«Citation for published item: Andreou, P. C., Louca, C. and Savva, C.S. Short-horizon event study estimation with a STAR model and real contaminated ...»
Short-Horizon Event Study Estimation with a STAR Model and
Real Contaminated Events
Citation for published item:
Andreou, P. C., Louca, C. and Savva, C.S. Short-horizon event study estimation
with a STAR model and real contaminated events. Review of Quantitative
Finance and Accounting, forthcoming.
View online & further information on publisher’s website:
Short-Horizon Event Study Estimation with a STAR Model and Real Contaminated Events Panayiotis C. Andreoua,b, Christodoulos Loucaa,b and Christos S. Savvaa,c a Department of Commerce, Finance and Shipping, Cyprus University of Technology, 3036 Lemesos, Cyprus b Durham University Business School, Mill Hill Lane, DH1 3LB, Durham, UK c Centre for Growth and Business Cycle Research, University of Manchester, Manchester, United Kingdom This version: February 2015 Abstract We propose a test statistic for nonzero mean abnormal returns based on a Smooth Transition Auto Regressive (STAR) model specification. Estimation of STAR takes into account the probability of contaminated events that could otherwise bias the parameters of the market model and thus the specification and power of the test statistic. Using both simulated and real stock returns data from mergers and acquisitions, we find that the STAR test statistic is robust to contaminated events occurring in the estimation window and in the presence of eventinduced increase in return variance. Under the STAR test statistic the true null hypothesis is rejected at appropriate levels. Moreover, it exhibits the highest levels of power when compared with other test statistics that are widely and routinely applied in short-horizon event studies.
JEL classification: G14; G34 Keywords: Event studies, test statistics, contaminated events, Markov switching regression model, Smooth Transition Auto Regressive model.
Panayiotis Andreou Email: email@example.com Cyprus University of Technology Dept. of Commerce, Finance and Shipping 115, Spyrou Araouzou Street 3036 Lemesos, Cyprus Tel.: +357 25002286 Fax.: +357 25002816 Electronic copy available at: http://ssrn.com/abstract=2516068
1. Introduction The short-horizon event study method, introduced in the seminal work of Fama, Fisher, Jensen and Roll (1969), has been one of the cornerstones of financial economics and accounting in the last few decades. Ever since, numerous research papers have endeavoured improvements in the basic empirical methodology.1 The most notable recent attempts focus on the introduction of test statistics for nonzero mean abnormal returns that are robust to event-induced increase in return variance (see, for instance, Harrington and Shrider 2007;
Kolari and Pennönen 2010 and references therein).2 More recently, Aktas et al. (2007a) emphasize the need to consider event study methods that mitigate the effect of contaminated (unrelated) events arising from corporate actions and announcements that may occur during the estimation window. It is reasonable to expect that company press releases or leakage of private information occurring in the estimation window could create cross-sectional variation in the abnormal returns. This would inevitably bias the estimation of the (true) returngenerating process parameters, in particular, the estimated variance of the parameters which could deteriorate the detection of abnormal performance in the event window.
In this study, we sought to make a dual contribution to the empirical corporate finance research. First, to tackle the estimation window contaminating-event problem, we estimate the widely applied event study market model as introduced by Sharpe (1963) by relying to regime switching approaches as a general method, highlighting at the same time the There is a vast amount of theoretical and empirical studies in the research realm of this topic such the ones of Ball and Torous (1988), Corrado (1989), Boehmer et al. (1991), Salinger (1992), Savickas (2003), Dombrow et al. (2000), Cyree and DeGennaro (2002), Harrington and Shrider (2007), Ahern (2009), Campbell et al. (2010) and Kolari and Pynnönen (2010), among others. The landmark work in this topic is that by Brown and Warner (1980, 1985) who investigate the specification (Type I error – rejecting the null when it is true) and power (Type II error – failing to reject the null when the alternative hypothesis is true) of several modifications of the shorthorizon event study by assuming that abnormal returns are intertemporally uncorrelated and there is no (significant) impact of event-induced variance.
Abnormal return (AR) is defined to be the difference between the actual return that is observed during the event day(s) (namely, the event window) and the expected return which is provided from a return-generating model estimated using stock returns data that precede the event (namely, the estimation window). Event-induced increase in return variance occurs when variance in the event window exceeds the variance over the estimation window; as a consequence, test statistics that ignore plausible implications of unexplained variation in true abnormal returns for the structure of heteroskedasticity may fail to detect event-related abnormal performance.
Electronic copy available at: http://ssrn.com/abstract=2516068 importance of the Smooth Transition Auto Regressive (STAR) specification; this is the first time that the STAR model is utilized to compute an event study test statistic for the detection of nonzero mean abnormal returns. STAR can be viewed as a statistical method that filters out firm-specific events that could otherwise induce unduly variance in the model’s generated returns (see Hansen 2011 for a detailed review in threshold autoregressive models and their applications). STAR is a regime-switching model that allows for two regimes, associated with extreme values of the transition function, where the transition from one regime to the other is done in a smooth or abrupt manner (see Terasvirta and Anderson 1992; Terasvirta 1994, among others). We employ the STAR method to better model the stock returns data generating mechanism by taking into account the probability of the occurrence of unrelated firm events. In this respect, estimated parameters of the market model should be less subject to the influence of contaminating events in the estimation period compared to more traditional rival choices. Further, the findings of this study highly support the application of the STAR specification in the event study framework since it fits the empirical stock returns data generating process much better than any other rival method. Thus, the STAR event study test statistic could allow the researcher to conduct valid large scale statistical analysis of abnormal returns that are much better specified, as well as more powerful in detecting the (true) size of the abnormal returns around a particular corporate announcement.
Second, we consider biases arising in short-horizon event study of test statistics for nonzero mean abnormal returns using real data from Mergers and Acquisitions (M&As) for the period 1980-2010. The majority of prior literature focuses on providing analytical and empirical evidence of the resulting test statistics biases using randomly selected firms with simulated induced contaminated events during the estimation period (e.g., Aktas et al. 2007a;
Harrington and Shrider 2007). Nevertheless, by carrying out specification and power tests on estimation periods using simulated returns data may not always be representative enough of the resulting cross-sectional variation in abnormal returns emerging from eventcontamination that is taking place in real situations. In the context of M&As, which by selection reflects a non-random sample (e.g., Fuller et al. 2002; Bhagat et al. 2005; Guo and Petmezas 2012), it is quite frequent for bidding firms to engage in several other unrelated corporate activities (e.g., earnings announcement, changes to dividend policy, etc) in the period preceding the deal announcement (see also, Bhagat et al. 2005; Aktas et al. 2009). In a vast amount of M&A deals, the number of major corporate events that may emerge in the estimation window is rather high: for example, Fuller et al. (2002) study a M&As sample where bidding firms complete bids for five or more targets within a three-year window.
Moreover, the nature and duration of an event may not be captured accordingly with simulated contaminated stock-return series. Company press releases or leakage of private information usually happens rather close to the event announcement (e.g., Harrington and Shrider 2007; Kolari and Pynnönen 2010; Guo and Petmezas 2012). Therefore, any contaminated events are more likely to cluster in a non-random manner in the period just before the announcement day (implying, for instance, the need of a right-skewed distribution for capturing the arrival of news in the market). Lastly, the widely-adopted simulation approach of Brown and Warner (1980, 1985) does not consider the possible endogeneity of announcement decisions in the presence of private or market information. Harford (2005), for example, documents that M&As cluster in time due to the market timing of industry shocks.
There is also evidence to support that market and firm-specific news-releases (Délèze and Hussain 2014; Laopodis 2010) and contagion effects caused by financial crises (Kenourgios et al. 2013) cause simultaneous price movements to different asset classes across different markets. This increases significantly the likelihood to observe great overlap on the event dates which could introduce contemporaneous correlations in the abnormal returns leading to incorrect inferences regarding the detection of abnormal returns.3 All abovementioned cases may be rather compelling to be modelled properly under the simulated data set environment, especially when corporate event announcements occur in extreme market conditions where, for instance, firm-specific mean stock return or volatility are particularly high. Therefore assessing any aberration between simulated and real events that could impact the specification and power of event study test statistics remains an open question which we investigate in this paper.
From the stand-point of both the academic researcher and the investment practitioner, it is crucial to know which cross-sectional abnormal return test statistic(s) should be employed to make inferences; this is especially important for many corporate events such as M&As that represent huge deals and associate with enormous market dollar values.4 To better answer this question under a general viewpoint, we undertake a horserace of returngenerating process and statistical tests. We identify the one(s) that are best suited using both, simulated stock returns in accordance to the traditional approach (Brown and Warner 1980 and 1985; Boehmer et al. 1991), as well as real stock returns coming from M&A deals. We focus on the most prominent cases of methods proposed so far in the literature. First, we include approaches that have been extensively used in prior empirical studies like the standardized cross-sectional test proposed by Boehmer et al. (1991), RANK test proposed by Corrado (1989) and GARCH test proposed by Savickas (2003). Second, we consider methods having greater flexibility to mitigate the presence of firm-specific (unrelated) events that precede corporate announcements like the BETA-1 approach, the two-state Markovswitching market model test (TSMM) proposed by Aktas et al. (2007a) and the STAR test Cam and Ramiah (2014) also discuss the possibility that researchers may reach different results depending on the financial econometrics adjustments and asset pricing model used when calculating expected returns.
Although we focus our analysis on M&As as a major corporate event, our inferences could easily be generalized for any other corporate decisions that exhibit similar market performance, such as season equity offerings, share repurchase, goodwill write offs, cross-listings, etc.
statistic introduced in this study. For the TSMM, we do not only consider two-state variance regimes as in Aktas et al. (2007a), but we also investigate two-state regimes in the market model parameters (i.e., stock return mean regression equation).5 Unlike their peers, the regime-switching family comprised by the TSMM and STAR models, postulate an adaptation of the event study methodology that automatically takes into consideration the probability of contaminated events that could otherwise bias the estimated parameters of the market model (affecting in this way the specification and power of test statistics). Therefore, they should exhibit superior robustness to event-induced increase in return variance caused by the crosssectional variation in the effects of a firm-specific event occurring in the estimation window.
In this respect, our null expectation is that test statistics computed from the regime-switching family, in particular from the STAR model, would perform significantly better than their rival methods.
Our results show that the traditional test statistics employed by prior researchers in the short-horizon event study are mis-specified and exhibit weak statistical power in the presence of contaminating events in the estimation period, especially in the presence of event-induced increase in return variance. On the contrary, the STAR methodology introduced in this study is the best choice since it is resilient in any type (simulated or real) of firm-specific contamination that may occur; moreover, its statistical power is high even under severe event-induced increase in return variance. The Markov switching regression models as proposed in Aktas et al (2007a) are found to be the second best choice, although not too behind from the performance of the STAR model. Nonetheless, with real stock returns subsamples which present either extreme mean stock returns or extreme stock return volatility, we find a clear superiority of the STAR method all over other rival ones. Previous research Letting the market model parameters being regime-dependent allow a more realistic representation of the return-generating mechanism since prior empirical research has revealed a significant time-variation in the slope parameter which depends on rising and falling market conditions (Hays and Upton 1986; Klein and Rosenfeld 1987; Chang and Weiss 1991; Chiang et al. 2013).