Coordination Costs, Institutional Investors, and. Firm Value

Transcription

1 Coordination Costs, Institutional Investors, and Firm Value Jiekun Huang First version: October 2011 This version: September 2012 Huang is from Department of Finance, National University of Singapore, phone: , fax: , I thank Yakov Amihud, Bernard Black, Paul Burik, Mark Chen, Abe de Jong, Yongheng Deng, Bing Liang, Alexander Ljungqvist, Angie Low, Roni Michaely, Vikram Nanda, Je rey Ponti, Qianru Qi, David Reeb, Bernard Yeung, Shan Zhao, Luigi Zingales, as well as conference and seminar participants at the 2012 European Finance Association annual meetings, the 2012 Financial Intermediation Research Society annual meetings, Rothschild Caesarea Center 9th Annual Academic Conference, the 2012 CGIO Academic Conference, the 6th Singapore International Conference on Finance Junior Faculty Workshop, Fudan University, National University of Singapore, Shanghai Advanced Institute of Finance (SAIF), and University of Adelaide for helpful comments and suggestions. Financial support from CGIO research grant award (N ) is gratefully acknowledged. I retain responsibility for any remaining errors.

2 Coordination Costs, Institutional Investors, and Firm Value Abstract Coordination costs among institutional investors have an important impact on corporate governance and rm value. We use two measures to proxy for coordination costs, one based on the geographic distance among institutional shareholders and the other based on the correlation in their portfolio allocation decisions. We nd that, after controlling for other factors, coordination costs are negatively associated with rm value as proxied by industry-adjusted Tobin s q. We exploit three exogenous shocks, namely, mergers of asset management rms, the 1992 proxy reform, and decimalization in 2001, and nd evidence consistent with a causal e ect of coordination costs on rm value. Furthermore, we show that the ease of coordination among institutions is associated with fewer anti-takeover provisions adopted by the rm, higher equity-based pay for CEOs, and improved CEO turnover-performance sensitivities. Overall, these ndings suggest that the ease of coordination improves rm value by enhancing the governance role of institutional investors. JEL Classification: G23, G32, G34 Keywords: Coordination costs; Institutional investors; Corporate governance; Firm value

3 1 Introduction While institutional investors collectively hold the majority of the U.S. equity market, their in uence on corporate governance and corporate performance remains unclear. Theoretical work suggests that institutional investors, as large shareholders, can discipline corporate managers through active monitoring and intervention (Shleifer and Vishny, 1986; Maug, 1998; Kahn and Winton, 1998) as well as through the threat of exit (Admati and P eiderer, 2009; Edmans, 2009). Empirical research, however, suggests that there is little evidence of improvement in the long-term rm value from institutional monitoring. 1 One major limitation of institutional monitoring is the free-rider problem, because institutional equity ownership is widely dispersed. As Figure 1 shows, the median value of an institution s equity holdings in a rm as a fraction of the rm s outstanding shares is 0:07% during 1980 to 2009 and decreases over the years. The di used institutional ownership structure suggests that, in the absence of coordination, the classical free-rider problem can prevail (Grossman and Hart, 1980). It has been recognized that institutions can play a more e ective corporate governance role through coordinated activities (see, e.g., Black, 1992). Recent survey evidence of McCahery, Sautner, and Starks (2010) shows that 59% of institutional investment managers consider coordinating their actions in disciplining corporate management. Of great importance, and so far largely unexplored, is the cost of coordinating a group of institutional investors, which includes information production costs (e.g., to identify trustworthy and cooperative peers), communication and other costs incurred to reach an agreement, as well as costs associated with monitoring and enforcement of the agreement. In this paper, we examine the impact of coordination costs on the role of institutional investors in improving corporate governance and rm value. We hypothesize that a low coordination cost improves rm value by facilitating a stronger governance role provided by institutional investors. On the one hand, coordination costs can impact the e ectiveness of institutional monitoring and intervention. Although it is not cost-e cient for a small shareholder to monitor managers because of the free-rider problem, low coordination costs enable dispersed institutional shareholders to conduct coordinated monitoring activities and mitigate managerial agency costs. For instance, 1 See, e.g., Gillan and Starks (2007) and Yermack (2010) for recent surveys of the literature. 1

4 institutions can form a shareholder coalition to sponsor proxy proposals to e ect changes in corporate governance (Gillan and Starks, 2000; Del Guercio, Seery, and Woidtke, 2008) as well as to engage in direct negotiation with corporate management seeking governance changes (Becht, Franks, Mayer, and Rossi, 2009). This predicts that a low coordination cost should enhance the monitoring role of institutions and lead to higher rm valuation. On the other hand, the ease of coordination can also intensify the threat of exit. Admati and P eiderer (2009) argue that the threat of exit by a large shareholder can have a disciplinary impact if the shareholder possesses private information about corporate managers extraction of private bene ts (and hence her trading can have an impact on the stock price on which managerial compensation is based). In the absence of coordination, institutions may be limited in using the threat of exit as a disciplinary device, because, as mentioned above, the individual equity stake by an institution is tiny and because information production is costly. Thus a low coordination cost enables institutional investors to share information and to conduct coordinated selling, which can strengthen the disciplinary e ect of the threat of exit. This again predicts that the ease of coordination should be related to improved corporate governance and rm value. Coordination costs are hard to observe or quantify. In this paper, we use data on institutional shareholders and construct two measures to capture the ease with which they conduct coordinated actions (in monitoring and selling). The rst measure is the geographic distance among a rm s institutional shareholders. If a rm s institutional shareholders are geographically close to one another, they are more likely to communicate and thus coordinate their actions in major corporate events such as takeovers. This arises because geographic proximity facilitates word-of-mouth communication among professional money managers (Hong, Kubik, and Stein, 2005) and because geographic proximity can promote cooperation among agents through repeated interaction and mutual trust (Leamer and Storper, 2001). The second measure is the correlation in portfolio allocation decisions among institutional shareholders. Institutional asset managers with similar portfolio allocations are likely to form strong ties among themselves because of the homophily e ect. A high portfolio correlation can be the consequences of social connections as well, because institutions in the same social networks have access to the same information sources (Cohen, Frazzini, and Malloy, 2008) and because they engage in direct communication with one another (Hong, Kubik, 2

5 and Stein, 2005; Stein, 2008). Thus, a high portfolio correlation indicates greater homophily and stronger social ties among institutional asset managers, which should facilitate coordination. Using a comprehensive sample of stocks from 1980 to 2009, we nd that rm valuation (proxied by an industry-adjusted Tobin s q) decreases with coordination costs among institutional shareholders. The economic magnitude of this e ect is meaningful: Moving from the 10 th percentile in the geographic distance (portfolio correlation) among institutional shareholders to the 90 th percentile decreases (increases) the industry-adjusted Tobin s q by 0:065 (0:198), as compared to the median Tobin s q of 1:29. The e ect is robust to controls for other institutional shareholder characteristics (such as aggregate institutional ownership, institutional ownership concentration, investment horizons of institutional shareholders, and the distance between institutional shareholders and the rm), rm size, growth opportunities, diversi cation, nancial performance, managerial ownership, and rm-speci c e ects. Furthermore, this e ect is driven mainly by independent institutions and non-transient institutions, both of which are more likely to play an active governance role. These results are consistent with the hypothesis that the ease of coordination among institutional shareholders enhances rm value. Institutional investors do not randomly invest in rms, which suggests that institutional ownership structure and hence our coordination cost measures may be endogenous. For instance, institutions that are located nearby to one another may share similar preferences and invest in high-q stocks. This will result in a reverse causality from rm valuation to the geographic clustering of institutional shareholders. To address this potential endogeneity concern, we apply a di erence-in-di erences approach that exploits exogenous shocks to shareholder coordination. The rst exogenous shock we consider is mergers of asset management rms, which result in selling of large positions that the target institution holds and hence an increase in coordination costs for these stocks (treatment sample). The mergers are reasonably exogenous to the treatment stocks, because these mergers are generally unrelated to the fundamentals of the treatment stocks and because the exit decision of the merged institution from the treatment stocks post-merger does not seem to be driven by the fundamentals. Using a di erence-in-di erences approach, we nd that the industry-adjusted Tobin s q of treatment stocks decreases relative to control stocks following mergers of asset management rms, by 0:075. 3

6 We further show that the increase in coordination costs after the mergers of asset managers is driven primarily by treatment stocks that have a low pre-merger coordination cost, and that the decrease in rm value post-merger is concentrated among these low coordination cost stocks. The second exogenous shock we consider is the proxy reform in 1992, which eases communication and coordination among shareholders (Choi, 2000; Bradley, Brav, Goldstein, and Jiang, 2010). Prior to the reform, shareholders were restricted in their ability to communicate among themselves information about voting decisions, because it would amount to proxy solicitation and was not allowed until the shareholders involved deliver a formal proxy statement to other shareholders. The 1992 proxy reform relaxed this communication restriction such that any shareholder communication not directly seeking the power to vote as proxy for other shareholders would not be considered as proxy solicitation. This predicts that the e ect of our rm-level coordination cost measures on rm value should become stronger post-reform. Consistent with this, we nd that, compared to otherwise similar rms, rms whose shareholders can coordinate with relative ease experience an increase in rm value post-reform. This result suggests that the e ects are driven by explicit communication and coordination, rather than by implicit coordination, such as homogeneous preferences and behavior, among shareholders as in Kandel, Massa, and Simonov (2011). The third exogenous shock we consider is decimalization in 2001, which signi cantly reduces trading costs for stocks (Bessembinder, 2003; Fang, Noe, and Tice, 2009). Because institutional investors are sensitive to trading costs (e.g., Wermers, 2000; Edelen, Evans, and Kadlec, 2007), a decrease in trading costs can enable coordinated institutions to more e ectively use the threat of exit as a disciplining device. Consistent with this, we nd evidence that the impact of coordination costs on rm value becomes signi cantly stronger after decimalization. Furthermore, consistent with the notion that trading costs drop more signi cantly for low-priced stocks post-decimalization, a triple-di erences analysis shows that the e ect of decimalization on the e ectiveness of the threat of exit by coordinated institutions is stronger for rms with a low stock price. Taken together, the results of the di erence-in-di erences approach provide evidence that coordination costs have a causal impact on rm value. We then consider how the ease of coordination among institutional shareholders might add value by focusing on corporate governance mechanisms and governance outcomes. We nd that 4

7 rms with low coordination costs are associated with better corporate governance, as proxied by the number of anti-takeover provisions (i.e., the G-index and the E-index). We also show that low coordination costs are associated with higher CEO equity-based pay and improved CEO turnoverperformance sensitivities. These results strengthen our interpretation that the ease of coordination enhances the role of institutional investors in corporate governance. This paper is related to two strands of empirical literature, of which the rst is the literature on institutional monitoring. A number of studies suggest that institutional investors in uence corporate policies through costly monitoring or intervention (see, e.g., Hartzell and Starks, 2003; Chen, Harford, and Li, 2007; Gillan and Starks, 2000; Del Guercio, Seery, and Woidtke, 2008) as well as through the threat of exit (see, e.g., Parrino, Sias, and Starks, 2003). Much of the literature, however, implicitly treats institutional investors (or certain types of institutions) as a monolithic entity. Our paper is the rst in the literature to study the impact of coordination costs on the role of institutional investors in improving corporate governance and rm value. 2 This paper also connects to the growing body of nance literature on geography. Hong, Kubik, and Stein (2005) show that mutual fund managers located close by make similar portfolio decisions, suggesting that geographic proximity facilitates communication among professional money managers. Coval and Moskowitz (1999, 2001) nd that mutual fund managers exhibit a strong bias towards locally headquartered rms and deliver superior returns on their local investments than distant investments, indicating an information transfer from rm managers to mutual fund managers located nearby. Gaspar and Massa (2007) show that mutual funds located near their portfolio companies play the role of informed monitors. Kang and Kim (2008) nd that, in partial block acquisitions, acquirer rms tend to pursue geographically proximate targets and play a strong monitoring role in such targets post-acquisition. Our paper adds to the literature by showing that the geographic proximity among shareholders matters by a ecting the governance role of shareholders. The rest of the paper is organized as follows. Section 2 describes the data and summary 2 Two papers related to ours explore the idea of coordination costs in other settings. Bradley, Brav, Goldstein, and Jiang (2010) show that the ease of communication among investors of closed-end funds facilitates activist arbitrage. Kandel, Massa, and Simonov (2011) contend that implicit coordination, proxied using age similarity, among small individual shareholders can play a corporate governance role. However, neither of these papers examines the coordination cost among institutional investors, which are the dominant shareholders in the U.S. 5

8 statistics. Section 3 presents the empirical results on the e ect of coordination costs on rm value. Section 4 presents the empirical results on the impact of coordination costs on corporate governance mechanisms and outcomes. Section 5 concludes. 2 Data and Summary Statistics We retrieve the data for our study from the Center for Research in Stock Prices (CRSP) database, COMPUSTAT, and Thomson Reuters 13F institutional ownership database. Our sample includes all common stocks listed on the NYSE, AMEX or NASDAQ during the period from 1980 to 2009 for which su cient information is available in the three databases. There are 104; 204 rm-year observations in the sample. We construct two measures to capture the coordination cost among institutional shareholders of a rm. The rst measure is based on the geographic concentration of institutional ownership. The premise is that asset managers located close to one another are more likely to come into direct contact (Hong, Kubik, and Stein, 2005) and hence to take coordinated actions. Moreover, geographic proximity can promote cooperation among institutional asset managers by facilitating repeated interaction and cultivating trust (Leamer and Storper, 2001). To construct the geographic distance measure, we rst manually identify the location (zip code) of institutional investors using the Nelson s Directory of Investment Managers and by searching the lings by institutional investors on the SEC Edgar website. We then calculate, for each rmquarter, the weighted-average geographic distance among institutional shareholders of the rm. In particular, for each institutional shareholder in the rm, we calculate the geographic distance between the institution and all institutions in the rm, weighted by their respective fractional holdings in the rm. This measure captures the average distance between an institutional shareholder and its peers. We then calculate a weighted-average of these distances across all institutional shareholders of the rm, again weighted by their fractional holdings. This weighting scheme ensures that institutions that are likely to be more in uential, i.e., those with larger holdings in the rm, receive greater weights in determining the distance among shareholders. Last, we take a simple average of the geographic distance among shareholders for each rm over the four quarters in a 6

9 year. Speci cally, Geographic distance among institutional shareholders for rm c = X 4 X w c;i;q X w c;j;q l Dist ij;q l A5, (1) 4 i2s j2s l=1 where S is the set of institutional shareholders in rm c, w c;i;t is the weight of institution i in the total percentage held by institutions in rm c at quarter q, and Dist ij;t is the geographic distance between institutions i and j at quarter q. To reduce the skewness of the variable, we use the logarithm of one plus the geographic distance among institutional shareholders, Log(1 + Shareholder distance), as an explanatory variable in the regressions. The second measure is based on the portfolio correlation among institutional shareholders of the rm. This variable is intended to capture the extent of homophily and social ties among institutional shareholders. A large body of literature on homophily in social networks suggests that individuals tend to build connections with others similar to themselves (see McPherson, Smith-Lovin, and Cook, 2001 for a review of research on homophily in social networks). Thus, institutional asset managers that share similar views about certain stocks, i.e., a high correlation in their portfolio allocations, are more likely to exhibit homophily and form strong ties among themselves, which should facilitate coordination. Moreover, a high portfolio correlation can be the consequences of social ties, because institutions in the same social networks have access to the same information sources, e.g., through shared educational ties (Cohen, Frazzini, and Malloy, 2008) and geographic proximity (Coval and Moskowitz, 2001), and because they engage in direct communication with one another (Hong, Kubik, and Stein, 2005; Stein, 2008). To the extent that the portfolio correlation measure captures homophily and social connectedness among institutional asset managers, it should be negatively related to shareholder coordination costs. To construct the portfolio correlation measure, we retrieve the entire portfolio holdings of all institutional shareholders of our sample rms in each quarter. For each pair of institutional shareholders, we calculate the correlation coe cient of the excess portfolio weights on common holdings, i.e., stocks that are held by both institutions. 3 The excess portfolio weights are calculated as 3 If two institutions have less than ve common holdings, we set the correlation to zero. The results are robust to 7

10 the actual portfolio weight assigned to a stock relative to the weight of the stock in the market portfolio. We use the excess portfolio weights, rather than the actual weights, to focus on active portfolio allocation decisions of institutional asset managers. Similar to the construction of the geographic distance variable, we rst calculate, for each institutional shareholder, the portfolio correlation between the institution and all institutions in the rm, weighted by their respective fractional holdings. We then calculate the weighted-average of these correlations across all institutional shareholders, again weighted by each institution s fractional holdings in the rm. We take a simple average of the institutional portfolio correlation for the stock over four quarters in a year. Speci cally, Portfolio correlation among institutional shareholders for rm c = X 4 X w c;i;q X w c;j;q l Corr ij;q l A5, (2) 4 i2s j2s l=1 where S is the set of institutional shareholders in rm c, w c;i;t is the weight of institution i in the total percentage held by institutions in rm c at quarter q, and Corr ij;t is the correlation coe cient of the excess portfolio weight (measured as the actual weight relative to the weight in the market portfolio) allocated to overlapping holdings between institutions i and j at quarter q. Panel A of Table 1 presents summary statistics for the two measures of shareholder coordination costs for all sample rms. The average geographic distance among institutional shareholders is 831:7 miles. 4 The average portfolio correlation among institutional shareholders is 0:29. Both of the coordination cost measures exhibit a fair degree of cross-sectional variation across rms. Panel A of Table 1 also presents summary statistics for other shareholder characteristics and rm characteristics. In particular, since institutions located close to their portfolio companies are likely to play a monitoring role (Gaspar and Massa, 2007; Chhaochharia, Kumar, and Niessen-Ruenzi, 2011), we calculate the weighted-average geographic distance between institutional shareholders using a di erent cuto or setting it to missing. 4 The magnitude of this distance appears large. However, it should be noted that it is measured across all institutions that hold shares in the rm. Consider a hypothetical rm with 50 institutional shareholders from the 50 states in the U.S. (assuming they are located in the state capitals), each holding 2% of the rm s outstanding shares. The shareholder distance for the rm would be 1; 822 miles. Thus, the seemingly large distance among institutions is driven mainly by the fact that the U.S. is geographically large. 8

11 and rms, weighted by institutions fractional holdings. The average distance between institutional investors and the rm is 943:8 miles. Institutional shareholders, in aggregate, own 33% of the outstanding shares of the average rm. Following Hartzell and Starks (2003), we calculate institutional ownership concentration as a Her ndahl Index of institutional ownership concentration based on the percentages of institutional holdings by all 13F institutions. The average institutional ownership concentration for the rms is 0:012. Following Gaspar, Massa, and Matos (2005), we calculate shareholder turnover of a rm as the weighted-average of the average total portfolio turnover rate of the rm s institutional shareholders. The average shareholder turnover rate for the rms is 0:27. We calculate Tobin s q as the ratio of market value of assets to book value of assets, where market value of assets is measured as the market value of common equity plus the book value of preferred stock (carrying value) plus the book value of long-term debt minus deferred taxes and investment tax credit. 5 The industry-adjusted Tobin s q is calculated as the di erence between the rm s Tobin s q and its industry median using the two-digit SIC code. The mean and median of Tobin s q are 1:82 and 1:29, respectively, and those of industry-adjusted Tobin s q are 0:36 and 0:02, respectively. Panel B of Table 1 presents a correlation matrix of the main variables. The two coordination cost proxies are highly negatively correlated, with a correlation coe cient of 0:78, suggesting that institutional shareholders located close to one another tend to have correlated portfolio allocations. This is consistent with the word-of-mouth e ect documented by Hong, Kubik, and Stein (2005). Furthermore, industry-adjusted Tobin s q is signi cantly negatively correlated with the geographic distance measure and signi cantly positively correlated with the portfolio correlation measure. These results give a preliminary indication that the ease of coordination may improve rm value. In addition, we retrieve various corporate governance-related variables, such as managerial ownership, board characteristics, and executive compensation, from RiskMetrics and ExecComp. We report the summary statistics as well as the correlation matrix for these variables in Table 1. [Insert Table 1 about here] 5 A more sophisticated approach to estimating Tobin s q is to calculate the replacement cost of assets (Lindenberg and Ross, 1981). We use the simple approach, instead of the more sophisticated one because the latter requires arbitrary assumptions about depreciation and in ation rates and because the two approaches deliver highly correlated estimates of Tobin s q (Villalonga and Amit, 2006). 9

12 3 Empirical Results: Coordination Costs and Firm Value In this section, we rst examine the relation between coordination costs and rm value using rm- xed e ects regressions. We then address endogeneity concerns by using a di erence-in-di erences approach to gauge the impact of a series of exogenous shocks on rm value. 3.1 Firm- xed E ects Regressions To examine the e ects of coordination costs on rm valuation, we run rm- xed e ects regressions of industry-adjusted Tobin s q on our coordination costs proxies and control variables. We lag all our explanatory variables by one year to mitigate any confounding e ects due to contemporaneous measurement. Speci cally, q j;t = + j + Coordination Costs j;t 1 + X i x i;j;t 1 + " j;t, (3) where q j;t is rm j s industry-adjusted Tobin s q at the end of year t, j is rm- xed e ects, Coordination Costs j;t 1 is one of the two measures of coordination costs for rm j in year t 1, and x i;j;t 1 includes standard control variables for Tobin s q, such as rm size, pro tability, capital expenditure, leverage, R&D expenses, and year dummies, and other institutional ownership characteristics, including institutional ownership, institutional ownership concentration, investment horizons of institutional shareholders, and the distance between institutional shareholders and the rm. We include rm xed e ects to capture unobserved rm-speci c time-invariant factors that in uence rm value. We cluster the standard errors at the rm level (Petersen, 2009). Since, as Panel B of Table 1 shows, the two coordination cost proxies are highly negatively correlated, we include them in the regressions one at a time. The regression results, reported in Panel A of Table 2, indicate that the ease of coordination has a positive e ect on rm value. The economic magnitude of this e ect is meaningful: Based on the full speci cation (i.e., the last two columns of Table 2, Panel A), moving from the 10 th percentile in the geographic distance (portfolio correlation) among institutional shareholders to the 90 th percentile decreases (increases) the industry-adjusted Tobin s q by 0:065 (0:198). These 10

13 numbers translate into a 5% to 15% di erence in rm value, based on the median Tobin s q of 1:29. Panel A of Table 2 also reveals a number of other interesting ndings. Institutional ownership is positively associated with industry-adjusted Tobin q, which is consistent with the ndings of McConnell and Servaes (1990). Institutional ownership concentration, however, is negatively associated with industry-adjusted Tobin q, possibly because high-q rms attract a more dispersed shareholder base. The distance between the rm and institutional shareholders does not signi - cantly a ect rm value. Shareholder turnover is positively related to industry-adjusted Tobin q, which may be because trading by short-term (high turnover) institutional investors improves stock price e ciency and hence rm valuation. Furthermore, high-q rms are generally smaller in asset size, have more debt, higher R&D intensities, and higher capital expenditures. These results are broadly consistent with the literature (see, e.g., McConnell and Servaes, 1990; Bebchuk, Cohen, and Ferrell, 2009). Since institutions may di er in their incentives and abilities to play a governance role, we partition institutional investors into groups in two ways. First, we classify institutions into independent institutions and grey institutions following Chen, Harford, and Li (2007). Independent institutions include investment companies, independent investment advisors, and public pension funds, which do not have business relationships with their portfolio companies and hence are more likely to engage in active monitoring. Grey institutions include insurance companies, banks, and private pension funds, which are less likely to play a governance role because of their business ties with the rms they invest in. Second, we divide institutions into transient and non-transient categories following Bushee (1998). Non-transient institutions are dedicated and quasi-indexer based on Bushee s de nition, which are likely to be more e ective monitors. We expect that the e ect of coordination costs on rm value should be driven mainly by independent institutions and non-transient institutions. We reconstruct the coordination cost measures separately for each category of institutions. We replace the aggregate coordination cost measures in Eq. (3) with separate coordination cost measures for each category of institutions, and re-estimate the regressions. Panel B of Table 2 reports the results. 6 Consistent with our expectation, the negative e ects of coordination costs on 6 Note that the sample size for these regressions is slightly smaller, because, to construct the coordination costs measures for a category of institutions, we require that a rm have at least one institution in that category. 11

14 rm value are driven mainly by independent institutions and by non-transient institutions. We further add control variables related to managerial ownership and board structure in our rm- xed e ects regressions to examine whether the negative relation between Tobin s q and coordination costs are driven by these factors. In particular, we include managerial ownership, managerial ownership squared, board size, board independence (the fraction of independent outside directors on the board), and CEO/Chairman duality. Because only about 10% of the sample has non-missing information on these variables, we create two indicator variables: one equal to 1 for observations with missing information on managerial ownership and the other equal to 1 for observations with missing information on board characteristics. The results, reported in the last two columns of Table 2, Panel B, suggest that the negative e ects of coordination costs on rm value are robust to adding these controls. The coe cient on managerial ownership is positive and signi cant, whereas that on managerial ownership squared is negative and signi cant. These results are consistent with an inverted U-shaped relation between Tobin s q and managerial ownership (e.g., Morck, Shleifer, and Vishny, 1988). [Insert Table 2 about here] We then conduct a series of robustness checks on the relation between coordination costs and rm value. First, we control for the e ect of local institutions on rm value by excluding these institutions in the construction of the coordination costs measures. Investors located close to their investments are likely to have an informational advantage (Coval and Moskowitz, 2001; Baik, Kang, and Kim, 2010) as well as to provide a strong monitoring role (Gaspar and Massa, 2007; Kang and Kim, 2008). To test whether the ndings are driven by local institutional shareholders, we reconstruct the two measures of coordination costs by excluding institutional investors located within 100 kilometers of the rm s headquarter and re-estimate Eq. (3). The results, reported in the rst two columns of Table 3, show that the e ects of coordination costs on rm value are qualitatively unchanged, suggesting that the results are not driven by local institutions. Second, we exclude cities with highly concentrated institutional investors in the construction of the coordination costs measures. Institutional asset management is highly geographically concentrated. One concern is that the ndings are driven by a few cities with a high concentration of 12

15 institutional investors. We thus repeat the analysis by excluding these cities. We use metropolitan statistical areas (MSAs) to de ne the location of institutional asset managers. For each MSA and each quarter, we calculate the total dollar value of equity holdings that are managed by institutions residing in that MSA. New York and Boston dominate the institutional asset management landscape, representing 19:3% and 16:0% of the total dollar holdings by all 13F institutions, respectively. We then construct the two measures of shareholder coordination costs by excluding the two MSAs and re-estimate Eq. (3). Columns 3 and 4 of Table 3 show that the results are again qualitatively unchanged. This nding suggests that the results are not driven by the two extreme cities per se. Third, we exclude foreign institutions. The fraction of the total institutional equity holdings in the U.S. managed by foreign institutions has increased signi cantly from 3% in 1980 to 15% in On the one hand, the presence of foreign institutions can increase the geographic distance among shareholders and, to the extent that they have di erent investment objectives from domestic institutions, decrease the portfolio correlation among the institutional shareholders of the rms foreign institutions invest in. On the other hand, foreign institutions might be less e ective in monitoring management than domestic institutions due to geographic distance (Kang and Kim, 2008). To test whether the results are driven by foreign institutions, we reconstruct the two measures of shareholder coordination costs by excluding foreign institutions and re-estimate Eq. (3). The results, reported in columns 5 and 6 of Table 3, are essentially unchanged compared to the baseline results reported in Table 2, Panel A, which suggests that foreign institutions do not drive the results. Fourth, we use OLS regressions with lagged dependent variables. We estimate OLS regressions adding lagged industry-adjusted Tobin s q as a control variable. Columns 7 and 8 of Table 3 report the results. As expected, the coe cient on the lagged industry-adjusted Tobin s q is positive and highly signi cant. The coe cients on our key variables, i.e., the coordination costs variables, remain signi cant and in the predicted directions. Furthermore, our results are robust to adding two or three lags of industry-adjusted Tobin s q in the OLS speci cation. Last, we use Arellano-Bond dynamic panel regressions. Since rm value is persistent, we add lagged industry-adjusted Tobin s q to Eq. (3) and reestimate the regressions using Arellano-Bond 13

16 (1991) dynamic panel estimator. The results, reported in the last two columns of Table 3, are qualitatively similar to our baseline results. [Insert Table 3 about here] 3.2 Addressing Endogeneity Concerns The panel regression results presented above may raise endogeneity concerns, because institutional investors do not invest randomly. For instance, institutions that are located nearby to one another may share similar preferences and invest in high-q stocks. This will result in a reverse causality from rm valuation to the geographic clustering of institutional shareholders. To address this potential endogeneity e ect, we use a di erence-in-di erences approach that exploits exogenous shocks to shareholder coordination. We consider three exogenous shocks. The rst is the mergers of asset management companies which leads to sales of large holdings by the target institution, and the other two are regulatory changes that can a ect shareholder coordination. We use a di erence-in-di erences estimator to compare the change in rm value for the treatment stocks after the shocks to the change for a control group of stocks that are not a ected by the shocks Mergers of Asset Management Firms We use mergers of asset management rms as an exogenous shock that increases coordination costs for stocks in which the target institution is a signi cant shareholder pre-merger. When two institutions merge, large holdings by the target institution typically get dissolved, resulting in a more dispersed ownership structure and hence an increased coordination cost among the shareholders of these stocks, which constitute the treatment group. The change in coordination costs induced by mergers of institutions is reasonably exogenous to the treatment stocks because of two reasons. First, these mergers are generally not driven by the performance of the underlying assets of the target institution but rather by strategic considerations of the acquiring institution such as to realign product o erings and to achieve economies of scale (Jayaraman, Khorana, and Nelling, 2002). Second, as we show below, post-merger sales of the target institution s large holdings are 14

17 driven primarily by the merged institution s portfolio considerations and are largely unrelated to the fundamentals of the stocks being sold. We use stocks in which the target institution owns more than 1% of the outstanding shares pre-merger as our treatment group. We face a trade-o in choosing the cut-o point to classify treatment stocks. On the one hand, choosing a too low threshold value can add noise to the tests because of the inclusion of small shareholders, whose trading decisions would have little impact on the shareholder base and the coordination costs. On the other hand, choosing a too high threshold value can sigi cantly reduce the sample size for our treatment group. We choose 1% as the cut-o point to focus on economically relevant shareholders and to obtain a reasonable sample size. The 1% holding is at the 88 th percentile of individual institutional ownership over the entire sample period. We retrieve mergers among nancial institutions from SDC s Mergers and Acquisitions database and manually match the acquirers and the targets to 13F les by name. We require that (1) the merger is announced during the period between 1980 and 2009; (2) the merger is completed within one year after the initial announcement; (3) the target institution stops ling 13F forms within one year after the completion of the deal; (4) the target institution has at least one equity position that exceeds 1% of the stock s outstanding shares. After applying the lters, we identify 127 mergers between asset management rms. We then partition the sample of stocks into treatment and control groups. The treatment group consists of stocks in which the target institution is a signi cant shareholder before the merger, i.e., the target institution s holdings in the stock premerger exceed 1% of the outstanding shares of the stock. We identify 5; 292 stocks as the treatment group. Panel A of Table 4 reports the number of asset management rm mergers and the treatment stocks by year in which the mergers are completed. A summary of the top 20 mergers ranked by the number of treatment stocks is provided in Panel B of the same table. As is seen in the table, many of the large mergers between institutional asset management rms are the result of bank mergers, such as the merger between Bank of America and FleetBoston Financial in 2004 and that between Chase Manhattan and J.P. Morgan Chase in [Insert Table 4 about here] 15

18 We gauge the impact of the merger on the merged institution s holdings of treatment stocks by exploiting variation in trading across institutions in the same stocks. Speci cally, we identify, for each pair of merging institutions in a treatment stock, a matched institution as the institution that remains independent, holds more than 1% of the treatment stock s outstanding shares, and has the closest equity portfolio size to the combined portfolio size of the target and the acquirer institution pre-merger. To the extent that the merged and the matched institutions respond in the same way to public information about the fundamentals of the treatment stocks, this matching institution approach e ectively eliminates the possibility that sales of the treatment stocks by merged institutions are driven by public news. 7 Speci cally, we compare the changes in the fractional ownership of the treatment stocks after the merger between merged and matched institutions. We choose the quarter immediately before the merger announcement as the pre-merger period, and the rst quarter-end one year after the merger completion as the post-merger period. Panel A of Table 5 reports the results. Both merged and matched institutions sell their holdings of treatment stocks post-merger, but merged institutions sell more aggressively. The di erence-in-di erences estimate shows that post-merger sales of treatment stocks by merged institutions are 0:67 percentage points (or 68%) more than that by matched institutions. This result provides evidence that the merger leads to a more dispersed shareholder base and hence may increase the coordination costs. Our identi cation strategy rests on the premise that post-merger sales of the treatment stocks by the merged institutions are unrelated to the fundamentals of the stocks. While the matching institution approach can rule out the possibility that public information about the fundamentals drives the merged institutions selling decision, it remains possible that the merged institutions sell the treatment stocks because of their superior private information about the fundamentals of the treatment stocks. We test this by evaluating the merged institutions trade performance postmerger and compare it with that of the matched institutions. In particular, we infer trades in the treatment stocks made by the merged and the matched institutions in each of the four quarters post-merger from the quarterly holdings reports. We regress subsequent abnormal returns of the treatment stocks on trading by the merged institutions, trading by the matched institutions, lagged holdings by the merged and the matched institutions, and control variables that can in uence 7 If anything, this matching strategy may bias against nding signi cant di erences in the trading of treatment stocks post-merger between merged and matched institutions, because selling of the treatment stocks by the merged institution and the ensuing decrease in the monitoring intensity may prompt other institutions to sell. 16

19 institutional preferences and stock returns (Gompers and Metrick, 2001; Baik, Kang, and Kim, 2010). If the merged institutions do not possess private information, we expect trading by the merged institutions to be uninformative in predicting subsequent stock performance. 8 The results, reported in Panel B of Table 5, show that the merged institutions trading in the treatment stocks does not positively predict subsequent abnormal returns of the stocks. In contrast, trading by the matched institution seems to be informative, which is consistent with the general nding in the literature (see, e.g., Gompers and Metrick, 2001). We conduct F -tests to examine whether the coe cients on the trading measures across the merged and the matched institutions are identical. The F -tests reject the null at the 5% level. These results provide support for the presumption that post-merger sales of treatment stocks by the merged institution are largely unrelated to the fundamentals and likely driven by portfolio considerations. [Insert Table 5 about here] To gauge the impact of mergers of asset management rms on rm value, we follow Hong and Kacperczyk (2010) to construct the control group and the benchmark-adjusted di erence-indi erences estimator. Speci cally, the control group consists all the remaining stocks that are matched to the treatment stocks based on market capitalization, book-to-market ratio, past returns, and institutional ownership. In particular, we rst sort stocks into terciles based on market capitalization. Within each size tercile, we then sort stocks into terciles based on book-to-market ratio. Within each size and book-to-market tercile, we further sort stocks into terciles based on past 12 month returns. Last, within each of the above portfolios, we further sort stocks into terciles based on institutional ownership. This sequential sorting results in 108 benchmark portfolios. We then compare the change in rm value for treatment stocks around institution mergers against that for the benchmark portfolios by constructing the following benchmark-adjusted di erence-indi erences estimator: 8 Note that for mergers of asset management rms to have an e ect on coordination costs and hence on rm value, it is not necessary that trading by the merged institutions positively predict subsequent returns because of two reasons. First, to the extent that the market correctly anticipates the e ect of the mergers on the treatment stocks shareholder base and the coordination costs, the decline in rm value can occur even before the merged institutions actually sell the treatment stocks. Second, in the cross-section of treatment sample, stocks that the merged institutions sell more patiently (i.e., in smaller trade sizes) should experience the greatest decline in rm value, because these stocks are likely to experience the greatest increase in shareholder dispersion and hence the coordination costs. These stocks, however, do not necessarily experience the largest magnitude of selling by the merged institutions. 17

20 BDID i = (q T;i;post q T;i;pre ) (q C;i;post q C;i;pre ) where q T;i;post and q T;i;pre are industry-adjusted Tobin s q for treatment stock i post- and premerger, respectively, and q C;i;post and q C;i;pre are the average industry-adjusted Tobin s q for the benchmark portfolios that are matched to stock i post- and pre-merger, respectively. We choose the scal year immediately before the merger as the pre-merger period, and the rst scal year-end at least 12 months after the merger completion as the post-merger period. We take the average of BDID across all treatment stocks to evaluate the average e ect. We expect the average BDID estimator to be negative and signi cant. Table 6 reports the results. Panel A shows that treatment and control groups have the same pre-merger rm value, as measured using industry-adjusted Tobin s q. Firm value drops for both treatment and control groups, possibly due to the trend toward increased industry competition, but rm value drops more signi cantly for treatment stocks. As a result, the di erence-in-di erences tests show that the rm value of treatment stocks decreases relative to control stocks following mergers of asset management rms, by 0:075 (or 5:8% based on the median Tobin s q of 1:29). We conduct similar di erence-in-di erences tests for the coordination costs variables. The results, unreported, suggest that the geographic distance for the treatment sample increases by 3:2% (signi cant at the 1% level), relative to that for the control sample. The result using the portfolio correlation measure is in the predicted direction but insigni cant. A caveat here is that the geographic distance and the portfolio correlation measures capture the coordination costs among institutional shareholders and do not re ect the overall coordination costs, i.e., those among all shareholders. Because the merged institution may sell the treatment stocks to individual investors with whom coordination may be particularly costly, the above estimated magnitude of changes in the coordination costs induced by the mergers should be considered as a lower bound of the true e ects. We conduct further tests by conditioning the di erence-in-di erences tests on pre-merger coordination costs. When a rm has a large, widely dispersed shareholder base, i.e., when the coordination cost among shareholders is high, the exit decision of one institutional shareholder 18

21 would have a relatively muted impact on the shareholder base and hence the coordination costs. In contrast, for rms with a single shareholder or a closely-knit group of shareholders, exit by one shareholder would have a greater impact on the coordination costs provided that the block is broken up and sold in pieces. 9 We test this by partitioning the treatment group into a high and low coordination cost subsamples. We sort the treatment stocks into quartiles based on the pre-merger coordination costs. The low coordination cost subsample consists of treatment stocks that are in the bottom quartile of geographic distance or in the top quartile of portfolio correlation, and the high coordination cost subsample consists of the remaining treatment stocks. Panel B of Table 6 shows that the coordination costs increase signi cantly for low coordination cost group. For instance, the geographic distance among institutional shareholders of the treatment stocks with low pre-merger coordination costs increases relative to control stocks, by 10:2%, and that for the treatment stocks with high pre-merger coordination costs decreases relative to control stocks, by 4:9%. 10 The triple-di erences estimate is highly signi cant. The results for the portfolio correlation variable are similar. Because mergers of asset management rms increase coordination costs more for treatment stocks with a low pre-merger coordination cost, the e ect of the mergers on rm value should be driven mainly by these stocks. The results, reported in the same panel, show that the di erencein-di erences estimate for the low coordination cost group is 0:103 and statistically signi cant, whereas that for the high coordination cost group is 0:042 and insigni cant. The triple-di erences estimate is economically large ( 0:061), but statistically insigni cant because of high standard errors associated with the high coordination cost group. These results provide further evidence that mergers of asset management rms have an e ect on rm value through coordination costs. [Insert Table 6 about here] 9 It is possible that the large shareholder sells the entire block to another shareholder, in which case the coordination costs would not change. Doing so, however, would entail large transaction costs for the selling shareholder (Holthausen, Leftwich, and Mayers, 1987). Thus, it is more likely that the large shareholder breaks up the block and sell it in small trade sizes to minimize price impacts. 10 Note that a decrease in the geographic distance among institutions for the treatment stocks with high pre-merger coordination costs does not necessarily imply a decrease in overall coordination costs for these stocks. Consider the case in which the merged institution sells its holdings of the treatment stocks to dispersed individual investors and all other institutions do not change their holdings in the treatment stocks. The coordination costs for the remaining institutional shareholders can decrease, but those for all shareholders can actually increase because it can be particularly di cult to coordinate with individual investors. 19