An Empirical Two-Sided Matching Model of Acquisitions: Understanding Merger Incentives and Outcomes in the Mutual Fund Industry

Transcription

1 An Empirical Two-Sided Matching Model of Acquisitions: Understanding Merger Incentives and Outcomes in the Mutual Fund Industry Minjung Park, University of Minnesota September 27, 2008 Abstract This paper examines the incentives of acquirers and targets in the merger market. Using data on acquisitions among mutual fund management companies from 1991 through 2004, I estimate a two-sided matching model of the merger market jointly with equations representing merger outcomes. I nd that companies that are prone to misaligned incentives between owners and managers are more acquisitive than others, yet have signi cantly worse post-merger operating performance. I also nd that these acquirers, despite their higher willingness to pay for targets, are not any more likely to match with high-quality targets due to targets incentive to avoid bad organizations. Keywords: M&A, Two-Sided Matching, Gibbs Sampling, Mutual Funds JEL Classi cation: L21, G34, G23, D21 Department of Economics, University of Minnesota, mpark@umn.edu. Tim Bresnahan provided much inspiration and encouragement through this project. I thank Dennis Carlton, Liran Einav, Jon Levin, Eric Zitzewitz, and Takeshi Amemiya for their guidance and insightful comments; Paul Riskind for helpful comments; Jiawei Chen for his help with code; and seminar participants at Stanford University, University of Minnesota, University of Wisconsin, Ross School of Business, University of Illinois, Urbana-Champaign, and Harvard Business School for helpful comments. I gratefully acknowledge support by Ewha University and the Koret Foundation through a grant to the Stanford Institute for Economic Policy Research. All remaining errors are my own. 1

2 1 Introduction A variety of motives may impel rms to pursue an acquisition. Companies may acquire to increase pro ts for shareholders or to obtain private bene ts for managers. This paper studies how merger behavior and outcomes di er between these two types of acquirers and how they are evaluated by targets in the merger market. 1 I address these questions in the context of acquisitions among mutual fund management companies in the US. If some companies acquire at least in part because an acquisition enables their managers to obtain private bene ts, we might expect the companies to possess the following two properties. First, because of the additional private bene ts their managers get, these companies might be more eager to pursue an acquisition than pro t-maximizing companies are, all else equal. Second, because of misaligned incentives between owners and managers at these companies, they might be more likely to have poor merger outcomes. For instance, Meyer, Milgrom, and Roberts (1992) note that mergers create new opportunities to transfer rents among units, which could increase in uence activities and organizational con icts, and Scharfstein and Stein (2000) show that such organizational con icts could be more severe for rms that su er from misaligned incentives. In the mutual fund industry, where big ego, highly skilled human capital is important, organizational con icts and in uence activities are likely to occur during post-merger integration, and they could take a signi cant toll on post-merger performance. These lead to the rst hypothesis of the paper: There are some companies in the mutual fund industry whose acquisition is managerially motivated. They are eager to make an acquisition, but have poor merger outcomes when they actually acquire. When there is a mix of e cient and ine cient acquirers in the merger market, an interesting question is whether targets prefer one type of acquirer to the other. 2 Consider a fund management company that wants to grow in order to market its funds more widely. Suppose the rm has two suitors, one e cient suitor and one ine cient suitor. The ine cient suitor might be more eager to make an acquisition due to the additional private bene ts the manager would obtain and thus have a higher willingness to pay for the target. This price e ect would make the ine cient suitor more attractive. However, the target might expect the ine cient suitor to be poor at managing the merged rm, and this could make the ine cient suitor less attractive. In this industry where human 1 In this paper, I use the term merger and acquisition interchangeably, as is typical in the economics literature. 2 In this paper, e ciency means e ciency from the perspective of rm shareholders. Hence, e cient acquirers are those who pursue an acquisition for pro t maximization. 2

3 capital is very important, one of the assets that an acquirer tries to buy is often the people from the target company. As a result, a high proportion of targets managers stay with the company after the merger, and they will share the success or failure of the merged rm, through earn-out contracts or employment contracts that link pay to rm performance. Hence, targets might have a strong incentive to prefer an acquirer who is expected to perform better post-merger. This leads to the second hypothesis of the paper: The poor post-merger management skill of ine cient acquirers would reduce their attractiveness as a merger partner. I empirically test these hypotheses by estimating a model of takeover market and equations representing the outcomes of mergers. I model the takeover market as a two-sided matching game in which pairings between acquirers and targets arise as part of a stable assignment. Since I want to study how the equilibrium matching in the takeover market is in uenced by both acquirers and targets preferences, a single agent model would not be su cient and an equilibrium model such as matching game is called for. Moreover, a matching game allows me to account for the exclusivity of transaction in the merger market. A target cannot be sold to more than one rm, and a rm cannot acquire more than a certain number of targets in a given period. This exclusivity of transaction makes a matching game a more suitable modeling framework for the merger market than a standard discrete choice model, because only the former accounts for the possibility that rm A chose target X instead of target Y not because rm A prefers X to Y, but rather because target Y had a better merger partner available. As a merger outcome variable, I study post-merger asset growth. Economies of scale in marketing and distributing funds are important in this industry, largely on the demand side, and post-merger asset growth is a good measure of the degree to which the newly merged rm captures such scale economies. I then jointly estimate the matching model and the outcome equations, allowing correlation between the errors of the matching model and the outcome equations. The joint estimation allows me to correct for selection bias in estimating the outcome equations (Sørensen, 2007), as discussed in Section 5. The interdependence among players in the matching model presents numerical di culties for the joint estimation. Bayesian methods using Gibbs sampling and data augmentation provide an elegant solution to this numerical problem. For estimation, I use data on acquisitions among mutual fund management companies in the US from 1991 to The mutual fund industry is an excellent area in which to address the questions posed by this paper. In most other industries, it is di cult to obtain detailed measures 3

4 of companies performance after acquisitions. In the mutual fund industry, however, such data are available, including measures of operating performance such as asset ows and fund returns, enabling me to study post-merger performance. Moreover, this industry likely has both e cient and ine cient acquirers, providing a good setting for my analysis. Since economies of scale in marketing and distribution are important, one would expect that many rms would engage in acquisitions to achieve a more e cient scale of operation. On the other hand, prior research on this industry such as Chevalier and Ellison (1997, 1999) has found an incentive gap between fund managers and fund investors, and one might expect similar agency problems between the shareholders and managers of fund management companies when it comes to acquisition decisions. The fact that the mutual fund industry is human capital intensive also makes it a great place to test my hypotheses for reasons discussed above. The mutual fund industry also merits study simply for its importance: The industry manages over $11 trillion of assets invested by almost 90 million Americans. 3 Since a company s acquisition motive is never directly observed, I employ a proxy for acquirers with managerial motivations in my empirical investigation. My proxy is based on the following two ideas: (1) companies that have performed poorly are likely to be more vulnerable to incentive problems, and (2) public rms are more prone to incentive problems due to the separation of ownership and control. Theories and prior literature that justify the proxy are discussed in Section 2.1. Based on the proxy, I identify a set of companies that are likely to have managerial motives for acquisitions, and empirically test whether their behavior systematically di ers from others in the three dimensions discussed earlier: tendency to pursue an acquisition, merger outcomes, and targets evaluation of them in the merger market. My estimation results provide an interesting picture about the merger market in the mutual fund industry. First, I nd that there are indeed many e cient acquisitions in this industry. For instance, my results indicate that rms are more likely to merge with other companies that use the same channel of distribution (selling funds directly to investors or indirectly through intermediaries), suggesting that many rms engage in an acquisition to bene t from economies of scale in marketing and distributing funds. And they do bene t from such scale economies post-merger, as the outcome equations show that the merged rm attracts larger asset in ows when the two merging rms use 3 Investment Company Institute, Examples of research on the mutual fund industry in the economics literature include Chevalier and Ellison (1997, 1999), Hortaçsu and Syverson (2004), Chen, Hong, Huang, and Kubik (2004), Berk and Green (2004), Mahoney (2004), and Khorana, Tufano, and Wedge (2007). 4

5 the same distribution channel. However, I reject the hypothesis that all acquisitions in this industry are driven by pro t maximization. In particular, I nd that the proxy for acquirers with managerial motivations predicts the following three things. First, all else equal, i.e. holding xed the amount of e ciency gains from mergers, companies with managerial motives have a higher willingness to make an acquisition. Second, they are much worse at achieving asset growth post-merger, even after controlling for di erences in the observed and unobserved characteristics of targets. Third, despite their higher willingness to pay for targets, they are not any more likely to match with high-quality targets due to targets dislike of bad organizations. Using the estimates of the model, I perform counterfactual analysis to investigate the role of targets incentive in the allocation of resources in the merger market. According to the analysis, targets dislike of badly run organizations is a powerful mechanism that discourages ine cient takeovers: Ine cient acquirers would buy many more targets in the merger market without such an incentive of targets. This paper makes three major contributions. First, unlike most prior work in the merger literature which examines either acquirers incentives or targets incentives but not both simultaneously, my paper explicitly addresses the fact that the equilibrium matching of the merger market is determined by both acquirers and targets incentives. Second, I borrow insights from the empirical matching literature (Sørensen, 2007; Fox, 2007; Akkus and Hortaçsu, 2007) and illustrate that a matching game can provide a very useful tool for the analysis of mergers. Third, economists have studied various mechanisms that could discourage the non-pro t-maximizing behavior of managers, such as product market competition, labor market competition, compensation schemes, and monitoring by the board of directors (see, for instance, Hart, 1983; Jensen and Murphy, 1990; Holmstrom and Kaplan, 2001). I show that targets dislike of badly managed organizations provides partial discipline for ine cient acquirers, and this is a new idea. In that regard, this paper is closely related to the paper of Mitchell and Lehn (1990), which shows that bad acquirers later become takeover targets. The main di erence between my paper and theirs is that I study targets preference for e cient acquirers as a possible discipline mechanism whereas they focus on e cient acquirers taking over rms who previously made ine cient acquisitions. A brief overview of the rest of the paper follows. Section 2 describes the mutual fund industry and my data. Section 3 presents empirical facts that support my hypotheses and also motivate 5

6 my model. I discuss my model of the merger market in Section 4. Section 5 then presents an econometric model of the merger market and discusses strategies for estimation. Section 6 provides empirical ndings from the model and discusses counterfactual exercises I performed. Section 7 concludes the paper. 2 Industry and Data 2.1 Industry A mutual fund pools capital from many people and invests it in stocks, bonds, or other assets. Mutual fund management companies such as Fidelity and T. Rowe Price o er wide ranges of mutual funds and retain professional portfolio managers to manage the funds. The set of mutual funds o ered by a fund management company is called a fund family. Examples of fund families include Fidelity funds, managed by Fidelity Management & Research Company, and American funds, managed by Capital Research & Management Co. Mutual funds are legal entities distinct from the management companies that manage them. They have their own boards of directors or trustees who owe duciary obligations to fund investors (also called fund shareholders). Mutual fund management companies register their funds as corporations or trusts, set up the funds boards, and sell the funds to investors. Hence, although mutual funds are legally separate from the companies that manage them, they resemble products produced by management companies and sold to consumers (fund investors). 4 Since fund management companies have their own shareholders and boards distinct from their funds shareholders and boards, I use the term shareholders to refer to the shareholders of fund management companies and fund investors to refer to the shareholders of individual funds. I use the term managers to refer to senior executives such as the CEO of a fund management company and fund managers to refer to the people who choose investments for individual funds. Of course, some managers are also fund managers. Many fund management companies are parts of larger nancial institutions. Broadly speaking, fund management companies fall into 5 categories: pure mutual fund management companies, fullservice brokerage companies, discount brokerage companies, banks, and insurance companies. Due 4 There is an ongoing debate about whether mutual fund investors should be viewed as owners of funds or consumers. See Tkac (2004) for an insightful discussion of this debate. 6

7 to data limitations, I do not incorporate into this paper s analysis information on fund management companies a liations with other nancial businesses. If many mergers in this industry are carried out for reasons unrelated to the mutual fund business, my ability to explain the merger pattern and merger outcome would be limited, and the estimation results would show this. The fact that I get very clear and striking patterns in the data and estimation results suggests that this is not the case. Legal precedents favorable to poison pills and other anti-takeover tactics made hostile takeovers rare in the period I study, 1991 through Hostile takeovers are especially rare in this industry because of the importance of human capital. In the words of Todd Ruppert, president and CEO of T. Rowe Price Global Investment Services, You just can t do a hostile takeover in this industry. The asset you are really buying is the people, and they can choose to walk out. So, you can only realistically deal with a motivated seller. 5 Some acquirers may primarily seek assets rather than people, but generally the importance of human capital discourages hostile takeovers. Hence, in my analysis I treat all mergers as friendly ones. Mergers may increase rm pro ts via: (1) economies of scale in production obtained by increasing the number of funds sharing research, back o ce, or brokerage functions; (2) economies of scale in marketing or distribution of funds, largely on the demand side, e.g. by providing one-stop shopping within a family of funds; and (3) skillful acquirers superior ability to manage assets or distribute funds. In this paper, I focus on (2) and (3) because cost savings do not seem to be a critical concern for mergers in this industry and I do not have a direct measure of costs. The mutual fund industry has low industry concentration, with the Her ndahl-hirschman Index below 400 throughout the sample period. 6 Moreover, mergers have a negligible impact on HHI. Hence, this paper does not focus on anti-competitive concerns associated with horizontal mergers, which is the main focus of much of the previous work on mergers, both empirical and theoretical, in the economics literature (see, for example, Farrell and Shapiro, 1990; Focarelli and Panetta, 2003). 5 Funds Europe report, 6 According to the Department of Justice Merger Guidelines, an industry is considered to be in the unconcentrated range if its HHI is below

8 2.2 Data I study data on U.S. mutual funds from the Center for Research in Security Prices (CRSP). The CRSP data set includes data on all open-end mutual funds 7 that have ever existed including: the amount of assets invested by the fund, the identity of the management company running the fund, the fund s investment objective, the fund s monthly returns, and the structure of the fund s fees. CRSP assigns each fund a unique identi er that stays the same even when the fund s management company changes. CRSP also assigns each management company an identi er but reuses the identi ers of extinct management companies so that identi ers are not necessarily unique. Thus, I constructed a unique management company identi er by checking the names and years of operation of management companies with the same identi er to determine whether they are the same companies. I also identi ed management companies as public or private using Thomson Financial s SDC Platinum and the web sites of management companies. Hereafter, I identify publicly traded companies and subsidiaries of publicly traded companies as public. I label all other management companies in the sample including the small numbers of non-pro t companies and companies owned by fund investors (such as Vanguard) as private companies. I identi ed mergers and acquisitions using the CRSP data rather than press releases. Identifying mergers using press releases would be excessively time-consuming and risk omitting many unreported mergers of small or private companies. Using the CRSP data presents some problems, 8 but enables me to identify all mergers between fund management companies, big or small, during the sample period. Note that my sample of mergers includes only mergers of two companies already active in the mutual fund industry. Because of the way I constructed the merger sample, it is possible that a combination of two subsidiaries of the same parent is captured as a merger. However, it is rare that a nancial institution owns two separate fund management companies. I say that company A acquires company B in year t if (1) during year t company A acquires more than 90% of funds that belonged to company B in year t 1 and survived into year t, and (2) company B dies during year t. By this de nition, a total of 266 mergers occurred during 7 Open-end mutual funds are a type of mutual fund that does not have restrictions on the amount of shares the fund will issue. If demand is high enough, the fund will continue to issue shares no matter how many investors there are. Open-end funds also buy back shares when investors wish to sell. ( This is the most common type of mutual fund. 8 It is hard to identify the exact month a merger occurs, and an actual integration could happen with some time lag after a merger announcement. 8

9 the period from 1991 through 2004 an average of 19 per year. For each merger case, I have information on the acquiring rm and target rm prior to the merger, and also how the merged company performs after the merger. Designation as acquirer or target is solely based on the data: The surviving rm is the acquirer and the disappearing rm is the target. For some cases, this distinction might be nominal rather than real. However, as the summary statistics in Table 2 below show, the average size of acquirers is about 10 times larger than that of targets, justifying my decision to conceptualize these merger cases as acquisitions rather than mergers of equals. My main interest lies in how merger behavior and merger outcomes di er between e cient acquirers and ine cient acquirers, but a company s acquisition motive is never directly observed. Therefore, I employ a proxy for acquirers with managerial motivations in my empirical investigation. My proxy is based on the following two ideas: (1) companies that have performed poorly are likely to be more vulnerable to incentive problems, and (2) public rms are more prone to incentive problems due to the separation of ownership and control. The idea that public companies are vulnerable to incentive problems is an old one, going back to Berle and Means (1932). There are theories and empirical works that support idea (1) as well. For instance, empirical works have shown that poor performance can increase the odds of manager dismissal (Chevalier and Ellison, 1999; Huson, Parrino, and Starks, 2001; Kaplan and Minton, 2006) and Edlin and Stiglitz (1995) present a model in which acquisitions can help managers to entrench their positions by conferring informational advantages on the incumbent managers who initiate the acquisitions (also see Shleifer and Vishny, 1989). In the context of the mutual fund industry, Chevalier and Ellison (1997) show that funds that trail the market have an incentive to gamble and try to catch the market. In addition, managers with empire-building motives would be more inclined to make an acquisition when the internal growth of their rms slows down. Therefore, I identify public companies with poor recent performance as a group that potentially has non-pro t-maximizing motivations for acquisitions, and test empirically whether their behavior di ers from others. In doing so, I separately control for public ownership and poor recent performance to ensure that the interaction between these two variables, which is my proxy, does not represent a systematic di erence between public and private companies or between companies with good and poor recent performance. To construct this proxy, I need a measure of rm performance, and I follow the mutual fund literature to do so. First, I measure the return of fund f during year t using objective-adjusted- 9

10 returns (OAR f;t ). OAR f;t of fund f is de ned as 12 Q m=1 Q12 (1 + R f;m ) 1 m=1 (1 + R o;m ) 1 (1) where R f;m is the return of fund f in month m, and R o;m is the average return of all funds in the market with the same investment objective as fund f in month m. This measure of returns implicitly adjusts for sector, industry, and style-speci c factors that may exogenously a ect all funds in the same investment objective category (Jayaraman, Khorana, and Nelling, 2002). Beginning in 1992, CRSP assigns each fund to one of 192 categories on the basis of its investment objective, 9 apparently enough categories to capture systematic di erences among di erent types of funds. To assess whether a company performed well or poorly prior to a merger, I calculate the weighted OAR of all funds o ered by the company in the year immediately preceding the merger (W OAR), using as weights the amount of assets each fund manages. A negative W OAR means the company s performance is below average. Then, my proxy for acquirers with managerial motivations is public companies whose W OAR is negative in the year prior to the merger. To minimize lost observations, I use all funds that have at least 3 months of return information in calculating W OAR for each company. I calculate OAR for each fund assuming that a fund earns the average return of its investment objective category in each month for which the fund lacks return data. My results do not change when I exclude funds that have less than 6 months or less than 9 months of return data. Table 1 presents descriptive statistics for all mutual fund management companies in the market. Table 2 presents summary statistics for the mergers in my sample and compares various attributes of acquirers and targets. 3 Empirical Facts In this section, I test the three implications of my hypotheses in turn via descriptive analysis. The empirical facts of this section provide support for my hypotheses and also motivate and inform my model of the merger market. Since I have not yet presented a uni ed empirical model within which to examine the three dimensions simultaneously, I examine one at a time in this section. A uni ed empirical framework is developed in Sections 4 and 5, and I will present the estimation 9 There are 27 categories of investment objectives prior to

11 results of such a model in Section 6. later sections. This section, hence, is best viewed as laying groundwork for 3.1 Incentives to Make an Acquisition First, I examine potential acquirers incentives to undertake an acquisition to see whether companies with misaligned incentives between managers and owners are more willing than other companies to make acquisitions, ceteris paribus, due to the presence of additional managerial gains. Table 3 presents the number of potential (top panel) and actual (bottom panel) acquirers in each of four groups: private companies with positive W OAR in the previous year, private companies with negative W OAR in the previous year, public companies with positive W OAR in the previous year, and public companies with negative W OAR in the previous year. I calculate the number of potential and actual acquirers in each group in a given year and pool them for all years in the sample period. I de ne potential acquirers in year t as including all companies not acquired or liquidated during or prior to that year. Comparing the two panels of Table 3 shows that public companies with bad recent performance are disproportionately more likely to make an acquisition. I conduct a probit analysis to control for rm size and other characteristics that might a ect a rm s likelihood of making an acquisition. The dependent variable of the regression is a dummy variable for making an acquisition; regressors include characteristics of rms such as size and age, and dummy variables representing the year. I measure all regressors in the year before the potential merger. Table 4 presents the estimation results. Table 4 shows that even after controlling for various rm characteristics, public companies with poor recent performance are much more likely to make an acquisition than other companies. Poor performance does not a ect a private company s probability of making an acquisition, but poor performance increases the probability that a public company will make an acquisition by about 2 percentage points. Since only about 2.8% of all potential acquirers make an acquisition in a particular year, poor performance increases a public company s probability of making an acquisition by around 71%. This result is consistent with the idea that poorly performing public companies, whose managers are more likely to pursue an acquisition that is not pro t-maximizing for shareholders, are more eager than other companies to make acquisitions. Acquisitions may help managers entrench their threatened positions, or managers of public companies might be pressured to solve their problems quick because of the sensitivity of public companies to short-run nancial results, e.g. quarterly 11

12 reports. However, Table 4 by itself does not allow us to determine whether poorly performing public companies high propensity to acquire is motivated by pro t maximization or managerial concerns. When we combine this information with merger outcomes, we might be able to say more. For instance, if we nd systematically worse merger outcomes for public companies with poor premerger performance, it weakens the argument that they tend to acquire more often because they have greater e ciency gains to realize. Accordingly, I examine merger outcomes in the next subsection. 3.2 Outcomes of Mergers This subsection analyzes whether non-pro t-maximizing acquirers have worse merger outcomes than other acquirers. I rst examine the di erence between (1) the pre-merger performance of funds previously owned by the target and (2) their performance one year and two years after the merger. My measure of the impact of a merger occurring in year t will be OAR f;t+1 OAR f;t 1 (change in OAR of fund f in the year after the merger) and OAR f;t+2 OAR f;t 1 (change in OAR of fund f in the two years after the merger). Targets funds change ownership after a merger, so acquirers skill in overseeing targets funds may account for changes in their performance. For instance, a skillful acquirer might be able to provide compensation packages that are better at incentivizing fund managers. Or a skillful acquirer might be better at minimizing in uence activities or other organizational costs of integration, which could have an impact on the performance of fund managers. My second measure of mergers impact on performance is net asset in ows into the merged company, as a percentage of the value of existing assets. An advantage of this measure is that it captures acquirers ability to exploit demand-side economies of scale in marketing and distribution, an important rationale for mergers in this industry. The second measure is inclusive of the rst measure since net asset in ows will be partly in uenced by post-merger fund performance Change in the Performance of Target Funds Table 5 compares changes in the return of target funds for di erent types of acquirers, with no controls. As before, I categorize acquirers on the basis of whether they are public and whether they have negative W OAR in the year preceding the merger, t 1. I analyze target funds that are 12

13 still alive by the end of year t + 2. The last column in Table 5 reports the number of target funds for each group. It is clear that poorly performing public companies tend to buy targets that own large numbers of funds. Table 5 shows that funds bought by public acquirers with poor recent performance experience the worst changes in performance post-merger. These funds have lower returns one year and two years after a merger than they did pre-merger, and only these funds returns consistently decrease post-merger. These funds cumulative returns in the two years following a merger are 0.9 percentage points lower than pre-merger returns. Returns of funds bought by other acquirers, however, increase post-merger. To see whether this pattern continues to hold when we control for other characteristics that might in uence the performance of target funds, Table 6 presents OLS estimation results controlling for characteristics of acquirers and targets and characteristics of the target funds. I measure all regressors in the year before the merger. The results in Table 6 show patterns similar to those of Table 5. The results indicate that annual returns of funds acquired by poorly performing public companies increase by percentage points less post-merger than returns of funds acquired by other companies. One puzzling pattern in Table 6 is that private acquirers with negative pre-merger W OAR do better in managing targets funds than private acquirers with positive pre-merger W OAR. Although I do not have a good explanation for the pattern, it might be that nancial returns are not the ultimate objective for some acquirers. Companies can bene t from acquisitions via demand economies of scale even if their nancial returns do not improve after mergers. Hence, I study asset in ows as a merger outcome variable in the next subsection Net Asset In ows into the Merged Company My second measure of post-merger performance is the net ow of assets into the merged company. Let T NA i;t be the total net value of assets managed by company i at the end of year t. 100 T NA i;t T NA i;t 1 T NA i;t 1 is then the net asset in ow into company i during year t as a percentage of the value of existing assets. Unlike in most papers on mutual funds, I do not calculate in ows net of reinvestments of dividends and distributions, because I view high reinvestments due to superior post-merger fund performance as part of merger outcomes. For each company formed by a merger in year t, I calculate net asset in ows during years t+1, t+2, and t+3. I use these three measures 13

14 of annual net asset in ows to compute the average annual net asset in ows after the merger for each of the four types of acquirers. Three results of my calculations of average net asset in ows, reported in Table 7, deserve mention. 10 First, all types of acquirers experience positive net asset in ows on average post-merger. The reason is that the mutual fund industry grew signi cantly during the sample period. Second, acquirers with below average pre-merger performance tend to attract less money from investors after a merger than do acquirers with above average pre-merger performance, re ecting investors tendency to chase recent past performance. Third, the discrepancy in the post-merger asset in ows of acquirers with positive and negative W OAR is much larger when the acquirer is public than when the acquirer is private (9.31 percentage points v percentage points). This suggests that the post-merger underperformance of public companies with negative pre-merger W OAR relative to public companies with positive pre-merger W OAR is not simply due to past performance in uencing future asset in ows. In other words, my results cannot be simply characterized as the continued poor performance of previously poorly performing companies. It is the interaction between poor pre-merger performance and public ownership that leads to bad merger outcomes. Table 8 presents OLS estimation results using net asset in ows as the dependent variable and controlling for characteristics of the acquirer and target that might a ect net in ows. The rst column includes a few basic controls, such as dummies for the year of the acquisition and the log of the value of existing assets. The second column adds expenses and loads, in the year before the dependent variable is measured, as controls. The third column adds a few more characteristics of the acquirer and target and match-speci c characteristics such as whether the acquirer and target use the same channel for distributing and marketing funds. The last column adds the growth rates of acquirers assets under management in the one and two years before the merger as controls, to distinguish the impact of pre-existing growth trends from the impact of mergers. Here, I simply The total number of observations is 489. The number of observations is less than 798 (=266 (number of acquisitions) 3 (for three years after the acquisition), because I exclude observations if the acquirer disappears (as a result of being acquired or liquidated) within three years after the merger, and the sample lacks at least one net asset in ow for mergers that occur in the last three years of the sample. Moreover, net asset in ows for an acquirer in a given year appear in the sample only once, even if the acquirer makes multiple acquisitions in that year (Recall that I analyze the net asset in ows at the fund management company level, so multiple acquisitions by the same acquirer do not generate distinct observations). Rather than double count observations corresponding with multiple acquisitions by the same acquirer, I treat the target as having the average characteristics of all targets acquired by the acquirer during the year. 2. I cap the maximum and minimum net asset in ows at 100% and -50% to ensure that a few outliers do not distort the results. The sample includes 12 observations with net asset in ows exceeding 100% and 6 observations with net in ows below -50%. 14

15 note that the coe cient on the dummy variable indicating poorly performing public acquirers is negative and signi cant in all speci cations, and reserve a full discussion of merger outcomes for Section 6, where I use similar outcome equations in the model estimation. 11 These results regarding the impact of mergers and the results presented in the previous subsection regarding incentives to make an acquisition are consistent with the hypothesis that a subset of fund management companies acquire for motives other than maximizing pro ts. Companies that are expected to be more vulnerable to agency problems according to the proxy are much more likely to acquire, but when they do acquire, they have much worse outcomes. If poorly performing public companies make acquisitions more often because they have greater e ciency gains to obtain, we would not expect to nd consistently worse post-merger performance by these companies. I do not know exactly why these ine cient acquirers perform poorly post-merger. One possibility is that organizational diseconomies of scale are part of the story (Meyer, Milgrom, and Roberts, 1992; Scharfstein and Stein, 2000), given that human assets are the key to success in this industry. These ine cient acquirers may be particularly likely to su er from organizational con icts and in uence activities during post-merger integration due to their incentive problems. The following quote suggests that the human-capital-intensive nature of the mutual fund industry makes it vulnerable to such problems during mergers. The focus can go away from asset management towards dealing with the corporate politics of the merger. In e ect, the threat is to both houses. M&A is a time-consuming distraction, and the corporate-political fall-out can hit buyer as well as seller. 12 One might be able to test this speci c theory, for example, by looking at fund managers turnover behavior after mergers, and I leave it for future work. 3.3 Targets Incentives In this subsection, I analyze patterns of matching between actual acquirers and targets to infer targets preferences regarding merger partners. I have shown above that poorly performing public 11 One might suspect that the poor outcomes of public acquirers with bad pre-merger performance might actually be an improvement over their pre-merger performance. If so, it is hard to argue that their mergers are not pro tmaximizing. However, I nd that their poor outcomes are in fact a further deterioration compared with their premerger performance. The post-merger growth rates of these acquirers are signi cantly lower than their pre-merger growth rates. In addition, for three years prior to a merger, public acquirers with poor pre-merger performance had about 3 percentage points lower growth rates than public acquirers with good pre-merger performance, but for three years after the merger, the former had about 10 percentage points lower growth rates than the latter. 12 Todd Ruppert, president and CEO of T. Rowe Price Global Investment Services, Harnessing talent post-merger, 15

16 companies are more willing to acquire than other companies, all else equal. In any plausible bargaining game, these companies greater willingness to acquire would translate into a higher willingness to pay for targets. Previous research found that acquirers seeking private bene ts for managers indeed tend to pay more (Morck, Shleifer, and Vishny, 1990; Slusky and Caves, 1991). Therefore, targets should prefer these companies to other acquirers (which I call price e ects), all else equal, unless targets have other incentives to avoid these companies. Targets, however, may care about factors other than the acquisition price. As mentioned in the introduction, the post-merger retention rate of targets management is relatively high in this industry, meaning that targets would care about what they expect to happen to the merged rm. As a result, targets would prefer an acquirer who is expected to perform better post-merger. The results from the merger outcome regressions showed poorly performing public acquirers di culty in attracting new money from investors post-merger. If targets incentive to avoid acquirers with poor post-merger management skill dominates the e ect of the higher willingness to pay of these ine cient acquirers, we would expect that these acquirers would be forced to match with less preferred targets. It is not easy to construct a measure of targets attractiveness, especially because di erent acquirers would evaluate the same target di erently depending on match-speci c synergies and there are multiple aspects that determine the overall attractiveness of a given target. The matching model in Sections 4 and 5 would enable me to account for match-speci c factors as well as many other target characteristics in rms merger decisions. Here, I simply use targets performance in the year before acquisition as a crude measure of target quality. This is unlikely to be the most important determinant of target quality, so one should take the analysis in this subsection at most suggestive. Table 9 shows the distribution of target quality across di erent types of acquirers. Each cell shows, for the speci ed type of acquirer, the percentage of acquirers who matched with a target that had a negative W OAR in the year before it was acquired. For instance, the number in the upper left cell means that 53% of acquirers that are private and had positive W OAR in the year before the merger matched with a target that had a negative W OAR in the year before it was acquired. Hence, the table shows that public companies with poor recent performance tend to match with low-quality targets more often. The di erence in the likelihood of matching with a low-quality target is not large, however, and the average W OARs of target companies for the four categories of 16

17 acquirers (not reported) show that while public acquirers with poor pre-merger performance tend to match with lower-quality targets than do public acquirers with good pre-merger performance, the same pattern is true for private acquirers. In addition to simple binary classi cations (above average v. below average), I also construct rankings to use in a regression. I rank actual targets in each year based on their W OAR in the year prior to the merger. A higher ranking indicates better recent performance. Next, I normalize the rankings by dividing them by the total number of actual targets in the year in question. example, if we have 20 actual targets, the worst-performing target receives a ranking of 1 20, and the best-performing target receives a ranking of 1. Similarly, I rank actual acquirers in each year based on their pre-merger W OAR and compute normalized rankings. For I then predict the ranking of the actual target matched with each actual acquirer using the acquirer s characteristics (including its ranking), the target s characteristics, and interactions between the two companies characteristics. As before, I measure all characteristics of the target and acquirer in the year preceding the merger. Table 10 reports the results of the regression. The results indicate that public companies with poor recent performance are more likely to buy low-quality targets than are public companies with good recent performance. If an acquirer is private, there is no such positive correlation (rather there is a negative correlation) between the acquirer s pre-merger performance and the quality of its matched target. In other words, public acquirers with poor recent performance are disadvantaged in the merger market compared to public acquirers with good recent performance, while for private acquirers, poor recent performance does not impose any disadvantage on them in the matching market compared to other private acquirers. One interpretation of this empirical pattern is that targets concern over being acquired by poorly run organizations mostly o sets the e ect of ine cient acquirers higher willingness to pay. 4 Model This section develops a model of the takeover market as a two-sided matching game (see Roth and Sotomayor, 1990) wherein potential acquirers and potential targets decide whether they want and with whom to merge. The primitives of the model are acquirers and targets preferences, which, together with the rules of the matching game, determine the equilibrium matching. In addition to providing a uni ed framework for my analysis, a matching model enables me to correct for selection bias in the outcome equations through joint estimation of the matching model 17

18 and outcome equations, as nicely illustrated by Sørensen (2007). If acquirers with a particular characteristic tend to match with targets that are desirable for unobservable reasons, the estimated coe cient for the acquirer s characteristic in the outcome equations will re ect the e ects of the unobserved target characteristics, biasing the coe cient upward. Following Sørensen (2007), and more generally Berry, Levinsohn, and Pakes (1995) and Bresnahan (1987), I use characteristics of other companies in the takeover market as an instrument to correct for such a bias. The rationale is that the characteristics of other companies in the market a ect the set of feasible merger partners for each company, providing exogenous variation that a ects matching, but not merger outcomes directly. The model also enables me to perform counterfactual analysis to investigate how merger market outcomes would be a ected when the behavior and incentives of merger market participants change. 4.1 Agents Market t has two non-overlapping sets of agents. The set of potential acquirers is I t and the set of potential targets is J t. The numbers of potential acquirers and potential targets in market t are ji t j and jj t j, respectively. Each potential acquirer can buy up to one target, 13 and each potential target can be sold to only one acquirer. Hence, the model is a one-to-one, two-sided matching model in which one side of the market consists of potential acquirers and the other of potential targets. Searching for matches is costless in this model, and players observe the lists of potential acquirers and potential targets. I use a two-sided matching game as my modeling framework instead of a roommate game. As I discussed in Section 2.2, I conceptualize mergers in my data set as acquisitions rather than mergers of equals, because the large size di erence between acquirers and targets suggests that there is more than a nominal distinction between the two. Hence, a two-sided matching model, which divides rms into either the acquirer side or the target side, seems to be a better representation of the merger market than a roommate model, which treats all rms symmetrically. Managers of potential acquirers and targets are the decision makers in my model. Each manager maximizes his own expected utility. If a manager s interests align perfectly with shareholders interests, maximizing the manager s utility is equivalent to maximizing shareholders utility. If 13 I relax this assumption later when I estimate the model. All results I obtain in this section hold for a many-to-one matching model if I assume responsive preferences. Preferences are responsive if for any two matchings that di er in only one target, an acquirer prefers the matching that contains the more preferred target. 18

19 these interests diverge, the manager maximizes his expected utility subject to constraints imposed by the shareholders, such as employment contracts or oversight by independent directors or significant shareholders. I further assume that targets do not have a problem of misaligned incentives. This assumption allows me to focus on con icts of interest at potential acquiring companies, and is partly justi ed by the following two reasons. Since hostile takeovers are rare in the mutual fund industry, management resistance to takeovers, such as poison pills or greenmail, which is intended to resist the sale of the rm is not relevant for my analysis. Shareholders and managers may disagree regarding by whom they should be acquired, but it seems to be of second order importance compared to the potential agency problems at acquiring companies. Moreover, almost 70% of target companies in the data are private companies which presumably have a much lower degree of agency problems, if at all. I assume a complete information game to make the model tractable. One implication of this assumption is that participants of the merger market know which acquirers are pro t-maximizing ones and which ones are not. It seems plausible that participants in the merger market have a fairly well-informed guess/knowledge about the management skills and motives of acquirers. What is harder to justify about the assumption is the following: If targets know which rms are managerially motivated, shareholders of the rms would know as well, and there should not be any non-pro tmaximizing acquisitions since shareholders would block them. I argue that even if shareholders of acquiring companies have as much information about their rms acquisition motives as targets do, they are not always able to block bad mergers. Unless an acquiring company issues more than 20% of the company s outstanding common stock for the transaction, it does not need to obtain shareholders approval. The company still needs the approval of the board, but it is well established in the corporate governance literature that boards often do not act in the best interest of shareholders. 4.2 Preferences Let S i;j;t be a potential acquirer i s valuation of a merger with target j in market t. The valuation measures the bene t of the merger to the potential acquirer s manager. S i;j;t is modeled as the sum of the e ciency gains from the acquisition and the private gains the manager obtains from the deal. This re ects managers need to pay some attention to shareholders interests even if they also pursue their own private interests. If a company pursues an acquisition purely for the sake of 19