Voting with their Feet or Activism? Institutional Investors Impact on CEO Turnover

Transcription

1 Current Draft: September 30, 2011 Voting with their Feet or Activism? Institutional Investors Impact on CEO Turnover Jean Helwege Moore College of Business 1705 College Street University of South Carolina Columbia, SC (803) Vincent J. Intintoli College of Business Southern Illinois University Carbondale Carbondale, IL (618) Andrew Zhang College of Business The University of Nevada Las Vegas Las Vegas, NV (702) We are grateful for helpful comments from an anonymous referee, Brian Bushee, Shingo Goto, Jason Greene, Kathy Kahle, Simi Kedia, Sandy Klasa, Karthik Krishnan, Y.C. Loon, Ernst Maug, Harold Mulherin, Charu Raheja, Laura Starks, Deon Strickland, D.H. Zhang and seminar participants at University of Georgia, University of Nevada at Las Vegas, Pennsylvania State University, University of South Carolina, and Southern Illinois University Carbondale as well as attendees at the 2010 Western Finance Association meeting, the 2009 Florida State Spring Beach conference, the 2009 New York Accounting and Finance Forum and the 2009 Northern Finance Association meeting. Vincent Intintoli is also grateful for financial support from the Office of Research Development and Administration at Southern Illinois University Carbondale. Andrew Zhang is grateful for financial support from the College of Business at UNLV for 2010 summer research support. Amine Khayati, Ana Mkrtchyan, Gang Xiao, and Yuan Wang provided excellent research assistance. Electronic copy available at:

2 Voting with their Feet or Activism? Institutional Investors Impact on CEO Turnover Abstract We examine the relation between institutional investors and management discipline over the last several decades to better understand how CEO turnover has increased. Using a sample of forced and voluntary turnovers, we investigate the changing roles of activism and exit among institutional investors between and We find evidence of activist investors throughout the sample period and their impact is consistently significant in multivariate analysis. In contrast, voting with their feet has declined to the point where it no longer affects turnover outcomes. Nonetheless, activism is fairly uncommon and does not explain the higher turnover observed over time. Block holdings of known activists have increased and are linked to improving target firms. However, other blocks merely reflect the increasing size of institutional money managers. Going forward, the increasing size of institutional investors seems likely to inhibit voting with their feet while activism remains an important vehicle for change. Electronic copy available at:

3 From October 2008 to March 2009 Bank of America s share price fell by 85%, reflecting both the effects of the subprime crisis and the impact on the firm s health of two acquisitions (Countrywide and Merrill Lynch) that brought the bank to the brink of insolvency. Shareholders blamed Ken Lewis, the CEO and Chairman, for the decline, arguing that he should have resisted government pressure to buy Merrill Lynch s toxic assets. By April 2009 he was ousted as Chairman after shareholders voted to separate the job from that of CEO. The close vote (49.7 to 50.3) was influenced by CalPERS, which had called for shareholders to vote against both duality and all 18 board members. A few months later Lewis took early retirement and relinquished the CEO title as well. While CalPERS has been targeting underperforming firms for several decades, a more common reaction by institutional shareholders to poor stock returns is to cut their losses and simply sell the stock (McCahery, Sautner, and Starks (2010)). Recent theoretical work suggests that when institutions choose to exit underperforming stocks, the sale of shares may serve as a governance mechanism (Edmans (2009), Edmans and Manso (2011), and Admati and Pfleiderer (2009)). These theories receive empirical support from Parrino, Sias, and Starks (2003), who find that institutional stock sales significantly increase the probability of forced CEO turnover, as well as from studies by Gopalan (2008), Gallagher, Gardner, and Swan (2009), Chen, Smith, and Swan (2009), Bharath, Jayaraman, and Nagar (2010), and Qian (2011). This raises the question of whether Lewis lost control of Bank of America because a few major shareholders pushed him out or because of the decline in share price as other institutional investors voted with their feet. CalPERS itself had mostly pursued a policy of exit in cases of poor returns before the mid 1980s, but it switched its strategy to fixing underperformers to avoid further losses from selling. According to manager Dale Hanson, voting by selling the stock was no longer an option: 1

4 It s simply not feasible for us to dump our stock in companies when we are dissatisfied with the management. Public employee pension funds are so large that the sale of such a huge block of a company stock would [not] be in our own best interest. 1 Carleton, Nelson, and Weisbach (1998) report that TIAA-CREF chose to index most of its stock holdings for the same reason. Since adopting this stance CalPERS has pursued activism via shareholder proposals, discussions with management, and focus lists. Its strategies also involve aligning with other institutions in order to gain a shareholder block with greater power. 2 The changes at CalPERS and TIAA-CREF call into question whether institutional investors will continue to view exit or the threat of exit as an option in disciplining management. As institutions grow and hold increasingly larger stakes, exit may become a more costly alternative to actively fixing the management problem. Thus, while activism has the problem that improvements in the firm benefit all shareholders while costs accrue only to agitators (Shleifer and Vishny (1986), Grossman and Hart (1980), Kahn and Winton (1998), and Admati, Pfleiderer, and Zechner (1994)), exit may lose its impact as a governance tool as institutions are increasingly constrained to hold large cap stocks regardless of performance. While both exit and voice may impact corporate policy, trends in the money management industry give rise to the possibility of a stronger role for activism over time. In this paper we examine how institutional investor s impact on forced CEO turnover has changed over time. We analyze institutional activism and voting with their feet using a sample of forced and voluntary turnovers from and compare institutional investor actions in the first half of the sample ( ) to their choices during later years ( ). We find that news of activism is significant in multivariate logits in both time periods, but voting with their feet loses 1 Washington Times, November 17, 1991 Who s at the Helm? Big business, big investors slug it out for corporate control. See Smith (1996), Karpoff (2001), and English, Smythe, and McNeil (2004) on CalPERS strategy. 2 Wahal (1996), Strickland, Wiles, and Zenner (1996), Opler and Sokobin (1998), Del Guercio and Hawkins (1999), Gillan and Starks (2000, 2007), and Del Guercio, Seery, and Woidtke (2008) analyze the effects of such activism. 2

5 significance in the later period. Thus we conclude that exit is unlikely to be a dominant force in management discipline going forward while institutional activism will likely continue to play an important role. Although activism by institutional investors is a significant factor in forced CEO turnover throughout the sample time period, it is a fairly uncommon occurrence. Moreover, its time trend is flat, so it is unlikely to explain the higher frequency of turnover observed over time. News of activist institutions appears in roughly 20% of forced turnovers in each half of the sample. Block holdings (5% or more) of known activist institutions and the percentage of firms with such investors have trended up over time, providing slight evidence that activism is on the rise. And while logit regressions suggest blocks are more often bought as part of a strategy to improve underperforming firms, we note that the vast majority of firms have an institutional blockholder by the end of the sample period and many of these investors may form blocks only because they have large assets under management. Thus, in multivariate logits explaining forced turnover the presence of an institutional blockholder is only significant in the earlier period. Likewise, we do not find evidence that voting with their feet explains the upward trend in turnover. In fact, exit, measured by the net change in institutional holdings, has declined at forced CEO turnover firms over the sample period. While univariate tests show that exit at forced CEO turnover firms is statistically different from that of voluntary turnover firms in both periods, significance in the later half merely reflects higher institutional holdings at other firms. The change in exit, however, does not reflect a change in gross sales by institutions a la CalPERS and TIAA-CREF as much as it does greater institutional purchases of stock. These purchases often create or add to blocks, which may be less costly as stock market liquidity 3

6 improves over time (Chordia, Roll, and Subrahmanyam (2001)) or may simply reflect the increasing size of mutual fund families in recent years. In summary, we analyze the trends in turnover, activism, institutional stock holdings, and the presence of blockholders to determine the relative roles of activism and exit in corporate governance. We find a diminishing role for exit while activism is fairly stable. While our results speak to the topic of the upward trend in turnover over time, they do not explain the trend beyond ruling out exit, activism, and blockholders as potential factors. This suggests that future research on turnover may be more fruitful if the focus turns away from institutional investors. The remainder of the paper is organized as follows. Section 2 describes the data on turnover and activism. Section 3 discusses trends in institutional ownership, turnover, and activism. Section 4 discusses evidence of exit and activism in the sample of forced CEO turnover firms compared to voluntary and matching firms. Section 5 concludes. 2. Data We create a sample of CEO turnovers at NYSE firms during by examining annual Forbes executive compensation surveys and searching for turnover announcement dates and explanations in the Wall Street Journal (WSJ). 3 We exclude cases involving mergers and those involving illegal activity. We identify an event as forced if (1) WSJ indicates it is, or (2) when the departing CEO is less than 60 years old and the departure does not owe to poor health, death, or acceptance of a new position (Parrino (1997)); otherwise, the turnover is voluntary. 4 Our sample consists of 227 forced and 924 voluntary turnovers, with slightly more than half of forced turnovers occurring in the period. We also compare our forced turnover 3 We thank Robert Parrino for graciously sharing his pre-1996 firm turnover data. 4 Jenter and Lewellen (2010) note that many departures that are classified as voluntary may actually be forced. In unreported results we redefine underperforming voluntary turnovers as forced and find similar results. 4

7 sample to a group of non-turnover firms that are matched on size and performance. We choose the NYSE firm with the closest one year stock return from among those with a market capitalization between 50% and 150% of that of the forced turnover firm, measured one year prior to the event and verify that each firm does not change its CEO from two years before to one year after the turnover. Quarterly holdings of institutions that file 13-Fs (those with at least $100 million in equity) are from Thompson Financial. We also identify blockholders from acquisitions of minority interest in the Securities Data Corporation (SDC) database during the three years before the event date. SDC data includes 13-D filers as well as investors that file 13-Fs. We match by CUSIP, taking into consideration name changes, and verify in LexisNexis that the blocks were not sold before the turnover date. We ignore blocks owned by employee stock ownership plans because they are unlikely to monitor the firm. Most block investors identified from SDC data also file 13-Fs, so while these data allow a comprehensive analysis of institutional investors most of the information in the data lies with the 13-F filers. We calculate the percentage of shares held by institutional investors by dividing 13-F holdings by the number of shares outstanding in CRSP. Some institutions incorrectly report quarter-end holdings, such as by forgetting about recent stock splits, which lead to outliers. We eliminate them with the following steps: First, whenever institutional ownership sums to more than 100% we investigate stock splits, equity issuance, and reverse splits. 5 Second, we investigate all observations where institutional ownership changes by more than 20% during the year before the turnover event, either cumulatively or in one quarter. Finally, we winsorize all of the continuous variables used in analysis at the 1 and 99 percentiles. 5 If the data in CRSP or 13-F filings appear to be updated with a lag we adjust the percentages to reflect these events. We delete two observations where we cannot determine why the total ownership exceeds 100%. 5

8 We identify institutional activism by searching LexisNexis and the WSJ for articles in the year before the event date for evidence of institutional investor pressure. We set the indicator variable news of activism to one when where there is (1) a threat or attempt of takeover, (2) an effort to replace all or part of the board of directors or the CEO, (3) an initiation or threat of a proxy contest, (4) an effort to amend anti-takeover provisions, or (5) a demand for asset downsizing. 6 When we find such news, we create a second variable (news of activism involving 13-F filer) for activists that file 13-Fs. Some activists use investment vehicles that are not named in news stories. We identify them from other articles in the press and the list of vehicles in Holderness and Sheehan (1985). 7 We also create an indicator variable based on known activists (known activists holding a block), which equals one whenever any known activist institutions individually or collectively have 13-F holdings of 5% or more in the target firm at t=0. 8 These activists include those found in news about our sample firms and investors identified by searching on the words activist and shareholder in LexisNexis over the period Trends in turnover, ownership, and activism We next turn to an examination of trends in turnover, ownership, activism, and block holdings over our 25 year sample. Columns (1)-(8) in Table 1 focus on our sample of turnover firms and columns (9)-(12) cover the sample of all CRSP firms. Consistent with data in Kaplan and Minton (2010), the percentage of firms with forced CEO turnover in our sample has increased over time (column (1)). 10 While the upward trend is not smooth, a regression of 6 These activities are not mutually exclusive and often occur simultaneously in a given instance of activism. 7 Holderness and Sheehan (1985) examine six major corporate raiders (Carl Icahn, Irwin Jacobs, Carl Lindner, David Murdoch, Victor Posner and Charles Bluhdorn) and list 35 companies controlled by these raiders that are used to make investments in the target firms. 8 We define t=0 holdings as those in the calendar quarter closest to but before the turnover date. 9 We eliminate Fidelity and Franklin in creating this variable to avoid overstating the incidence of activism. 10 Our turnover figures differ from theirs because we ignore mergers. 6

9 turnover on time reveals a coefficient with a t-statistic of If increasing management discipline owes to a larger role for voting with their feet, it must be that firms with forced turnover experience far different trends in institutional ownership than the typical public firm because institutional ownership rises sharply from Column (9) shows that the fraction of institutionally owned shares in the universe of CRSP stocks increase from about 13% to nearly 50%, suggesting that more institutions find themselves in CalPERS and TIAA-CREF s position as time goes on. Column (2) shows that our sample of voluntary and forced CEO turnover firms also experience a large increase in institutional ownership, but because our firms are large enough to be included in the Forbes surveys, institutions owned a large fraction of their shares even as early as As can be seen in columns (3) and (4), the growth in institutional share holdings over the last several decades owes in large part to the increasing size of the mutual fund industry. Greater institutional holdings will not imply more difficulty in exiting stocks if it is the result of a larger number of institutions holding portfolios of roughly the same size. Results in Table 1 show that this is not the case, as the incidence of block holdings by institutions has risen along with their number. Columns (5), (6), (10), and (11) show that institutional block positions have increased over the sample period, whether measured by the holdings or the percentage of firms with such blocks. Institutional block holdings increase over twofold from 5.4% in 1982 to more than 13% by 2006 in our sample, and rise sharply for CRSP firms as well. While some of these blocks reflect an increased desire to exert control in the firm (Bethel, Liebeskind, and Opler (1998), Denis and Serrano (1996) and Barclay and Holderness (1991)), others simply 7

10 reflect the large holdings of mutual fund families, such as Fidelity, whose shares are aggregated in the 13-F data. 11 Kaplan and Minton (2010) link the rise in block holdings over time to increased CEO turnover, but we note that nearly every firm in our sample has an institutional blockholder by For the CRSP universe, institutional block holdings increase in every year but one, while the upward trend in CEO turnover is far weaker. Given that some of the blocks are held by mutual fund families that do not pursue activist strategies, the presence of a blockholder may be a poor proxy for the level of monitoring. Further evidence that the trend in monitoring may be flat is in the frequency with which we find activism among our sample firms. While this news occurs more often later in the sample, the regression coefficient is not significant, casting doubt on activism as an explanation for the increase in turnover over time. However, if we consider the presence of institutions that have been known to pursue activist strategies (known activists holding a block), both their holdings and the fraction of firms targeted has increased steadily over the period (columns (7) and (8) and (11) and (12)). One reason why the incidence of reported activism has not increased since 1982 is the changing nature of activists over time. During the first half of our sample ( ) activists were mostly corporate raiders, who usually made block purchases as initial steps in takeover threats (Holderness and Sheehan (1985)). As Pound (1992) notes, many of these takeover attempts were viewed by politicians and the general public in a negative light, resulting in laws and regulations that left the corporate raiders tactics ineffectual after the late 1980s. The subsequent demise of the hostile takeover market would suggest a decrease in institutional 11 We note that exit may become less costly for these blockholders if liquidity improves over time, as Chordia et al. (2001) suggest. However, the same factors that improve liquidity may also reduce the price impact of exit and thus its effect on corporate policy. 12 For a contrary view see Mikkelson and Partch (1997). 8

11 activism over time absent greater efforts by other investors, especially, as Black (1992) argues, because these other institutions face legal hurdles that considerably limit their influence on governance. But in the early 1990s regulations changed, making it easier for pension funds and others to step in where corporate raiders had left (Holmstrom and Kaplan (2001)). For example, in 1992 the SEC eased restrictions on communications between shareholders, allowing investors with smaller holdings to band together more freely in monitoring the firm, as in the case of CalPERS and the Council of Institutional Investors (Opler and Sokobin (1998)). In 1994, the Department of Labor explicitly encouraged monitoring by pension funds (Gillan and Starks (2007)). Prudent man standards changed so that alternative allocations, including those to activist hedge funds, became the norm. Hedge funds have come to play an increasingly important role in governance since the early 1990s (Klein and Zur (2006, 2009), Brav, Jiang, Partnoy, and Thomas (2008), Becht, Franks, Mayer, and Rossi (2009), Greenwood and Schor (2009), Boyson and Mooradian (2011) and Aggarwal, Erel, Ferreira, and Matos (2011)). For example, Greenwood and Schor (2009) show that activist hedge funds went after dozens of target firms per year from 1997 on, with 141 events in 2005 alone. Thus, despite the many changes in the types of institutions that pursue activist strategies, forced CEO turnover may reflect the efforts of activism by institutional investors in both earlier decades and more recently. While the aggregate data suggest that exit is less likely to play a role in turnover, forced turnover is confined to a small fraction of the firms in CRSP and voting with their feet may be more important in this group. We investigate exit and activism at these firms next. 9

12 4. Institutional trading and activism in forced CEO turnover firms We examine the trading patterns of institutional investors and activism in the sample of firms that have forced CEO turnover during the year leading up to the turnover event (quarters t=-4 and t=0), comparing them to voluntary turnover firms and matches. In unreported results we find qualitatively similar patterns when we analyze institutional holdings starting in t=-8 and also if we examine trading patterns in separate quarters leading to the turnover date. 13 To determine how these patterns have changed over time, we split the sample period into two halves ( and ). This allows us to work with fairly large samples of forced CEO turnover firms in each period (91 in the first half and 137 in the second period). We then estimate multivariate logit regressions for each sample period to compare exit and activism while controlling for other factors. We focus on net changes in institutional shares because exit-asgovernance theories rely on the assumption that institutional investor stock sales causes downward pressure on the stock price, which would not occur if institutions both sold and purchased shares in equal amounts. However, we also include some evidence on gross sales and purchases. Next, we discuss control variables, followed by summary statistics for the sample and then present data and tests on voting with their feet and activism. A. Control variables Previous research shows conclusively that CEOs are more often forced out when their firms perform poorly. 14 We control for firm performance using return on assets (ROA) and unadjusted stock returns in the year before turnover. ROA is defined as earnings before interest and taxes over total assets less its median 2-digit SIC code industry ROA. We focus on raw 13 Institutional data are only reported quarterly, so the year before the event is actually the prior four quarters. 14 See Coughlan and Schmidt (1985), Warner, Watts, and Wruck (1988), Weisbach (1988), Gilson (1989), Denis and Denis (1995), Kang and Shivdasani (1995), Huson, Parrino, and Starks (2001), Conyon and Florou (2002), Huson, Malatesta, and Parrino (2004), Fee and Hadlock (2004), Jenter and Kanaan (2008), and Kaplan and Minton (2010). 10

13 returns to measure stock performance because CEO turnover is more sensitive to them than to adjusted returns (Jenter and Kanaan (2008)). 15 We also create a variable (distress indicator) to capture situations where the stock is so beaten down as to no longer be considered an appropriate investment for institutions. Distress equals one for firms where in the year before the turnover event: (1) the firm s debt is downgraded from investment-grade to junk, (2) the firm is delisted from the S&P 500 index, (3) the stock trades below $5, (4) the firm files for bankruptcy protection, or (5) the firm cuts or omits a dividend payment. Deletions from the S&P 500 are identified from CRSP. Rating changes are identified using the Fixed Income Securities Database (FISD), Compustat, LexisNexis, and Standard and Poor s Creditweek. We identify bankruptcies using FISD, Compustat, CRSP delisting codes, and the Capital Changes Reporter. To control for nonlinear effects on performance, we consider whether sudden drops (or increases) in stock prices lead shareholders and the board of directors to replace management by using the standard deviation of stock returns in the year prior to turnover. Stockholders may blame poor performance on the CEO in some cases but not in others and we cannot readily discern the two situations. We posit that the CEO is deemed to be at fault more often when there are write-offs, as they should increase after bad acquisitions and projects that confirm the poor decision-making of the CEO. We create an indicator variable for write-offs that is set to one whenever Compustat item 17 is negative and has an absolute value that is more than 1% of assets (Elliot and Shaw (1988)). Managers may be difficult to oust even when their firms perform poorly if they have effective control over the board of directors (Denis, Denis, and Sarin (1997)). We measure entrenchment by the percentage of outside directors, duality, and an indicator variable for family- or founder-run firms (Weisbach (1988), Huson, Parrino, and Starks (2001), McNeil, 15 In unreported estimates we use NYSE-adjusted and industry-adjusted stock returns and find similar results. 11

14 Niehaus, and Powers (2004) and Parrino (1997)). Duality and the percentage of outside directors are collected from annual proxy statements, where outside directors are defined as non-employee directors who have no association with the firm. For the early part of the sample we obtain these data from the Million Dollar Directory and America s Corporate Families. 16 Information for our founding family indicator variable is obtained from proxy statements and the WSJ. We also collect board ownership from proxy filings and Value Line (see Kole (1995)). As is common in the literature, we control for the age of the CEO. The age variable is set equal to one if the CEO is older than 59 and zero otherwise. B. Summary statistics Table 2 provides descriptive statistics for our sample. Both forced and voluntary turnover firms are large, with book assets in the billions of dollars. As expected, departing CEOs who are forced out tend to be younger and have a shorter tenure at the top post than those that turn over voluntarily. Instances of forced CEO turnover on average involve severe underperformance. In comparison, the stock returns for voluntary firms are more typical of those for the S&P 500. Our matched sample firms have similarly poor one year stock returns when compared to the forced turnover sample, as expected given how we select matches. Accounting profits at forced CEO turnover firms lag those of both voluntary and matched firms and the fraction of firms with write-downs is significantly higher at forced turnover firms as well. Distress is sharply more likely among forced turnover firms, both because they lost substantial value before the turnover and because they were once large enough and sufficiently profitable to pay dividends, hold an investment grade rating, and belong in the S&P For the 12 cases where board composition data are unavailable we assume the percentage of outside directors equals the sample mean. When we remove these observations the results are qualitatively the same. 12

15 Governance among the non-turnover matched firms differs markedly from that at the forced turnover firms. Since both sets of firms are similar in size and both have exceptionally poor performance, one would expect both groups to experience above average turnover yet all of the CEOs in the forced turnover set lose their jobs while not one CEO in the group of matches is ousted. This difference is likely related to the latter s greater ability to maintain control as matched firms are far more often run by founders or their families, their boards have considerably fewer outside directors, and the CEO is more often also the Chairman. Governance at forced CEO turnover firms is similar to that of voluntary turnover firms, if not stronger. Their percentage of outside directors is slightly higher and the fraction of family firms is somewhat lower. Forced CEO turnover firms also have higher board ownership but these figures include CEO ownership and may not reflect ownership by other members. C. Voting with their feet in the year leading up to the turnover event Table 3 shows the extent of institutional exit for firms with forced CEO turnover compared to voluntary turnover and non-turnover matched firms. The results reveal a significant amount of voting with their feet in cases of turnover in both halves of the sample, but while there is a drop-off in institutional holdings at forced turnover firms in the second half of the sample, the decline is quite small. Nevertheless, because institutional ownership has trended up for most firms, even this low level of exit is significantly below the change at voluntary turnover firms. Panel B also shows that in the case of the matched firms, institutions do not dump losers on average in the later time period. Notably, changes in institutional investor holdings at the matching firms are similar to those at forced CEO turnover firms in all time periods, suggesting that exit alone does not lead to turnover. 13

16 Theories of exit-as-governance rely on price pressure on the stock as informed investors sell the shares and price pressure is expected to be greater the larger the number of shares sold (Scholes (1972)). So while the net change in institutional holdings at underperforming firms is not large overall, net stock sales by blockholders could still drive changes in corporate policy in these models. In Table 4 we examine voting with their feet according to the size of the institution s holdings at t=-4 and t=0, splitting institutional investors into those with small stakes (less than 1%), intermediate size holdings (1%-5%) and blocks (5% or more). The change in small stakes at forced CEO turnover firms is negative and significantly different from those at the voluntary turnover firms, and the pattern is equally strong in both halves of the sample. Moreover, although not statistically significant, the institutions with small stakes actually vote with their feet to a greater extent in the latter half of the sample (-1.31%) than in the first half (-1.25%). Theory, however, suggests that sales of small stakes, which are not considered informed, will not lead to price pressure, and empirically such a small net change in their positions is unlikely to impact prices. Thus, we focus more on the figures for blockholders, who are more likely to be informed investors. Block investors, in contrast, do not exit firms where the CEO is forced out, instead increasing their net holdings slightly as the year progresses. The accumulation of block positions is somewhat stronger in the period, although not significantly so. Institutions with intermediate stakes in forced turnover firms tend to sell their shares on net in earlier years but maintain their positions in the later period, similar to the change in strategy at CalPERS. In fact, CalPERS does not hold a block position in any of our sample firms, so to the extent its strategy of low turnover affects our data, it would show up in the intermediate category. 14

17 As before, forced turnover firms and their matches have similar patterns of exit despite their different turnover rates, again suggesting that exit does not cause turnover. Overall, the larger their equity stakes, the less likely investors are to vote with their feet and this finding is stronger in the later time period. The declining overall amount of institutional exit, especially given its increasing concentration among smaller stakeholders, seems unlikely to have led to a rise in forced CEO turnover over the sample period. While these results are consistent with the idea that institutions sell their shares less often as they increase in size (i.e., they follow the CalPERS strategy), the figures in Table 4 represent net changes in positions, not gross sales. Table 5 presents more detailed data on the trading patterns of institutional investors to determine whether the decline in exit reflects lower gross sales or whether the net positions change over time because of purchases. We investigate this question in two ways: First, in Panel A, we look at the net change in institutional investor holdings for investors who might have left the firm (those with stakes at t=-4) and those that might be new buyers (investors who held stakes at t=0). We show these net changes in ownership by size of the stake. 17 Second, we investigate the gross sales and gross purchases, again by size of the holding, with the results summarized in Panel B. If the decline in exit over time reflects institutions decision to buy and hold more often, then the reduction in exit seen in Table 4 should show up in Panel A of Table 5 as a decrease in the sales by t=-4 investors. Alternatively, the net figures in Table 4 could be driven by increases in holdings, including new purchases, of the t=0 investors. Likewise, if the net figures change because of a decline in gross selling the gross sales shown in the left side of Panel B should fall in the latter half of the sample, but if exit declines because of purchases this should be evident on the right side of Panel B. 17 Some firms do not have blockholders at either t=-4 or t=0 so we set the change in block holdings for them to zero. 15

18 The data in Panel A of Table 5 suggest that small and intermediate investors at t=-4 sell fewer shares of all types of firms (voluntary turnover, forced turnover and matches) in the later period than in the earlier period, consistent with a switch to lower share turnover. However, block sales actually increase over time, with the aggregate net sales climbing from 2.82% to 3.15% for the forced turnover firms. Given this result, the only way for the block ownership figures in Table 4 to increase is if some institutions purchased blocks in the year leading up to the turnover event. This can be seen on the right side of the panel, where block holdings for t=0 investors rise by 4.40% in the latter half of the sample, which is an increase of 1.38% over block holdings in the earlier period. When comparing early to late period holdings, Panel B shows that gross sales by institutions do not decline for any type of firm or any size investor, indicating that many institutions did not change to a CalPERS-type governance strategy. These results hold whether we look at investors who might have sold off their entire stake (t=-4 investors) or those who maintain some stake in the firm a year later (t=0 investors). Notably, gross sales by t=-4 investors are higher for forced CEO turnover firms than for voluntary turnover firms in most cases in both early and late periods. However, purchases of stock are also rising over time. Purchases in Panel A are best evaluated by examining the t=0 investors, as completely new investor purchases are ignored in the t=-4 figures. Purchases are higher in the later period for all size categories of institutions and the purchases of blocks are significantly larger for forced turnover firms compared to voluntary turnover firms. Rather than a switch to a buy and hold strategy to avoid the effects of price pressure, these data suggest that institutions actually sell (and buy) shares more often now than in the past. 16

19 Is the CalPERS strategy the exception in our sample? CalPERS gross sales of forced turnover firm stock drops from.29% to.16% in the later period, with the largest decline occurring among sales of its intermediate-sized stakes. In unreported results, we find that most other institutions follow a similar strategy. For example, if we look at the number of institutions that sell shares as a fraction of the institutions that hold at stake at t=-4, the number of selling investors has fallen over time from 72% to 68%,68% to 64%, and 71% to 66% for forced turnover, voluntary turnover and matched firms, respectively. In fact, if we split the data into the six types of institutions reported in the 13-F data, only the mutual funds (type 3 institutions) exhibit a strong increase in gross selling and their trades drive much of the net changes. 18 The other five types of institutions either sell about the same amount in each period or have lower gross sales in the later period. Not only did mutual funds disagree with CalPERS strategy, but their impact on the figures in Table 5 is quite large in the second half of the sample because these investors represent a larger portion of the sample in later years (recall columns (3) and (4) of Table 1). The large amount of gross sales in Panel B, especially by blockholders, raises the question of whether this form of exit causes higher CEO turnover. First, we note that block sales at matched firms are as high for matched firms as they are at forced turnover firms in both periods but this pressure does not lead to turnover. Second, we note that block sales and purchases by mutual funds may not represent the exit and entrance of the informed investors that drive price pressure theories. In particular, the sales and purchases of blocks over time may merely reflect the fact that 13-F data treat mutual fund families as a group but the individual 18 In unreported analysis, we examine changes at grey and independent institutions (Brickley, Lease, and Smith, (1988) and Chen, Harford, Li (2007)) and also at dedicated, transitory and quasi-indexers (Bushee 1998). We find that buying and selling is more common at independent and transitory institutions, consistent with results on mutual funds. 17

20 funds owned by the families may not trade in concert. Thus they may form blocks accidentally when one fund s purchases push them over the 5% mark and then disintegrate when another fund decides to reduce its position the next quarter. Such mutual fund investment policies imply that purchases of blocks are not part of an activist strategy either. To test whether blocks in the later period represent an increase in monitoring over time, we consider the formation of new blocks at forced turnover firms. Panel A shows that the institutional block investors that are present at the time of the forced turnover increased their holdings on net by 3.02% in the first half of the sample, offsetting net sales of 2.82% by blockholders who had been at the firm at t=-4. The comparable figures for block investors are even higher in the second half of the sample period (purchases of 4.40% relative to sales of 3.15%). Looking at gross purchases in Panel B, we again see that block holdings increase sharply among investors who joined the firm s shareholder base in the year leading up to the turnover event. It seems likely that some of the is activity represents that of new investors who are willing to monitor and discipline the firm since most of the blocks were formed in the year leading up to the turnover event and block formation is more likely as time goes on, and the new blocks are more common at forced turnover firms. In the next section we consider more direct evidence on activism at forced CEO turnover firms. D. Activism in the year leading up to the turnover event Tables 4 and 5 suggest that there are higher block holdings at forced CEO turnover firms compared to voluntary firms, especially in the second half of the sample period. If institutional blockholders buy their shares after the firm s troubles begin, it is more likely that they do so with an eye toward making improvements compared to blockholders who have been invested in the firm for some time. The latter group of blockholders may be mutual fund families or other large 18

21 investors that do not find it profitable to try to influence the company. If institutional blockholders gain these large stakes in the year leading up to the ouster of the CEO because they intend to improve the returns through activist strategies, we should see that new blockholders are more likely to appear when the firm is underperforming and the chance of reform is higher. Likewise, activists should be common among firms where the payoff to improvement is higher so their targets should have lower performance and less entrenched management. Table 6 shows the presence of institutions that might undertake reforms and the incidence of activism at forced CEO turnover firms, while Table 7 reports the results of logit regressions to predict activism and block holdings. Panel A of Table 6 shows that forced turnover firms attract more new blockholders (first column) and are more likely to have a blockholder at t=0 than voluntary turnover firms. This is true whether we count blockholders using 13-F data only (second column) or by combining 13-F data and SDC data (third column). 19 However, the fraction of firms with blockholders is only significantly higher in the earlier period. New institutional blockholders appear at forced CEO turnover firms as frequently in the second half of the sample period as they do in the first half. While these results suggest that block holdings lead to increased monitoring at underperforming firms, the high presence of blocks in the second period at all firms suggests that the set of institutional blockholders and institutional activists overlaps less over time. Undoubtedly, this owes in part to the rise of very large funds that simply hold large amounts of all stocks. Direct evidence of activism in the financial press indicates a large role for institutional monitoring at forced CEO turnover firms, shown in Panel B of Table 6. Institutional investor activism is far more common in cases of management discipline than otherwise: Based on news 19 Note that the addition of the SDC data on blockholders scarcely affects the number of firms with blockholders, as most of the investors in the SDC data also file 13-Fs. 19

22 articles (first column), the incidence of activism when the CEO is ousted is over 20%, compared to about 2% for the voluntary turnover and matched firms. We find similar results when we examine the fraction of firms with 13-F investors that are identified in the news articles (news of activism involving 13-F filer), shown in the second column of Panel B. In fact, most of the activists mentioned in news articles are also in the 13-F data. For example, panel B shows that 72% of the news about forced turnovers in the second half of the sample involves 13-F filers. The other 28% of the institutions are exempt from filing 13-F reports (or do so under an unknown name) and also avoid 13-D filings by holding less than 5% in the firm. Defining an activist more broadly (known activists holding a block), we also find that forced CEO turnover firms are more likely to be targeted, although the differences are not always significant. However, we find that activism by these three measures is about as strong in the first half of the sample period in cases of forced turnover as it is in the second half. This result supports our view that the nature of institutional activism has changed over the last several decades, moving away from hostile takeover threats and corporate control by a small cadre of raiders to less extreme discipline measures by a wider group of institutions that focus on improving the operations of the firm. Table 7 presents estimates from four logit equations to determine if institutional blocks and activism are more common at firms where the expected payoff to activism should be stronger; where performance is poor and entrenchment is low. We measure firm performance with stock returns, accounting profitability, and write-offs. We control for management entrenchment with the percentage of outside directors, duality, and a family firm indicator. Further, since blocks are considered illiquid in general, the ability to exit later may affect the decision to buy the block in the first place (Maug (1998)). Thus, we also include the Amihud 20

23 (2002) illiquidity measure to control for the influence of price pressure (See Goyenko, Holden, and Trzcinka (2009)). Although the coefficients are not always significant, the results suggest that blockholders and activists tend to target underperforming firms where change is more likely. Poor stock returns, low accounting profits and high write-offs are more often associated with attack and activists are somewhat more likely to avoid firms where the management would be difficult to oust. We find that the types of firms that are targeted by activists and institutional blockholders are similar in both halves of the sample period. However, illiquidity only affects the probability of known activists holding a block in the early period, consistent with the idea that liquidity is less of a factor over time. 20 E. Multivariate analysis of turnover Univariate results show that both exit and institutional activism, measured by news of pressure on the CEO and institutional block holdings, are higher in cases of forced CEO turnover. Since these are not mutually exclusive factors, we next estimate logit regressions to determine the relative importance of each in the turnover decision. We again split the sample into two periods to test whether the roles of either exit or activism have changed over time. In Table 8 we compare the probability of forced CEO turnover and voluntary turnover controlling for standard measures of performance (ROA, past year stock returns) and governance (family firms, duality, board ownership, and fraction of outside board members). 21 Model (1) measures voting with their feet with the gross sales of institutional holdings and activism with the news of activism variable. 22 In unreported results we also measure exit by the net change in 20 In untabulated results, we find that average illiquidity in our sample declines from the early period to the late one. 21 We also examine whether voting with their feet or activism has influenced turnover-performance sensitivity over time by interacting previous year stock returns with exit and news of activism. In results not shown, we find insignificant coefficients on the interaction variables in both time periods. 22 In results not shown (for brevity) we also separately control for the concentration of blockholders (all institutional investors) and the change in the number of blockholders (all institutional investors). Ownership concentration is 21

24 institutional holdings, the net change by size of holdings (i.e., small, intermediate, and block), and the percentage of institutions that sell their t=-4 holding. Results are similar or weaker in both time periods for each of the three specifications. 23 Panels A and B present the same models for the early and later periods, respectively. Panel A shows that exit is more important in the earlier half of the sample, as the magnitude of the coefficient on gross selling is sharply higher before 1995 and it is only significant in the first half of the sample. In contrast, news of activism is significant in both time periods. The slightly higher coefficient in the later time period is suggestive of a stronger role for activism as time goes on. Control variables are consistent with the results in Table 6, as they show that CEOs are more likely to be forced out when performance is poor and the management is not entrenched. The coefficient on the duality variable is much larger in magnitude after 1994, supporting Kaplan and Minton s (2010) conclusion that governance has become more important in recent years. In models (2) and (3) we include variables for the presence of institutional blockholders. The arrival of a new blockholder in the 13-F data (model (2)) is not significant in either time period, and its coefficient is even smaller in the second half, likely reflecting the increasingly large number of institutions that hold blocks merely as a byproduct of their demand for shares. In models (3) through (6) we include the presence of an institutional blockholder at t=0 (identified in 13-F or 13-D filings) and find that such investors make turnover more likely, consistent with the results in Denis, Denis, and Sarin (1997). However, the variable loses significance (and flips sign) in the later period. Given the finding in Table 1 that the fraction of firms with an measured by the Herfendahl index of holdings and the change in the number of holders is scaled by the number of holders at t=-4. Our main findings are robust to the inclusion of these controls. 23 However, we find that the net change in institutional holdings is significant in the period. 22

25 institutional blockholder in this time period is at least two thirds and as high as 85%, the variable is unlikely to be useful in distinguishing the degree of monitoring going forward. In model (4) we restrict the cases of activism identified in the press to those involving 13-F institutions. The results are similar regardless of which measure of activism news is used. Model (5) replaces news of activism involving 13-F filers with the variable known activists holding a block. While the latter has the correct sign, its impact is not significant. Finally, in model (6) we consider three additional variables to more precisely determine the impact of performance on turnover. While we do not find a nonlinear effect of performance, the distress and write-off variables help explain turnover. Distress is only significant in the earlier period, but the write-off variable is significant in both periods. Note that the activism variable remains significant even with the additional controls for performance. 24 This last result is important when considering whether activism causes turnover. It could be that activists target underperformers but boards ignore the activists and simply focus on performance when determining whether to oust the CEO. If so, we would incorrectly attribute causality to the activism variable when none exists. We note that the significance of activism even with additional controls for performance suggests that activism does cause turnover. We consider this issue further when we analyze the set of matching firms. We next estimate a logit where the dependent variable is one for forced CEO turnover firms and zero for their matches, shown in Table The specifications in Panels A and B report 24 We also add year fixed effects to each of the models in Tables 8 and 9 and find that the results are qualitatively similar. 25 We do not estimate a multinomial logit where the dependent variable includes all three outcomes (forced CEO turnover, voluntary turnover, and matched firms) because of econometric problems. Specifically, such a multinomial logit would not yield unbiased coefficients because the matched sample is created based on the outcome of the dependent variable (choice-based sampling) and the number of matched observations is determined by the number of forced CEO turnover cases (and hence is not random). Instead, we deal with these two statistical issues in the matched sample by estimating a conditional logit, where the intercept is not estimated in order to avoid estimation error from the non-random sample size. 23