(Job Market Paper) Anticipating Takeovers and their Payment Methods: A New Approach Using U.S. Acquisitions

Transcription

1 (Job Market Paper) Anticipating Takeovers and their Payment Methods: A New Approach Using U.S. Acquisitions Mohammad Irani First version : September 08, 2013 This version : November 10, 2015 ABSTRACT This paper introduces a new approach for identifying the dates on which the market anticipates both takeovers and their payment forms, prior to their announcement dates. This approach predicts that the variance-covariance of the stock returns shifts when the market is informed by signals about potential takeovers with synergistic gains. Using a sample of acquisitions between U.S. public firms, I find that 86% of takeovers (62% of payment forms) are anticipated on average nine (six) months in advance. This is much earlier than reported by previous studies (two months). Moreover, when I add anticipation dates to the cross-sectional determinants of payment methods, some of the preceding results change. The reason for this finding is that the firm characteristics are not only related to the choice of payment method but also to the extent to which takeovers are anticipated. JEL classification: G14; G17; G34 Keywords: Mergers and acquisitions, payment method, prediction, structural breaks, variance, covariance The author is a PhD candidate in Finance at Stockholm Business School, Stockholm University, SE Stockholm, mir@sbs.su.se, I would especially like to thank my supervisors Lars Nordén and Rickard Sandberg for their helpful comments and numerous discussions. I am grateful for comments from Chris Adcock, Thomas Bates, Mariassunta Giannetti, Michael Halling, Martin Holmén, Cătălin Stărică, Johan Stennek, and Joakim Westerlund. I also thank seminar discussants and participants at the second National PhD Workshop in Finance (Stockholm, 2013), the 3rd PhD Conference (Stockholm University, 2014), the 63rd Annual Meeting of the Midwest Finance Association (Orlando, 2014), the 23rd Annual Meeting of the European Financial Management Association (Rome, 2014), Stockholm Business School (2015), and the Annual Meeting of Financial Management Association (Orlando, 2015). I am gratefully acknowledge the Jan Wallander and Tom Hedelius foundation for research support. Any errors are my own.

2 1 Introduction Jensen and Ruback (1983) and Martynova and Renneboog (2008) review takeover studies and report that the target shareholders gain large abnormal returns while the returns to the acquiring firms are negligible (or even negative). Travlos (1987) and Schwert (1996) also document that gains in cash-financed offers exceed those in equity-financed and mixed-financed offers. These differential returns induce investors to try to profit from anticipating takeovers and their payment forms. The extant studies provide cross-sectional predictions, which provide no guidance about anticipation dates. 1 However, identifying those dates can have important implications for our understanding of takeovers. Edmans, Goldstein, and Jiang (2012) explain that stock prices of merging firms are endogenous as they are contaminated with takeover anticipations. The dates can hence be used to remove the effect of takeover anticipation on prices, which in turn can improve measurement of explanatory variables that use stock prices (e.g., Tobin s Q, price-earnings ratio, the discount rate, the acquisition stock returns, return variance, covariance and correlation of merging stocks). The aim of this paper is to investigate whether the market can anticipate mergers and acquisitions (M&As) and their payment methods prior to their announcement dates. consequences for the choice of payment method in M&As? If so, when do those anticipations occur, and what are their Detecting the dates on which M&As are anticipated is a challenging task. Li and Prabhala (2007) argue that M&As occur neither randomly nor unexpectedly, but takeover studies provide indirect evidence regarding ex-ante predictability of M&As. Schwert (1996) shows that the cumulative average abnormal returns (CAARs) to the target shareholders start to accumulate 42 trading days before the announcement date. He finds that there is a one-for-one relation between this pre-announcement run-up and the total premium paid to the target shareholders (the sum of run-up and post-announcement markup returns). Based on this result, Schwert concludes that the target pre-announcement run-up is not affected by takeover anticipations, otherwise it should not be added to the total premium paid by acquirers. However, Betton, Eckbo, Thompson, and Thorburn (2014) illuminate that the relation is significantly smaller after they account for the anticipation effect on the target run-up. Overall, findings of event studies suggest that the market anticipates the target firms on average two months in advance, but the acquirers are unpredictable because their CAARs do not show any trend in the preannouncement period. 2 Furthermore, the event-study methodology focuses on just one of the pair of firms at a time (either the target or the acquirer firm), thereby neglecting the additional information contained in their joint return distribution (e.g., the covariance), which might be useful for anticipating M&As. To deal with the unpredictability issue, this paper introduces a new time-series and direct approach for identifying both the deal-anticipation (a pair of target and acquirer firms) and payment-form-anticipation dates. This approach is based on analyzing 1 Amihud, Lev, and Travlos (1990), Hasbrouck (1985), and Martin (1996), among others, employ a crosssectional (binary discrimination) approach to examine investors predictions of acquirer and target merger candidacy and their predictions of payment methods. 2 Similarly, Smith and Kim (1994) find that the CAARs of the target firms in their sample start accruing 60 days before the announcement day; however, the returns to the acquiring firms do not deviate from zero in this period. I find comparable CAARs to theirs for the target and acquirer firms in my sample. 1

3 breaks in the variance-covariance structure of the joint target and acquirer daily return series. Betton et al. (2014) suggest that M&As are anticipated when the market receives informative signals about potential takeovers with synergistic gains. Accordingly, when the market anticipates a potential deal for the first time during the pre-announcement period, the merging likelihood should increase significantly. Simultaneously, the market also discounts the expected synergy value of that merger and incorporates it into the price process of the merging firms, which in turn can affect the moments of the stocks returns (including levels, variances, covariance, etc.). The merging likelihood and the expected synergy are generally unobservable, while shifts in the second-order moments can be estimated, allowing the anticipation dates to be identified. Not every shift during the pre-announcement period, however, is related to takeover anticipation. Restrictions are imposed on the sign of shifts in the variance-covariance structure to disentangle consistent from inconsistent ones. A break date is interpreted as an anticipation date if shifts around that date are consistent with the expected ones. The expected shift for the anticipation of a deal is a significant decline in the target s variance and any significant changes in the rest of the moments (the acquirer s variance, the acquirer-target covariance, or the acquirer-target correlation) during the pre-announcement period. Anticipating the payment form requires the correlation and (or) the covariance to shift upwards in the case of equity offers and shift towards zero in the case of cash offers. This paper contributes to the M&A literature by documenting anticipation shifts. Using a sample of 125 completed acquisitions of U.S. public companies between 2003 and 2006, the proposed approach finds that a takeover is anticipated on average 187 trading days before the announcement date in 86% of the deals. The market anticipates the payment form in 62% of the deals on average 123 trading days in advance. The anticipation dates that are identified are much earlier than in previous event studies (e.g., Schwert, 1996) that find only the target (and not the acquirer) firms to be anticipatable, and just two months prior to the announcement date. The results suggest that breaks in the variance-covariance structure reveal significant information about future M&As. Moreover, splitting the sample based on the offered payment form (54 cash, 33 equity, and 38 mixed-payment deals) indicates that the market anticipates the cash offers almost at the same time as it anticipates the deals themselves, but on average there is a lag of 90 trading days in the equity and mixed subsamples. This evidence reveals that the announcement of cash (equity and mixed) offers contains the least (most) unexpected information for market investors. The market is less successful in anticipating mixed offers because this requires the forecasting of an additional parameter, i.e. the portion of cash in the total bid payment. Additionally, controlling the deal-anticipation and payment-form-anticipation dates in this paper provides new insights into the choice of payment method in M&As. 3 The first and secondorder moments of the target and acquirer stock returns are used as cross-sectional predictors in the payment-method regressions. Among them, the correlation plays an important role in explaining the choice because it is a proxy for the relatedness of the target and acquirer 3 Various competing hypotheses (the asymmetric information, tax consideration, capital structure, managerial control, and behavioral motives, among others) explain this choice. See, for instance, Martin (1996) and Betton, Eckbo, and Thorburn (2008) and the references therein. 2

4 businesses. However, previous findings are mixed. On the one hand, Houston and Ryngaert (1997), Officer (2004), and Bhagwat and Dam (2014) find that, if the stand-alone target and acquirer stock prices co-move (are highly correlated), offering equity as the medium of payment reduces the risk of an unfair merger, i.e., one in which the terms of the merger become unfair to one party at the deal completion date. On the other hand, Faccio and Masulis (2005), Rhodes-Kropf and Viswanathan (2004), and Aggarwal and Baxamusa (2013) argue that the acquirer should pay in equity when buying uncorrelated target firms if those targets have an informational disadvantage in that the acquirer s stock is overvalued. Consistent with the first of the above strands, I find that the greater the pre-merger correlation, the larger is the fraction of the acquirer s equity in the bid payment. However, adding new variables (i.e. the deal-anticipation and payment-form-anticipation dates) to this regression changes some of the previous results. Namely, the significance of the correlation disappears, the coefficient of the average target return becomes significant, and the coefficient of the target volatility is attenuated. The above evidence suggests that assuming the unpredictability of M&As in paymentmethod regressions may result in some endogeneity issues. First, the moments are usually estimated from the pre-announcement period, but the anticipation of M&As can affect them. The moments hence need to be estimated from a period in which the takeovers are totally unexpected (i.e. from the pre-anticipation segment); otherwise, they might be mismeasured, which in turn may lead to a measurement error in the regressions. In fact, the evidence here indicates that the pre-announcement target and acquirer variances are underestimated compared with those from the pre-anticipation period. Similarly, the covariance and correlation are underestimated in cash offers and overestimated in equity offers. Second, the target, acquirer, and acquisition characteristics are used as predictors in those regressions. My findings demonstrate that the anticipation variables are significantly correlated with some of those characteristics, making them relevant for inclusion in the regressions. In other words, excluding the anticipation variables may cause omitted-variable bias. My results indeed confirm that this is the case, at least for my sample of takeovers. I find robust support in the data for two theoretical factors (merging likelihood and expected synergy) that generate the anticipation shifts. Using a measure for the takeover probability (from Samuelson and Rosenthal, 1986), the merging likelihood of the target and the acquirer firms become significant for the first time during the pre-announcement period after the detected deal-anticipation dates, and then trend upwards. Moreover, the CAARs of both the target and acquirer shares also show positive and significant trends after those anticipation dates, suggesting that the market discounts part of the expected synergy of future M&As around those dates. These findings confirm the assumptions in the theoretical model of Betton et al. (2014), in which takeovers are anticipated when the market is informed by strong signals about potential takeovers with synergistic gains. I perform robustness tests in an attempt to rule out several alternative explanations of the results: (1) If those anticipation shifts are caused by firm-specific events (and not anticipation of M&As), then the portfolio theory would suggest that the variance-covariance structure of the 3

5 portfolios of merging firms should be stable due to the diversification effect. However, I also find similar shifts in the variance-covariance structure of the portfolios, indicating that firm-specific news is not the source of anticipation shifts in the individual bivariate return series. (2) If some market-wide events during the sample period of this paper cause those anticipation shifts, then similar shifts should also be observed in a benchmark sample (i.e., a random sample of non-m&a firms). The anticipation shifts occur much less frequently and in a less expected way in the benchmark sample than in the M&A sample, implying that the likely mechanism for the later shifts is anticipation of takeovers and not market-wide events. 4 I also apply cross-sectional analyses to determine the significant predictors that explain why some takeovers are anticipated and some not, and why some are anticipated earlier than others. I find that some of the cross-sectional characteristics of merging firms in the pre-anticipation segment are significant in discriminating anticipated from unanticipated deals and in explaining cross-sectional variation in the anticipation dates. While Meulbroek (1992) and Schwert (1996) consider leakage of information by insiders as the main source of M&A anticipation, my results propose another source: the existence of M&A anticipators in the market who use public information that is available prior to the deal-anticipation date to predict likely M&As. The rest of the paper is organized as follows: Section 2 reviews the merger arbitrage literature to identify expected shifts that signal the anticipation of takeovers. Section 3 presents the methodology and the anticipation hypotheses. Section 4 provides data and descriptive statistics. Section 5 documents the anticipation results. Section 6 discusses the robustness of the results. Section 7 reports the cross-sectional results, and finally Section 8 concludes. 2 Anticipation Shifts According to the efficient market hypothesis (EMH), all relevant information should be instantly incorporated into stock prices. Therefore, if the market anticipates a takeover and its payment method, this information should be reflected in the price process of the target and the acquirer stocks at the anticipation time. The merger arbitrage literature documents changes in the second-order moments of stock returns after the announcement of takeovers. If the market is efficient, one can expect to observe similar changes during the pre-announcement period when a deal and (or) its payment form are anticipated. The expected shifts that would indicate the anticipation of the deal and payment form can be summarized as follows: a significant decline in the target volatility indicates the anticipation of the target firm while any significant shift in the acquirer s volatility would imply the anticipation of the acquirer firm. The existence of a significant decline in the target s variance and of any significant changes in the rest of the moments (the acquirer s variance, the acquirer-target covariance, or the acquirer-target correlation) during the pre-announcement period signal the anticipation of the takeover. If the market anticipates equity offers, the covariance should 4 The anticipation shifts exist regardless of the use of alternative approaches. The alternatives include the following: three univariate tests as opposed to one multivariate test for detecting breaks in the variance-covariance structure, the use of raw instead of winsorized returns series, and the employment of returns from the preannouncement period rather than from both the pre- and post-announcement periods in the above tests. 4

6 increase and (or) the correlation should become one. In the case of cash offers, the covariance and (or) the correlation should converge to zero. A mixed-payment offer will follow the rule for equity (cash) offers if the equity (cash) is the dominant portion in the total bid payment. 2.1 Target Return Volatility Bhagat, Brickley, and Loewenstein (1987) report an anomaly in the relation between risk and return in targets stocks during the post-announcement period. They find significant declines in both the beta and the sample unconditional volatility of the target of a cash bid, but rising returns. Bhagat et al. (1987) propose that the price of a target share during that period is the sum of the value of common stock and the value of a put option. The target shareholders have the option to sell their shares to the acquirer firm in the post-announcement period. Bhagat et al. (1987) then show via option theory that a portfolio containing a stock and a put option has a lower standard deviation than the stock itself. This conjecture explains the observed decline in the target s volatility. Hutson and Kearney (2001, 2005) extend this analysis to the targets of both cash and equity offers and to completed and failed ones. They document that, regardless of the payment form and the final outcome of the pending bid-offer, the conditional volatility of the target s shares declines significantly after the bid announcement. The greatest (smallest) decline is in the cash (equity) subsample. Hutson and Kearney (2001) claim that this observation is due to a fundamental change in the price formation process, i.e. traders opinions about the value of the target stock converge during the post-announcement period. Overall, the decline in the target volatility regardless of the payment method and the offer outcome indicates an increase in the likelihood of a firm being the target of an acquisition. 2.2 Acquirer Return Volatility Various competing hypotheses have been developed about the post-acquisition (i.e., longrun) risk profile of the acquirers. First, the portfolio effect hypothesis predicts that the risk of the acquiring firm in this period is nothing more than the risk of a market-value-weighted portfolio formed from the shares of the two firms in the pre-announcement period (e.g., Langetieg, Haugen, and Wichern, 1980), thereby suggesting a decline in its risk. Second, the leverage effect hypothesis expects acquisitions that worsen the leverage (debt-to-equity ratio) of the acquirer firm to induce an increase in the risk of the consolidated firm (Hamada, 1972). Third, the integration risk hypothesis conjectures an increase in the risk of the consolidated firm if the acquirer s management is inefficient in merging the two firms into a single corporation (Bharath and Wu, 2006). Fourth, the merger wave effect hypothesis posits that industry shocks trigger inter-industry acquisitions (Mitchell and Mulherin, 1996; Harford, 2005). Since completed inter-industry acquisitions have stabilizing effects, this hypothesis predicts a decline in the risk of the consolidated firm. Finally, the synergy effect hypothesis predicts a decline in the acquirer s risk since the synergistic gains increase the value of the consolidated firm due to the following reasons: cost reductions due to economies of scale, economies of scope, more di- 5

7 verse corporate skills, more efficient redeployment of the combined assets, and enhanced market power, among others (e.g., Bradley, Desai, and Kim, 1988; Chatterjee and Lubatkin, 1990). Given the opposing predictions, the net impact of acquisitions on the risk of the acquirer firm is an empirical question. The empirical findings are indeed mixed. For example, Langetieg et al. (1980) show an increase in various unconditional risk measures of the acquirers in the post-acquisition period. They relate this result to aggressive management of the acquirer firms together with an increase in their leverage. Geppert and Kamerschen (2008) find a statistically greater implied volatility of the acquirer firms than the value predicted by the portfolio hypothesis in the post-announcement period. They argue that the integration risk and uncertainty about potential efficiency gains explain this result. However, after controlling for the systematic risk of target firms, Chatterjee et al. (1990) find the acquirers systematic risk to decline over both short and long horizons after the acquisitions due to synergistic gains. Hutson and Kearney (2005) find that the average unconditional volatility of the acquiring firms declines significantly after the announcements, but that the reduction is only significant in cash offers. Declining shifts are more consistent with the proposed anticipation mechanism in which a takeover is anticipated if the market perceives a synergistic gain from the merger. A decline in the acquirer risk during the pre-announcement period hence suggests that the diversification and synergy effects dominate the leverage and integration impacts. Overall, the opposing hypotheses and mixed evidence predict changes in the acquirer risk in different directions. Therefore, any change in the pre-announcement period is interpreted in this paper as an indicator of the anticipation of the acquirer firm. 2.3 Acquirer-Target Return Correlation Houston et al. (1997), Officer (2004), and Bhagwat et al. (2014) report a positive relation between the fraction of the acquirer s equity in the bid payment and the pre-announcement correlation. Bhagwat et al. (2014) propose that the risk of overpayment is reduced by an equity offer if the acquirer and target returns are highly correlated but the risk increases with weak correlation. This result suggests that shifts in the correlation can be used to anticipate the payment method in M&As. Subramanian (2004) constructs a theoretical model and provides empirical evidence in which the announcement of an equity offer causes the correlation to shift towards perfect correlation. He argues that if the market assigns a high likelihood to the success of a merger attempt, the two stock price processes must be perfectly correlated after the bid announcement since the acquirer offers a constant equity-exchange ratio to acquire each share of the target firm in an equity bid. This paper extends the work by Subramanian (2004) by documenting the behavior of the correlation around the announcements of cash and mixed-payment offers. When the likelihood of the success of a cash offer is high, the target volatility should drop significantly due to the offered bid premium. In this case, the two return processes will become uncorrelated. Moreover, the proportion of cash in a mixed payment will determine whether the correlation should shift towards zero or one. If the cash portion is dominant, the above argument predicts that the correlation should converge to zero. Otherwise, it should shift towards perfect correlation. 6

8 2.4 Acquirer-Target Return Covariance The behavior of the acquirer-target return covariance around the announcement of cash, equity and mixed-payment offers has not been studied previously. This paper does so for the first time. Both the covariance and the correlation measure the degree of covariation between the target and acquirer returns. However, standardization via individual volatilities in the denominator of the correlation can mean that useful variations necessary for anticipating the payment form are discarded. It is likely that the correlation will be stable, while the covariance will shift in response to the market s anticipation of the payment form. This suggests that shifts in the covariance can be used along with shifts in the correlation to detect the payment-formanticipation dates. Overall, the expected shifts in the correlation apply to the covariance as well. 3 Methodology and Anticipation Hypotheses First, each return series is prepared (i.e., adjusted for outliers, and for breaks in the mean, and then demeaned) before the structural break test is applied to investigate whether a deal and its payment form are anticipated or not. Second, if the test detects break(s) in the variancecovariance structure of a bivariate return series, the significance of shifts in its second-order moments are examined individually around each break date. Finally, an inference about the existence of anticipation dates is made based on the signs of the significant shifts. 3.1 Data Preparation The daily log returns (henceforth, the returns) of the acquirer and target stocks are used in all the tests in this paper and are computed in the following way: ( ) Pi,n,t r i,n,t = ln, (1) P i,n,t 1 where i =Acq, Trg; n is the index for the deals in the sample; t=(-379,..., 0, 1,..., C ) is the daily subscript; r Acq,n,t and r T rg,n,t represent the realized returns to acquirer and target shareholders involved in deal n on day t; and P Acq,n,t and P T rg,n,t are their adjusted closing prices on day t. Similarly to in Schwert (1996), the sample observation period for each of the target and acquirer return series starts 379 days prior to the announcement date (t = 0) and ends at the delisting date of the target s shares, which is C days after the announcement. 5 The pre (post)-announcement period runs from Day -379 (Day 0) to Day -1 (Day C ). The largest absolute returns of each series are identified and winsorized at 99%. The mean of the returns should be stable during the whole observation period in order that breaks in the second-order moments can be identified. The structural breaks methodology developed by 5 The cross-sectional regressions (e.g., Amihud et al., 1990; Hasbrouck, 1985; Martin, 1996) use predictors from the last year to discriminate merging from non-merging firms. Starting the sample observation period at Day -379 (one and a half years before the announcement day) here corresponds to the idea in those papers that the relevant information about future M&As can be revealed one year in advance. 7

9 Bai and Perron (1998, 2003, 2006) detects shifts in the mean of each return series, each series then being adjusted for the detected shifts. See the Data Preparation section in the internet appendix for further details about the adjustment of outliers and the structural break test. All tests used in this paper assume the sample mean of each return series to be zero. The following transformation is therefore applied to each series: where r i,n,t = C C t= 379 the mean-adjusted return series. r i,n,t = r i,n,t r i,n,t, (2) r i,n,t is the sample mean of the observed return series and r i,n,t is 3.2 Detecting Breaks in the Variance-Covariance Structure I use a cumulative sum (CUSUM) type test proposed by Aue, Hörmann, Horváth, and Reimherr (2009). This test is suitable since it does not impose any normality or parametric assumptions, which are usually assumed in parametric and cross-sectional tests. However, it requires the finiteness of the fourth sample moment of the bivariate series. The 99% winsorization is thus necessary and used to fulfill this requirement. The appealing feature of CUSUM-type tests is their ability to use a non-parametric HAC (heteroskedasticity- and autocorrelationconsistent) type estimator to capture the dependence structure in the data. As recommended by Aue et al. (2009), the Bartlett estimator is used as a proxy for the asymptotic covariance matrix in the testing procedure. 6,7 Since shifts in the variance-covariance structure can occur in both the pre- and postannouncement periods, I use the multiple break detection version of Aue et al. (2009), which is based on the binary segmentation approach. When this test detects a significant break date, it is reapplied separately across the two new segments that are obtained from splitting the data into two subsamples around that break date. This approach ends when the test can no longer detect any significant breaks in the new segments. Henceforth, the identity subscript of the deals is excluded from the notation for simplicity, but the following procedure is applied for each deal in the sample. Let (y 379,..., y 0,..., y C ) be a sequence of two-dimensional random return vectors over the sample observation period of a deal. For example, y 379 = ( racq, 379 r T rg, 379 is the bivariate vector of mean-adjusted realized returns for the target and acquirer shareholders 6 Berkes, Horváth, Kokoszka, and Shao (2005) show that the Bartlett estimator to be a consistent estimator of the asymptotic covariance matrix since it converges almost surely. 7 The data-dependent approach of Newey and West (1994) is also applied to determine an optimal truncation lag in the Bartlett estimator. Rodrigues and Rubia (2007) show this truncation lag to be able to improve the finite-sample performance of CUSUM-type tests. ) (3) 8

10 on Day ( σ 2 Acq, 379 ) ( ) Cov(y 379 ) = σ 2 AcgT rg, 379 σ 2 T rg, 379 = r 2 Acq, 379 r Acq, 379. r T rg, 379 r 2 T rg, 379 (4) is the realized variance-covariance matrix of the bivariate returns on Day -379, which contains the realized target and acquirer variances and their realized covariance on that date. The test statistic of Aue et al. (2009, p. 4050) detects structural breaks in the variance-covariance structure of a bivariate return process by examining the following hypotheses: H 0 : Cov(y 379 ) =... = Cov(y C ) (5) H A : Cov(y 379 ) =... = Cov(y K1 ) Cov(y K1 +1) =... = Cov(y K2 )... =... Cov(y Km+1) =... = Cov(y C ). (6) While the null hypothesis indicates the constancy of the variance-covariance structure of a bivariate return series in the sample observation period, the alternative allows several changepoints (m 1). The test itself identifies the unknown number of change-points (m) and their unknown locations (-379<K 1 <... <K m <C ). I do not impose any restrictions on these unknown parameters so as to capture all informational events of each bivariate series. The merger arbitrage literature usually compares the cross-sectional average pre- and postannouncement second-order moments by assuming the announcement day to be the only break date (e.g., Bhagat et al., 1987; Hutson and Kearney, 2001; Subramanian, 2004). To the best of my knowledge, Jayaraman, Mandelker, and Shastri (1991) is the only study that documents predictability for the target firms by assuming three uniform break dates in their implied volatility during the pre-announcement period. Gelman and Wilfling (2009) apply a Markov-switching GARCH model to detect shifts in the conditional volatility of target stocks during the postannouncement period. However, the use here of the test provided by Aue et al. (2009) extends the above studies because it relaxes any assumptions about the number and location of break dates. 3.3 Tests for Equality of Second-Order Moments around the Break Dates Since Aue et al. s (2009) test is a joint test, it does not identify which of the secondorder moments changes significantly after each break date, knowledge of which is necessary in order to examine the anticipation hypotheses. Therefore, I perform tests for the equality of those moments to determine significant shifts after each break. Since the sample second-order moments are locally stationary in each segment, I use them as estimates for the population moments across the detected segments. Suppose, e.g., that the joint test finds one break in the variance-covariance structure at time K 1. Then the estimates of the target s variance in the 9

11 pre- and post-break segments are ˆσ 2 T rg,pre = ˆσ 2 T rg,post = 1 (379 + K 1 ) 1 (C K 1 ) K 1 t= 379 C t=k 1 +1 r 2 T rg,t (7) r 2 T rg,t. (8) Other sample second-order moments (i.e., acquirer variance, covariance and correlation) in the pre- and post-break segments are estimated in a similar way. The modified version of Levene s (1960) F-test, proposed by Brown and Forsythe (1974), examines whether the sample variance is the same across two adjacent segments. This test is proper since it uses the median instead of the mean in computing the absolute deviations, and thereby robust against non-normality. 8 For example, this test examines whether the target variance is constant around a break date: H 0 : σ 2 T rg,pre = σ 2 T rg,post (9) H A : σ 2 T rg,pre σ 2 T rg,post. (10) The F-statistic is always positive, but the covariance can be negative. Thus, the absolute value of the covariance is used here for computing the common F-test for the equality of covariances across two adjacent segments. However, using absolute values can lead to a more conservative test in detecting heterogeneity when the sign of the covariance is different in the two segments. The number of significant changes might hence be underestimated if the absolute values are small. Jennrich s (1970) test examines the equality of sample correlations across two adjacent segments: H 0 : ρ AcqT rg,pre = ρ AcqT rg,post (11) H A : ρ AcqT rg,pre ρ AcqT rg,post, (12) where ρ AcqT rg,pre and ρ AcqT rg,post are the sample acquirer-target return correlation in the preand post-break segments. 3.4 Anticipation Hypotheses Previous studies document that information about the pair of firms (e.g., the size of the target relative to the size of the acquirer) explains the choice of payment method in M&As. Thus, first, the deal anticipation should be examined and then the payment-form anticipation. If the deal is unanticipated, the procedure for detecting the payment-form-anticipation date is abandoned. Moreover, the payment form can be anticipated either at the deal-anticipation date or at a later date during the pre-announcement period. 8 Lim and Loh (1996) compare seven existent tests for the equality of variances in a simulation exercise and find that the modified Levene test is the most powerful. 10

12 The following conjectures help to interpret the results of the structural break test, and in turn to detect the anticipation dates. First, the existence of multiple consistent breaks will indicate that expectations about the potential deal are updated sequentially. Such breaks will occur due to the release of new information about the deal, the reinterpretation of existing information, and (or) the reassessment of the perceived synergy of the deal and its division between the target and acquirer shareholders. Second, assuming a threshold of 50% shows whether a mixed offer is anticipated or not. Thus, if more than 50% of a deal value is paid in cash (equity), then cash (equity) is the dominant payment form in the mixed-payment offer. Table 1 summarizes possible outcomes of the test (by Aue et al., 2009), which are inferred to identify break date(s) consistent with the deal and payment-form-anticipation hypotheses. First, if there is no break during the pre-announcement period of a bivariate (target-acquirer) return series, then the deal and its payment form are unanticipated. Furthermore, non-m&a reasons can increase the volatility of the target shares. Thus, any increasing shift in this moment will lead to a break date not being recognized as a deal and payment-form-anticipation date, regardless of the observation of expected shifts in other moments (see, e.g., outcomes 2 and 5 in Table 1). Insert Table 1 here Second, suppose that the test detects only one break date (K 1 ). If at least one of the second-order moments (other than the target variance) changes significantly after this break, then K 1 is the deal-anticipation date (outcomes 3 and 4). A decline in the target variance may only indicate that the market anticipates the target firm and not the deal (which is a pair of target and acquirer firms). Therefore, in addition to a declining target variance, the existence of at least one significant shift in another moment is required for a break to be considered as the deal-anticipation date. The deal-anticipation date is the first candidate for assessing the payment-form-anticipation hypothesis. If at least one of the covariance and the correlation shifts significantly towards its expected level after this date, then K 1 is also the payment-form-anticipation date. An expected shift would be a rise in the case of equity offers, or a shift towards zero in the case of cash offers. A mixed offer would follow the rule for an equity (cash) offer if equity (cash) made up the dominant portion of the total payment. Finally, let us assume for simplicity a two-break case (K 1 and K 2 ), though this can easily be generalized to a case with more than two breaks. In the multiple break case, the procedure starts from the closest break to the announcement date (K 2 ) since it captures the most recent expectations of the market about the M&A activity of the pair of firms. If there are any inconsistent shifts (e.g., increasing target variance), then the deal and its payment form are inferred to be unanticipated (outcomes 5 and 6). However, if K 2 contains consistent shifts, then the anticipation procedure considers the farther break (K 1 ) as well. If there are consistent break(s) around this date, then the market starts to anticipate the deal and (or) its payment form from the date of the distant break, i.e., K 1 (outcomes 9 and 10), otherwise the deal and (or) its payment form are anticipated at K 2 (outcomes 7 and 8). Moreover, a reversal in 11

13 the co-movement moments across breaks can cause the deal-anticipation and payment-formanticipation dates to differ, and can even cause the payment form to be unanticipated (if an inconsistent break is detected at the closest break date). If the dates are different, then the deal is anticipated (K 1 ) earlier than the payment form (K 2 ), e.g., outcomes 11 and 12. The choice of payment method at the acquisition time can differ from the anticipated one. If so, then the payment form is found to be unanticipated (anticipated incorrectly). For instance, according to possible outcomes 3 and 9 (7 and 12), the market anticipates at date K 1 (K 2 ) that the acquirer will use its own shares to finance the acquisition, but if instead it uses cash at the acquisition time, then the payment form will have been unanticipated. This example shows cases in which the deal is anticipated but not its payment form. Section 6.1 below provides an example of how the above procedures are performed so as to identify the anticipation dates. 4 Data and Descriptive Statistics 4.1 Sample Selection I sample takeovers from the Bureau Van Dijk Zephyr database using the transaction form of merger or acquisition. The sample consists of all completed acquisitions between U.S. publicly listed target and acquirer firms. The sample period is from June 2003 to June 2006, which corresponds to the sixth M&A wave in which both the equity and the takeover markets are stable (Martinova and Renneboog, 2008). Macroeconomic and industry factors are less likely to generate breaks in this period than is the release of information about firms (e.g., takeover transactions). This definition leads to 1647 deals being identified. Table 2 summarizes how this sample is then reduced based on the following filters. Insert Table 2 here I focus on the sample of deals in which the following hold: (i) An acquirer gains entire control of a target firm by acquiring 100% of the target shares in one takeover transaction. The most relevant cases in which to verify the anticipation mechanism are those in which the ex-ante merging likelihood is trivial. I thus exclude acquisitions in which an acquirer has acquired some stake in the target firm before the current bid, because previous bids (of even a minority or toehold stake) can raise the likelihood to a significant level. (ii) The method of payment is allcash, all-equity or a mixture of cash and equity payments. (iii) A bid offer takes between 19 and 253 trading days from its first announcement date to complete. According to the William Act of 1968, only bid offers for subsidiaries of U.S. public targets or private targets can be completed in a shorter period (Officer, 2004; Bhagwat and Dam, 2014). The daily prices (and thus the returns) are usually unobservable for these firms, and so the anticipation mechanism would be unverifiable for these types of deals. (iv) The deal value exceeds $50 million. (v) Neither the acquirer nor the target firm is a bank since the latter are highly leveraged and subject to different regulations. The noise that would be added to the estimation of the cross-sectional regressions due to their different characteristics is the primary reason for ignoring banks. (vi) An acquirer has only one bid record in the sample period. If an acquirer has multiple bids, a 12

14 consistent break would not be identifiable as the anticipation break of a specific bid, as it might be related to another bid made by that acquirer. (vii) The target has a stock price exceeding $2 on Day -42. Schwert (1996) argues that the returns on low-priced stocks could be imprecise as they are more heavily exposed to frictions in the market microstructure. (viii) Both firms have more than 120 adjusted daily-closed stock prices during the pre-announcement period in Thomson Financial DataStream. The final sample contains 125 deals with enough return data, and splits into 54 all-cash, 32 all-equity and 38 mixed-payment deals. This sample size might be considered small compared to the sample of the usual takeover study. The reason is that more restrictive filters are used here to construct the sample, which is necessary for testing the anticipation hypotheses. For example, filter (i), requiring bids for a 100% stake of the target, reduces the sample significantly by 1120 deals, and other studies do not impose this requirement. The target and acquirer firms accounting and deal information are retrieved from Thomson Financial DataStream and the Bureau Van Dijk Zephyr database. The internet appendix provides further details of the data analysis, including the definition of the variables for the cross-sectional analysis, descriptive statistics of the deal, target and acquirer characteristics, the sample higher-order moments before and after winsorizing the return series, and the results for the breaks in the mean returns. 4.2 Summarized Statistics of Sample Second-Order Moments Table 3 provides descriptive statistics for the sample second-order moments of the acquirer and target returns in the pre- and post-announcement periods. I use the sample moments and not the realized moments in order to compare the results with those of previous studies. The cross-sectional average (median) annual target volatility for the full sample declines by a highly significant 44.1% (47.1%), from (0.463) to 0.27 (0.245), from the pre- to the post-announcement periods. 9 This average decline is similar to Hutson and Kearney s (2001) report of an average decline of 46% after the bid announcement. While the analysis of variance (ANOVA) indicates that the average target volatility is comparable across payment subsamples during the pre-announcement period, its large F-value (16.28) for the post-announcement period suggests the opposite. Consistent with Hutson and Kearney (2001), the largest (smallest) risk reduction is seen for the targets of cash (equity) offers. Overall, the announcement of a takeover significantly reduces the target s volatility regardless of the payment method offered. This is consistent with that part of the deal-anticipation hypothesis that partially anticipates a takeover due to a significant decline in this moment. Insert Table 3 here The decline in the average acquirer volatility is significant at the 1% level. It changes from to between the pre- and post-announcement periods. Again, this average decline 9 I will report the volatilities (and not the variances) throughout the paper for comparability of the results with previous literature, and annualize all moments (over 252 trading days) to match them with the annual cross-sectional data used in the regressions. 13

15 (17.4%) is similar to the 21% drop reported in Hutson and Kearney (2005). This average reduction in the risk of acquirers after bid announcements is less than that found for the targets (44.1%) since the above-mentioned opposing factors affect the acquirer s volatility. Moreover, the average reduction is significant across all payment subsamples, ranging from 15% to 19%. Overall, the synergy and diversification effects dominate the leverage and integration effects in this sample, such that the bid announcements lead to a decrease in the acquirers volatility. The difference between the average post- and pre-announcement covariance is insignificant, which is surprising at first glance. However, the unreported results of the individual equality tests indicate that the covariance changes significantly in 110 out of 125 (88% of) deals after the announcement. The decreasing and increasing changes cancel each other out in this case, causing the average change to be insignificant. Furthermore, the significant ANOVA results (with F- values of 5.72 and 22.93) imply that both the average pre- and post-announcement covariance differ significantly across payment subsamples. The firm-pairs that are more linearly dependent during the pre-announcement period use more equity as the medium of exchange. Moreover, consistent with the payment-form-anticipation hypothesis, announcing an equity (cash) offer significantly raises (declines) the average covariance by 65.7% (99.1%), from (0.029) to 0.11 (0.0003), from the pre- to the post-announcement periods. Overall, the covariance changes significantly after the bid announcement, dependent on the offered payment method. Consistent with Bhagwat et al. (2014), I find that the more correlated is the pair of firms in the pre-announcement period, the more shares are used by the acquirer to finance the M&A (F-value of 8.32). Announcements of equity offers significantly increase the average correlation from to This result supports the evidence in Subramanian (2004) in which it is proposed that the correlation of firm-pairs involved in equity deals should shift upward after the announcement. The average correlations decline significantly by 84.4% (from 0.17 to 0.026) after cash announcements. Similarly to in the covariance case, the results for the correlation are consistent with the expected changes for the anticipation of the payment method. All in all, the sample post-announcement moments are different from the pre-announcement ones. The differences are consistent with the previous findings in the literature, showing this sample to be comparable with previous ones. Moreover, the announcements move the secondorder moments towards the aforementioned expected levels. 5 Evidence on the Anticipation of Deal and Payment Form 5.1 Breaks in the Variance-Covariance Structure Table 4 and Figure 1 summarize the results of performing Aue et al. s (2009) test on the 125 bivariate return series. There is at least one significant break in 87.2% of the deals (109 out of 125) and multiple breaks in 60% of them (Table 4). However, this test does not detect any breaks in 16 of the deals, indicating that they are unanticipated. The market may not perceive a substantial synergy in their mergers; otherwise there should be shift(s), at least after their announcements. Insert Table 4 here 14

16 The findings imply a significantly skewed location of breaks towards the pre-announcement period, since 108 (out of the abovementioned 109) deals have shifts in this period (Table 4). Moreover, Panel A of Figure 1 illustrates a considerable variation in the distribution of break dates during the sample observation period; however, its mass is located in the pre-announcement period. The merger arbitrage literature assumes a break only on the announcement date of each bid, and the spike around the announcement (Day 0) in Panel A of Figure 1 supports this assumption. However, 27 out of 28 deals with break(s) during the post-announcement period indeed have break(s) in the pre-announcement period as well. The existence of these additional post-announcement breaks indicates that the market only revises its pre-announcement predictions after the bid offers have been made. If we assume that a shift in the variance-covariance structure reveals some relevant information about M&As, the results here indicate that much of the information is leaked during the pre-announcement period. Insert Figure 1 here 5.2 Deal Anticipation Table 5 presents the deal-anticipation results. I find that 108 deals are anticipated. The rest of the deals (17) are unanticipated due to the absence of breaks in the pre-announcement period. The anticipated deals consist of 46 cash, 30 equity and 32 mixed-payment deals. The fraction of anticipated deals is similar across the payment subsamples, implying that a deal is anticipated regardless of the payment form offered. Overall, the existence of consistent shifts in a large fraction of deals (86.4%) indicates that anticipation is a key characteristic of the deals in this M&A sample, and that the anticipation mechanism is successful in identifying them. Insert Table 5 here Deal-Anticipation Dates A deal is anticipated on average 187 trading days prior to the announcement day (Table 5). The interquartile range of the deal-anticipation dates is between Day -252 and Day -133, implying that 75% of deals are anticipated at least six months prior to their announcement. Panel B of Figure 1 demonstrates relatively similar distributions of deal-anticipation dates across the cash, equity and mixed subsamples. A very small F-value of the ANOVA test (0.05) in Table 5 also indicates the similarity of the average anticipation dates across payment subsamples. This leads me to conclude that the market starts to anticipate a potential takeover on average almost nine months prior to the announcement date. This result indicates that takeovers are anticipatable much earlier than has been documented in the previous event studies (i.e., two months prior to the announcement as reported by Schwert, 1996) Sample Moments around Deal-Anticipation Dates The unreported results of the equality tests indicate that a significant decline in the acquirers volatility (in the correlation) leads to the anticipation of five deals (one deal). The majority 15