Event studies have been accepted among those empirical techniques, which are widely used both in accounting and finance. Barakat & Terry (2009, p. 2) note that the studies have been useful as a methodology in the assessment of how an event would affect the kind of returns that a firm would be able to gain from the common price of its stocks. Their design is normally done in a way that enables the detection of any abnormal changes in the financial assets during a specific period of time, as related to specific events. However, because of the many controversies that accompany its use, it has been argued that any research using the event-study technique must have knowledge of an abnormal performance for any given setting within an institution. Irrespective of full development and the wide use of event study methods in the testing of the various financial theories, concerns have been raised as to whether the method is efficient enough and, therefore, its reliability.
The methodology has, thus, become the standardized methods of measuring the extent to which the prices of securities will react to a given event or even an announcement. There is, thus, vast information that has accumulated over the time enriching the field of financial economics. This has since made it easy to understand several issues that are considered basic to understanding corporate finance. According to Barakat & Terry (2009, p. 2), event studies have been utilized in testing the null hypothesis that information is efficiently incorporated by the market. It has also been used in determining the extent to which an event would affect the wealth of the security holders of a specific firm given that the market has efficiently incorporated the information that is available to the public as at that time. This points out to its usefulness in the corporate world as it enables the analysts to understand the policy decisions taken by the corporate organizations.
According to Kothari and Warner (2004, p. 4), event studies also help in the testing of how efficient the capital market is. They note that whenever a particular corporate event causes the persistence of a systematic security returns that is nonzero, it is an indication of lack on consistence in the efficiency of the market. They further add that the application of event studies is never limited to the financial economics. They gave an example with the attention that the impact of the announcement of earnings the prices of stocks has continued to receive in accounting literature. Similarly, Sundin & Horowitz (2003, p. 870) note that the study has also been widely applied in law and economics where they help in examining how regulation affects the corporate market as well as the kind and extent of damages that result from the legality of different liability cases.
Irrespective of the identified weaknesses, it is good to note that event studies are widely accepted as one of a well established and mature study in terms of the available literature. Over 500 studies have been done on this field and results published (Seiler, 2000 p. 2). There is also a good level of understanding on the methods used in the study with the increasing amount of study now focusing on the methods’ statistical properties. Researchers who seek to use this method can therefore readily obtain message on when and how to appropriately use the study. However, there is still some controversy with some scholars pointing out certain problems related to the event induced variance studies, which they insist must be solved if the method is to be efficient. Such problems are related to method’s statistical properties (Kothari and Warner, 2004, p. 5). This study seeks to identify the problems related to the event studies, while also providing a review of literature on how these problems can best be solved.
Critiques of the Problems of Event Induced Variance on Event Studies and their Solutions
According to Barakat & Terry (2009, p. 2), whenever the method of event study failed in the consistency of its reliability, any condition that might result not into its invalidation must be ruled out. He, however, cautions that this is not to mean that every study involving event methodology has errors, but instead notes that under given conditions, the test is likely to give wrong results based on its reliance on the CAR methodology as given by Brown and Warner. Barakat & Terry (2009, p. 2) argue that this problem results from the kurtosis and the effect of clustering, which is common with the event-study portfolios’ returns. The study will therefore proceed to identify and critique the specific problems associated with this technique, while proposing their possible solutions.
Methodology Related Problems
Sundin & Horowitz (2003, p. 872) observe that even though literature has continued to show that there is no change that has occurred in the event studies’ statistical format, there are two main methodological changes that have occurred generating certain controversial debates. First is the common usage of daily data on security returns instead of the initially used monthly data. This is said to have allowed an increased level of preciseness attained in measuring of abnormal returns as well as permitting a more in-depth study of the effect of announcement on securities. Secondly, there has been an improvement on the methods used in the estimation of the abnormal returns and calibrating their statistical significance. Pynonen & Kolari (2010, p. 3397) add that this change has solved the problems that were initially being encountered in studying the long-horizon types of events. Its effectiveness was demonstrated by the 1990s research findings on the statistical properties related to returns on long-horizon security.
Even with these improvements on the long-horizon methods, the methodology is still faced with several limitations. According to (Kothari and Warner, 2004 p.5), any inference made from long-horizon test must be done with extreme cautions. Cowan & Sergeant (2001, p. 742) further emphasize that even the use of the methods aiming at analyzing the long-abnormal returns, which is perceived as the best method, can be treacherous in this case. Such revelations have triggered earlier warnings by some scholars on the doubtfulness of the reliability of the methods that are based on the long-horizon.
However, this problem does not present a major setback to the studies since it can be solved by promoting the use of the tests, based on the short-horizon methods. Kothari and Warner (2004, p. 6) argue that these tests are still the most efficient based on the fact that their evidences are the cleanest. This has solved the problems related to difficulty in interpreting results received via the long-horizon methods. They further note that apart from being vulnerable to the problems related to the joint tests, the long-horizon tests’ power is also very low.
Noting the generality of such statements, certain scholars have noted the need for a more meaningful premise upon which the usefulness of the event studies can be assessed irrespective of whether it is based on long or short horizon methods. For instance, Kothari and Warner (2004, p. 6) emphasize the need to have a framework that can identify the specific statistical and economic hypothesis to be utilized in an event study. Equally important is the reconsideration of basis through which the performance of the various event study method can be measured and compared.
Problems related to the Biasness in the Sample distribution of the Test Statistics
The even distribution normally computes tests statistics to allow a comparison between it and its assumed distribution. This is normally done in cases where the mean abnormal performance is at zero. Dombrow, Rodriquez & Sirmans (2000, p. 366) note that in cases where the test statistic goes beyond a critical value, the null hypothesis will be rejected. This means that the test statistics variable is random, since there is an error in the measuring of the abnormal returns.
This error is normally caused by the impreciseness of the predictions made on the unconditional expected returns on the securities and when the return realized by the firm during the period of the occurrence of the event are affected by unrelated reasons (McKinly, 1997 p. 30). This is especially so in cases where the component fails to average to a literal zero even in the cross-section. They note that another problem results from the use of an incorrect independence assumptions, concerning the abnormal returns based on event time clustering. This shows that the end result would be a biased standard deviation estimate and the statistic of the test. The standard deviation will go downward, while the test statistic will go upward.
Equally, this problem does not imply that the method is never efficient, since it can be solved by gauging the significance of the period in which the event occurred averaged by the abnormal return. This is done through the use of the differences in the event portfolio returns’ time series just before or immediately after the date of the event. Kothari and Warner (2004, p. 14) argued that the portfolio returns’ standard deviation can be useful in assessing average between the significance of the even window and the abnormal return. This method accounts for the cross-sectional dependence based on the fact that the portfolio return’s variability ultimately takes into account all cross dependence that existed among the returns realized on the individual even securities.
However, this approach has also been criticized, because it increases the uncertainty of the event period. The critics have stated that the evaluation done on the basis of the portfolio returns of an event –firm can lead to an over-estimation of statistical significance of the abnormal performance of the event window. Kappel, Schmidt & Ziegler (2009, p. 2) note that other scholars have also contradicted this critic, arguing that it is possible to estimate the extent of the increase in the returns of the event-periods’ variability. Leemakdej (2009, p. 2) identifies one of such ways as the estimation of the cross-sectional variability of returns, which had occurred during both the non-event and event periods.
According to Kothari and Warner (2004, p. 15), the estimate of the extent to which variability of returns has increased throughout the event period can successfully be estimated by obtaining the ratio of the variance for the event period and that for the non-event period. The result can be used in the adjustment of the bias that had been realized during the calculation of the test statistic. This allows one to ignore the uncertainty, which may have resulted from the increase in the event period.
Problems related to Non-Normally Distributed Data.
In his demonstration of the limitations of non-normally distributed data, Corrado (2011, p. 11) cited the result of a study by Brown and Warner in which they had tested for the robust test’s performance, which was not based on an assumption of normally distributed returns. This made it hard for them to obtain correct specifications. However, Corrado noted in such cases non-parametric sign and rank could be used to improve the efficiency. Compared to other tests, the experiment by Brown and Warner revealed that the non-parametric sign and rank were the most successful (Kolari & Pynonen, 2011 p. 960).
In their study, Brown and Warne (1984, p. 4) had concluded that besides being well specified, sign and rank test greatly improve the power of any test than the standard parametric tests. Brown and Warne (1984, p. 4) note the growing concerns that non-normally distributed data of security returns can mislead the researcher to poorly specify the assumptions of normality. This results in inferences, which can never be précised. Corrado & Truong (2008, p. 494) note that even though many stock exchanges have not considered this as a matter of concern, the issue has proved to be of serious concern with the exchanges from certain return data.
The problem can also be solved through the use of rank test, which has been proved to have the potential of performing better than the parametric tests. This was evidenced with data from Toronto and Copenhagen stock exchange. Another study carried out by Corrado and Truong with data from the Asian-Pacific stock exchange revealed the possibility of having the standard parametric event study being specified poorly, while being well specified in cases where non-parametric rank and rank tests were used (Corrado, 2011 p. 11).
Problem of Event Induced Volatility
Another matter of concern identified by Brown and Warne (1984, p. 4) is the issue of event-induced volatility. They argue that evidence has shown a notable increase in a security return’s variance during certain days close to the time of occurrence of certain events. Some scholars have noted the possibility of the variance increase by almost a factor of two. The effect of this is the excess rejection of the null hypothesis, which means that, the mean excess return will not be zero. The solution to this would be to acknowledge that all the events have a way of increasing the cross-sectional variance and that a researcher has to estimate and make the necessary adjustments in every test that has been used in assessing the event date abnormal return’s statistical significance (Bremer, Buchman & English, 2011 p. 514).
However, Corrado (2011, p. 18) has warned that the adjustment of crossectional variance are never practical in the event studies involving a small and medium sample sizes. In fact, according to Hilliard and Savickas as reported by Corrado, the unreliability involved in the estimation and making of inferences on event-induced volatilities even when the techniques used are of the advanced econometrics. Bathodoldy, Oloson & Peeare (2006, p. 231) on their part maintains that the problem can equally be solved by resorting to the rank test. According to their argument, the test has a high degree of robustness enabling it to counter the effect of the event-induced variance.
Problem of the Effect of the Subsequent Events
Kothari and Warner (2004, p. 5) note that even though the CAR of the event stocks’ portfolio may be able to capture post event responses correctly, the averaging process utilized in the creation of portfolio, CAR normally renders their effect insignificant based on the fact that the events of post-event for a specific stock are normally randomly distributed. The end result is, therefore, a bias in the portfolio resulting from underestimation. This affects its relationship with the individual securities. Giving example with the case of Turkey, Batchelor & Oracciglu (2003, p. 296) note that the bias would be more acute in a case where a researcher wants to infer concerning the efficiency of the market or where the testing was focused in obtaining the possibility of experiencing post event trading.
Their carried out in Turkey had revealed that there is an occurrence likelihood of extreme bias of under-estimation in any sample having substantial opportunities for post-event trading. Solibakke (2011, p. 4) added that the daily CAR’s volatility tends to be increased by the fact that the post-event events for the portfolio’s securities are neither aligned nor synchronized. This is contrary to the case of the events occurring on the date of an event. The end result is a significant reduction of the portfolio CAR’S t-values’ by the increased daily CAR.
Davaney (2012, p. 225) emphasizes that this means that the effect of the subsequent methods must not be allowed to affect those of the event being studied in order to be able to obtain inferences that are credible. Kothari & Warner (2004, p. 6) went further and presented a three-fold solution to the problem. According to them, the first solution is to employ the use of either non-parametric methods or parametric methods in the determination of post-event events. The second solution is the calculation of the impacts that the post-event events have on the returns realized from the stocks. The third solution is the isolation of the effects of all post events from the ones that were caused by the initial events. They argue that the use of this process can allow one to make credible conclusions, regarding the efficiency of the market through the study event. However, it is only possible where hourly returns can be obtained.
Problems Associated with the Use of Daily Data
According to Brown and Warner (1985, p. 4), the usage of data collected on a daily basis comes with a variety of problems. First, the problem concerns the departure of stock return collected for a specific security on a daily basis. They note that it is evidenced that daily returns’ distribution are normally fat tailed compared to the normal distribution. The second problem is related to the estimation of the parameter of the market model and non-synchronous trading. Brown and Warner (1985, p. 4) note that the market-model parameters’ ordinary least squares normally have some form of biasness and inconsistency. They further note that the level of this biasness increases whenever the daily data has been used. According to them, for the problem to be solved, the researchers have to employ the use of a number of techniques to help in estimating the parameter.
Third is the issue concerning the problems relating to the issues of variance estimation. Brown and Warner (1985, p. 4) identified a number of issues relating to daily data’s time-series properties. They note that it is possible for daily excess returns to be exhibited by serial dependence. This is a consequence that results from a non-synchronized trading. Another complication is the fact that the daily variance is stationary. Chung (2011, p. 147) notes that this is in contrast with the available findings that the variance of stock returns do increase some days after the occurrence of an event like the announcement of an expected change in earnings. Finally, it is the test static that is always being taken as the inference basis for any event studies. Binder (1998, p. 115) notes that the only solution to this problem is that anyone using the method must possess a good knowledge of the excess returns’ distributional properties. The researcher must be vast with good knowledge of both the cross-section and the time-series.
It is, thus, evidenced that event studies has a number of weaknesses that makes the method’s efficiency and reliability to be questionable. However, irrespective of all the identified weaknesses of the study, it has contributed enormously to research in the field of capital market. The study is today applicable in a number of other disciplines and not just in accounting and finance. It is also evident that the study has continued to rely on deviations obtained through the prediction of the models in the measurement of the abnormal returns. Equally, the study has also attracted many researchers, who continue to improve the methods, which are used in analyzing the abnormal returns to come up with statistical inferences. However, there is a need for the method to be used alongside other methods based on the fact that the study is applicable in many different settings.