THE RETURNS TO REPEAT ACQUIRERS. This Version: 14 September 2008

Transcription

1 THE RETURNS TO REPEAT ACQUIRERS KENNETH R. AHERN UNIVERSITY OF MICHIGAN ROSS SCHOOL OF BUSINESS Abstract Why are repeat acquirers early abnormal returns higher than those from later acquisitions? This paper argues that acquirers minimize integration and transaction costs by choosing an optimal target size. As acquirers get larger, they optimally choose targets of larger absolute size, but smaller relative size, leading to declining returns. Using a panel dataset of repeat acquirers over two decades, empirical tests support this argument. In contrast, I find no support for alternative explanations of declining returns based on hubris and diminishing opportunity sets, and only weak support for agency explanations. In addition, using new econometric techniques, I reject the popular theory that declining returns result from market anticipation of later deals. This Version: 14 September 2008 JEL Classification: G30, G32, G34 Keywords: Mergers and acquisitions, repeat acquirers, corporate governance, agency, hubris This paper is an extension of chapter two of my doctoral dissertation completed at UCLA. I am extremely grateful to Antonio Bernardo, Jean-Laurent Rosenthal, and J. Fred Weston for advice and support. I also especially thank David Robinson, Karin Thorburn, Roni Michaely, and Katrina Ellis. Comments provided by Raffaella Giacomini, Marc Martos-Vila, MP Narayanan, Han Kim, Geoffrey Tate, Uday Rajan, Mike Stegemoller, Liu Yang, and seminar participants at the 2008 AFA Annual Meeting, 2006 FMA Annual Meeting, the 2006 US and European FMA Doctoral Seminars, the Anderson School at UCLA, UCLA Department of Economics IO Workshop, London Business School, Penn State, the University of British Columbia, Virginia Tech, Michigan, Purdue, Maryland, and Vanderbilt improved this paper significantly. I gratefully acknowledge the financial support from the Research Program on Takeovers, Restructuring, and Governance at the Anderson School, UCLA. Please direct correspondence to Kenneth R. Ahern, Ross School of Business, University of Michigan, Ann Arbor MI Telephone: (734) Fax: (734) kenahern@umich.edu.

2 THE RETURNS TO REPEAT ACQUIRERS Abstract Why are repeat acquirers early abnormal returns higher than those from later acquisitions? This paper argues that acquirers minimize integration and transaction costs by choosing an optimal target size. As acquirers get larger, they optimally choose targets of larger absolute size, but smaller relative size, leading to declining returns. Using a panel dataset of repeat acquirers over two decades, empirical tests support this argument. In contrast, I find no support for alternative explanations of declining returns based on hubris and diminishing opportunity sets, and only weak support for agency explanations. In addition, using new econometric techniques, I reject the popular theory that declining returns result from market anticipation of later deals. JEL Classification: G30, G32, G34 Keywords: Mergers and acquisitions, repeat acquirers, corporate governance, agency, hubris

3 1. Introduction The vast majority of research on mergers and acquisitions assumes M&As are singular firm events. Yet in a sample of 12,942 mergers from 1980 to 2004, I find that only 38% of deals are made by first-time acquirers. Furthermore, the most active ten percent of the firms in the sample account for 35% of all deals. Repeat acquirers are the norm, not the exception. The few studies that account for repeat acquisitions find that a firm s announcement returns decline substantially over subsequent deals (Fuller, Netter, and Stegemoller, 2002; Aktas, de Bodt, and Roll, 2007b). In my sample, acquirers have an average three-day abnormal announcement return of 3.19% on the first deal, declining to -0.11% for fifth and later deals. This paper empirically addresses the question raised by this pattern: Why do repeat acquirers have declining returns? Whereas academic studies have concentrated on strategic fit and over-payment as key determinants of merger success, consulting firms and the business press have emphasized the role of target size on the costs of due diligence and integration strategies. For example, the 2002 merger of equals between Hewlett-Packard and Compaq was widely presumed to be doomed to failure at its announcement. However, 16 months after the merger, costs had been reduced by $3.5 billion, roughly 5%. By 2005, HP s stock had risen 46% compared to rival Dell s rise of 2% and IBM s decline of 23%. Compaq s CFO, Jeff Clarke, attributed the merger s success to careful integration, stating, We believe this was the most thoroughly planned merger in history (Harris, 2003, 2006). In contrast, poor integration led to high profile failures in the ATT-NCR and Daimler-Chrysler mergers, even though the economic motivations for the deals were clear. Though research has found that firm size affects acquirer returns, it is still not understood what drives this result (Moeller, Schlingemann, and Stulz, 2004). In this paper, I argue that returns decline by deal order because the costs of M&As increase as acquirers get larger. I develop a simple model to illustrate my hypothesis. I assume that transaction costs depend upon target size and integration costs are determined by relative size. Acquirers choose an optimal target to maximize profits taking these costs into consideration. Because larger targets incur greater integration costs, acquirers have an incentive to buy small targets. However, targets that are too small do not provide enough value to offset the search and transaction costs of the acquisition. Thus both the absolute and relative size of a target 1

4 2 THE RETURNS TO REPEAT ACQUIRERS influence the acquirer s outcome. The model predicts that larger firms will optimally choose larger targets, though both relative size and returns exhibit a hump-shaped pattern. Above a critical size, as acquirers get larger they optimally buy targets of decreasing relative size, though increasing absolute size, leading to declining returns. Since firms grow through acquisitions, firms later in a deal sequence are larger than firms earlier in a sequence. Thus the model predicts a pattern of declining returns to repeat acquirers even when firms are maximizing profit. My empirical tests support this argument. Nonparametric kernel regressions reveal a positive relationship between the absolute sizes of the acquirer and the target as well as a hump-shaped pattern of both returns and relative size as a function of acquirer size. The data confirm that acquirers get larger over deal sequences and since most first-time acquirers are large enough to be on the declining side of the hump-shaped pattern of returns, later deals generate lower average returns than do earlier deals. In addition, by analyzing the relationship between transaction size and advisor fees, I find evidence supporting the assumption that transaction costs increase with the size of the target. I also find that the relative size of target to acquirer is positively related to integration costs, proxied by industry-relatedness and geographic distance between bidder and target. Next, I use a large panel dataset of repeat acquirers to test the predictions of the cost minimization hypothesis directly against three alternative explanations: agency, hubris, and diminishing opportunity sets. The agency hypothesis predicts that management interests become less aligned with shareholder interests as a firm matures. Thus later deals may be made to generate private managerial benefits, not shareholder wealth gains (Moeller, Schlingemann, and Stulz, 2004). The hubris hypothesis predicts that early success leads to managerial overconfidence and thus overbidding in later deals (Aktas, de Bodt, and Roll, 2007a). Finally, the opportunity set hypothesis predicts that the best targets are acquired first and worse targets later (Klasa and Stegemoller, 2007). I test these hypotheses by first identifying the cross-sectional determinants of abnormal returns for a fixed deal number. Then I determine if these factors are changing systematically over a deal sequence. Both conditions are necessary to explain significant declines in returns. I measure agency using the Gompers, Ishii, and Metrick (2003) g index of managerial entrenchment

5 THE RETURNS TO REPEAT ACQUIRERS 3 and outside monitoring by independent blockholders. Hubris is measured by premiums paid, and opportunity sets are measured by an acquirer s market-to-book ratio, Tobin s q, and prior year returns. Regression analysis supports the cost minimization hypothesis. Controlling for a host of factors, the absolute and relative sizes of the target affect returns in the cross-section and also change systematically over deal sequences. In contrast, I find only weak support for the agency theory, and none for hubris and opportunity set explanations. Agency variables affect cross-sectional returns, but vary only slightly across deal sequences. In contrast, premiums and growth option proxies change substantially over deal sequences, but do not affect returns in the cross-section. The above results rely on independent cross-sectional and longitudinal tests. In contrast, the dynamic process of market anticipation of future deals at the announcement of earlier deals could explain declining returns as well: when later deals are announced there is no stock price effect because the value of the deal has already been capitalized. Though anticipation is widely cited 1, prior direct tests find mixed results, suffer from small samples, and do not account for the dynamic endogeneity between the likelihood of future deals and current returns (Schipper and Thompson, 1983; Asquith, Bruner, and Mullins, Jr., 1983). To verify the robustness of my results to an anticipation effect, I conduct a series of novel empirical tests designed to overcome limitations of prior studies. First, to address endogeneity, I estimate a simultaneous equations model of the interaction between current M&A returns and the likelihood of future deals. I find that markets do not capitalize the expected value of later deals at the announcements of earlier acquisitions. Though repeat acquirers have higher first announcement returns than firms that do not make subsequent acquisitions, these higher returns are not related to the likelihood of future acquisitions. Second, in a new econometric approach, I use quantile regression to identify the effect that deal order has on information revealed by an announcement. If markets anticipate future mergers, less information will be revealed at the announcement of later deals compared to earlier deals. I find that information, 1 Fuller, Netter, and Stegemoller (2002, p. 1764) assume that markets anticipate mergers for repeat acquirers, allowing them to control for much of the information about bidder characteristics contained in the returns at the announcement of the takeover. Other recent empirical studies that refer to anticipation as a possible effect on acquirer returns include Song and Walkling (2000), Wulf (2004), Bhagat, Dong, Hirshleifer, and Noah (2005), and Song and Walkling (2008).

6 4 THE RETURNS TO REPEAT ACQUIRERS as measured by the dispersion in returns for a cross section of acquisitions, controlling for other factors, is constant for the first six deals in a sequence, contrary to the anticipation theory and the assumptions made in prior studies. These results are robust to restricting the analysis to cases where anticipation is most likely, namely samples of large transactions and of the most frequent acquirers. Thus I find no evidence supporting anticipation using two independent and unique empirical tests. These results are relevant in their own right, but also validate my main results. Prior research has investigated other issues concerned with repeat acquirers. Loderer and Martin (1990) is perhaps the earliest paper to recognize that firms make multiple acquisitions. More recently, Fuller, Netter, and Stegemoller (2002) examines repeat acquirers to investigate how the public status of the target affects acquirer returns. Moeller, Schlingemann, and Stulz (2005) notes that large loss deals are usually made by a repeat acquirer following a series of successful deals, and Song and Walkling (2008) investigates the anticipation of future acquisitions following industry shocks. To the best of my knowledge, the only other paper to attempt to directly understand why returns decline is Aktas, de Bodt, and Roll (2007a). They posit that a self-interested manager learns through repeat acquisitions to increase premiums in order to capture larger private benefits, which leads to declining returns. More generally, my results contribute to a growing body of research that is concerned with corporate decisions in a dynamic, rather than static setting. See for example Leary and Roberts (2005) on dynamic capital structure, Helwege, Pirinsky, and Stulz (2007) on the evolution of insider ownership, and DeMarzo and Fishman (2007) on the dynamic interaction between agency conflicts and investment. The remainder of the paper is organized as follows. Section 2 presents the theoretical model of M&A cost minimization. The data are described in Section 3. Empirical tests of the cost minimization hypothesis and alternative theories are described in Section 4. Section 5 presents tests of market anticipation. Section 6 concludes.

7 THE RETURNS TO REPEAT ACQUIRERS 5 2. The cost minimization hypothesis In this section I present a simple model where acquirers maximize profit by choosing an optimal target size given transaction and integration costs. The model generates empirically testable predictions about the relations between acquirer and target size and returns. First I discuss how integration and transaction costs are related to target size. Though acquiring larger targets might generate more synergies, transaction costs are certainly higher. Larger targets require greater costs of due diligence, valuation analysis, and more complex financing. Consistent with this argument, a positive relation between advisory fees and transaction size has been documented by McLaughlin (1990, 1992) and Ma (2006). This means that bidders have an incentive to purchase small firms to reduce costs. However, targets that are very small may not provide enough value to offset the search and transaction costs of the acquisition. Anecdotal evidence supports this idea. In the 1984 Berkshire Hathaway Chairman s Letter to Shareholders, Warren Buffet details the qualities of an attractive takeover candidate. The first requirement is (1) large purchases (at least $5 million of after-tax earnings). In 1985, the size requirement was listed as $10 million in earnings (Buffet, 1984; Buffet, 1985). Clearly the actual size of a large acquisition depends on the size of the acquirer. This implies that the relative size of target to acquirer is an important decision criteria as well. In particular, integration is likely to be more costly when the target is large relative to the acquirer. Though there is no direct empirical evidence on integration costs, it is commonly accepted that mergers-of-equals produce greater post-merger frictions, compared to mergers where the acquirer dominates the target. These relations between target size and costs imply that even if there is a strong economic motivation for a deal, the successful realization of net gains will depend upon target size. I next present a simple model of M&As where acquirers trade off these types of costs and choose targets based on size in order to maximize their profits. I assume that benefits from a merger are linearly increasing in target size, T, at constant rate β. Benefits here are the synergies from the deal net of the purchase price. If the marginal benefits were decreasing in target size, as might be expected, the qualitative results of this model would remain, so linearity is assumed for convenience. Transaction costs are denoted g(t) and integration costs are denoted h(t/a),

8 6 THE RETURNS TO REPEAT ACQUIRERS where A is the acquirer size. Thus transaction costs are a function of absolute target size, and integration costs are a function of relative size as argued above. Search costs are fixed and denoted F. It is plausible that search costs decrease with target size, since smaller firms are less visible. However, this assumption would strengthen the qualitative results of the analysis, so fixed search costs are assumed for simplicity. The acquiring firm chooses target size T = T to maximize absolute profit: π(t(a),a) = βt F g(t) h(t/a) (1) I assume that costs increase monotonically, g (T) > 0 and h (T/A) > 0. The returns are defined as profit scaled by acquirer size, R(T(A),A) = π A. The following two propositions define how the shape of the cost curves lead to changes in the optimal target size and relative size derived from maximizing the firm s objective function in Equation 1. Proposition 1. If T A h (T/A) h (T/A) < 1, then T (A) > 0. Proposition 2. If g (T) > h (T/A) A, then T A (A) < 0. All proofs are in Appendix A. Proposition 1 states that if integration costs are not too concave, than the optimal target size will increase monotonically as acquirer size gets larger. Thus larger acquirers optimally choose larger targets. 2 Proposition 2 states that the optimal size of the target grows at a slower rate than acquirer size when the change in the marginal transaction cost is greater than the marginal integration cost. Intuitively, as acquirers get larger, the optimal relative size of target to acquirer will decrease if transaction costs are increasing faster than integration costs. For a fixed target size, integration costs will fall less for a large acquirer than for a small acquirer for the same increase in acquirer size. For size increases in small acquirers, the optimal relative value of target to acquirer may actually increase until the transaction costs outweigh the integration costs. Proposition 2 also implies that transaction costs must be strictly convex for relative size to decrease with acquirer size because h (T/A) > 0 2 One can interpret the necessary degree of convexity by noting that T A h (T/A) h (T/A) is the same form as the relative risk aversion coefficient for utility functions. For values of this coefficient between zero and one, costs are concave, but the maximization problem will still lead to increasing target size as acquirers get larger. Thus the shape of the integration cost curve is not restricted to convexity for this result.

9 THE RETURNS TO REPEAT ACQUIRERS 7 by assumption. This restriction is plausible given that convex transaction costs have been documented for activities less complex than mergers, notably the issuance of new securities (Altinkiliç and Hansen, 2000). The third proposition gives the necessary condition for returns to decrease as a function of acquirer size. Proposition 3. If T A h (T/A) A < π A = R(A), then R (A) < 0. This proposition states that when the marginal cost of integration scaled by the relative size of target to acquirer is less than the return, then returns are declining in A. In other words, an increase in acquirer size will lead to a less than proportional increase in profits when the savings from reduced integration costs are not large. Thus, as in Proposition 2, the same increase in size for a small acquirer, will lead to higher returns than for a large acquirer, because the marginal reduction in integration costs is larger for the small firm. Conversely, since the marginal benefit of reduced integration costs to large firms is small, returns will be lower for larger acquirers. To illustrate the implications of the general model I solve for the optimal target size, relative value, and returns using the following specification: π(t,a) = βt F 0.5δT 2 0.5γ (T/A) 2 (2) where the parameters β, δ, and γ measure the intensity of net synergy, transaction costs, and integration costs, respectively. The profit function is concave and the first-order condition yields an optimal target size T and relative size T A as follows: T = A2 β A 2 δ + γ (3) T A = Aβ A 2 δ + γ (4) The conditions of Proposition 1 and 2 are satisfied because both cost functions are strictly convex. Thus the optimal target size, T, is increasing in the acquirer size A for all A, but the optimal relative size decreases as acquirers get larger only after acquirers reach a certain critical

10 8 THE RETURNS TO REPEAT ACQUIRERS size where increases in transaction costs outweigh increases in integration costs. Substituting the optimal T in the profit equation yields, and the returns from the acquisition are, π(t (A),A) = A 2 β 2 2(A 2 δ + γ) F (5) R(T (A),A) = π A = Aβ 2 2(A 2 δ + γ) F A. (6) To illustrate the results of this model, I plot the optimal target size, relative size, and return as functions of acquirer size in Figure 1. Parameter values are β = 1, F = 0.5, δ = 0.1, and γ = 5. Since γ measures the integration cost based on relative size and δ measures the transaction cost based on the absolute size, γ is much larger than δ. Figure 1 shows that as acquirers get larger they optimally choose larger targets though of decreasing relative size. The relative size and returns depend upon the size of the acquirer in similar ways. When acquirers are small, the integration costs are large relative to the transaction costs. As the acquirer size increases, the savings from the integration costs more than offset the additional transaction costs and so returns and relative size increase. At a critical point, the transaction costs overcome the integration costs and increasing acquirer size leads to decreasing returns and relative size. In summary, under the following assumptions, 1. Acquirers maximize profits by choosing an optimal target size. 2. Transaction costs are increasing in the absolute size of the target. 3. Integration costs are increasing in relative size and are not too concave. the model generates the following empirically testable predictions, 1. Target size grows monotonically with acquirer size. 2. Relative size and returns follow a hump-shaped pattern, increasing for small acquirers and decreasing for acquirers larger than a critical size. The implication for repeat acquirers is that if there is a positive relationship between deal order and acquirer size, as would be expected, this model predicts that we will observe that returns

11 THE RETURNS TO REPEAT ACQUIRERS 9 decline over a deal sequence for large enough firms even when firms are maximizing profit from acquisitions. 3. Data and methodology To test theories of repeat acquirers it would be ideal to have returns data and complete acquisition histories of all acquiring firms. However, comprehensive merger data begins in 1980 and returns data are only available for public firms. Thus to produce the most complete acquisition histories I limit my sample to firms that publicly list after This may produce two types of bias. First, firms may have extensive acquisition histories as private firms that would not be captured in my data. However, it is likely that acquisitive private firms also will be acquisitive public firms and this bias will affect all firms equally. Second, the post-1980 listing restriction may bias my sample toward firms in certain industries. I address this problem below and find little bias. The following presents a detailed description of the data. The sample data are taken from Securities Data Corporations s (SDC) U.S. Mergers and Acquisitions database. Only acquisitions worth at least $1 million announced between 01/01/1980 and 12/23/2004 that were completed within 1,000 days are included in the sample. 3 Because repeat acquirers may be more likely to acquire many small firms, rather than fewer large firms, no restriction is placed on the relative value of the target to the acquirer as is commonly done in prior studies. Also, acquirers have to own less than 50% of the target before the acquisition, and 100% after the acquisition. This prevents the inclusion of repeat partial acquisitions of the same target. Acquirers have to be public firms with data available on the Center for Research in Security Prices (CRSP) and CompuStat databases. Targets are restricted to public, private, or subsidiaries of a public or private firm. Also, multiple acquisition announcements by the same firm within five days of each other are excluded. Finally, as noted above, to ensure acquisition deal histories are correctly measured, I exclude all acquirers that were listed on CRSP before 01/01/1980. This exclusion is not typically done in prior research on multiple acquirers but provides a solid benchmark from which to order 3 I restrict attention to completed deals because data on incomplete deals will likely be biased toward public targets. However, only using completed deals may lead to a misproportional small number of hostile deals, since hostile deals are more likely to fail (Walkling, 1985).

12 10 THE RETURNS TO REPEAT ACQUIRERS acquisitions. Of course acquisition histories are still likely to be incomplete as pre-ipo firms make acquisitions. However, if no benchmark is used, acquisition data limitations will lead to a downward bias in the measurement of acquisition experience for older firms. Using this restriction also avoids defining the beginning of a merger program by an arbitrary no-acquisition hiatus of between two and eight years, as is commonly done in prior studies (Loderer and Martin, 1990; Conn, Cosh, Guest, and Hughes, 2004). This sampling procedure produces 12,942 acquisitions made by 4,879 acquirers. The prototypical repeat acquirer, Cisco Systems, completed 50 acquisitions, the largest number in the sample, though the average firm completed 2.7 deals over the sample 15-year period. If a 1% relative value restriction had been placed on the sample, Cisco would only have 10 deals in the sample. A 5% cutoff would have left only one deal in the sample for Cisco. Thus, imposing relative value restrictions may alter the sample significantly. Table 1 presents a summary description of the sample by year. Total deals peaked in 1997 with 1,437 announcements, though total transaction value peaked in 2000 with $615,382 million. The median transaction value for all years is $25.38 million, considerably less than the average value of $571 million, reflecting the positive skewness of the distribution of transaction values. Though I limit the sample to firms not listed before 1980, the distribution of deals by industry shifts only slightly toward high-technology industries. In a sample where acquirers are not restricted to being listed after 1980, using the 49 Fama French Industry Classifications, 4 banking accounts for the largest number of deals without restricting acquirer listing dates (13.9% of all deals). Computer software (9.9%), business services (6.9%), electronic equipment (5.8%), and communication (5.5%) round out the top five industries which together account for 42% of all deals in the unrestricted sample. The top five industries for the sample used in this paper, where acquirers must be first listed after 1980, are software (13.9%), banking (10.8%), business services (8.6%), communication (6.5%), and electronic equipment (6.2%), totalling 46% of all deals. Thus the industry clustering in merger activity reported in prior work is confirmed here, and relatively unchanged by my sample restrictions (Mitchell and Mulherin, 1996; Harford, 2005). This suggests that the 1980 listing requirement will not produce extensive bias in my results. 4 Generously provided on Kenneth French s Web site. library.html

13 THE RETURNS TO REPEAT ACQUIRERS 11 Because prior acquisitions may affect any event study prediction method which estimates abnormal returns using firm historical returns, I calculate abnormal returns using a marketadjusted model with the equally weighted CRSP index as a market proxy. For each day in the event period, market returns are subtracted from firm returns (Brown and Warner, 1985). Cumulative abnormal returns (CARs) are computed over the five days surrounding the announcement because the announcement dates listed on SDC are not always accurate, especially for the small deals in my sample. Significance tests of CARs are conducted with a sign test (Corrado and Zivney, 1992). Table 2 reports CARs grouped by total number of deals in a firm s series, acquisition order in series, and target organizational status. There are 2,212 firms that made only one acquisition in the sample period, while there are 503 with over five acquisitions. These 503 firms account for 10% of all firms in the sample, but complete 35% of all the deals. The average CAR for all firms and all deals is a significant 1.98%. Subsidiary targets yield higher returns (3.09%) than do private targets (2.30%), which yield higher returns than public targets, which are statistically negative (-0.86%). Consistent with prior studies, CARs are declining with deal order, regardless of target status. Public targets yield an insignificant 0.35% for first deals declining to a significant 2.06% for sixth and later deals. First acquisitions of private targets generate 3.39%, declining to 0.72% for sixth and later deals. 5 The sample sizes reported for each CAR in Table 2 also reflect that the type of target is changing with deal order. For the first through third deals, roughly 16% are public targets, 55% are private, and 29% are subsidiaries. For fourth and higher deals, public targets become more prevalent, increasing to 25% of the sample for the sixth or later deals, while private targets decrease to 48%. Subsidiary targets first increase to 33% then decline to 27% for the later deals. This supports the idea that changing firm characteristics may explain part of the decline in abnormal returns with deal order. 5 In unreported calculations, average (median) abnormal dollar returns are $19.5 million ($0.4 million) in 2005 dollars, consistent with the same calculations in Moeller, Schlingemann, and Stulz (2004). However, the abnormal dollar returns do not exhibit a strong pattern of decline over deals, as do the equally weighted abnormal returns. Nevertheless, I replaced CARs with dollar returns in all of the following analyses as a robustness check and found no qualitative difference in the results.

14 12 THE RETURNS TO REPEAT ACQUIRERS 4. Empirical tests of the cost minimization hypothesis To test the cost minimization hypothesis, I first empirically examine the plausibility of its assumptions and then compare the model predictions to the data. The model assumes that transaction costs affect the absolute size of the target, whereas integration costs affect the relative size. To measure transaction costs I retrieve the total acquirer financial advisor fees and the number of acquirer advisors per deal from SDC. Larger deals are predicted to have larger transaction costs. To proxy for integration costs I record whether the target and bidder are in the same Fama French 49 industry classification. Second, I calculate the geographic distance between the location of bidder and target headquarters measured at the zipcode level. 6 I hypothesize that targets that are in different industries and located farther away from the acquirer will have greater frictions and thus higher integration costs. To test the relationship between these cost measures and target size, I run log-log regressions to estimate elasticities between the variables. These results are presented in Table 3. This analysis should not be interpreted as causal evidence. Instead the results record whether larger transactions are associated with higher costs, controlling for other factors. First, a 1% increase in acquirer size is associated with a 0.78% increase in the transaction size and a 0.63% decrease in the relative size of target to acquirer. This result is consistent with model prediction 1. Second, larger deals are associated with larger transaction costs measured both by total fees and by the number of advisers. Higher fees are also associated with deals of larger relative values. Finally, the proxies for integration costs are positively related to relative value. Higher relative values are associated with higher integration costs as measured by distance and industry-relatedness. These results provide credibility to model assumptions 2 and 3 on the relationships between costs and target sizes. Assumption 1 states that firms maximize profits. The plausibility of this assumption is directly tested against other alternative motives for M&As in later analyses. Next, I estimate the model predictions about the relationship between firm size and returns. Econometrically, I want to estimate E(X Acquirer Size), where X is either returns, target size, or relative size. Since the model makes distinctly non-linear predictions, I do not impose 6 The zipcode is taken from SDC. Using the US Census Bureau s database of zipcode longitudes and latitudes, I calculate the surface distance in statute miles.

15 THE RETURNS TO REPEAT ACQUIRERS 13 a functional form on this expectation, but instead use nonparametric kernel regression to plot the relationships. 7 These estimated expectations are plotted in Figure 2 along with scatterplots of the data. The kernel regression estimates closely follow the model s predictions. This evidence is particularly strong because the unique hump-shaped pattern predicted by the model is replicated in the data. In particular, both returns and relative size increase over a range of small acquirer values and then decrease toward zero. Moreover, consistent with the model, the marginal change in relative value is much larger than the marginal change in returns for increases in acquirer size. Finally, transaction size is also increasing in acquirer size, as predicted. This evidence shows that returns are related to acquirer and target sizes. Thus, if acquirers are getting larger with subsequent deals, than returns will decline over deal sequences. Though the nonparametric estimations are strong evidence in support of the cost minimization hypothesis, they do not control for other factors that may explain declining returns. In particular, the cost minimization hypothesis assumes firms are maximizing profits by choosing an optimal target size. Alternative theories of M&As include agency, hubris, and diminishing opportunity sets. The next set of tests explicitly controls for a host of variables and investigates these alternative theories Tests of the alternative theories To test the alternative theories, I first identify the factors that significantly affect returns in the cross-section and then test whether these factors are changing over deal sequences. Only factors that both explain cross-sectional variation and that vary systematically over a deal sequence can explain the pattern of declining returns. In contrast to the efficiency-based size effect in my model, Moeller, Schlingemann, and Stulz (2004) hypothesize that the size effect reported in their study is likely due to agency problems of larger firms, though they provide no formal tests. I test this hypothesis directly by including measures of internal monitoring and managerial entrenchment/antitakeover provisions 7 In particular I use the leave-one-out Nadaraya-Watson estimator with a Gaussian kernel. Cross-validation is performed by minimizing the estimated prediction error in order to find the optimal bandwidth. See Härdle (1990) for more details on kernel regression estimates.

16 14 THE RETURNS TO REPEAT ACQUIRERS in regressions on acquirer returns. As a measure of internal monitoring I use the number of non-officer directors that are blockholders in the firm. These data on 1,913 firms over come from the Blockholders database maintained by Wharton Research Data Services (WRDS) and described in Dlugosz, Fahlenbrach, Gompers, and Metrick (2006). Entrenchment is measured using the Gompers-Ishii-Metrick (GIM) governance index of the data in the Investor Responsibility Research Center s (IRRC) Governance database. This data set provides data on 24 antitakeover provisions, such as staggered boards, poison pills, and others, for a sample of predominately large firms for selected years starting in For further information see Gompers, Ishii, and Metrick (2003). The agency theory hypothesizes that more non-officer director blockholders will be associated with higher returns and more antitakeover provisions will be associated with lower returns. Since internal monitoring and the market for corporate control may be substitutes, I also look at the interaction between the two. To investigate hubris, I look at premiums paid by the acquirer. Premiums are defined as the transaction value from SDC divided by the market value of the target 50 trading days before the announcement date. The relation between premiums and CARs is not well defined. The learning model of Aktas, de Bodt, and Roll (2007a) states that higher premiums drive down abnormal returns from acquisitions made later in a deal sequence. However, in contrast to this theory, it is possible that high premiums may indicate the possibility of high synergies between bidder and target, which could lead to higher abnormal returns. Finally, to test the diminishing opportunity set theory, I would like to be able to measure the number and quality of remaining targets for a given acquirer. However, I am unaware of any direct measure of these factors. Instead, I use acquirer book-to-market, Tobin s q, and prior year returns as proxies of opportunity sets. These variables have been used in other studies to measure growth options, and I assume that a component of these growth option measures includes acquisition options (Billett, King, and Mauer, 2007). Table 4 presents firm fixed effect regressions designed to test the cost minimization hypothesis against the alternative explanations. The first column regresses the five-day CAR on acquirer, target, and deal characteristics, controlling for unobserved firm heterogeneity and time effects. First, deal number is not significantly related to abnormal returns. This means that other

17 THE RETURNS TO REPEAT ACQUIRERS 15 determinants of returns must be changing over time to explain declining returns. Moreover, experience does not appear to influence merger returns. Second, acquirer size is negatively and significantly related to CARs, consistent with the cost minimization hypothesis, but also with hubris and agency. In addition deals/year is also negatively related to CARs, though time elapsed since the prior deal is positively related. Firms that make many acquisitions quickly have lower CARs than firms that do not. Song and Walkling (2008) use this as evidence of market anticipation of later deals. A short duration between deals may instead be related to the indigestion hypothesis of Conn, Cosh, Guest, and Hughes (2004), where integration between the target and bidder is hampered by a subsequent acquisition. Moreover, in various explicit tests reported in Section 5, I do not find support for anticipation as a determinant of returns. Third, the results in Table 4 show that public targets and particularly those purchased with stock, generate significantly lower returns, consistent with the liquidity premium shown in Officer (2007). All these results are consistent with prior studies (Fuller, Netter, and Stegemoller, 2002; Moeller, Schlingemann, and Stulz, 2004). Finally, note that relative value is positive and significant at the 10.4% level, though transaction value is insignificant. Next, I include the variables measuring agency costs in column (2) under the Governance heading in Table 4. Outside director blockholders is significant and positive as hypothesized, the entrenchment index is negative, but not significant, and the interaction term is significantly negative. The negative sign of the interaction term indicates that the benefit of internal monitoring is eroded with more entrenchment provisions. These results are consistent with Masulis, Wang, and Xie (2007) who show greater shareholder control is positively related to acquirer returns in a static cross-section analysis. Also of note is that the inclusion of these agency variables does not change the insignificant acquirer size effect between regressions (1) and (2). This does not support an agency explanation of the size effect as suggested in Moeller, Schlingemann, and Stulz (2004), but neither is it convincing evidence against this hypothesis, since the firms with observed agency variables tend to be much larger than those firms omitted from the IRRC database. This size constraint on the sample also affects the estimated effect of the relative value on the CARs. Since larger firms have lower CARs, larger relative sizes decrease

18 16 THE RETURNS TO REPEAT ACQUIRERS the returns, as opposed to the positive relationship found in the full sample where returns are higher on average. Next, I test the hubris story, where I restrict my sample to acquisitions of public targets in order to calculate premiums. The results in column (2) under the Public Targets heading in Table 4 suggest that there is no relationship between premiums and CARs. However, target size has a significant negative relation to acquirer CARs, consistent with the cost minimization hypothesis. In unreported tests, for robustness I also include the number of bidders in a takeover contest and find no qualitative differences in the results. Finally, none of the acquirer opportunity set measures are significant in any specification in Table 4. This suggests that diminishing opportunity sets do not explain declining acquirer returns either. In summary, though no evidence is found to support the hubris or opportunity set hypotheses, the above results show that both target size and more managerial entrenchment with less oversight significantly reduces acquirer returns in the cross-section. Other factors also explain announcement returns. Public targets generate lower returns, as does using all stock financing. Both relative value and the number of years since the last acquisition have a positive effect in the large sample, but a negative effect in the public target samples. However, to explain declining acquirer returns, it is not enough that a variable affects CARs in the cross-section alone. It also must be the case that the level of the variable changes systematically over deal sequences. To determine which of these variables are consistently changing over deal number, I calculate means and medians of firm and deal characteristics by deal number for all firms in the sample as well as slope coefficients for both a linear and squared term similar to the procedure in Aktas, de Bodt, and Roll (2007b). These results, presented in Table 5, provide more evidence in support of the cost minimization hypothesis. Average acquirer size continues to grow over subsequent acquisitions, but the average relative size of the target declines at a declining rate over deal sequences. Thus later deals are dominated by acquisitions of large targets, though of a small relative size. Because relatively smaller targets create smaller abnormal returns, acquirer returns are decreasing over deal sequences. In summary, the empirical relation between costs and sizes, the nonlinear kernel regression results, and the longitudinal decline in relative value

19 THE RETURNS TO REPEAT ACQUIRERS 17 and increase in target absolute size support the basic assumptions and predictions of the cost minimization hypothesis. Returning to the results in Table 5, agency problems appear to have a weak negative relation to declining acquisition returns. First, though the number of outside director blockholders is significantly related to CARs, they are unchanging over deal sequences, a surprising result considering the large increase in the average acquirer size. Second, though managers are significantly more entrenched in later deals than in earlier deals in a statistical sense, the actual change in the average number of antitakeover provisions over the first ten deals is very small. Since these entrenchment changes only affect returns significantly in the interaction with the outside director monitoring variable, the final effect of increased entrenchment on CARs is very small. For robustness, other measures of agency might have been used, but they would likely suffer from the same time invariance. For example, inside ownership may affect merger returns, but both Zhou (2001) and McConnell, Servaes, and Lins (2007) report that inside ownership changes are extremely small over time within the same firm. Sample attrition may explain the deal-series variation if the firms completing later deals are significantly different than those completing earlier deals. To account for this potential bias, in unreported tests I examine deal-series variation using only observations from the 503 acquirers with more than five deals in the sample. The results are unchanged using this smaller sample. In addition, I control for firm fixed effects by looking at within-firm changes in variables over deal numbers and find results that are qualitatively the same as those presented above. In summary, returns decline because acquirers get larger and the relative size of targets to acquirers gets smaller. This pattern is shown to be consistent with a firm s objective to minimize the costs of M&As, after controlling for variables related to agency, hubris, and opportunity set theories. 5. Theory and prior evidence of market anticipation of mergers Though the above results are consistent with the cost minimization hypothesis, if investors anticipate later deals at the announcement of earlier deals, the empirical patterns of the returns to repeat acquirers could also be the consequence of an entirely different effect which would

20 18 THE RETURNS TO REPEAT ACQUIRERS not be detected in the above analyses. Schipper and Thompson (1983) propose a capitalization theory where markets reflect the entire benefit of an acquisition sequence in the first announcement of the program. Later acquisition returns only reflect surprises, which are zero on average. A related signaling theory proposed in Asquith, Bruner, and Mullins, Jr. (1983) suggests that each acquisition announcement provides less information to the market about the true value of the firm than the preceding announcement. Since the signaling theory is equivalent to the capitalization theory with uncertainty, I group them together in a theory called the anticipation theory. This theory predicts that acquisition returns will be declining as uncertainty is resolved, and later deals will reflect less new information. Since the dynamic effect of anticipation could distort any cross-sectional theory explaining declining returns, it is crucial that we determine its effect, if any Capitalization theory There is an endogenous relationship between current returns and future expected returns. A high first deal return of a repeat acquirer may simply reflect a survival bias, where a successful firm will continue to make acquisitions, rather than reflect the present value of future deals, as suggested by the anticipation theory. To explicitly control for this endogeneity problem, I use a simultaneous equations framework with panel data which allows me to control for the likelihood of future acquisition activity at the current deal. I define the following simultaneous equations model, CAR ia = α 1 EV F ia + X 1ia β 1 + c 1i + u ia a = 1,...,A (7) EV F ia = α 2 CAR ia + X 2ia β 2 + c 2i + v ia a = 1,...,A (8) where EV F = Expected Value of Future Deals a = Order number of acquisition.

21 THE RETURNS TO REPEAT ACQUIRERS 19 This model allows for a simultaneous relationship between the present CAR and the expected value of future acquisitions. The c 1i and c 2i terms capture assumed time-invariant unobserved firm heterogeneity that may affect returns and the value of future deals. This would include such attributes as corporate culture and organizational ability. The variables in the X s reflect other explanatory variables in the equations including size, valuation, deal number, and time elapsed between deals. To estimate the expected value of future deals (EV F) I must account for both the probability of completing more deals and the value of the deals. First, even after controlling for numerous factors, cross-sectional studies of returns usually report R 2 measures of less than 10%, indicating that much of the variance in returns is unexplained. Thus, to reduce noise, I assume all firms would realize a common gain if they carried out a future deal. Second, the probability of making a subsequent deal is much higher than the probability of making ten more deals. Compounding probabilities implies that the likelihood of the immediately subsequent deal captures the greatest portion of the uncertainty of future M&A activity. Thus the uncertainty of the value and likelihood of future deals motivates the following simplifying assumption, EV F ia = P ia V a+1 (9) where the value of the future deal, V a+1, is common to all firms, but the probability of making a subsequent deal, P ia, varies by the firm and deal characteristics of the current deal, a. According to the CARs presented above, V a+1 is non-negative on average, and so there should exist a positive relationship between EV F ia and CAR ia in Equations (7) and (8). 8 8 One could argue that the likelihood of a successful deal is inversely related to the value of the deal. Hietala, Kaplan, and Robinson (2003) show that Viacom won the takeover battle for Paramount in 1994, but overpaid substantially. Thus, due to a winner s curse, highest bidders are most likely to succeed in an acquisition, but destroy value. I do not think this is a large concern in my analysis. The probability I measure is the likelihood of making a future acquisition as measured at the time of a current announcement. This incorporates both the likelihood of making an offer and the likelihood of success. Only the second likelihood might be negatively related to deal value and it is arguable less important than the fundamental decision to make an acquisition or not.