Mutual Fund Peer Benchmarking and Corporate Governance

Transcription

1 Mutual Fund Peer Benchmarking and Corporate Governance Yijun Zhou * INSEAD This Version: Jan 14, 2018 Abstract The performance of mutual funds is benchmarked against their peer funds of the same style (i.e. peer benchmarking) and companies held by many peer funds become benchmark companies for mutual funds. This paper empirically examines how the heterogeneous degree of mutual fund peer benchmarking affects corporate governance in companies. First, benchmark companies are found to receive more attention from investors and have better corporate governance. Second, I show that mutual funds are more reluctant to sell their shares on the proxy record date and value their voting rights more in benchmark companies. Finally, I provide direct evidence of active monitoring by mutual funds in benchmark companies by examining their actual voting behavior. Incentives of an individual fund to monitor a company are affected by its peer funds of the same style, and I exploit exogenous fire sales and purchases as instruments to identify the causal effect. Overall, my findings suggest that mutual fund peer benchmarking contributes to governance in benchmark companies. Keywords: Corporate Governance, Mutual Funds, Peer Benchmarking, Proxy Voting, Trading JEL Classification: G23, G30, G34 * Finance Department, INSEAD, Boulevard de Constance, Fontainebleau, France yijun.zhou@insead.edu

2 1. Introduction In the mutual fund industry, investment objectives and stylistic classifications are widely used to characterize differences between funds and to facilitate the portfolio allocation decision of mutual fund investors. Every mutual fund has its style documented in its prospectus and is required by the SEC to invest in accordance with the investment style. More importantly, the performance of mutual funds is measured relative to the performance of their peer funds of the same style. This peer benchmark is determined by the performance of all mutual funds in the style, and thus their holdings in portfolio companies. Companies that are commonly held by peer funds in the style determine the peer benchmark to greater extents than companies held by few peer funds in the style. Therefore, companies vary in the degree of relevance to mutual fund peer benchmarks. 1 In this paper, I focus on active equity funds and define companies that are commonly held by same style peer funds as benchmark stocks or benchmark companies for mutual funds. I study how the heterogeneous degree of mutual fund peer benchmarking affects corporate governance in portfolio companies. Some recent theoretical papers (e.g. Cuoco and Kaniel, 2011; Basak and Pavlova, 2013; Breugem and Buss, 2017) study the impact of benchmarking on institutional investors portfolio allocation and information acquisition decisions and the consequent asset pricing implications. With benchmarking, institutional investors are found to optimally tilt their portfolios toward stocks 1 For example, both Addus HomeCare Corporation and Brightcove, Inc. are small companies of similar size and have similar number of mutual fund shareholders. However, Addus HomeCare is held by almost 40% of small cap funds while only 7% small cap funds are shareholders in Brightcove. Thus, Addus HomeCare determines the benchmark return for small cap funds to much larger extents than Brightcove. 1

3 that co-vary positively with the benchmark due to hedging needs and will acquire less private information, which will result in decreased price informativeness in the financial market. Yet, the questions relating to corporate governance have not been answered by the literature and are worthwhile to examine. Will corporate governance in different companies differ due to companies heterogeneous degrees of relevance to mutual fund benchmarks? If so, where does the difference come from? Will mutual funds behave differently in monitoring the benchmark companies compared with non-benchmark companies? On the one hand, the role of mutual funds in corporate governance is of great significance since mutual funds collectively own 24% of U.S. corporate equity (ICI, 2017) and are in a position to significantly affect the future of corporations (SEC, 2003). On the other hand, mutual funds are traditionally considered as shareholders that are reluctant to challenge management and will just exit the company by selling shares when they are dissatisfied. Moreover, mutual funds are found to be constrained by their limited attention and thus have to decide how to allocate their attention across portfolio companies (Fang, Peress, and Zheng, 2014; Kacperczyk, Nieuwerburgh, and Veldkamp, 2016). Due to the common holdings in benchmark companies by mutual funds, incentives of peer funds to pay attention to monitor or sell these companies may be different because of possible interactions among mutual funds belonging to the same style, which will affect corporate governance in companies. To empirically study how mutual fund peer benchmarking affects corporate governance in companies, I first examine the aggregate level of attention or monitoring received by companies. Chen, Cohen, Gurun, Lou and Malloy (2017) show that mutual fund managers only follow and download information on a very particular subset of companies. Companies that received more 2

4 attention from investors are more likely to be monitored and therefore have better corporate governance. Therefore, I measure the aggregate attention and monitoring in a company by the number of aggregate searches in the SEC EDGAR database which hosts all mandatory filings by public companies. If a company is searched more and its company filings are downloaded more in the EDGAR database, the company is considered to receive more attention and monitoring from investors and shareholders. I find that the quintile of companies with the highest level of mutual fund peer benchmarking receives 253 EDGAR searches per day on average, while the quintile of companies with the lowest level of mutual fund peer benchmarking receives only 29 EDGAR searches per day on average. The regression result also shows that companies of greater significance to mutual fund peer benchmarking receive a higher level of aggregate attention or monitoring after controlling for general institutional ownership, mutual fund ownership and other company characteristics. Knowing that benchmark companies receive more attention and monitoring, I then investigate the outcome of monitoring and the level of governance in companies with different levels of mutual fund peer benchmarking. Kempf, Manconi and Spalt (2016) find that firms with decreased monitoring from distracted shareholders are more likely to be involved in corporate actions related to the agency problem, such as diversifying and value-destroying acquisitions. I use acquisition decisions to reveal monitoring and look at the acquirer abnormal returns following Chen, Harford and Li (2007). In companies with more monitoring and better governance, the acquisition decision is less likely to be related to agency problems (e.g. CEO empire building), and thus the market is predicted to react more positively to the acquisition announcement. I find that benchmark companies make better M&A decisions compared with non-benchmark companies. When the extent to which a company relates to mutual fund peer benchmarking is larger, the 3

5 acquirer abnormal return is also higher. The results suggest that corporate governance in benchmark companies is better than that in non-benchmark companies. However, the greater level of attention or monitoring received and better corporate governance in benchmark companies may not be entirely due to mutual funds. To further understand the mechanism and the role of mutual funds as shareholders, I analyze the trading behavior of mutual funds on the proxy record date of shareholder meetings. Shareholders that hold company shares on the proxy record date are entitled with voting rights in the company. Aggarwal, Saffi and Sturgess (2015) examine the securities lending market and find that investors restrict lendable supply and/or recall loaned shares on the proxy record date to exercise voting rights. In addition, they document a spike of loan fees on the record date. If shareholders do not value their votes and are reluctant to actively monitor the company, they are likely to sell or lend their stocks on the record date. To examine whether mutual funds value their votes differently in benchmark companies compared with non-benchmark companies, I take advantage of a trading database of institutional investors and look at the daily abnormal trading behavior of mutual funds on each individual company. I find that mutual funds are more likely to sell their holdings in nonbenchmark companies, but are more likely to retain their shares in benchmark companies on the proxy record date which will result in keeping their voting rights in the company. The most important empirical evidence provided in this paper relates to the voting behavior of mutual funds, which helps to understand how mutual funds actually contribute to corporate governance in benchmark companies. Since July 2003, the SEC has required mutual funds to report their votes on all shares held in their portfolios, addressing that the enormous voting power gives them significant ability collectively to affect the outcome of shareholder votes and influence 4

6 the governance of corporations. In the survey of McCahery, Sautner and Starks (2016), 53% of the respondents as institutional investors report voting against management as a shareholder engagement channel. Outside of behind the scenes negotiations, voting is one of the most direct and observable monitoring actions that mutual funds can take to influence corporate decisions (Cai, Garner and Walkling, 2009; Gramlich, Miller and White, 2009). Thus, I examine the voting behavior of mutual funds in benchmark companies compared with non-benchmark companies. The company-level empirical results show that mutual funds are more likely to challenge contentious proposals sponsored by managers in benchmark companies compared with nonbenchmark companies, which suggests that active monitoring from mutual funds disciplines the behavior of managers and contributes to the better governance in benchmark companies. This is robust after controlling for a series of company characteristics and with year and company fixed effects. Finally, I scrutinize the voting behavior at the individual fund level to shed light on the incentives of mutual funds and the underlying mechanism of the peer benchmarking effect. Because there is no unique fund identifier common to the ISS voting database and CRSP mutual fund database that contains the information of fund styles and holdings, I hand-match the fund names of 2,567 US domestic equity funds from 2004 to I estimate a linear probability regression of fund votes and the fund-level empirical results show that when a fund has more holdings in a company, the fund is more likely to challenge contentious proposals in the company because its incentive to monitor the company increases (Shleifer and Vishny, 1986). More importantly, an individual fund is more likely to challenge contentious proposals in a company when more of its peer funds in the same style are also holding the company and when its peer funds have more holdings in the company, controlling for its own holdings in the company. The 5

7 incentive of one fund to monitor a company increases with the presence of its peer funds and the holdings of its peer fund. The strategic complementarity in the monitoring decision among the same style funds leads to the positive effect of mutual fund peer benchmarking on corporate governance. To address endogeneity concerns, I instrument the holdings of the peer funds (other mutual funds in the same style) in each company by exploiting exogenous fire sales and purchases by peer funds. Funds in the top decile experience an average inflow of 23.7% whereas funds in the bottom decile experience outflows of 12.5%. This large variation is driven by the overall portfolio performance of mutual funds and many market-wise factors, and thus is unlikely to be affected by an individual fund and a company in the portfolio. Funds experiencing large inflows and outflows tend to increase or decrease existing positions proportionally (Coval and Stafford, 2007). Edmans, Goldstein and Jiang (2012) and Khan, Kogan and Serafeim (2012) also use the mutual fund flowdriven trading pressure as exogenous price shocks affecting managerial actions (of seasoned equity offering and M&A, respectively). I find that the likelihood of active monitoring from an individual fund in a company significantly increases with the instrumented holdings of peer funds in the company. When its peer funds hold more shares in the company due to exogenous fire purchases, the fund is more likely to exert monitoring effort in the company. The instrumental variable probit regression results provide evidence of the causal effect from mutual fund peer benchmarking. This paper mainly contributes to four blocks of literature. The first block of literature studies the effect of shareholder composition on company value (e.g., Morck, Shleifer and Vishny, 1988; McConnell and Servaes, 1990; Himmelberg, Hubbard, and Palia, 1999; Holderness, Kroszner, and Sheehan, 1999; Franks and Mayer, 2001; Franks, Mayer, and Renneboog, 2001; Kandel, Massa 6

8 and Simonov, 2011) and the impact of institutional investors on corporate governance (e.g. Aggarwal, Erel, Ferreira, and Matos, 2011; Appel, Gormley and Keim, 2016; Chung and Zhang, 2011; Crane, Koch and Michenaud, 2016; Dasgupta and Piacentino, 2015; Duan and Jiao, 2016; Edmans and Manso, 2011; McCahery, Sautner and Starks, 2016), CEO pay sensitivity (Hartzell and Starks, 2003), shareholder proposals (Gillan and Starks, 2000), and corporate policies (e.g. Michaely, Popadak, and Vincent, 2015; Grinstein and Michaely, 2005). This paper studies how a prominent and significant feature in the mutual fund industry benchmarking affects the incentives of mutual funds as shareholders in their portfolio companies and the aggregate consequences on the corporate governance in companies. The second block of literature studies the limited attention of investors and looks at the asset pricing implications (e.g. Hong and Stein, 1999; Hirshleifer and Teoh, 2003; Kacperczyk, Nieuwerburgh, and Veldkamp 2016). Similar to Kempf, Manconi and Spalt (2016), this paper also studies the implications of limited shareholder attention on corporate actions, but with a focus on mutual fund peer benchmarking. The third block of literature examines strategic mutual fund behavior such as window dressing (O'Neal, 2001), risk shifting (Brown, Harlow, and Starks 1996; Chevalier and Ellison 1997; Koski and Pontiff 1999, Goetzmann, Ingersoll, and Welch 2007, Huang, Sialm and Zhang, 2011), cross-fund subsidization (Gaspar, Massa and Matos, 2006), voting with business ties (Davis and Kim, 2007), and name changing (Cooper, Gulen and Rau, 2005). This paper also looks at possible strategic behavior of mutual funds, but focuses on their monitoring behavior. The fourth block of literature is related to the implication of benchmarking in the financial market (e.g. Cuoco and Kaneil, 2011; Basak and Pavlova, 2013; Baffa, Vayanos and Woolley, 2014; Basak and Pavlova, 2016; and Buss and Breugem, 2017). In contrast to these papers that focus on asset pricing, this paper examines 7

9 the implication of benchmarking on corporate governance and provides empirical evidence of such implications. The remainder of this paper is structured as follows. In section 2, I discuss two competing views on how mutual fund peer benchmarking affects corporate governance based on different theories. In Section 3, I describe the empirical approach. In section 4, I describe the data used and provide descriptive statistics. The main results are provided and discussed in Section 5. A brief summary follows in Section Two Competing Views Some recent theoretical papers (e.g. Cuoco and Kaniel, 2011; Basak and Pavlova, 2013; Breugem and Buss, 2017) have examined asset pricing implications of the impact of benchmarking, however, the questions relating to corporate governance have not been answered. Even though incentives and behavior of mutual funds in benchmark companies may be affected by possible interactions among mutual funds belonging to the same style, there is no clear answer on whether mutual fund peer benchmarking contributes to or impairs governance, and the outcome of governance in companies. Mutual fund peer benchmarking may contribute to corporate governance for various reasons. Firstly, due to incentives to avoid falling behind the peer benchmark (Lakonishok, Shleifer and Vishny, 1992), and hedging needs (Breugem and Buss, 2017), the demand from mutual funds of the same style on these benchmark companies will be higher and they are less likely to sell these benchmark companies. Gaspar, Massa and Matos (2005) suggest that weak monitoring from short- 8

10 term investors will allow corporate managers to proceed with value-reducing acquisitions. On the contrary, mutual funds are more likely to be long-term investors in benchmark companies and are thus more likely to monitor mangers and deter value-decreasing behaviors. Therefore, companies of greater significance to mutual fund benchmarks will receive more monitoring from mutual funds, suggesting positive impacts of mutual fund peer benchmarking on governance. Secondly, when shareholders behave or monitor in a coordinated fashion, the discipline effect on corporate managers will be stronger while the monitoring cost for each shareholder will be lower. Given that the coordination cost among shareholders is potentially high, Kandel, Massa and Simonov (2011) find that age similarity among small shareholders can act as an implicit coordinating device for their actions. As same style mutual funds have higher similarity, they are more likely to think in a similar way and behave in a coordinated fashion. When one mutual fund s incentive to monitor a company increases, other mutual funds in the same style may also have increased incentives to monitor. This strategic complementarity in the monitoring decision among mutual funds will lead to a multiplier effect on corporate governance. Thus, the monitoring effect from mutual funds in benchmark companies will be stronger. Finally, there are positive payoff externalities that are specific to mutual funds belonging to the same style. Investors with categorical thinking consider risky assets as of different styles (Barberis and Shleifer, 2003). Due to style investing, assets in the same style comove too much, and assets in different styles comove too little. Similarly, investors move capital across different fund styles based on their past performance. Cooper, Gulen and Rau (2005) document that mutual fund investors allocate more capital to the current hot investment style, and some mutual funds take advantage of the irrational behavior by changing the name of the fund to become associated with 9

11 the current hot style and attract more flows. The better performance of these benchmark companies and other funds in the same style contributes to the overall performance of this style, which will then attract investor flows to all the funds in this style. This mechanism implies that there are positive payoff spillover effects among different mutual funds belonging to the same style, indicating stronger incentives for mutual funds to monitor and improve the governance of these benchmark companies. On the contrary, there are theories suggesting negative effects of mutual fund peer benchmarking on corporate governance. The free rider problem each individual shareholder bears the entire cost of his monitoring effort but only shares a fraction of the benefit thus each of them will have little incentives to monitor managers has been considered as the biggest obstacle of good corporate governance. Given a higher similarity between same style funds, each individual fund may expect its peer funds to voice out their common beliefs/objectives and do the monitoring so that it can just free ride. Therefore, the outcome may be that no peer fund will do the monitoring in benchmark companies because of their similarity and increased incentives to free ride. Moreover, the existence of competition between mutual funds in the same style may exacerbate the free rider problem. The literature has studied the impact of tournaments on some strategic behaviors of mutual funds, such as excessive risk taking when they are behind others (e.g. Brown, Harlow, and Starks 1996; Chevalier and Ellison, 1997). When allocating limited monitoring efforts or attention, mutual funds may have incentives to devote monitoring away from benchmark companies. This is because improving the governance of these benchmark companies will contribute to the performance of all peer funds that commonly hold these. Each peer fund is trying to outperform the others, and monitoring benchmark companies will not help an individual fund 10

12 outperform other peer funds under peer benchmarking. Lastly, there is a body of literature that examines how the threat of exit affects corporate governance (e.g. Admati and Pfleiderer, 2009; Edmans, 2009). When shareholders exit the company by selling their shares, the stock price in the company will go down and the CEO and managers in the company will be punished because of market-based performance evaluation and stock-based compensation. With the threat of exit, shareholders direct monitoring on the company will be more effective. However, if the exit threat is not credible, the effect of direct monitoring will also be weakened because it is difficult to actually punish company CEOs. As mentioned earlier, mutual funds are less likely to sell shares in benchmark companies due to hedging needs, the threat of exit in benchmark companies is less credible. Therefore, the effect of monitoring (voice) will also be weaker and these benchmark companies may suffer more from agency problems. To summarize, in the presence of benchmarking, many forces are acting simultaneously, which affects incentives of mutual funds to pay attention to monitor and sell each company in different directions. Therefore, it is of great significance to empirically examine the real outcome and understand the underlying mechanism. 11

13 3. Empirical Approach 3.1 Measuring the degree of mutual fund peer benchmarking relevance As mentioned earlier, mutual funds are benchmarked against their peer benchmarks and companies that are held in common by many peer funds in that style become benchmark stocks for that style of mutual funds. To measure a company s degree of relevance to mutual fund peer benchmarking, I first calculate the percentage of funds in each style that have holdings in the company. I divide the number of mutual funds in a style that have holdings in the company by the total number of mutual funds belonging to that style. As more mutual funds in a style hold the company, the more the company affects the benchmark of the style of funds. Since a company may be held by several different styles of funds, I use the average percentage of funds in each style that have holdings in the company (Equation 1) to measure the aggregate degree of benchmarking relevance in a company. Equation (1) illustrates the first benchmark relevance measure for company i held by mutual funds in style s at time t. I also consider the mutual fund style that a company is most relevant to and use the percentage of funds in that style with holdings in the company as the second measure for the company s relevance to mutual fund peer benchmarking (Equation 2). Since different styles of funds invest in different sets of companies with few overlaps, it is relatively rare for a company to be benchmark by many different styles of funds. Thus the number of styles that are holding the company is negatively related to the benchmarking relevance in the company. # 89 9:;<6.; 60=>? 6 B8CDA;=.!"#$h&'() &"'+,(" 1.,0 = mean 6 # 89 9:;<6.; 60=>? 6 # 89 9:;<6.; 60=>? 6 B8CDA;=.!"#$h&'() &"'+,(" 2.,0 = max 6 # 89 9:;<6.; 60=>? 6 (1) (2) 12

14 3.2 Aggregate monitoring and corporate governance in the company To examine how mutual fund peer benchmarking affects attention and monitoring in the company, I analyze the searching behavior of investors in the SEC EDGAR database following Chen, Cohen, Gurun, Lou and Malloy (2017). The SEC s EDGAR server hosts all mandatory filings by public companies and the SEC maintains a log file of all activity performed by users on EDGAR, which records each unique user request to acquire a specific filing. The more a company is searched in the EDGAR database, the more attention the company receives. So for each company i, I count the number of searches in EDGAR (excluding web crawlers) everyday and calculate the average daily searches for every year t. Then I look at the how companies heterogeneous degrees of relevance to mutual fund peer benchmarking affect the aggregate level of attention received in companies by conducting multivariate regressions as in Equation (3). G.,0HI are the company level controls such as company size, Tobin s Q, institutional ownership and mutual fund ownership. The dependent variable is the standardized average daily EDGAR searches in each company each year. If benchmark companies receive more attention, J I will be positive and vice versa. 'K"('L" M'NOP +"'($h"+.,0 = J I &,R,'O S,#M!"#$h&'()N#L.,0HI + U G.,0HI + V.,0 (3) Following Chen, Harford, and Li (2007), I use acquisition decisions to reveal governance and look at acquirer announcement returns. When the announcement returns are high, the acquisition decision is less likely to have been initiated as a result of agency problems and poor governance, so the market reacts positively to the deal; When the announcement returns are low, the acquisition decision is more likely to have been initiated as a result of agency problems, so the market reacts negatively to the deal. I look at both GWX( 1, +1), the abnormal announcement-period return over days 1, +1, and GWX 0, +2, the abnormal announcement-period return over days 13

15 (0, +2), where day 0 is the date of initial acquisition announcement by the sample firm. Daily abnormal stock returns are computed using the market model and the value-weighted CRSP index. The estimation window is days ( 200, 60) prior to the acquisition announcement date. Combining 3.1 and 3.2, I estimate the impact of mutual fund peer benchmarking relevance in the company on the acquisition performance by conducting multivariate regressions as in Equation (4). G A,0HI are the controls of the acquirer a. '$^,N("( '##_,$"&"#R ("R,(#+ A,0 = J I &,R,'O S,#M!"#$h&'()N#L A,0HI + U G A,0HI + V A,0 (4) If mutual funds exert more active monitoring in the benchmark companies compared with nonbenchmark companies, governance will be improved and the agency problem should be less severe in benchmark companies, and thus these companies should make better M&A decisions, which will be perceived more positively by the market and J I will be positive. On the contrary, if mutual funds allocate monitoring efforts and resources away from benchmark companies, the poor governance in these companies will lead to M&A decisions initiated due to agency problems, which will be perceived more negatively by the market and J I will be negative. 3.3 Abnormal sell by mutual funds on the record date in the company A main challenge of studying the role of mutual funds in corporate governance is to directly observe actions of mutual funds in companies. The behavior of mutual funds is unobservable with the exception of buying and selling shares. Therefore, I look at the trading behavior of mutual funds in each company on the proxy record date and examine how much they value their voting rights in the company. Shareholders that own a company s shares on the proxy record date are 14

16 entitled to vote at the subsequent shareholder meeting. Aggarwal, Saffi and Sturgess (2015) examine the securities lending market and find that investors restrict lendable supply and/or recall loaned shares on the proxy record date to exercise voting rights. In addition, they document a spike of loan fees on the record date. If shareholders do not value their votes and are reluctant to actively monitor the company, they are more likely to sell or lend their stocks on the record date. To examine whether mutual funds value their votes differently in benchmark companies compared with non-benchmark companies, I make use of a trading database of institutional investors that has been used by the asset pricing literature that accounts for 8% of CRSP trading volume and 10% of institutional trading volume (Puckett and Yan, 2011). Since mutual funds take mostly long positions in companies, I focus on their selling behavior on the record date to capture their willingness to actively monitor the company. I aggregate the daily trading of mutual funds in each company and calculate the abnormal sell by mutual funds in each company on the record date. Following Hu, McLean, Pontiff and Wang (2013), the abnormal sell is the actual daily sell volume in the company minus the baseline average daily sell volume in the company, all divided by the baseline average daily sell volume in the company (Equation 5). The baseline daily sell volume is calculated in an estimation window of 60 trading days and a gap window of 10 days before the record date t. Then a multivariate regression is estimated as in Equation (6). If mutual funds do not value their voting rights in portfolio companies, they are more likely to sell their shares in the company on the record date when the market demand for the stock is higher. However, if mutual funds do value their voting rights in benchmark companies and want to retain monitoring power in these 15

17 companies (compared with non-benchmark companies), they will be reluctant to sell their shares in spite of the higher market price, so J I will be negative and vice versa. '!#_(&'O +"OO.,0 = <A.>= 6?>>`,aH Ab?cAd? <A.>= 6?>>`,aefg ah aeig Ab?cAd? <A.>= 6?>>`,aefg ah aeig (5) '!#_(&'O +"OO.,0 = J I &,R,'O S,#M!"#$h&'()N#L.,0HI + U G.,0HI + V.,0 (6) 3.4 Direct monitoring action by mutual funds in the company Since July 2003, the SEC has required mutual funds to report their votes on all shares held in their portfolios. In the survey of McCahery, Sautner and Starks (2016), 53% of institutional investor respondents report voting against management as a shareholder engagement channel. Moreover, Cai, Garner, and Walkling (2009) find that poor election outcomes lead to reduced CEO compensation and increased CEO turnover, along with removal of anti-takeover mechanisms. Similarly, Fischer, Gramlich, Miller, and White (2009) show that lower voting outcomes lead to subsequent CEO and director turnover, along with changes in investment and acquisition policies of the firms. Since there are many regular proposals to be voted in shareholder meetings that will have less impact on the company, I focus on contentious proposals sponsored by management when the Institutional Shareholder Services (ISS), a leading proxy advisory service company recommends voting against (Illiev and Lowry, 2015). It has been shown that ISS s voting suggestions contain information about the merits of proxy proposals (Alexander, Chen, Seppi and Spatt; 2010), and the literature has used the recommendation of ISS as a standard proxy for the quality of voting proposals (e.g. Cai, Garner and Walkling, 2009; Crane, Koch and Michenaud, 2016). If mutual funds are active in corporate governance, they are expected to vote against 16

18 proposals of poor quality (Duan and Jiao, 2016). Examining these proposals can also control for the impact of ISS. Moreover, mutual funds may care about the current or potential future business ties with companies and are reluctant to vote against management (Davis and Kim, 2007). Therefore, mutual funds are more likely to be actively monitoring the company when the management receives a lower support rate from mutual funds in contentious proposals. Empirically, I use the support rate from mutual funds in contentious proposals in the company to measure the monitoring effort of mutual funds in each firm i at time t. Then I conduct the company level empirical tests by estimating multivariate regressions as in Equation (7), where the dependent variable is the percentage of support votes and the benchmarking relevance measure in each company is the independent variable. If mutual funds exert more active monitoring in the benchmark companies compared with non-benchmark companies, the management will receive a lower support rate, and thus J I will be negative and vice versa. &LR +,jj_(r ('R".,0 = J I &,R,'O S,#M!"#$h&'()N#L.,0HI + U G.,0HI + V.,0 (7) To further understand the incentives of mutual funds, I conduct fund-level empirical tests by estimating a linear probability regression as the following equation, where on the left hand side I have an indicator whether a fund j votes against management in the contentious proposal in company i. I regress this indicator on the peer fund holdings in the company, its own holdings in the company (included in the fund level controls S k,0hi ), and company level controls. There are two measures for peer fund holdings. The first measure is the percentage of company holdings owned by peer funds. The second measure is the percentage of same style funds that hold the company, which is the crucial component of benchmarking as in Equations (1) and (2). If the incentive of an individual fund is affected by its peer funds, the coefficient J I will be significant. 17

19 If the coefficient J I is positive, it means that there is strategic complementarity in the monitoring decision between same style funds, which will result in the positive effect of mutual fund peer benchmarking on corporate governance. 1 NS K_R" 'L'N#+R &L&R.,k,0 = J I j""( S,#M h_omn#l+.,k,0hi + l S k,0hi + U G.,0HI + V.,k,0 (8) 3.5 Exogenous fire sales and purchases To address endogeneity concerns of mutual fund ownership, I exploit exogenous mutual fund flows and fire sales and purchase. Funds in the top decile experience an average inflow of 23.7% whereas funds in the bottom decile experience outflows of 12.5%. This large variation is driven by the overall portfolio performance of mutual funds and many market-wise factors, and thus is unlikely to be affected by a company in the portfolio and an individual fund. Funds experiencing large inflows and outflows tend to increase or decrease existing positions proportionally (Coval and Stafford, 2007). Edmans, Goldstein and Jiang (2012) and Khan, Kogan and Serafeim (2012) also use the mutual fund flow-driven trading pressure as exogenous price shocks affecting managerial actions (of seasoned equity offering and M&A, respectively). I estimate a two-step instrumental variable probit regression. In the first stage regression, I instrument the holdings of the peer funds (other mutual funds in the same style) in each company by exploiting exogenous fire sales and purchases by peer funds in the same style. Following Edmans, Goldstein and Jiang (2012), I consider only mutual funds that experience inflows or outflows of at least 5% of total assets and define the flow induced traders as fire purchases or sales. Then I estimate the two step probit regression with the indicator variable whether a fund j votes 18

20 against management in the contentious proposal in company i as the independent variable as in Equation (9). 1 NS K_R" 'L'N#+R &L&R.,k,0 = J I N#+R(,&"#R"M j""( S,#M h_omn#l+.,k,0hi + l S k,0hi + U G.,0HI + V.,k,0 (9) 4. Data 4.1 Data source The first data source in this paper is the CRSP mutual fund database with information of fund holdings in each company, the net asset value and return of each fund, and other characteristics. The information on fund style also comes from the CRSP mutual fund database, which combines the mutual fund objective codes from three different sources and provides a strategic insights objective code for each mutual fund. I use this strategic insights objective code to define the style of mutual funds. I focus on domestic equity funds since I limit my sample to US domestic companies/stocks. For domestic equity funds, there are 22 different styles, including 13 sector based styles (e.g. Utilities funds, Real Estate funds etc.), 4 cap based styles (e.g. Large cap funds, Mid cap funds etc.), and 5 investment objective styles (e.g. Growth funds, Income funds etc.). To obtain the most accurate measure of mutual fund ownership and benchmarking in each company, I gather latest holdings for each portfolio before the M&As and shareholder meetings events and aggregate portfolio holdings of mutual funds in each company. Accounting information in each company (e.g. ROA, CAPEX etc.) is from the COMPUSTAT database. Regarding the searching data, I collect all the log files from SEC s website which records each 19

21 unique user request to acquire a specific filing in the company. For each company, I count the number of searches in EDGAR that do not come from web crawlers everyday and calculate the average daily searches every year from 2004 to Then I match the EDGAR searches on companies with accounting information from COMPUSTAT and mutual fund holding data from CRSP, and obtain 32,537 observations. For the M&A data, I collect all announced US mergers with announcement dates between January 1, 2002 and December 31, 2015 as identified by the Mergers and Acquisitions database of Securities Data Company (SDC). Following Chen, Harford, and Li (2007), I exclude divestitures, repurchases and deals that are only rumored. I require acquirers to have available price and return data from the CRSP and accounting information from COMPUSTAT, and then I match the acquirer companies with the CRSP mutual fund holding data. Finally, I obtain 13,915 observations with acquirer announcement returns, mutual fund holding information and acquirer company data. The trading database of institutional investors comes from Abel Noser Solutions, a consulting firm that helps institutional investors monitor execution costs and thus directly gets transaction data from institutional investors trading systems. This data accounts for 8% of CRSP trading volume and 10% of institutional trading volume, and has been used widely by the asset pricing literature (e.g. Goldstein, Irvine, Kandel and Weiner, 2009; Puckett and Yan, 2011 etc.). Detailed daily transaction data is reported from a sample of institutional investors (including mutual funds) in each stock from 1999 to I match the trading of mutual funds to the record dates of 15,633 shareholder meetings of 3,282 unique companies from 2003 to Since July 2003, the SEC has required mutual funds to report their votes on all shares held in 20

22 their portfolios. Institutional Shareholder Services (ISS) has compiled the votes by the top 250 mutual fund families into its Voting Analytics database. This voting database contains the fund name, portfolio company name, ticker and cusip, meeting date, agenda times, management voting recommendation, ISS voting recommendation, and all votes cast by the fund from 2003 to The voting data is combined with portfolio holding information of mutual funds from CRSP, and my sample contains 5,446 unique companies and 33,220 contentious shareholder meetings with mutual fund voting and holding information in each meeting. To make use of the fund level detailed information, I match each individual fund in the ISS voting database that contains their voting records with the CRSP mutual fund database that contains the information of fund styles and holdings. I first match the portfolio company in the voting data to CRSP mutual fund data by ticker and cusip. Second, I collect the fund names that have voting records in each company each year from the voting data and the fund names that have holdings in each company each year from the CRSP mutual fund holding data. Third, for each fund in the voting data, I calculate the Jaccard similarity score between its fund name and all the fund names from the CRSP mutual fund database that have holdings in the same company in the same year that the fund has voting records. The idea is that a fund must have holdings in the company to have to voting right in one company. Fourth, for each fund in the voting database, I keep three funds names from the CRSP mutual fund database with the highest Jaccard similarity scores and hand-match the fund names. My final sample contains 2,567 US domestic equity with fund style and holding data in each company from 2004 to

23 4.2 Descriptive statistics Table 1 provides the descriptive statistics on the benchmarking relevance measures in the company as in Section 3.1. The average benchmarking measure ranges from 17.82% to 27.75% in the sample period, measuring the maximum percentage of mutual funds in a style that are holding the company. The standard deviation of benchmarking measure ranges from 17.05% to 30% in the 13 years. Some companies are more relevant to mutual fund peer benchmarking while some are less relevant. Table 2 describes the number of M&A events and the acquirer announcement returns as in Section 3.2. Table 3 reports the average percentage of mutual funds that vote against management in shareholder meetings, from which we can also observe non-trivial cross sectional variations among companies. 5. Main Results 5.1 EDGAR searches and mutual fund peer benchmarking relevance Based on searching behavior of investors in the SEC EDGAR database which hosts all mandatory filings by public companies, I examine the aggregate level of attention or monitoring received by companies. In Table 4, I first group companies into quintiles each year based on the mutual fund peer benchmarking relevance measures from low to high and examine the average daily number of EDGAR searches on companies in each quintile. In panel A, the quintile is based on benchmarking measure (1) while in panel B the quintile is based on benchmarking measure (2). For companies in each quintile, the table presents the 25th percentile, median, and 75th percentile of the average daily number of EDGAR searches. As shown in Panel A, the median number of 22

24 daily EDGAR searches in the fifth quintile (companies with the highest level of mutual fund peer benchmarking) is 253, while the median number of daily EDGAR searches in the first quintile (companies with the lowest level of mutual fund peer benchmarking) is only 29. There is a monotonic increase in the average daily number of EDGAR searches on companies from quintiles one to five. The regression result in Table 5 also supports that companies that are most relevant for mutual fund peer benchmarking receive a higher level of aggregate attention or monitoring. The dependent variable is the standardized average daily number of EDGAR searches on the company, and the main variables of interest are the two benchmarking relevance measures. Both benchmarking measures are positive and significant even after controlling for general institutional ownership, mutual fund ownership and other company characteristics. 5.2 Acquirer announcement returns and mutual fund peer benchmarking relevance To examine the impact of mutual fund peer benchmarking on the company, I use acquisition decisions to reveal governance of the company and look at the acquirer abnormal returns. The results show that companies that are more relevant to mutual fund peer benchmarking exhibit higher acquirer abnormal returns (Tables 6 7). In Table 6, I look at the impact of the average degree of mutual fund peer benchmarking relevance (benchmarking measure 1) and the number of styles of mutual funds that have holdings in the company on the acquirer announcement returns. The first variable is positive and the second variable is negative, both significant at the 1% level for GWX 1, +1 and GWX 0, +2. When the mutual fund peer benchmarking relevance in the 23

25 company is one standard deviation higher, the acquirer announcement return will increase by 0.24%. When the average percent of same style funds holding the company is higher in the company, the company affects mutual fund benchmarks more strongly, and mutual funds will be more likely to monitor the company. With more monitoring from mutual funds as shareholders, the acquisition decision is less likely to be related to poor governance and agency problems (e.g. CEO empire building). Thus the market is predicted to react more positively to the acquisition announcement. In Table 7, I use the second benchmarking relevance measure as in Section 3.1, which is also found to be positive and significant. Since the second measure only considers benchmarking from the style of funds that the company is most relevant, the result suggests that there is a significant monitoring effect from a single style of funds. The result is robust after controlling for the overall level of mutual fund ownership, institutional ownership, other company and deal characteristics, and year and industry fixed effects. Moreover, the overall level of mutual fund ownership is insignificant, suggesting that my measures of mutual fund peer benchmarking relevance can better capture the effect of mutual fund ownership on governance in companies. 5.3 Abnormal selling and mutual fund peer benchmarking Shareholders that own a company s shares on the proxy record date are entitled to vote at the subsequent shareholder meetings. Institutional investors that value their voting rights are found to restrict lendable supply and/or recall loaned shares on the proxy record date even though there is a spike of loan fees on that day (Aggarwal, Saffi and Sturgess, 2015). To examine whether mutual 24

26 funds that mostly have long positions value their votes differently in benchmark companies compared with non-benchmark companies, I use their abnormal selling behavior on the record date to capture their willingness to actively monitor the company instead of exiting. The results in Tables 8 and 9 show that mutual funds will sell much less on the proxy record date in benchmark companies compared with non-benchmark companies. The large magnitude of coefficients suggests a strong effect of benchmarking relevance on the selling behavior of mutual funds, indicating that mutual funds are more reluctant to sell their shares on the proxy record date and value their voting rights more in benchmark companies. 5.4 Voting and mutual fund peer benchmarking I use proxy voting as the direct monitoring action of mutual funds and examine votes of mutual funds in contentious management proposals in each company. It has been shown that ISS s voting suggestions contain information about the merits of proxy proposals (Alexander, Chen, Seppi and Spatt; 2010), and the literature has used the recommendation of ISS as a standard proxy for the quality of voting proposals (e.g. Cai, Garner and Walkling, 2009; Crane, Koch and Michenaud, 2016). If mutual funds are active in corporate governance, they are expected to vote against management on proposals for which ISS recommends doing so, and thus management will receive a lower support rate (Duan and Jiao, 2016). Examining these proposals can also control for the impact of ISS, a leading proxy advisory service company. In the multivariate regressions in Tables 10 and 11, the dependent variable is the support rates from mutual funds in contentious management proposals. The main variables of interest are the two benchmarking relevance measures in the two tables respectively. Both benchmarking measures are negative and significant. A one standard deviation increase in benchmarking 25

27 relevance measure (1) leads to a 2.2% lower support rate for management, which indicates a negative relationship between mutual fund peer benchmarking relevance and management support rates in proposals of poor quality. Even though mutual funds may care about the current or potential future business ties with companies and are therefore reluctant to vote against management (Davis and Kim, 2007), they are more likely oppose management in benchmark companies. In addition, when more styles of funds are holding a company, the company is less likely to become benchmark stocks for all styles, and thus receives less monitoring from mutual funds, which is consistent with the positively significant coefficient of the second independent variable in both tables. Controls of various company characteristics such as size, ROA and Tobin s Q are included, and the effect of mutual fund peer benchmarking on companies is still significant when company fixed effects are added. Moreover, this finding is consistent with previous results and provides consistent evidence that mutual funds value their voting rights and engage in active monitoring in benchmark companies, leading to better governance in these benchmark companies. In the probit regressions in Table 12, the dependent variable is an indicator whether a fund j votes against management in the contentious proposal in company i. The main variables of interests are the percentage of company holdings owned by peer funds, the percentage of same style funds that hold the company, and the holding of fund j in the company. All three variables are found to be positive significant. When a fund has more holdings in a company, the fund is more likely to challenge contentious proposals in the company because its incentive to monitor the company increases (Shleifer and Vishny, 1986). More importantly, an individual fund is more likely to challenge contentious proposals in a company when more of its peer funds in the same style are also holding the company and when its peer funds have more holdings in the company, controlling for its own holdings in the company. The incentive of one fund to monitor a company 26