You are on page 1of 25

Long-term abnormal returns in event studies

Tatiana Khabarova (647768) t.khabarova@uvt.nl BSc International Business Thesis Tilburg University June 2007

Supervisor: Jrmie Lefebvre Words: 6,207

Tatiana Khabarova Tilburg University June 2006

Abstract Results from various long-term event studies have led researchers to become dissatisfied with the use of a short-term event window. There was a lot of doubt cast upon the efficient market hypothesis. Researchers, therefore, have been advocating the use of longterm event windows to supplement the results from the traditional short-term event studies. This paper analyses the reasons for and problems with the use of long-term event studies.

Researchers mention the complexity of the event, insider trading and information leakage, slow release of relevant information, and the size of the news among the reasons in favour of using the long-term events. In addition, cognitive limitations of investors and impossibility to exploit large riskless profits quickly have been mentioned as obstacles to rapid market reactions. The problems with long-term studies can be divided into two general categories: (1) the bad model problem, which advocates that neither the market model, nor the popular BHAR approach should be used, and (2) econometric problems. These problems result in long-term event studies being unreliable and inferences drawn from them misleading.

The paper is concluded by noting that long-term event studies methods have improved significantly since their inception and long-term windows in event studies should indeed be used in some situations but the results should be treated with caution.

Tatiana Khabarova Tilburg University June 2006

1. Introduction
Event studies containing long-term event windows have first appeared in the 1960s. Since then these studies have rapidly grown in popularity and experienced wide applications in many financial, economic, legal and other areas. Many authors have debated whether long-term windows should be used and there seem to be no unilateral agreement on this issue. A lot of the debate has centred on whether the stock markets are efficient. Stock market efficiency is a very powerful concept and has many implications, in particular for the event studies methodology. Before elaborating further on market efficiency, this section of the paper motivates the research presented further in this paper and briefly talks about the importance of using a long event window in an event study.

1.1 Event studies and research motivation An event study, in various research fields, is an analysis of stock price reaction to an occurrence, such as an announcement of a given type, which is thought to affect public firms' market values. In an event study a researcher defines the event, the research sample containing firms that have been affected by the event, the event window, and measurement of normal and abnormal returns. An example of a time line for an event study is presented in figure 1. The researcher uses the estimation window to calculate the normal or expected returns if there was no event. An event window usually includes the event itself and is used to estimate the cumulative abnormal returns (true returns from the stock market minus expected returns). Post event window data is usually not used in an event study. This paper is concerned with the definition and use of an event window in an event study, in particular in a study of mergers and acquisitions.

Traditionally event studies have used short event windows, such as of 3 or 5 days. A study of horizontal mergers by Yan (2006) uses various, though still rather short, event windows to demonstrate the robustness of the findings. The authors choice of an event window is 7 days centred on the announcement date; but the paper also presents results using other event windows 3 days, 5 days, 13 days, and 25 days centred on the announcement date. The use of short/medium term event window is justified by the view

Tatiana Khabarova Tilburg University June 2006

that the markets initial reaction accurately and fully reflects the economic impact of the event. This may not necessarily be so.

Event studies of mergers and acquisitions announcements have found that the targets' stock prices normally react positively to the merger announcement, although the bidders' reactions are only marginally positive or close to zero on average. An example may be provided from a paper by Gong and Firth (2006). The authors investigate stock price reactions to merger announcements in the US airline industry in the period from 1985 to 2001. They find that on the event day 0, the average abnormal return to bidders is slightly negative at -0.24% but is statistically insignificant. The cumulative abnormal return during day -1 and day 0 is marginally significant at 1.50%, which indicates that the market expects the mergers to slightly increase the bidders' future profitability. Target firms on average experienced a high and positive abnormal return of 10.5% on the first announcement of the mergers. The cumulative abnormal return during the two days leading up to the announcement is 11.18%. Results of other event studies suggest that the shareholders of the target firms achieve substantial short-term gains as a result of successful takeovers. Ross et al. (2005) present data from numerous studies reporting that on average target firms experience 20% abnormal stock-price increases after a merger while bidders experience no change in the stock-price. However, the authors also report the abnormal five-year1 stock returns of the acquiring firms. On average, the acquirers experience abnormal stock returns of -6.5%. This is different from the initial findings of close to zero abnormal returns for the acquiring firms.

Other authors present similar findings. Andrade et al. (2001) say that several long-term event studies measuring negative abnormal returns over the three to five years following a merger completion provide reasons to doubt the use and findings of short-window event studies. According to these studies, as the paper authors claim, investors systematically fail to assess quickly the full impact of corporate announcements. They continue by saying that some researchers find that the long-term negative abnormal returns in acquiring firm stock prices is greater than the positive combined (target and acquirer)
1

from 1970 to 1989

Tatiana Khabarova Tilburg University June 2006

stock price reaction at announcement. Andrade et al. (2001) give an example of an event study by Loughran and Vijh (1997), in which the two mentioned authors find that acquiring firms using stock financing have abnormal returns of -24.2% over the five-year period after the merger, whereas the abnormal return is 18.5% for cash mergers. Harrison et al. (2005) confirm the findings by investigating horizontal acquisitions. The authors show that although market participants bid up the price of an acquiring firms stock shortly after the announcement of an acquisition, the long-term stock price movements are negative.

It may, therefore, be suggested that financial markets do not (always) fully and quickly reflect the new information that comes to the market or possibly react incorrectly to that information. Researchers that rely on short-run event studies only may at times come to incorrect conclusions. Hence, the use of (only) short-term event windows is questionable and long-term event windows should (also) be employed in event studies. This paper reviews the arguments supporting the use of the long-term event window and problems that arise when using a long-term window.

1.2 Paper structure The remainder of this Section talks about the origins, definition and use of the efficient market hypothesis as it provides an important foundation of an event study thinking. Next, the paper covers some arguments supporting the use of long-term event studies and, finally, Section 3 moves onto the discussion of problems related to the use and methodology of long-term event studies.

1.3 Efficient Market Hypothesis The origin of the ideas central to this hypothesis can be traced back to the early work by Bachelier (1900). He studied the behaviour of securities prices on La Bourse (the Paris Stock Exchange) and concluded that the price changes were identically and independently distributed, so that the future price changes could not be predicted from an examination of previous movements (Williams, 2005). The origin of the efficient market hypothesis (EMH) is most commonly attributed to the publication of Eugene Famas

Tatiana Khabarova Tilburg University June 2006

paper in 1965, Random Walks in Stock Market Prices. In his paper the author investigated correlations in daily price changes of 30 stocks and concluded that daily changes had a very small (approaching zero) positive correlation. He then formulated the definition of an efficient market.

In an efficient market, on the average, competition will cause the full effects of new information on intrinsic values to be reflected instantaneously in actual prices (Chuvakhin, 2007). This means that taking advantage of mis-priced stocks is very difficult as any mis-pricing is adjusted very rapidly. The different forms of the hypothesis were developed later.

EMH has wide applications in empirical finance and more recently in legal and regulatory issues. Event studies use the Efficient Market Hypothesis (EMH) as their basis. As Cox and Portes (1998) write, EMH, in its various forms, is one of the most important concepts in modern financial economics. The semi-strong version of the EMH states that in an efficient market, market prices reflect all publicly available information and respond quickly and without bias to new information (Cornell and Morgan, 1990). Thus, if new (public) information becomes available to the market, such as an announcement of a merger, a stock split, earnings, a change in an executive position etc., then the stock price(s) of the company/companies affected by the event will adjust.

Cox and Porter (1998) continue by stating that there is now a vast body of evidence supporting the semi-strong EMH and that it is now a well accepted tenet of financial economic research. In addition, there have been literally hundreds of finance papers confirming the general conclusion that the United States stock market is semi-strong efficient.

Another justification of the Efficient Market Hypothesis has been provided by Spyrou et al. (2007). The researchers examine short-term (daily) investor over- and under-reaction

Tatiana Khabarova Tilburg University June 2006

to extreme events (market shocks2) focusing on major and directly observable equity portfolios that contain the same firms on every shock day. In particular, the authors study four indices: the FT30 and the FTSE100, which consist of the largest market capitalisation stocks; the FTSE250, which consists of medium market capitalisation stocks; and the FTSE SmallCap, which consists of small market capitalisation stocks. Spyrou et al. find that whether the EMH holds or not depends on capitalisation. Their results indicate that investors in large capitalisation stocks in the UK react efficiently to information contained in market shocks. That is, the EMH hypothesis holds if the market capitalisation is large and no riskless profits can be made on the day(s) following the event. However, according to the findings, investors in medium and small capitalisation stocks in the UK show different behaviour. The authors find stock price under-reaction to extreme events. They even report that in some cases small and medium cap firms reaction is half as strong as that for large cap firms. Therefore, it may be concluded that the EMH is only justified for the large cap firms, where large means firms of the capitalisation size included in the FT30 and the FTSE100 indices that are traded on the London stock exchange.

Whether the EMH holds is important for determining the length of the event window. If it does then it is sufficient to use a short-term event study because the inferences drawn from such a study will come from the incorporation of all the relevant information. However, because some studies have allegedly found inconsistencies with the EMH a researcher may need a long-term event window (at least in some cases) to come to correct conclusions about the effects of an event. Next Section will review literature advocating the uses of a long event window.

The authors define an extreme event as an increase or decrease of the index return of magnitude bigger or smaller than two standard deviations the average daily index return.

Tatiana Khabarova Tilburg University June 2006

2. Importance of the use of long-term event windows


Harrison et al. (2005) advocate the use of long-term event windows in addition to more traditional short-term windows. The authors say that results from short-term event studies are not wrong but that there is an opportunity to expand the insight into the outcomes of events, and this opportunity should be exploited. This section of the paper reviews various arguments presented in favour of the use of long-term event windows. It also indicates in which situations it may be crucial that a long-term event window is used to complement a short-term window.

Most event studies implicitly assume that the markets respond fully, completely and in an unbiased manner to new information. Harrison et al. (2005) argue that these assumptions are more likely to be valid when the information revealed is easy to understand and interpret by market participants. However, when an event is complex, when its implications are not widely understood, or when the event is otherwise misunderstood by the market, early reactions may not be an accurate or complete assessment of the events economic impact. In these cases, a long time frame is necessary for the full economic implications to be realized.

The authors claim an acquisition or a merger to be an excellent example of a complex event. In acquisitions firms are radically changing their operational and financial characteristics. Combining firms with various often diverging organisational, financial, cultural, and other attributes can result in an almost infinite set of possible successes or failures. To fully assess the economic impact of mergers and acquisitions (and other complex events), Harrison et al. (2005) argue that long-term measures should be used in addition to short-term market reactions.

The authors suggest an explanation for the findings that while short-term abnormal returns are positive, long-term abnormal returns to the acquiring firms have been found to be negative. In the short-term many investors hold a popular belief that the acquisition has resulted in an increase in the market power of the acquirer and the obtainment of economies of scale, and, thus, lower unit costs, efficiency and higher bargaining power.

Tatiana Khabarova Tilburg University June 2006

However, in the long-run it becomes apparent as more information is released about the event and its performance that the market was unable to make an accurate and complete assessment at the time of an announcement as it takes longer for market participants to process complex information as compared to simpler information.

Why does it take longer to process and incorporate complex information? First, Harrison et al. (2005) find an answer in cognitive limitations. Humans are thought to be boundedly rational; they have limited abilities to process information especially as it becomes more and more complex, abstract or ambiguous. As a result some resort to simple rules of thumb and intuition. Limited cognition, therefore, creates a bias in event studies as stock prices do not react correctly or completely to the newly available information.3

Second, as the authors continue, there will be some investors that make use of sophisticated computer tools to compensate for bounded rationality. These investors can therefore try to do arbitrage to earn riskless profits from trading in mis-priced securities. However, in practice, short-term stock price reactions may not be favourable for the better-equipped investors and these investors will face large margin4 calls. If these investors do not have enough liquidity (such as cash) they will not be able to take full advantage of the mis-priced securities. Therefore, the arbitrage will not always rapidly bring securities to their true value and it will take time to correct stock prices.

A complex event may make a researcher to extend the event window into the post-event period. Third reason for using a long-term event window named by Harrison et al. (2005) may make a researcher to extend the window into the pre-event period. Despite the assumption of the market being semi-strong efficient, sometimes private information about an upcoming event becomes available to some participants in the market before the official announcement of an event. Then, information leakage and insider trading take place. Also, some investors may anticipate a merger or other event coming up due to
3

See Harrison et al (2005) a the review of some of the studies that applied concepts such as limited cognition and bounded rationality. 4 A margin refers to funds kept in a margin account for the purpose of covering losses. In the event of the unfavourable price movements, the margin account can reach the maintenance margin, in which case an investor will be required to add funds (liquid money) to the account (Chance, 2004).

Tatiana Khabarova Tilburg University June 2006

other events taking place one after another, which when taken together signal the intentions of a company. Event studies that cannot account for information leakage, i.e. studies that use short event windows, may be biased.

Harrison et al. (2005) say that the EMH does not consider the complexity of information. Often there are only three forms of efficiency and researchers consider a market as either efficient or not. The authors suggest a different kind of thinking: the concept of market efficiency is better viewed as a continuum: the market is more efficient in some settings than in others.

In their conclusion, Harrison et al. (2005) recognise some limitations to long-term event studies in general and their own study in particular. They say that evaluations over the long term can be problematic and contain noise due to influences of exogenous variables. The authors recommend the use of large samples, adequate control variables, and a control group consisting of similar firms in the same industries for calculation of abnormal returns.

Subramani and Walden (2002) provide another example of a complex event, the occurrence of which may require the use of a long-run event window. The information content in such an event needs not be complex or point to cognitive limitations. Rather, due to the nature of the event information is released over time. Subramani and Walden (2002) investigate stock price reactions to the announcements of initiatives to use ecommerce technologies. The authors employ different length event windows: 3, 11, and 21 days centred on the announcement day; and longer event windows of 6, 9 and 12 months after the announcement. Their findings vary as well. The average cumulative abnormal returns for the shorter event windows are: 0.44% over the 3-day window, 0.71% over the 11-day window and -1.13% over the 21-day window. Over the longer-run windows the returns are very similar to each other: 8.1%; 9.7%; and 9.5% - 6, 9 and 12 months after the event day. The authors use the results from the 12 months window to come to conclusions about the value created for the firms with the ecommerce initiatives. They motivate their choice by saying that not all the relevant information is available at

10

Tatiana Khabarova Tilburg University June 2006

the time of the announcement and, therefore, short-term price reactions are rather inconsistent and unrepresentative of the true value of the firm. In a complex event, such as a merger or acquisition, not all relevant information may be available at the announcement date. Similarly to announcements in ecommerce, information about mergers may be released over time making the deployment of a long-term event window more appropriate.

The event investigated by Subramani and Walden (2002) can also be considered as complex and the authors provide an explanation as to why not all the relevant information is available at the announcement day. Novel information technologies such as ecommerce technologies often come with considerable ambiguity related to benefits from these technologies. In the short-run it is not immediately clear what their effects on (other) industries, markets and firms will be. At the time of the announcement, as Subramani and Walden (2002) claim, relevant information is not available on complementary initiatives influencing value such as changes to business processes, willingness of supply chain partners etc. Therefore, adjustments of stock prices over long-run windows are likely to reflect more informed assessment of the ability of firms to generate benefits and to appropriate value created by novel information technologies than those within a few days of the event. In other words, the authors propose that the length of the event window used in an event study should be positively related to the time period in which relevant information becomes available. For example, if talks about the finalisation of a merger continue for several months after the merger announcement, then an event study should also include these months as part of the event window.

Another finding in the Subramani and Walden (2002) paper is the fluctuation of short-run cumulative returns. The authors notice a correspondence between the AMEX Internet Index5 and the variation in the short-term returns in the same period. This suggests that in the short-run investors react strongly to industry-wide Internet shocks rather than firmspecific information content in announcements. Once again, this creates a bias in the
5

The index used in this study is produced by the American Stock Exchange (AMEX), New York. Inter@ctive Week Internet Index is the mostly widely known index of internet companies produced by the AMEX (Wikipedia).

11

Tatiana Khabarova Tilburg University June 2006

short-term return and makes predictions and conclusions drawn from these returns less reliable.

A paper by Krivin et al. (2003) suggests that long event windows should be used in case of unusual news and large surprises. They investigate larger surprises as a standardised measure with regards to earnings announcements. The authors suggest that larger surprises will take longer for the market to process since these are more likely to contain some form of unexpected information. The authors also find that the period of abnormal price movements is positively correlated with the size of the news. Hence, the more surprising, unusual, and the more information the announcement contains, the longer should the event window be.

Therefore, this section may be concluded by saying that some of the reasons for the use of a long window in an event study are specific to the reality of the markets (such as liquidity constraints when trying to perform arbitrage transactions); other relate to human limitations explained by behavioural sciences; and yet other relate to the event itself (such as the inescapably slow availability of information).

In long-term event studies, many researchers use the buy-and-hold (BHAR) approach in which they allegedly take into account the potential statistical and model problems that have been indicated by other researchers. These authors insist on finding that their evidence of long-term abnormal performance and their study results are very robust. The final Section of this paper will present the problems arising with the popular BHAR approach and other critiques of the use of long-run event windows.

12

Tatiana Khabarova Tilburg University June 2006

3. Critiques of long-term event windows


Often researchers opt for short-term event windows because those provide statistically most reliable outcomes, uncomplicated by various outside, unaccounted for, factors. However, there may be other reasons for resorting to the traditional short-term event studies. Here the paper reviews the literature concerning the problems associated with long-term event studies.

Even if the use of long-term event studies may sound natural is some cases, there is a number of methodological concerns with the economic models used in the long-term event studies and econometric concerns with the statistics involved in these studies. These concerns may make the deployment and drawing conclusions from such studies questionable.

3.1 The bad model problem Various papers question every aspect of the statistics and assumptions in long-term event studies (see Andrade et al., 2001, Barber and Lyon, 1997, and Kothari and Warner, 1997). There seems to be a lot of concern coming from testing and estimation of longterm abnormal performance. Event studies may use different models, such as the constant mean return model and the market model, for estimating the normal performance returns (MacKinlay, 1997). As Andrade et al. (2001) explain, in short-window (e.g. 3 days) event studies the expected returns are virtually zero regardless of what model of expected returns is used. Therefore, there is no argument about the rejection of normal returns in the event window if the expected return is, for example, 0.05% and the observed (abnormal) return is about 1 to 3%. The story becomes different in long-term event studies. Andrade et al. (2001) say three-year expected returns can easily range from 30% to 65%, depending on the model used. This makes it very difficult to determine whether an abnormal return of, for example, 15% is statistically significant. It may be significant if one model to estimate normal returns is used and insignificant if a different model is used. Hence, it may be concluded that estimation of long-term expected returns is very rough and, therefore, estimates of long-term abnormal returns are also imprecise (Andrade et al., 2001). Imprecise estimates of the long-term returns stem from the design

13

Tatiana Khabarova Tilburg University June 2006

of the economic model that is used in long-term studies. Often these studies use the buyand-hold (BHAR) approach to estimate the abnormal returns.

Short-term event studies, on the other hand, often use the market model mentioned above. Should not this model also be used in long-term studies? Barber and Lyon (1997) believe it should not. The authors claim that cumulative abnormal returns (CARs), which are part of the methodology in the market model, become biased predictors of true abnormal returns in the long-run. Barber and Lyon calculate CARs for a sample of firms that all have a zero annual buy-and-hold abnormal returns (BHARs). They find that the CARs are 5% on average. The authors calculate that researchers using random samples for their event studies would draw different conclusions using CARs than using BHARs in roughly 4% of all samples. In addition, even if the inferences from the use of two models do coincide the two methodologies will still produce results of different magnitude. An incorrect prediction of the magnitude of abnormal returns would largely disappoint an investor who has, for example, invested some money in a (target) firm prior to a merger hoping to make money after the stock market has incorporated all the effects and possible synergies of the merger. Therefore, long-term event studies often resort to a variant of the buy-and-hold approach.

Some long-term event studies employ a technique called the characteristic-based matching process. In an event study a comparison needs to be made between the firms performance after the change and what that performance would have been if there had been no change. When the characteristic-based matching process is used in an event study this comparison is done by comparing the post-event performance of the firms that announced a change with firms that have characteristics that are believed to be similar to them and that had not experienced the event. Firms that have not made an announcement but have been selected for comparison are referred to as matching or control firms (Bremer et al., 2007).

Bremer et al. (2007) review literature that has found that characteristic-based matching process has a poor performance which relates to the low power of the test statistic that is

14

Tatiana Khabarova Tilburg University June 2006

used in this process to test for the presence of long-term abnormal returns. The power of a test statistic (in an event study) is its ability to correctly statistically detect the change in performance of a firm following an event and the announcement. Among other things, this ability of the test statistic relies directly on the comparability of the match.

The authors investigate several components of the characteristic-based matching process and conclude that the causes of low power levels include the selection criteria used to select matching firms, matching that is based on criteria not directly related to the variable of interest, and the failure of matching firms to satisfy matching criteria after the event period. Therefore, it may be said that the characteristic-based matching process that is important in the long-term event study methodology and is also known as the BHAR approach is a complicated process requiring a lot of attention and since it is impossible to find identical firms some of which had the event and others did not, the flaws in the characteristic-based matching process can not be eradicated.

3.2 Econometric concerns Frequently, event studies use a t-test (to identify the presence of abnormal returns), in which the test statistic has a Student's t distribution. The probability density function of this distribution has a mean of zero. A problem for the t-test is that long-term abnormal returns tend to be right-skewed (Khotari and Warner, 2006). Students t distribution is then asymmetric with the mean smaller than zero. This is called the skewness bias as it introduces an error into the test statistic for the null hypothesis that mean abnormal performance for the portfolio of event firms is zero. Fortunately, the extent of skewness bias in the test statistic is expected to decline with sample size (Khotari and Warner, 2006). Therefore, there is a possible skewness problem if the sample size is small as the Central Limit Theorems does not apply in the case the sample size is smaller than 30.

In addition, it is often thought that the skewness problem can often be cured by introducing a bootstrapping procedure for statistical inference, which relaxes the assumption of normality. This procedure was presented in a study by Ikenberry, Lakonishok, and Vermaelen in 1995. In their study the authors examine long-run firm

15

Tatiana Khabarova Tilburg University June 2006

performance following open market share repurchase announcements during the years 1980-1990. The bootstrapping procedure has since then seen popular uses in long-term event studies (Trainor, 2005).

Another statistical concern with long-term event studies is presented in Andrade et al. (2001). They claim that many long-term event studies assume that abnormal returns are independent across firms. However, that needs not be true. Major corporate actions like mergers are not random events. For example, from 1963 to 1964 there were 3,311 total acquisition announcements, while from 1968 to 1969 there were many more acquisition announcements, namely 10,569. Similarly, two other periods from 1979 to 1980 and from 1990 to 1991 there were approximately 4,000 acquisition announcements; while the late 1980s and late 1990s there were 9,278 announcements in 1999 alone (RhodesKropf and Viswanathan, 2004). These are known as merger waves. If events, like merger announcements are not independent then a sample in an event study is unlikely to consist of independent observations. Thus, there is positive cross-correlation of abnormal returns, which in turn means that test statistics that assume independence are severely overstated (Andrade et al., 2001). Moreover, the homogeneity of the sample firms examined (e.g., sample firms clustered in one industry) amplifies the problem as it increases the degree to which firms returns are correlated. Moreover, unlike the assumption of normal distribution of returns which is frequent in event studies and which becomes more plausible with sample size increase; the dependence problem becomes more severe with sample size (Mitchell and Stafford, 1999).

To investigate the magnitude of the problems arising from the skewness bias and crosscorrelation of returns Mitchell and Stafford (1999) employ three large samples of major managerial decisions, which have received attention in other studies. In their paper the three samples investigated are (1) seasoned equity offerings consisting of 4,911 announcements; (2) share repurchases that consisted of 2,421 announcements; and (3) 2,193 acquisitions announcements. In their study Mitchell and Stafford assume normality of the mean buy-and-hold returns (BHAR) and after calculating three-year abnormal returns they conclude that that was a reasonable assumption for the three large event

16

Tatiana Khabarova Tilburg University June 2006

samples. Therefore, according to this study it seems that the problem of skewness is not severe and was corrected for by the use of large samples.

Unlike the problem of skewness, the problem of independent returns results to be significant. Mitchell and Stafford (1999) found that frequently the test statistic (tstatistics) fell from over 6.0 to less than 1.5 after the authors used a covariance matrix to account for cross-correlations. This finding implies that at 5% significance the results change from rejecting the null hypothesis (i.e. that long-term average BHAR returns are zero) to accepting it. Later the authors find that a three-year BHAR of 15% is not statistically different from zero. Moreover, after Mitchell and Stafford had taken into account the positive cross-correlations of individual firm BHARs, they concluded that there was no reliable evidence to suggest the existence of long-term abnormal returns for any of the three event samples. Therefore, if the popular long-term buy-and-hold approach is corrected for the dependence of individual firm abnormal returns then it may be claimed that the long-term abnormal returns do not exist.

In contradiction with Bremer et al. (2007), Mitchell and Stafford (1999) stress that their findings result from accounting for the econometric problem (cross-correlation problem) and not from accounting for the construction of the matching firm or control firm portfolios (the bad model problem).

3.3 A different model - the calendar-time portfolio approach Evidently, all the complications with the BHAR approach mentioned above prove too much and Fama (1998) argues against the BHAR procedures altogether because this particular approach produces systematic errors. As has been mentioned in the beginning of this Section, imperfect model design (the bad model problem) results in imprecise abnormal returns that get compounded over a long-term window and aggravate the problem of drawing conclusions from statistical tests. In addition, the procedures produce overstate test statistics due to cross-sectional dependence (cross-correlation problem) of event firm abnormal returns. However, the BHAR procedure is not the only one that can be used and it can be argued that a statistical model can be designed in such a way as to

17

Tatiana Khabarova Tilburg University June 2006

make more precise estimates. For example, the calendar-time portfolio approach6 advocated by Fama (1998) can be used instead of the BHAR approach. Fama presents three reasons for the use of the calendar-time portfolio approach. First, monthly returns are less susceptible to the bad model problem. Second, the model automatically takes into calculations all cross-correlations of event-firm abnormal returns. Third, the distribution of the estimated abnormal returns can be approximated by the normal distribution, which forgoes that problem that returns are often situated much more to the right of the mean and allows for classical statistical tests.

Mitchell and Stafford (1999) confirm that the calendar-time portfolio approach is robust to the most serious statistical problems. The two authors use both Famas calendar-time portfolio approach and a BHAR approach that they adapt to account for what they believe to be a serious statistical drawback the positive cross-correlation of abnormal returns of event firms. The conclusions drawn from the two methods are found to be quite similar and therefore, Mitchell and Stafford (1999), just like Fama (1998), suggest that the calendar-time portfolio approach should be used in long-term event studies. In addition, the authors find that the calendar-time portfolio procedure is even better than their own adjusted BHAR approach as it has more power to identify reliable evidence of abnormal performance. But the model is still not perfect and results collected from the calendartime portfolio approach can still be susceptible to minor and less known statistical problems.

Even when methodological concerns with long-term event studies have been identified and new models have been developed, it still remains unclear whether the apparent abnormal returns are due to mispricing, or simply the result of measurement problems. Long-term event studies methods have improved significantly since their inception in the late 1960s and they now include rather complex statistical techniques that try to account for all the biases mentioned above. However, serious limitations remain and the challenge among financial economists is to continue to refine long-horizon methods
6

See Jegadeesh and Karceski (2004) for another example of long-term event study methodology that takes into account the problems of cross-correlation and also heteroskedasticity (a situation when variables have different variances).

18

Tatiana Khabarova Tilburg University June 2006

(Khotari and Warner, 2006). And if the abnormal returns are due to not yet incorporated measurement problems then the application of tests that are currently used can result in mistaken inferences in practical applications.

19

Tatiana Khabarova Tilburg University June 2006

Conclusion The more traditional event studies use short-term event windows. A window of 3 or 5 days centred on the announcement date is common in these studies. Results from various event studies have led researchers to become dissatisfied with the use of a short-term event window. Examples of these confusing studies can be found in the studies of mergers and acquisitions. For instance, some authors report the abnormal five-year stock returns of the acquiring firms to be -24.2% is the firm uses stock financing and 18.5% for cash mergers. This is quite different from the initial findings of close to zero abnormal returns for the acquiring firms. These and other similar findings have made researchers to suggest that despite the vast body of evidence supporting the semi-strong efficient market hypothesis (EHM) and the wide acceptance of United States and London stock markets being semi-strong efficient (at least for the large capitalisation firms), the EMH may be invalid. By this they mean that financial markets do not (always) fully and quickly reflect the new information that comes to the market or possibly react incorrectly to that information. Researchers, therefore, have been advocating the use of long-term event windows, such as 1 to 5 years after the announcement, to supplement the results from the traditional short-term event studies.

The debate around the alleged failures of the EMH has stimulated researchers to look for reasons of such occurrence and, therefore, for reasons advocating the use of long-term event studies. Investigation into the topic provides several instances when it is favourable to use a long-term event window. First, difficult to understand and interpret information related to the event may take time to become reflected in stock market prices. Second, insider trading and information leakage will lead to stock price reactions prior to the event announcement, which indicates that the event window should be extended both ways and not only into the post-event period. Third, information relevant for announcements and related to novel technologies such as those used in ecommerce can be partially unavailable at the announcement date and is therefore released over time. Forth, it was found that in the short-run investors sometimes react strongly to industrywide shocks rather than firm-specific information content in announcements. Fifth, unusual news and large surprising news may take time to become correctly reflected by

20

Tatiana Khabarova Tilburg University June 2006

stock prices. In addition, cognitive limitations of investors and the fact that large riskless profits cannot be exploited right away due to margin calls provide an obstacle to quick market reaction to announcements.

Notwithstanding the apparent need to use long-term event studies at least in some cases, there have been many critics of the methodologies and statistics deployed in these event studies. The problems related to long-term studies can be divided into two categories: (1) the bad model problem, which advocates that neither the market model which uses CARs, nor the BHAR approach should be used; and (2) econometric problem, which often focus on statistical problems related to the skewness and cross-correlation of longterm abnormal returns. Some studies have even denied the existence of long-term abnormal returns altogether.

These apparent problems with long-term event studies imply that inferences drawn from these studies are unreliable, incorrect and largely misleading. As a result, investors, financial planners, analysts and others relying on long-term studies can be hugely disappointed by the performance of their portfolios in particular and stock markets in general. There are, however, cases likes those documented in the Section 2 of this paper when the use of a long-term event window can be argued to be inevitable because the short-term event window very apparently does not or cannot reflect all the relevant information. Such cases may include particularly complex mergers and acquisitions. Long-term event studies methods have improved significantly since their inception in the late 1960s and they now include rather complex statistical techniques that try to account for all the known statistical biases. And the calendar-time portfolio approach seems to show a significant improvement in the methodology. Therefore, this paper is concluded by saying that long-term event windows in event studies should indeed be used in some situations but the conclusions of these studies should not be relied upon for 100% and the numerical results should be treated with caution.

21

Tatiana Khabarova Tilburg University June 2006

References Andrade, G., Mitchell, M., & Stafford, E. (2001). New evidence and perspectives on mergers. Journal of Economic Perspectives, 15, 2, 103120.

Barber, B. M., Lyon, J. D. (1997). Detecting long-run abnormal stock returns: The empirical power and specification of test statistics. Journal of Financial Economics, 43, 341-372

Bremer, R., Buchanan, B., & English II, P. C. (2007). Poor power properties in long-term event studies resulting from employing characteristic-based matching. Retrieved June 19, 2007, from http://www.fma.org/SLC/Papers/powerltevent.pdf

Chance, D. M. (2004). An introduction to derivatives and risk management. United States: Thompson Learning.

Chuvakhin, N. (2007). Efficient market hypothesis and behavioral finance is a compromise in sight? Retrieved June 19, 2007, from http://ncbase.com/papers/EMHBF.pdf

Cox, A. J., Portes, J. (1998). Mergers in regulated industries: the uses and abuses of event studies. Journal of Regulatory Economics, 14, 281-304.

Fama, E. F. (1998). Market efficiency, long-term returns, and behavioral finance. Journal of Financial Economics, 49, 283-306.

Gong, S. X. H., Firth, M. (2006). The competitive effects of airline mergers and acquisitions: more capital market evidence. Journal of Air Transportation, 11, 1, 88-106

Harrison, J. S., Oler, D., & Allen, M. R. (2005). Event studies and the importance of longer-term measures in assessing the performance outcomes of complex events. Retrieved June 19, 2007, from

22

Tatiana Khabarova Tilburg University June 2006

http://www.indiana.edu/~aisdept/people/files/Oler/Event%20Studies%20and%20the%20I mportance%20of%20Longer-Term%20Measures.pdf

Harrison, J. S., Oler, D., & Allen, M. R. (2006). Over-interpretation of event study findings in management research: an empirical illustration of the problem. Retrieved June 19, 2007, from http://www.indiana.edu/~aisdept/people/files/Oler/EvJMSsubmittedJuly31.pdf

Ikenberry, D., Lakonishok, J., Vermaelen, T. (1995). Market underreaction to open market share repurchases. Journal of Financial Economics, 39, 181-208.

Kothari, S. P., Warner, J. B. (1997). Measuring long-horizon security price performance. Journal of Financial Economics 43, 301-339

Kothari, S.P., Lewellen, J., Warner, J. B. (2006). Stock returns, aggregate earnings surprises, and behavioral finance. Journal of Financial Economics, 79, 537568

Krivin, D., Patton, R., Rose, E., & Tabak, D. (2003). Determination of the appropriate event window length in individual stock event studies. NERA Economic Consulting. Retrieved June 19, 2007, from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=466161

MacKinlay, A. C. (1997). Event studies in economics and finance. Journal of Economic Literature, 35, 13-39 Mitchell, M. L., Stafford, E. (1999). Managerial decisions and long-term stock price performance. Retrieved June 19, 2007, from http://papers.ssrn.com/sol3/papers.cfm?abstract_id=94137

Rhodes-Kropf, M., Viswanathan, S. (2004). Market valuation and merger waves. The Journal of Finance, 59, 6.

23

Tatiana Khabarova Tilburg University June 2006

Ross, S. A., Westerfield, R. W., & Jaffe, J. (2005). Corporate finance. New York, NY: McGraw-Hill/Irwin.

Spyrou, S., Kassimatis, K., & Galariotis, E. (2007). Short-term overreaction, underreaction and efficient reaction: evidence from the London stock exchange. Retrieved June 19, 2007, from http://www.efmaefm.org/efma2005/papers/243spyrou_paper.pdf

Subramani, M. R., Walden, E. A. (2002). Employing the event study to assess returns to firms from novel information technologies: an examination of ecommerce initiative announcements. Retrieved June 19, 2007, from http://www.misrc.umn.edu/workingpapers/fullpapers/2002/0201_100802.pdf

Trainor, Jr. W. J. (2005). Long-range confidence interval projections and probability estimates. Financial Services Review, 14, 7384

Wikipedia. American stock exchange. Retrieved May 30, 2007, from http://en.wikipedia.org/wiki/American_Stock_Exchange

Williams, L. V. (2005). Information Efficiency in Financial and Betting Markets. Retrieved May 28, 2007, from http://www.cambridge.org/uk/catalogue/catalogue.asp?isbn=0521816033&ss=exc

Yan, J. (2006). Theory and evidence of horizontal mergers. Retrieved June 19, 2007, from http://finance.wharton.upenn.edu/department/Seminar/2006Fall/micro/jinghuayanmicro-090706.pdf

24

Tatiana Khabarova Tilburg University June 2006

Annex

Figure 1 An example of a time line for an event study


E st im a t ion Win dow T0 -250 T1 -6 E ven t Win dow 0 P ost -E ven t Win dow

T2 250 T3

25

You might also like