Passive Investors are Passive Monitors. Davidson Heath David Eccles School of Business University of Utah

Transcription

1 Passive Investors are Passive Monitors Davidson Heath David Eccles School of Business University of Utah Daniele Macciocchi David Eccles School of Business University of Utah Roni Michaely Tech Cornell University Matthew C. Ringgenberg David Eccles School of Business University of Utah First Draft: March 15, 2018 This Draft: March 15, PRELIMINARY DRAFT - Please do not cite or distribute Comments welcome. We thank Russell for providing index data. We also thank Wei Wei and Alex Young for discussions about Russell research designs. All errors are our own. c 2018.

2 Passive Investors are Passive Monitors First Draft: March 15, 2018 This Draft: March 15, PRELIMINARY - Please do not cite or distribute ABSTRACT We develop a new research design based on post-2007 Russell index reconstitutions to examine monitoring by active versus passive mutual funds. We find that passive investors are more likely to vote with firm management, especially on contentious shareholder proposals. Higher passive ownership also leads to lower analyst coverage of the firm, consistent with passive funds monitoring less. As a consequence of less monitoring, we also find that passive ownership causes changes in firm payout policy: higher passive ownership leads firms to reduce share repurchases and pay out less cash to shareholders. Overall, we find that passive ownership has important negative consequences for monitoring, governance and payout. Keywords: index investing, passive investing, shareholder voting, monitoring, governance, payout, repurchases JEL Classification Numbers: G12, G14

3 By design, passive funds invest in all securities included in the index they track. Unlike active investors, they cannot express their disagreement with the decisions of individual issuers by selling their holdings. A higher share of passive investors could therefore weaken market discipline and alter the incentives of corporate and sovereign issuers to act in the interest of investors. -Bank for International Settlements Quarterly Review (March 2018) I. Introduction In recent years U.S. public corporations have experienced a dramatic increase in ownership by passive index funds (see Figure 1). 1 The rapid growth of passive investing raises questions about the consequences for monitoring and governance, since it is unclear if passive investors have the incentives and resources to monitor firms in their portfolio (Shleifer (1986)). Do passive investors monitor their portfolio firms more or less closely than active investors? Do information intermediaries produce the same quantity of information when a firm has higher passive ownership? And how do changes in scrutiny due to passive ownership impact managerial decision making? We document that passive investors are significantly more likely to vote in agreement with firm management. To our knowledge ours is the first paper to show this correlation. However, the main challenge in estimating the effect of passive ownership is the endogeneity of firms ownership structure. Firm characteristics affect both the firm s ownership structure and governance policies. Moreover, a firm s policies may attract certain types of investors. 2 To examine the causal effects of passive ownership, we develop a new research design using Russell index reconstitutions in the post- banding era from 2007 onward. After a stock switches into the Russell 2000 index, we find that it experiences a mechanical increase in passive ownership and a decrease in active ownership, and vice versa for stocks that 1 Data from Financial Times show that while outflows from actively-managed mutual funds have hit a cumulative $1.2tn since 2007, overall inflows into index-trackers and ETFs have topped $1.4tn over that period. 2 Grinstein and Michaely (2005) find that higher firm payouts attract institutional holdings, while Brav, Jiang, Partnoy, and Thomas (2008) and Aghion, Van Reenen, and Zingales (2013) find that active investors target firms with weak governance. 1

4 switch out. We find that those passive owners are more likely than the active owners to vote with management, especially on contentious shareholder proposals. Thus, both in the broad cross-section, and after exogenous shocks to their holdings, passive investors are more passive monitors. In response to increased passive ownership, information intermediaries decrease information production: analyst coverage of the firm declines significantly. We find that the reduction in monitoring has real effects, as higher passive ownership leads firms to reduce share repurchases and pay out less cash to shareholders. Overall, our results show that passive ownership has important negative consequences for monitoring, governance, and payout. Previous empirical studies provide evidence that active institutional investors influence firm policies (e.g., Brav et al. (2008); Aghion et al. (2013)). These studies document that active investors demand policy changes and may exit their positions if a firm underperforms. However, more recent studies suggest that passive investors actually strengthen firm governance and increase payouts to shareholders (see, e.g., Appel, Gormley, and Keim (2016); Crane, Michenaud, and Weston (2016)). Their argument is that passive investors could be effective at widespread monitoring of firms compliance with the best governance practices. Further, managers might be more inclined to listen to passive investors over active investors, as the latter are more transient investors that exhibit higher turnover rates (Del Guercio and Hawkins, 1999). However, it is unclear how passive institutional investors actually effect change. Theory suggests that investors can affect governance by voting their shares (the voice mechanism) or by selling their shares (the exit mechanism). It follows that passive investors are constrained in their ability to influence managerial decisions, because they are unable to exit a position unless there is a change in index membership. Furthermore, while passive investors may have a fiduciary duty to monitor, they may have weak incentives to monitor each individual firm in their portfolio. Finally, given that passive index funds are much more diversified on average, they could have insufficient resources to monitor each stock in their portfolio (Fich, 2

5 Harford, and Tran (2015)). In June of each year, Russell Investments reconstitutes their popular Russell 1000 (largecap) and Russell 2000 (small-cap) indexes. Starting in June 2007 Russell implemented a new assignment regime ( banding ) which broke the well-known discontinuity in index membership around the rank-1000 index threshold. This paper proceeds from the insight that banding replaced the yearly discontinuity in index membership with two yearly discontinuities in index switching. We use these yearly discontinuities from 2007 to 2014 to construct a stacked-cohort difference-in-differences estimator, which compares firms just above and just below each band in each reconstitution year, for three years pre- versus post-treatment. We find that ownership by passive mutual funds increases by a net 1.07% of market cap for stocks that switch into the Russell 2000 and falls by a net 1.33% of market cap for stocks that switch into the Russell In both cases, an opposite change in ownership by active funds mirrors the change in ownership by passive funds, suggesting that active mutual funds sell to passive funds that enter and buy from passive funds that exit a given stock. Consistent with our findings in the cross-section that passive funds vote more passively, for contentious agenda items, the fraction of shares that vote in the direction recommended by firm management rises for stocks with increased passive ownership. Intriguingly, in both cases the change in shareholder passivity is larger than the direct effect of passive funds voting. For example, for stocks that switch to the Russell 2000 and thus have higher passive ownership, 2.4% more of voting shares vote with management, only 1.08% of which is due directly to passive fund votes. This finding is consistent with passive funds lower demand for information imposing a negative externality, by raising the cost of monitoring for the active funds that remain. Consistent with our predictions about how information providers should react to exogenous shocks to passive ownership, we find that firms that switch into the Russell 2000 (i.e., firms with more passive ownership) experience a significant reduction of 1.9 analysts covering them on average (a 20% drop relative to the pre-period average coverage). Conversely, firms 3

6 switching from the Russell 2000 to the Russell 1000 experience an increase of 1.7 analysts on average, 18% of their pre-period average. More passive ownership leading to more passive voting and less analyst coverage are all consistent with less monitoring and external scrutiny. Accordingly, we further investigate whether changes in passive ownership have real effects on firm policy. We examine firm payout as a classic policy choice that is subject to agency problems between managers and owners. La Porta, Lopez-De-Silanes, Shleifer and Vishny (2000) propose that payout policy can be either a substitute or a complement to other mechanisms of external monitoring. Similar to the findings in La Porta et al. (2000), we find evidence that payouts and external monitoring are complements. In particular, we find that firms that switch from the Russell 1000 to the Russell 2000 (i.e., firms with more passive ownership) decrease their total payout by 1.6% of lagged total assets. On the other hand, firms that switch to the Russell 1000 increase their payout by 1.6% of lagged total assets. These changes in firm payout are entirely driven by stock repurchases rather than dividends. These findings differ from those of Crane et al. (2016) in the period up to 2007, but are consistent with the payout literature (see Farre-Mensa et al for a review) documenting that stock repurchases have become more popular in recent years and represent a weaker commitment than dividends. Our paper contributes to the literature in several ways. First, we reexamine the effect of passive investors on firm governance with a specific focus on institutional investor voting behavior, the information environment of firms, and payout policy. Understanding the monitoring role of passive investors is of primary importance since passive investing via index funds and ETFs has become increasingly popular and economically important in recent years. Consistent with economic theory (e.g., Edmans (2009, 2014); Maug (1998); Shleifer (1986)), our findings suggest that passive owners exercise weaker monitoring, which weakens a firm s information environment, leads to more passive shareholder voting, and reduces the use of payouts as a disciplining mechanism. We contribute to the ongoing debate as to whether payout policy may be a substitute or a complement to other mechanisms of exter- 4

7 nal monitoring; our findings are consistent with complementarity between governance and payouts, as posited by La Porta et al. (2000). Second, we analyze the effects of index investing in the post-banding period from 2007 onward. We focus on the post-2007 period for several reasons. First, the banding regime has received less attention in the Russell literature and so our results are novel and largely out-of-sample to previous results (up to 2006, see Appel et al. (2016); Crane et al. (2016); Boone and White (2015), among others). Second, the amount of passive investment linked to the Russell indexes has increased rapidly in recent years, which strengthens both the treatment effects and the relevance and external validity of our results. We develop a cohort difference-in-differences estimator, which has two main advantages to analyze changes in corporate policy relative to previous approaches that compare outcomes between treated and control firms in the year post-index-assignment. First, changes in corporate policy in response to index assignment can plausibly take longer than one year to implement. Our approach lets us examine firm policy for multiple years before and after index assignment. Second, it lets us use individual firm fixed effects in all of our estimates which remove any time-invariant differences between firms. As a result, our new methodology mitigates any selection bias, which is a potential concern in previous studies (see Wei and Young (2017) for a discussion of these issues). We confirm this with extensive balance and robustness checks. Overall, our paper is the first to show that passive investing directly leads to changes in voting behavior at the firm level, which in turn impacts the information environment and managerial decision making. A. Related Literature The theoretical literature on institutional ownership and corporate governance is premised on two main mechanisms: voice and exit Edmans (2014). Studies that analyze the voice mechanism posit that ownership concentration provides a blockholder with an incentive to overcome the free-rider problem and thus monitor its claim (Shleifer (1986). While active 5

8 funds may hold a large fraction of a single firm s ownership and therefore may have incentives and the ability to monitor (Fich et al. (2015)), evidence shows that passive index funds have a well-diversified portfolio and usually hold a small fraction of equity in a single firm. Some empirical studies also caution about the monitoring ability of passively managed funds. Roe (1990), for instance, argues that many mutual funds and pension funds have several regulatory constraints that could prevent them from buying large stakes and thus monitoring. All in all, the voice mechanism seems not to be a suitable governance mechanism for passive index funds that are the subject of our study. A second effective governance mechanism that large shareholders can use is the threat of exit, i.e. the Wall Street walk (Edmans (2009); Dasgupta and Piacentino (2015); Levit (2017)). If agency problems between managers and shareholders are severe, the threat of exit can force managers to adopt governance policies that are more welcomed by large shareholders. However, the objective of passively managed index funds is to track an index with the lowest error possible. Hence, the exit threat seems to be a less credible governance mechanism for index funds, given that they are required to invest in each firm listed on a specific index. In sum, theoretical work suggests that monitoring is more costly and less credible for passive index investors than for active investors. Empirical evidence on the relationship between institutional ownership and corporate governance also indicates that active investors, as opposed to passive investors, have the incentives and the resources to effectively monitor their firms. Aghion et al. (2013) show that corporations with more active shareholders invest more in innovation, whereas the opposite is true for firms with more passive institutional investors. Brav et al. (2008) find that activist hedge funds have a positive impact on corporate governance policies, while Cronqvist and Fahlenbrach (2008) find that active investors monitoring becomes more effective if the investor s holdings are less diversified, if the investor holds a large stake off a firm s share, and if the investor has board representation. Consistent with the studies discussed above, we predict that an increase of ownership by 6

9 passive index funds leads to negative governance consequences. We test this prediction by looking at (i) institutional investors voting behavior (i.e., active vs. passive voting), (ii) at changes in the information environment and (iii) at changes in payout policies. We first study the voting behavior of institutional investors (both active and passive investors), with a specific focus on investors tendency to vote with or against management s recommendations. We predict that passive index investors are less likely to vote against management s recommendations because that would imply more research and, therefore, more resources employed on monitoring each specific firm in their portfolio. Morgan, Poulsen, Wolf, and Yang (2011) find that a low expense ratio and low turnover are associated with a greater tendency to vote with management s recommendations in a cross-section of shareholder proposals from However, they do not examine voting by passive versus active funds. Next, we examine whether an exogenous increase in passive institutional ownership of a firm reduces the demand for information and therefore the production of information by analysts. We predict that passive investors demand less information than active investors. Hence, our prediction is that information suppliers will respond and produce less information (O Brien and Bhushan (1990)). We also examine whether firms respond to reduced external scrutiny by decreasing payouts. How firms choose their payout policy is debated since the Modigliani and Miller s seminal work (Modigliani and Miller 1958, 1961). Although the Modigliani and Miller s model establishes that, in a world with no frictions and when the firm s investment policy is held constant, a dividend policy has no consequences for shareholders wealth, public firms have shown heterogeneous payout policies (see Farre-Mensa, Michaely, Schmalz 2014 for a recent review of the literature). Dividends are usually paid by firms that are large and mature (DeAngelo, DeAngelo, and Skinner 2004), have slower growth (DeAngelo, DeAngelo, and Stulz 2006), have larger past earnings (Benartzi, Michaely and Thaler 1997), and have lower systematic risk (Grullon, Michaely, and Swaminathan 2002). 7

10 Prior research has offered two major motivations for firms payout policies. 3 The first motivation hinges on the idea that firms can signal future profitability by paying dividends or repurchasing shares (Bhattacharya 1979; Miller and Rock 1985; John and Williams 1985; Allen, Bernardo, and Welch 2000). The economic intuition behind this signaling models is centered on the information asymmetry between corporate insiders and outside shareholders. Managers have information about their firms profitability that the market participants do not have. Hence, managers of undervalued firms have incentives to reveal such information to the market. A rise in the payout policy signals that a firm will do better (i.e., higher future cash-flows or lower cash-flow volatility, see Lintner 1956 and Grullon, Michaely and Swaminathan 2002). These models can explain why the market reacts positively to announcement of unexpected payouts increase. Furthermore, a second implication of the signaling models is that increases in payouts will be followed by improvements in operating performance. However, empirical studies present conflicting evidence on this point (e.g., DeAngelo, DeAngelo, and Skinner 1996; Benartzi, Michaely, and Thaler 1997; Grullon and Michaely 2004). A second motivation for payout policies closer to our research question is related to the agency conflicts between corporate insiders and outside shareholders. In an agency framework, profits that are not paid out to shareholders may be employed in unprofitable projects that provide private benefits for managers. As a result, shareholders may prefer dividends or stock repurchases over retained earnings. It is important to notice that the agency model for payouts moves away from the Modigliani and Miller s model since it implies that paying out dividends may reduce the inefficiency of marginal investments (which is contrary to the Modigliani and Miller s assumption that the firm investment policy is independent of its dividend policy). Furthermore, the agency model of dividends posits that, even holding the investment policy constant, corporate insiders may get preferential treatment through asset diversion and transfer prices. 3 A third motive for payout policies usually advocated by the literature is related to the taxation of dividends and capital gains. However, empirical results indicate that although taxes have some effect on payouts, they do not seem to be the first-order explanation for observed variation in payouts (Farre-Mensa et al. 2014). 8

11 Consistent with an agency model of dividends, La Porta, Lopez-De-Silanes, Shleifer and Vishny (2000) propose the idea that payout policy can be either a substitute or a complement to other mechanisms of external monitoring. Payout policy and external monitoring are complements if strong external monitoring allows shareholders to impose a payout policy that limits managers opportunistic behavior. On the contrary, if managers increase payouts to establish a reputation for a good treatment of minority shareholders then payout policy and external monitoring can be viewed as substitutes. The first view predicts that stronger external monitoring should increase payouts, whereas the second view predicts the opposite. Findings in La Porta et al. (2000) support the argument that payouts and external monitoring are complements. Similar to La Porta et al (2000), Allen, Bernardo and Welch (2000) propose a framework centered on the idea that some institutional investors that prefer dividends for tax reasons may act as external monitors and influence a firm s dividends policies. A consequence of the presence of such institutional investors is greater firm value because of the monitoring role they play. Accordingly, the board has an incentive to induce these shareholders to take a position in the firm, especially if the firm is likely to have excess cash. Several papers have recently examined this relationship empirically providing mixed results (e.g., Grullon and Michaely 2012; Hoberg, Phillips and Prabhala 2014; Crane, Michenaud and Weston 2014). In particular, some studies find evidence consistent with the notion of the substitution model. For example, Officer (2011) and John and Knyazeva (2006) find that firms with strong governance mechanism pay lower dividends, and Crane, Michenaud and Weston (2014) find that high passive institutional holdings increase dividends but have no effect on stock repurchases. However, Mullins (2014) documents that the positive relationship between passive institutional holdings and payout disappears and is even reversed in some specifications. Furthermore, Grullon and Michaely (2012) find that firms in more concentrated industries pay lower dividends compared to firms in less concentrated industries and conclude in favor of complementarity between corporate governance and payout policies. 9

12 We speak to this literature by exploiting exogenous variation in passive investors ownership of a firm. Consistent with economic theory on the monitoring role of passive investors discussed above, we predict that an increase in passive investors ownership decreases a firm s payouts since passive investors have weaker incentives and ability to monitor a firm than active investors do. II. The Voting Behavior of Passive Funds Our main question of interest in this paper is whether passive funds monitor the firms whose stock they own more or less relative to active funds. Since passive funds are unable (or less able) to exit from the stock of firms that do not act as the funds wish, there are two possibilities. If voice and exit are substitute technologies for monitoring, then passive funds should exercise their voice option over management more and be more likely to vote in an idiosyncratic way. On the other hand if voice and exit are complements, the lack of an exit option means passive funds will exercise their voice option less, and are likely to monitor less overall. To investigate the fundamental question of how passive funds exercise their voice option, we examine funds voting behavior. We use the ISS Fund Voting data, which records the votes cast by individual funds on each agenda item at shareholder meetings for most publically traded U.S. firms starting in The dataset contains 58.9 million fund vote observations for 313,635 agenda items at 41,391 shareholder meetings for 6,470 U.S. firms from 2004 to We find that passive funds vote more passively. Across the entire set of shareholder meeting votes, passive funds voted in line with management s recommendation 92.3% of the time compared to 91.1% of the time for active funds (difference-of-means t-stat = 160.3). Many of the votes were largely procedural, however, such as renewing the board of directors each year. We focus on two subcategories of agenda items that were likely to be more 10

13 contentious: items proposed by shareholders, and items that were opposed by management i.e. management s recommendation was a No vote. For the subset of shareholder proposals, passive funds voted with management 58.5% of the time compared to 57.5% for active funds (t-stat = 15.7). For the subset of proposals that were opposed by management, passive funds voted with management 58.7% of the time compared to 57.3% for active funds (t-stat = 20.9). However, the set of stocks held by passive funds differs significantly from the set of stocks held by active funds, as passive funds most often hold a simple portfolio that mimics some value-weighted index. If the firms that passive funds hold are therefore different (say, better-run) on average, the selection bias in holdings could explain the differences we find in voting. To examine this possibility Table I runs regressions of funds voting behavior on an indicator for whether the fund is passive versus active plus year and firm fixed effects. The year fixed effects absorb changes in average voting behavior over time, while the firm fixed effects absorb any differences (such as whether a given firm is better or worse-run) that are constant over time within each firm. Across the full sample, the regression estimates are similar to the overall comparison, suggesting no major differences between firms or over time. However, within the subsamples of contentious agenda items controlling for selection bias via firm fixed effects increases the discrepancy between passive and active funds voting. Comparing within the same firm over time, on shareholder proposals, passive funds are 5% more likely than active funds to vote with firm management; on proposals opposed by management, passive funds are 5.5% more likely to vote with management. The latter two findings suggest that cross sectional comparisons understate the differences in funds voting behavior, consistent with passive funds holding firms that are better-run and less contentious on average. Table I shows that passive funds vote more passively after we control for differences 1) across time and 2) across firms. However, there is still the potential for bias due to 11

14 endogeneity in these estimates. If changes in firm governance produce changes in passive ownership, or changes in other firm characteristics drive changes in both governance and passive ownership, then we cannot interpret the results of Table I as causal effects in nature. To assess the effects of passive ownership more cleanly we turn to our main research design, which isolates mechanical changes in the composition of firm ownership by passive versus active funds. III. Russell Index Reconstitution In June of each year Russell Investments reconstitutes their popular Russell 1000 (largecap) and Russell 2000 (small-cap) indexes. To determine index assignment, Russell ranks all qualifying U.S. common stocks by their market capitalization as of the last business day in May. Before June 2007, index assignment followed a simple threshold rule: stocks ranked from were assigned to the Russell 1000 while stocks ranked from were assigned to the Russell Starting in June 2007, Russell implemented a new assignment regime ( banding ). After sorting stocks by their market cap, Russell computes an upper and lower band relative to the market cap of the stock with rank 1000, where the width of each band is 2.5% of the total market cap of the Russell Stocks within the bands do not switch index assignment. That is, if a stock that ranks above the threshold but below the upper band was in the Russell 2000 last year, it will stay in the Russell 2000 the next year and similarly, if a stock that ranked below the rank-1000 threshold but above the lower band was in the Russell 1000 last year, it will stay in the Russell 1000 the next year. Russell s data suggest that banding was successful in reducing the uncertainty in index membership. In the first seven years of banding ( ), the total number of stocks that switched between the Russell 1000 and 2000 fell to 430 compared to 872 in the last seven years prior to banding ( )

15 From a research design perspective, from 2007 onward the banding regime broke the discontinuity around the rank-1000 index threshold. Thus, a simple RDD around the index threshold is infeasible after This paper proceeds from the insight that the banding regime replaced the discontinuity in index membership with two separate discontinuities in index switching. IV. Research Design Figure 2 plots index assignments for 2006, the last year before banding, and 2007, the first year of the banding regime. The solid vertical line denotes the main index threshold, between the stocks ranked 1000 and 1001 according to their unadjusted May market cap. 5 In 2006 we see that there was a sharp discontinuity in index assignment at the threshold, which lends itself to a single regression discontinuity design (Chang et al. (2015); Crane et al. (2016); Appel et al. (2016)). In 2007, by contrast, the banding regime eliminated the discontinuity near the threshold. Hence an RDD around the threshold is no longer feasible. However, we see that there are two new discontinuities at the upper and lower bands (dashed vertical lines). These discontinuities are due to a switch in status across the bands, affecting whether nearby stocks switched indexes or stayed in their previous index, in the corresponding direction (From the Russell 2000 into the Russell 1000 for stocks near the upper band, and from the Russell 1000 into the Russell 2000 for stocks near the lower band). Consider a stock that was a member of the Russell 2000 as of May 2007 and was nearby the upper band when the indexes were reconstituted. This stock s new index assignment depended on whether it ranked just above the upper band, in which case it would switch to the Russell 1000, or just below in which case it would stay in the Russell Russell does not disclose its initial rankings on unadjusted May market cap, and because they use their own proprietary calculations we do not observe the true rankings. We compute proxy market capitalization and rankings at the end of May each year using CRSP and Compustat data following Chang, Hong, and Liskovich (2015). Our results are robust and very similar when we use alternative methods of imputing the Russell initial rankings. Details are in the Internet Appendix. 13

16 The stock s index assignment in June 2007 thus depended on five parameters: 1) The stock s overall ranking in the Russell 3000, which is sensitive to small fluctuations in the market cap of both the focal stock and all the neighboring stocks in the ranking 2) The market cap of the rank-1000 stock, which determines the index threshold 3) The total market cap of the Russell 3000 as calculated by Russell, which determines the width of the bands 4) The cumulative market cap as calculated by Russell of all the stocks ranked above the focal stock, which determines where the stock sits relative to the bands 5) Whether the stock was in the Russell 1000 or 2000 last year, which was set by Russell 12 months prior. The first four parameters were difficult to predict ex ante (indeed, Russell does not make their unadjusted market cap numbers available ex post). All five parameters were difficult or impossible to manipulate. This line of reasoning suggests that within a close window of each band in each year, whether a stock ranked above or below the band and therefore switched or stayed was plausibly as good as random. The post-banding discontinuities in index switching drive our research design. Specifically, for each Russell index reconstitution in June post-banding, we select a cohort of treated and control stocks. Specifically, we select all stocks that were potential switchers, in windows of rank W around the upper and lower bands. The following steps describe the selection of the cohort for June May 2008: 1) Rank all qualifying U.S. common stocks by their unadjusted market cap as of the last business day in May ) Select all stocks that i) ranked within +/-W ranks of the upper band and ii) were members of the Russell 2000 as of May This is the set of potential switchers near the upper band. 3) Select all stocks that i) ranked within +/-W ranks of the lower band, and ii) were members of the Russell 1000 as of May This is the set of potential switchers near the 14

17 lower band. 4) For each selected stock, collect its information from CRSP-Compustat over the 3 years prior to index assignment (here, ) and the 3 years post index assignment (here, ). Add all collected firm-years to the sample. Figure 3 shows the treated and control stocks in our sample for the year 2007 for a window of +/- 200 ranks. Figure 4 shows that relative to the universe of all Russell 3000 stocks, both the upper and lower-band samples represent narrow slices of mid-cap stocks. Our estimator of the causal effects of switching indexes is then: Y ict = β 1 R1000 R2000 ict I {yearict c}+β 2 R2000 R1000 ict I {yearict c}+λ ic +γ t +ɛ ict (1) where Y is a firm-year outcome variable, c is the cohort (treatment) year, i is the firm, t is the year, λ ic, γ t are stock-by-cohort and year fixed effects and 1 if R2000 ict = 1 and R2000 ic,t 1 = 0 R1000 R2000 ict = 0 if R2000 ict = 0 and R2000 ic,t 1 = 0 1 if R2000 ict = 0 and R2000 ic,t 1 = 1 R2000 R1000 ict = 0 if R2000 ict = 1 and R2000 ic,t 1 = 1 Thus, our research design is a stacked cohort difference-in-differences estimate that uses the change in outcome Y for stock i in cohort c from three years pretreatment to three years posttreatment. The pre-vs-post changes are compared between (1) stocks near the upper band that switched out of the Russell 2000 versus those that stayed (coefficient β 1 ) and (2) stocks near the lower band that switched into the Russell 2000 versus those that stayed (coefficient β 2 ). Importantly, this means that β 1 and β 2 the effects of switching from the R2000 to the R1000 and vice versa are identified from entirely disjoint subsets of treated and control stocks. 15

18 The stock-by-cohort fixed effects sweep out any non-time-varying differences between treated and control stocks, while the year fixed effects remove aggregate trends in firm behavior or ownership. The research design thus exploits the panel nature of the data and allows us to examine the effects of index assignment on firm behavior over a longer horizon post-assignment than a simple RDD using outcomes in the year after treatment. Notably, we do not employ Russell index assignment as an instrumental variable (IV) for the effects of passive ownership. The IV approach requires the exclusion restriction (that the effects of index assignment are only due to changes in passive ownership) to hold in addition to the requirement that treatment status is as good as randomly assigned. Rather, our diff-in-diff design allows us to identify multiple direct effects of index assignment. Indeed, we find suggestive evidence that index assignment has multiple direct effects, and it is thus a challenge to disentangle which effect causes which. In this setting the requirement for an unbiased causal estimate is parallel trends absent the index reconstitution, outcomes for treated and control stocks would have evolved in parallel. We conduct extensive pretrend checks, robustness checks and falsification tests which suggest that the parallel trends condition is plausibly met. A. Data Russell index membership data come directly from Russell. Stock and firm accounting data are from CRSP and merged CRSP-Compustat respectively. We use the most recent data for each firm from June 1 to the following May 31 of each year. Mutual fund ownership data comes from the Thomson Reuters S12 database. We compute the ownership of each sample stock by every fund in December of each year. We use the number of sole-voting shares held where available, otherwise the total shares held. We classify mutual funds as passive or active using their active share following Cremers and Petajisto (2009). Our measures of fund ownership are defined below. All measures are for each stock i as of December in year t, and are expressed as a percent of the stock s market 16

19 capitalization. T otalf undown it : The fraction of stock i s market cap held by all S12 mutual funds P assiveown R2000 it : The fraction of stock i s market cap held by mutual funds with active share of 0.6 or less relative to the Russell 2000 P assiveown R1000 it : The fraction of stock i s market cap held by mutual funds with active share of 0.6 or less relative to the Russell 1000 P assiveown Other it : The fraction of stock i s market cap held by mutual funds with active share of 0.6 or less relative to a major non-russell stock index such as the S&P500 ActiveOwn it : The fraction of stock i s market cap held by mutual funds with active share greater than 0.6 to all major stock indexes Table II presents summary statistics for our sample, which consists of firm-years from 2004 through 2016 for Russell cohorts (treatment years) from 2004 through By construction the sample consists of a tight grouping of mid-cap U.S. firms; the 10% and 90% quantiles for market capitalization are $1.0 billion and $3.7 billion respectively. V. Results A. Effects on Fund Ownership Table III presents estimates of the effects of Russell index assignment on mutual fund ownership. Column 1 shows the effects of index assignment on ownership by Russell 2000 index funds. We see that ownership by Russell 2000 index funds rises by an average 1.68% of market cap for stocks that switched into the Russell 2000 relative to similar stocks that stayed in the Russell At the same time, ownership falls by 1.67% of market cap for stocks that switched out of the Russell 2000 relative to similar stocks that stayed in the Russell Importantly, as Figure 3 illustrates, the two coefficients are estimated from disjoint cohorts of stocks; potential switchers around the lower band in the first case, and 17

20 potential switchers around the upper band in the second case. Column 2 shows the effects of index assignment on ownership by Russell 1000 index funds. We see opposite effects on ownership, relative to the change in ownership by Russell 2000 funds. However, the coefficient is smaller for Russell 1000 fund holdings, because both indexes are value-weighted and thus the index weights of stocks near the bottom of the 1000 are much lower than the index weights of stocks near the top of the Russell Column 3 shows the effects on ownership by index funds that replicate another index than the Russell 1000 or (On a dollar basis this category is mostly populated by S&P500 index funds.) This category represents a useful placebo test. Russell index assignments should be irrelevant to the holdings of index funds that replicate other indexes, and indeed they are: the changes in holdings due to Russell index assignment are tiny and statistically indistinguishable from zero. Table III Column 4 shows the effects on ownership by active mutual funds. Because active fund holdings are much more volatile, the standard errors on the change in active fund holdings are larger. However, the point estimates are similar in magnitude and opposite in sign to those for passive ownership in both cases, consistent with active mutual funds selling to passive funds that enter a given stock and buying from passive funds that exit. As a result, total holdings by all mutual funds in the S12 data (Column 5) do not change noticeably around either of the Russell bands. Our results are robust to changing the sample bandwidth; as we widen or tighten the bandwidth from 200 firms to 250, 150 or 100 firms on either side of each band, the estimates remain both similar in size and statistically significant at conventional levels. Overall, our estimates are strikingly symmetric on either side of the two yearly Russell bands, and suggest that switching into the Russell 2000 leads to higher passive fund ownership and lower active fund ownership, and vice versa for switching into the Russell

21 B. Effects on Shareholder Voting Table IV present estimates of the effects of index assignment on shareholder voting. The outcome variable is the percentage of all voting shares that voted with management s recommendation on each item. Column 1 shows that across all agenda items voted on in our sample firm-years, the changes in the composition of fund ownership documented in Table III were not accompanied by any significant change in voting behavior. However, as we mentioned earlier, the vast majority of these votes are procedural in nature and there is little scope for managers and shareholders to disagree. We next examine voting behavior in two subsets of contentious agenda items: items proposed by shareholders, and items opposed by firm management. Looking at shareholder proposals (Column 2), for firms that switched to the Russell 2000 and thus had higher passive ownership, 1.78% less of the firms shares voted with management s recommendation. At the same time, for firms that switched to the Russell 1000 and thus had lower passive ownership, 1.71% more of the firm s shares voted with management s recommendation. Because we have a relatively small set of shareholder proposals in our sample (N=500), neither coefficient is statistically significant but the point estimates are symmetric and opposite in sign and their magnitude is in line with our findings on the changes in passive ownership. Column 3 looks at shareholder voting on proposals that were specifically opposed by management, i.e. the most contentious items on the agenda. For firms that switched to the Russell 2000 and thus had higher passive ownership, 2.4% less of the firms shares voted with management s recommendation. At the same time, for firms that switched to the Russell 1000 and thus had lower passive ownership, 2.89% more of the firm s shares voted with management s recommendation. The coefficients for management-opposed proposals are thus symmetric and opposite in sign, and are both strongly statistically significant. Thus, the increase (decrease) in passive ownership across the lower (upper) Russell band is accompanied by a clear increase (decrease) in the overall passivity of shareholder voting. 19

22 Table IV Column 4 again looks at shareholder voting on management-opposed proposals. This time, the dependent variable is the fraction of the firm s market cap that 1) voted with management s recommendation and 2) was owned by passive funds. Thus, these estimates show the direct effect of passive funds voting. We see that indeed, passive funds voted more passively in our setting. Intriguingly, however, the magnitude of the direct effect of passive fund voting is smaller than the change in shareholder voting overall. This finding suggests that there might be spillovers in monitoring between passive and active funds. A clear potential channel for this is the availability of public information. If passive funds demand less information about the stocks they hold, then passive holdings may reduce the amount of information supplied about the firm. Since information such as analyst coverage or media scrutiny is nonrival in use, indeed usually public, this effect might actually raise the costs of monitoring for active funds, imposing a negative spillover on active funds monitoring activity. To shed further light on this possibility, we next examine the effects of passive ownership on analyst coverage in our setting. C. Effects on Analyst Coverage Table V presents estimates of the effects of index assignment on analyst coverage. For each firm-year in the sample we count the individual analysts in the I/B/E/S database that issue at least one forecast for that firm in that year. Column 1 shows that firms moving from the Russell 1000 to the 2000 (that is, that became small-caps) on average lost 1.9 analysts while firms moving from the 2000 to the 1000 on average gained 1.7 analysts over the 3 years post-treatment. In logs this represents an average fractional decrease of 20% for firms that became small-caps and an increase of 18% for firms that became large-caps; this corresponds closely to the mean (median) analyst coverage of 8 (9) analysts per sample stock. Again, the symmetry of the estimated effects is striking, and in both cases our estimates are consistent with passive funds demanding less information about the firms they hold, 20

23 which could impose higher costs of monitoring on the active owners that remain. D. Effects on Payout Policy Last, we investigate whether the changes in passive ownership, shareholder voting and monitoring that we document have real effects on firm policy. Table VI presents estimates of the effects of index assignment on firms payout. Column 1 shows that firms that switched to the Russell 2000 and thus had higher passive ownership reduced their total payouts by 1.6% of lagged book assets on average relative to firms that stayed in the Russell At the same time, firms that switched to the Russell 1000 and thus had lower passive ownership raised their total payouts by 1.6% of book assets on average relative to firms that stayed in the Russell The symmetry of these estimates is again striking. Table VI Columns 2 and 3 show that the changes in payout come entirely from changes in equity repurchases, which fell for firms switching into the Russell 2000 and vice versa. By contrast we see no effect of passive ownership on dividends around either Russell band. Note that the standard errors on firm dividends are even smaller than those for equity repurchases; that is, there is sufficient power in our setting to detect a change in dividend policy, but we see no such effect. VI. Conclusion We use post-2007 Russell index reconstitutions to test whether passive investors are passive monitors. We find that passive investors are passive monitors. Over our 2007 to 2013 sample period, we find that passive investors are significantly more likely to vote in agreement with managers. Consistent with a governance mechanism, we also find that an exogenous increase in passive ownership is associated with changes to the firm s information environment and managerial decisions. After a firm is added to the Russell 2000, it experiences a significant increase in passive ownership which leads to significant decrease in analyst cov- 21

24 erage and news coverage. In other words, information intermediaries respond to an increase in passive ownership by decreasing information supply. In equilibrium, this suggests that passive investors have lower demand for information. As a result of this weaker external scrutiny, we also find that passive ownership causes changes in firm payout policy. Passively owned firms are less likely to repurchase shares. Overall, we find that passive ownership has important negative consequences for monitoring, governance and payout. 22

25 References Aghion, P., Van Reenen, J., & Zingales, L. (2013). Innovation and institutional ownership. American economic review, 103 (1), Appel, I., Gormley, T. A., & Keim, D. B. (2016). Passive investors, not passive owners. Journal of Financial Economics, forthcoming. Boone, A. L., & White, J. T. (2015). The effect of institutional owners on firm transparency and information production. Journal of Financial Economics, 117, Brav, A., Jiang, W., Partnoy, F., & Thomas, R. (2008). Hedge fund activism, corporate governance, and firm performance. The Journal of Finance, 63 (4), Chang, Y.-C., Hong, H., & Liskovich, I. (2015). Regression discontinuity and the price effects of stock market indexing. Review of Financial Studies, 28, Crane, A. D., Michenaud, S., & Weston, J. P. (2016). The effect of institutional ownership on payout policy: Evidence from index thresholds. Review of Financial Studies, 29, Cremers, K. M., & Petajisto, A. (2009). How active is your fund manager? a new measure that predicts performance. Review of Financial Studies, 22, Cronqvist, H., & Fahlenbrach, R. (2008). Large shareholders and corporate policies. The Review of Financial Studies, 22 (10), Dasgupta, A., & Piacentino, G. (2015). The wall street walk when blockholders compete for flows. The Journal of Finance, 70 (6), Edmans, A. (2009). Blockholder trading, market efficiency, and managerial myopia. The Journal of Finance, 64 (6), Edmans, A. (2014). Blockholders and corporate governance. Annual Review of Financial Economics, 6 (1), financial Retrieved from doi: /annurev-financial Fich, E. M., Harford, J., & Tran, A. L. (2015). Motivated monitors: The importance of institutional investors portfolio weights. Journal of Financial Economics, 118 (1), Retrieved from 23

26 S X doi: Levit, D. (2017). Soft shareholder activism. Maug, E. (1998). Large shareholders as monitors: is there a trade-off between liquidity and control? Journal of Finance, 53 (1), Morgan, A., Poulsen, A., Wolf, J., & Yang, T. (2011). Mutual funds as monitors: Evidence from mutual fund voting. Journal of Corporate Finance, 17 (4), O Brien, P. C., & Bhushan, R. (1990). Analyst following and institutional ownership. Journal of Accounting Research, Roe, M. J. (1990). Political and legal restraints on ownership and control of public companies. Journal of financial economics, 27 (1), Shleifer, A. (1986). Do demand curves for stocks slope down? Journal of Finance, 41,