in a manner similar to Hong and Kacperczyk (2010). Most papers that have looked at mergers in the context of mutual funds have focused on mergers between funds (e.g. Khorana, Tufano, and Wedge, 2007), and not mergers at the family or adviser level. An important exception is Allen and Parwada (2006) who look at a subset of parent company mergers for mutual funds in Australia from 1995 to 1999, and also find evidence of negative outflow reactions. However, their focus is on excessive size and its negative effect on performance as the main culprit for investor adverse reaction to ownership changes. In contrast, my paper focuses on 6

8 a much larger set of U.S. firms, and tests the hypothesis that trust and relationships between investors and the advisor explain the empirical findings. In summary, my paper uses an exogenous shock to the investor-advisor relationship to measure investor reaction, and thus provides evidence of the significant value that investors attach to this relationship. It underlines the notion that past (and expected future) performance and expense ratios are not the only factors in how investors, especially retail investors, make mutual fund investment decisions. 2. Data The main data sources for this paper are the CRSP Survivor-Bias-Free US Mutual Fund Database, annual Morningstar Principia CDs, and the SDC Platinum M&A database. Additional data on fund advisers is downloaded from the SEC Investment Adviser Public Disclosure (IAPD) database 2 and SEC EDGAR, and stock-level data is collected from the CRSP/Compustat database. Fund inflows and outflows are collected directly from NSAR filings on EDGAR. Data on primary distribution networks is from Strategic Insight. Mutual Fund Sample: The sample consists of all domestic, diversified, actively-managed, equity mutual funds operating from 1995 through I construct this sample by merging CRSP and Morningstar, using ticker symbols and (when ticker symbols are missing) fund names. I then exclude all funds outside the nine main style boxes (e.g., smallcap value, largecap blend, etc.) leaving only domestic diversified equity funds. Finally, I eliminate index funds by removing all funds with the words index, S&P, Dow Jones, and NASDAQ in the fund name, and by excluding all funds in the Dimensional Fund Advisors (DFA), Direxion, Potomac, ProFunds, 2 The website for this service is: 7

9 and Rydex fund families. ETFs are also excluded by removing all observations with the word ETF in the fund name or funds with ticker symbols of four or fewer characters. I aggregate funds across fund classes using the Morningstar portfolio identifier (PORTCODE) or MFLinks variable (WFICN). I remove incubated funds by excluding funds that were not contemporaneously reported in Morningstar or had a blank CRSP fund name at the start of the calendar year. I also drop funds with less than $10 million in assets under management, as flows in these funds are highly volatile and contaminated by seeding from the fund family. This leaves 351,120 portfolio-month observations with the number of funds growing from 945 funds in January 1995 to 1,665 funds in December Fund Advisers and Adviser Ownership Changes: Morningstar is the main source for mutual fund advisers. I crosscheck the Morningstar adviser with the CRSP Management Company identifier and find that they match for over 80% of observations. However, CRSP sometimes reports the fund distributor as the management company, which is why I rely on Morningstar for this variable. I find the parent companies of mutual fund advisers by entering each adviser s name into the SEC s IAPD online database and looking up the Schedule A of Form ADV, which lists all direct owners and executive officers. For example, for Dreyfus Corporation, adviser to the Dreyfus funds, the Form ADV Schedule A shows that Bank of New York Mellon is the sole shareholder of this company. IAPD includes defunct fund advisers but it only began operations in 2000 so fund advisers that went defunct prior to 2000 are not included. Therefore, I gather ownership information on these companies by looking through mutual fund proxy documents on SEC EDGAR. 8

10 The IAPD database only includes current ownership information (or last reported ownership information for defunct firms). Therefore, I manually look up all fund advisers and their parent companies in the SDC Platinum M&A database, and note any changes in ownership and the announcement and effective (closing) dates for each ownership change. I also gather information from SDC Platinum on the identity, public status, and industry of the acquiring company, as well as the purpose or purposes for the merger/acquisition. Whenever Morningstar shows that a fund or fund family changes advisors or is merged into another fund or fund family and I can find no corresponding ownership change in SDC Platinum, I examine mutual fund proxy documents on SEC EDGAR to determine the reason for the change. Overall, I find a total of 185 parent company changes that were announced from July 1995 through December Initial public offerings and management buyouts are not included because they also coincide with the decisions to go public or private, which might have their own implications for fund flows. I use SDC Platinum to identify publicly traded parent companies (of acquirors and targets) and match them to CRSP using CUSIPs. I also look up several foreign parent companies on Google Finance to determine their public status. I define a fund as privately owned, with Private firm (dummy) set to one, if that company is not in CRSP, it is not traded on a foreign exchange, and it is not a mutual insurance company or non-profit organization. Privately owned advisory companies make up approximately 40% of the funds in the sample, but manage over half of the assets under management. This disparity is due to the fact that extremely large mutual fund advisers such as Fidelity Management & Research (Fidelity Funds) and Capital 3 Ownership changes announced in the first six months and last twelve months of the sample period are exclude because those months are required for studying fund flows around announcement dates. 9

11 Management & Research (American Funds) are privately held. These estimates are consistent with earlier research on public vs. private ownership of mutual fund management companies. For public firms that are available in CRSP/COMPUSTAT, I gather data on market capitalization and past stock returns from CRSP, and total revenues and segment revenues data from COMPUSTAT. Overall, 78 of the 185 ownership changes involve acquisitions of public (parent) companies. In mergers of equals such as the 1998 deal between Citicorp and Travelers, the company that is delisted in CRSP (in that case, Citicorp) is the one that is deemed to have a change in ownership. In order to avoid possible endogeneity concerns, I also run tests on firms whose main line of business is not in asset management, and whose change in ownership is therefore less likely to be related to anything happening at the mutual fund family. For each fund advisor whose parent company has revenues data in Compustat, I calculate the total estimated annual revenues of its mutual fund family 4 and divide by the parent company s revenues in the same fiscal year to calculate the Mutual fund revenues (%) variable. Non-asset management parent companies are defined as having less than 10% of total revenues coming from estimated mutual fund revenues. In addition, I check the Compustat Segments database to exclude all firms whose entire asset management segments produce revenues greater than 20% of total revenues. In total, 70 of the 78 ownership changes involving public parent M&A happen at non-asset management firms, mostly commercial and investment banks. Table 1 presents summary statistics on ownership change announcements for mutual fund advisory firms. Panel A displays the number of events, number of funds involved in each event, and the assets under management of those funds, for each year. Columns 1 through 3 show a 4 Estimated mutual fund revenues are defined as (1/12 Annual Expense Ratio Assets under Management) across all fund-month observations of a fund family in a particular fiscal year. 10

12 total of 185 announced ownership changes from July 1995 through December 2011, involving 843 funds with $587 billion in assets under management at the time of the announcement. M&A activity was strongest in the first six years of the sample period when the stock market was booming in the late 1990s and the asset management business was also experiencing significant growth. Columns 4 through 6 only include ownership changes due to public parent M&A, and show a total of 78 events involving 391 funds managing $203 billion. Finally, Columns 7 though 9 show summary data on ownership changes due to non-asset management public parent M&A. Among this subgroup, there are 70 events involving 295 funds managing $145 billion. Panel B of Table 1 shows a breakdown of the merger types that make up the events used in this paper. Among the entire sample of events, the merger types are fairly evenly distributed between banks acquiring other banks, securities firms acquiring other securities firms, and banks/insurance companies buying other securities firms. On the other hand, public parent M&A in Columns 3 through 6 is mostly dominated by bank mergers, with over two-thirds of events consisting of this merger type. Banks are larger and are therefore more likely to be publicly traded than asset management firms, which is why there is such a dramatic change in merger types. Appendix A shows fifteen examples of mergers used in this paper, with detailed information on the acquiror and target, as well as the announcement date and effective date for the merger. Fund Characteristics: The main variable of interest for this paper is monthly mutual fund flows. In order to calculate flows, I download data from the CRSP Mutual Fund database on monthly assets under management and net returns. Fund flows ($mil) in month t is defined as: (Eq.1) Fund flows ($) = Assets (end of t) Assets (start of t) (1 + Net Returns (over month t)) 11

13 Since this quantity is usually proportional to fund size, I standardize it by dividing by assets: (Eq.2) Fund flows (%) = Fund flows ($)/Assets (start of t) I adjust flows to eliminate assets added from another fund that was merged into the fund. Finally, in order to eliminate the effect of outliers, fund flows (%) is winsorized at the 1% and 99% level. Table 2 reports time-series averages of cross-sectional summary statistics for fund flow variables. Flows over this sample period are fairly close to zero. Although firms had average monthly inflows of approximately $0.3 million or 0.4% (4.8% on an annualized basis), the median firm experienced slight outflows due to the fact that inflows tend to be concentrated among the funds with the best past performance. Standard deviation of monthly flows, even after winsorizing, is 4.5%, which highlights the significant cross-sectional variation in fund flows. In most of the tests in the paper, I control for a number of fund variables that have been shown in the past to predict fund flows. These variables include past fund performance, fund assets under management (fund AUM), family assets under management (family AUM), fund age, and expense ratio. The prior literature on fund flows found a non-linear relationship between past performance and fund flows. In order to capture this non-linear relationship, I sort firms into deciles for each year s style-adjusted return from the past five years, and include five sets of past return decile dummies as controls in all specifications. Newer funds that weren t around for all five years and therefore don t have returns for a particular prior year are placed in a separate bucket (in addition to the 10 decile groups) for that year, which has its own dummy variable. The summary statistics in Table 2 show that the distributions of Fund AUM, Family AUM, Fund age, and Expense ratio, are positively skewed so I transform them with the natural logarithm and use the transformed variables as predictive variables in regression tests. Because 12

14 the paper looks at shocks to the parent company of the fund s adviser, I also collect parent company characteristics. Table 2 shows that 42% of funds have a privately owned adviser, 35% of funds have a publicly owned adviser whose parent company is not primarily an asset manager. Public parent companies have average monthly returns over the prior year of just under 1%, and derive 8% of revenues from mutual fund fees. Fund Inflows and Outflows: For part of the analysis in the paper, I decompose fund flows into inflows (dollar value of purchases of fund shares) and outflows (dollar value of sales of fund shares). Data on inflows and outflows is included in the semiannual NSAR filing made by each fund family. I use a script to download all NSAR filings from the EDGAR database, and match them to funds using fund name. Because there are often slight variations in fund names, I attempt to manually match any unmatched observations. NSARs also include assets under management so I confirm matches using this variable. Using machine and manual matching, I obtain inflow/outflow data for nearly 90% of the fund-month observations in my sample. Table 2 includes summary statistics on inflows and outflows. The monthly inflows for a typical fund are 3.8% of its assets under management at the start of the month, but 10% of funds have inflows exceeding 8% of assets, confirming the skewed nature of inflows as investors put new funds into the top past performers. Monthly outflows average 3.2% of assets under management and are less skewed with the 90 th percentile at 6%. Distribution Channels: My main source for distribution channels is a dataset provided by Strategic Insight. The Strategic Insight dataset includes current distribution channel data on each fund class as well as archival data on distribution channels for defunct funds and families. Generally, fund classes labeled A, B, C, and R are sold through brokers, fund classes labeled I (Institutional) or Retirement are sold to institutions, and fund classes labeled N or Retail or with 13

15 no class label are sold directly to investors (Direct). Brokers can be affiliated with the management firm and are then classified depending on whether the parent company is a bank (Bank Proprietary), insurance company (Insurance), or securities company (Proprietary), or not affiliated with the management company at all (Non Proprietary). Families generally use only one of these four distribution channels for non-direct and non-institutional sales. Finally, some funds are sold to members of the fraternal, religious, or non-profit organization that runs the fund (Other). I aggregate the total assets for each type of distribution channel across fund classes and then designate a fund portfolio s main distribution channel as the channel that has the most assets under management. Most portfolios (and fund families) distribute a significant proportion of assets using one distribution channel. The average amount distributed by a portfolio s top distribution channel is 95% with a median value of 100%. Non-Proprietary is the most common distribution channel used by approximately one-third of funds, followed by Direct distribution used by one-quarter of funds, and Institutional used by about 20% of funds. Other Variables: I collect a number of additional variables in order to test different theories for the paper s main results. For each fund class, I collect data from Morningstar on whether it is only open to institutional investors or whether retail investors are also allowed to invest in the class. 5 About 20% of fund classes in the sample are only open to institutions. Morningstar also reports manager names and tenure dates and is my source for the dates of manager changes. I use CRSP for fund closure dates. 3. Main Results 5 CRSP also has an institutional dummy variable but it is only available after 2000, which is why I use the Morningstar variable. 14

16 Before studying how flows are affected by ownership change announcements, I examine whether such announcements are actually exogenous events or whether they can be predicted by (and are correlated with) past fund performance or other fund characteristics. In Table 3, I run PROBIT regressions with an event announcement dummy, which equals one if there was an announced ownership change of the fund s adviser in the current month and zero otherwise, as the dependent variable, and fund/parent company characteristics as the explanatory variables. The announcements include completed mergers and one uncompleted merger, Zion s Bancorp attempted purchase of First Security Corp. (manager of the Achievement fund family) that was rejected by Zion s shareholders in As with all the tests in this paper, I first run regressions for the sample of all ownership changes of fund advisers (Columns 1 and 2), then restrict the sample to public ownership of fund advisers where we have publicly available data on the parent companies (Columns 3 and 4), and finally include only publicly owned advisers whose parent companies main line of business is not asset management (Columns 5 and 6). The main advantage of the sample restrictions is that the events are more likely to be exogenous to what s happening at the mutual fund level, while the main disadvantage is a reduction in the number of observations. Columns 1, 3, and 5, of Table 3 include a simple measure of past performance, the average style-adjusted returns over the past year. Columns 2, 4, and 6, use a more comprehensive measure, five sets of return decile dummies for each of the past five years. The main takeaway from Table 3 is that mutual fund and parent company characteristics are generally not predictive of event announcements. This indicates that the announced ownership changes are in fact exogenous and the windows around the changes can be used as a laboratory to study the effect on flows. We can see that past performance measures (returns and 15

17 flows) are not significant predictors of ownership changes. In fact, the coefficient on past styleadjusted returns is actually positive (although insignificant) indicating that better-performing funds are more likely to be subject to an adviser ownership change. The only significant (at the 5% level) predictive variables in any of the specifications are family size and parent company size, with the negative coefficients indicating larger fund families and parent companies are less likely to have an adviser ownership change. This finding may be due to capital constraints since there are very few investors or firms who can buy the largest asset management firms or parent companies. After confirming that ownership changes are largely unrelated to mutual fund characteristics, I next calculate average flows in the event window around the ownership changes. I define PREANN and POSTANN dummy variables for the timing of each observation around the event window. PREANN is set to one for all fund-month observations in the 6 months prior to the announcement date of an ownership change of the fund adviser, and zero otherwise. POSTANN is set to one for all fund-month observations on the announcement date and for one year after the announcement date of an ownership change of the fund adviser, and zero otherwise. Panel A of Table 4 shows the average value of monthly Fund flows (%) for all observations in each event window. As in Table 3, I start with the entire sample of events in Column 1 and restrict the sample to public parent companies in Column 2 and public non-asset management parent companies that undergo M&A in Column 3. Prior to the announcement date (PREANN=1), flows are not statistically different from zero. After the event announcement date (POSTANN=1), we can see statistically significant outflows. For example, in the sample of all events, the average value of Fund flows (%) in the post-announcement window is % 16

18 (~6.6% on an annualized basis) with a t-statistic of The results are very similar in the sample of all public parent mergers (Column 2). Finally, in the smaller sample of non-asset management mergers (Column 3), the average value of fund flows (%) in the post-announcement window is % (~8.8% on an annualized basis) with a t-statistic of Across all three specifications, the pattern of statistically insignificant flows prior to the announcement and strong outflows after the announcement is repeated. Two possible explanations for the results in Panel A are that the announcements happen to funds that are, for other reasons, likely to experience outflows, or that they are clustered in periods prior to fund outflows such as market peaks. In order to test these explanations, I construct a matched sample for each event-window observation and then calculate the average of match-adjusted flows (fund flows relative to matched sample). The matched sample flows for a particular fund-month observation are a weighted average of flows across all funds in the same month and in the same fund style (and in the same restricted sample in Columns 2 and 3), where the weights are proportional to closeness based on differences in size, past five years of returns, and fund age. Appendix B at the end of the paper describes the matching algorithm. Panel B of Table 4 presents average values of monthly match-adjusted Fund flows (%) for observations in each event window. As in Panel A, flows are indistinguishable from zero prior to the announcement date, but turn lower after the announcement date. The average flows are higher for both the pre- and post-announcement periods (due to the match adjustment) but the difference between pre- and post- remains very similar suggesting that it is not unique timing or differences in characteristics that explain the downturn in flows after the announcement dates of adviser ownership changes. 17

19 Next, I use a difference-in-difference approach, comparing match-adjusted fund flows in the post-announcement period to the pre-announcement match-adjusted fund flows, for each fund involved in an event. Using only observations in the event window (from 6 months before the announcement date to 12 months after the announcement date), I regress match-adjusted fund flows on POSTANN and also include firm fixed effects. Panel C of Table 4 presents estimated coefficients on the difference-in-difference estimator, POSTANN, in this regression. The coefficients range from % (~6% annualized) in the sample of all events to % (~8% annualized) in the sample of public non-asset management merger events, and all coefficients are statistically significant at the 1% level. We can calculate economic significance by looking at the total assets in funds undergoing events from Table 1. Across all funds, there were 185 events with $587 billion in assets so 6% outflows means about $190 million (($587b/185) 6%) of outflows per event. Across the sample of public non-asset management merger events, there were 70 events with $145 billion in assets so 8% outflows means about $165 million of outflows per event. It is also instructive to look at a graphical representation of monthly match-adjusted fund flows from 6 months prior to the announcement date to 12 months after the announcement date. Figure 1 provides this graphical representation (with each month s average flows and confidence intervals) for the entire sample of parent company changes, while Figure 2 depicts the same results for ownership changes of public non-asset manager parent companies. 6 Note that the flows in the two figures are not cumulative. The figures show that match-adjusted flows are near zero prior to the announcement date, and then decline slowly after the announcement date before accelerating downward in the final six months. Clearly, this is not the type of picture we are used 6 The graph for Column 2 of Table 4, which includes all public parent company mergers, looks very similar and is available upon request. 18

20 to seeing for event studies measuring market price reaction to events. However, it is not surprising that fund flows (predominantly due to retail flows, as I will show later) are much slower to react to new information than market prices, due to limited retail investor attention. It is also possible that fund investors wait under the deal is closed (usually 3 to 6 months after announcement) before reacting, or learn about the deal from intermittent fund disclosures. I compare flows prior to and after the effective date of the ownership change in the context of a regression in Table 12. Another method of analyzing the effect of ownership changes on fund flows is by using a multi-variable panel regression to control for an array of fund and parent company characteristics. I regress monthly Fund flows (%) on event window dummy variables (PREANN and POSTANN), fund-level controls, parent-level controls, style dummy variables, prior return deciles dummy variables, and time dummy variables. Table 5 presents estimated coefficients from this OLS regression for all funds (Column 1 and 2), funds whose advisers are owned by public parent companies (Column 3 and 4), and fund whose advisers are owned by non-asset management public parent companies (Column 5 and 6). The findings are broadly consistent with those in Table 4 and Figures 1 and 2. Fund flows (%) are close to zero prior to announcement dates, and negative and significant after the announcement dates. Table 5 also allows us to examine the effects of other characteristics on fund flows. There is a negative coefficient on fund size (Log fund AUM), since many large funds close to new assets because they are unable to trade without a large and costly price impact. The coefficient on family size (Log family AUM) is positive, perhaps because large fund families have more exposure and have bigger advertising budgets. The coefficient on Fund age is negative as newer funds attract more flows, including seed money from the fund family itself. Surprisingly, the 19

21 coefficient on expense ratios is positive, although it is not significant in Columns 4 and 6. Expense ratios include 12b-1 expenses (for advertising and promoting the fund) so these results do suggest that spending more on marketing works in attracting flows. There are two parent-level variables that also have explanatory value in predicting fund flows. Funds with privately held advisers have significantly higher flows: 0.27% per month, or approximately 3.2% per year. This might be due to the fact that privately-held firms are more likely to focus on asset management while public parent companies are mostly banks or insurance firms that also offer mutual funds. In addition, Ferris and Yan (2009) find that mutual funds with public parents underperform and suffer from more agency issues, while Adams, Mansi, and Nishikawa (2012) find more management changes at public parents. Both of these papers might explain the lower level of flows at funds managed by public companies. In addition, past stock returns (of public firms) also positively predict flows, as was previously documented by Sialm and Tham (2011). 4. Motivation for Decline in Fund Flows Tables 4 and 5 provide evidence that a mutual fund management company ownership change causes a decrease in fund flows. The next step is to examine why investors might behave in this way. One possible explanation is that fund investors trust the previous management firm (due to advertising or their experience with the firm) and this relationship is weakened or destroyed as a result of the ownership change, leading to an increase in the redemption of fund shares. Alternatively, investors might be reluctant to invest new money in the fund as a result of uncertainty arising from the change in management ownership. The ownership change could also 20

22 be temporarily accompanied by less emphasis on marketing, advertising, and distribution, leading to a decline in new fund purchases. I test these competing hypotheses by decomposing fund flows into inflows (purchases of shares) and outflows (sales of shares) and examining whether the aggregate flow effects are due to a decline in inflows or an increase in outflows. In Table 6, inflows (Columns 1, 3, and 5) and outflows (Columns 2, 4, and 6) are regressed on event window dummy variables (PREANN and POSTANN), fund-level controls, parent-level controls, style dummy variables, prior return deciles dummy variables, and time dummy variables. A comparison of columns makes it obvious that the aggregate results are entirely driven by increasing outflows. For instance, in Column 1, the coefficient changes from 0.181% prior to the announcement to 0.323% after the announcement, indicating an increase in inflows around the announcement date. However, for the corresponding outflows (Column 2), the coefficients rise even more, from 0.442% prior to the announcement to 1.332% after the announcement. The large increase in outflows overwhelms the small increase in inflows leading to the aggregate decline in fund flows seen in Table 5. The same results can be seen for the restricted samples in Columns 3 through 6. Overall, any changes in inflows are small while the outflows increase dramatically, providing support for the hypothesis that current fund investors are reacting to the ownership change by redeeming their shares. Still, while the increase in outflows is consistent with the trust hypothesis, it is not conclusive as to why investors are selling shares in the fund. One way to get at their motivation is to look at how different clienteles react to event announcements. The importance of trust and relationships should be more important for less sophisticated investors who don t have the skills 21

23 or resources to monitor the fund s management. Therefore, I test the trust hypothesis by looking at how investors of different levels of sophistication respond to the events in question. Morningstar provides an institutional dummy variable for each fund class, which equals one if only institutions are allowed to invest in that class, and zero otherwise. In order to use the heterogeneity in fund classes, I run regressions with observations at the fund class level (not aggregated to the portfolio level). For each fund class, I calculate flows using Equations 1 and 2, and also take the size, age, and expense ratio of the fund class instead of the portfolio-level weighted average used in previous tests. Finally, since some funds have multiple fund classes, I attach a weight to each observation equal to the assets of the fund class divided by the total assets of the entire fund, which ensures that each fund portfolio has the same weight in these regressions. I regress fund class flows on event window dummy variables, fund-level controls, parentlevel controls, style dummy variables, prior return deciles dummy variables, and time dummy variables. Table 7 shows estimated coefficients for different event samples using retail versus institutional fund classes. The dependent variables are flows of retail classes in Columns 1, 3, and 5, and flows of institutional classes in Columns 2, 4, and 6. It is easy to see from Table 7 that the aggregate results in Table 5 are being driven by retail class flows. The coefficients on POSTANN for institutional investors (Columns 2, 4, and 6) are small and statistically insignificant, while the coefficients on POSTANN for retail investors (Columns 1, 3, and 5) are two to four times larger and statistically significant at the 1% level. These results are somewhat surprising because retail investors are thought to be less attentive and more prone to inertia than institutional investors. In contrast, Table 7 shows that retail investors reaction is much stronger to this particular news, suggesting that they attach 22

24 more value to their relationship with their fund advisor than institutional investors. For instance, they might be clients of the parent company, obtaining banking, insurance, or other financial services from that company, and then deciding to move their fund portfolio elsewhere if the parent company is acquired by another financial services firm. Another way to understand the motivation for investor reactions to ownership changes is by looking at funds with different levels of expense ratios. Therefore, I next regress fund flows on event window dummy variables and controls, separately for funds with above-median expense ratios (high expense funds) and below-median expense ratios (low expense funds). If investors are willing to pay higher expense ratios because they have a special (trusting) relationship with the management company, we would expect to see more outflows at highexpense funds than low-expense funds after ownership changes. Table 8 presents estimated coefficients from OLS regressions for different event samples using high-expense vs. low-expense funds. Columns 1, 3, and 5 only include high-expense funds, while the remaining three columns only include low-expense funds. In Table 8, we can see significantly larger declines in flows from pre-announcement to post-announcement for highexpense funds compared to low-expense funds. For instance, in Column 5 (high-expense funds), we see a decrease of about 0.8% from the PREANN dummy to the POSTANN dummy variable, while the corresponding decline in Column 6 (low-expense funds) is only about 0.4%. The results in Table 8 confirm that investors willing to pay higher expenses (because they place some intangible value on investing in the fund) are also the ones more likely to pull out as a result of an announced change in ownership. 23

25 5. Cross-Sectional Variation in Fund Outflows in the Post-Announcement Window The results so far suggest that the average fund undergoing a change in ownership suffers a decline in flows, but there is significant cross-sectional variation in post-announcement flows. There are also significant differences in acquiring companies, funds managed by the acquiror, and purposes for acquisitions, leading to an intriguing question: Are investor reactions to ownership changes affected by the acquiror or the reason for the acquisition or are they simply reactions to the news that the prior management firm will be acquired? For instance, if the acquiror is (or owns) a mutual fund management company with a strong track record, one might expect more inflows to (or less outflows from) the target s funds. Alternatively, if the acquiror has little expertise in the mutual fund industry, investors might be even more likely to sell their assets in response to the ownership change announcement. I test these competing hypotheses in Table 10. Before discussing the regressions, Table 9 presents the relevant summary statistics for the explanatory variables used in Table 10. Panel A shows that there is no purpose available (in SDC Platinum) for over 56% of the events in the paper. Of the remaining events, the top three reasons for mergers are to strengthen the firm s operations, create synergies, and expand presence into new markets. These are standard explanations for M&A activity, so it does not seem like the financial industry acquisitions used in this paper are very different from most mergers. Some events have multiple purposes so the percentages do not add up to 100%. Panel B shows the breakdown for the type of acquiring firms. Almost half are banks, onethird are securities firms (broker-dealers, investment banks, asset managers, etc.), and the rest are insurance companies. More than 6 in 7 acquiring companies are publicly traded, but of these public firms, only 85% are in CRSP while the other 15% are traded over-the-counter or overseas. 24

26 Panel C shows some additional summary statistics for the acquiring firms. Their industryadjusted stock returns prior to the acquisition announcement are positive: 0.82% per month over the prior year, and 0.39% per month over the prior three years. This is consistent with successful firms being more likely to enter into acquisitions than struggling firms. On average, acquiring firms get only 5% of revenues from mutual fund expenses, consistent with most of them being banks and insurance companies. Finally, the style-adjusted returns of their mutual funds prior to acquisition are not statistically different from zero. In Table 10, I regress post-announcement match-adjusted fund flows for funds whose management undergoes a change in ownership on acquiror characteristics and dummy variables representing acquisition purposes. In Column 1 of Panel A, the predictive variables are indicators for whether the acquiring firm is private, an insurance company, or a securities company (banks are the omitted category). Interestingly, the coefficients on insurance and securities dummy variables are positive and statistically significant, indicating that investors react more negatively to acquisitions by banks than other financial firms. One possible explanation for this result is that investors see banks as having less expertise in money management. Alternatively, since banks are most likely to buy other banks, bank mutual fund investors might place a higher value on trust than investors in funds run by insurance companies or securities firms. In Columns 2 and 3 of Panel A, I also include past stock performance of the acquiring firm, and in Column 4, I include the proportion of firm revenues from mutual funds. None of the coefficients are significant so neither the past financial success of the acquiring firm, nor its concentration on asset management, seems to matter for investors in funds run by the target firm. Finally, in Column 5, I add dummies for the various possible purposes of the transaction. Only COR (concentrate on core businesses) and ISV (increase shareholder value) are significantly 25