Event Study in Economics and finance

American Economic Association Event Studies in Economics and Finance Author(s): A. Craig MacKinlay Source: Journal of Economic Literature, Vol. 35, No. 1 (Mar., 1997), pp. 13-39 Published by: American Economic Association Stable URL: http://www.jstor.org/stable/2729691 . Accessed: 14/10/2011 08:03 Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to Journal of Economic Literature. http://www.jstor.org Journal of Economic Literature Vol. XXXV (March 1997), pp. 13-39 Event Studies in Economics and Finance A. CRAIG MACKINLAY The Wharton School, University of Pennsylvania Thlanks to John Can ipbell, Briuce G-tr udly, Andrti-ewv Lo, and twvo anonymnotus referees for helpful comments andcl discussion. Research suLpport from the Rodiney L. WVhite Center for Finiancial Research is gratefiully acknzowvledged. 1. Introduction ECONOMISTS are frequently asked to measure the effects of an economic event on the value of firms. On the sur- face this seems like a difficult task, but a measure can be constructed easily using an event study. Using financial market data, an event study measures the impact of a specific event on the value of a firm. The usefulness of such a study comes from the fact that, given rationality in the marketplace, the effects of an event will be reflected immediately in security prices. Thus a measure of the event's economic impact can be constructed using security prices observed over a relatively short time period. In contrast, direct productivity related measures may require many months or even years of observation. The event study has many applica- tions. In accounting and finance re- search, event studies have been applied to a variety of firm specific and economy wide events. Some examples include mergers and acquisitions, earnings an- nouncements, issues of new debt or eq- uity, and announcements of macro- economic variables such as the trade deficit.1 However, applications in other fields are also abundant. For example, event studies are used in the field of law and economics to measure the impact on the value of a firm of a change in the regulatory environment (see G. William Schwert 1981) and in legal liability cases event studies are used to assess damages (see Mark Mitchell and Jeffry Netter 1994). In the majority of applications, the focus is the effect of an event on the price of a particular class of securities of the firm, most often common equity. In this paper the methodology is discussed in terms of applications that use common equity. However, event studies can be applied using debt securities with little modification. Event studies have a long history. Per- haps the first published study is James Dolley (1933). In this work, he examines the price effects of stock splits, studying nominal price changes at the time of the split. Using a sample of 95 splits from 1921 to 1931, he finds that the price in- I The first three examples will be discussed later in the paper. Grant McQueen and Vance Roley (1993) provide an illustration of the fourth using macroeconomic news announcements. 13 14 o ournal of Economic Literature, Vol. XXXV (March 1997) creased in 57 of the cases and the price declined in only 26 instances. Over the decades from the early 1930s until the late 1960s the level of sophistication of event studies increased. John H. Myers and Archie Bakay (1948), C. Austin Barker (1956, 1957, 1958), and John Ashley (1962) are examples of studies during this time period. The improve- ments included removing general stock market price movements and separating out confounding events. In the late 1960s seminal studies by Ray Ball and Philip Brown (1968) and Eugene Fama et al. (1969) introduced the methodology that is essentially the same as that which is in use today. Ball and Brown consid- ered the information content of earn- ings, and Fama et al. studied the effects of stock splits after removing the effects of simultaneous dividend increases. In the years since these pioneering studies, a number of modifications have been developed. These modifications re- late to complications arising from viola- tions of the statistical assumptions used in the early work and relate to adjust- ments in the design to accommodate more specific hypotheses. Useful papers which deal with the practical importance of many of the complications and adjust- ments are the work by Stephen Brown and Jerold Warner published in 1980 and 1985. The 1980 paper considers imple- mentation issues for data sampled at a monthly interval and the 1985 paper deals with issues for daily data. In this paper, event study methods are reviewed and summarized. The paper begins with discussion of one possible procedure for conducting an event study in Section 2. Section 3 sets up a sample event study which will be used to illus- trate the methodology. Central to an event study is the measurement of an ab- normal stock return. Section 4 details the first step-measuring the normal performance-and Section 5 follows with the necessary tools for calculating an abnormal return, making statistical in- ferences about these returns, and aggre- gating over many event observations. The null hypothesis that the event has no impact on the distribution of returns is maintained in Sections 4 and 5. Section 6 discusses modifying this null hypothesis to focus only on the mean of the return distribution. Section 7 presents analysis of the power of an event study. Section 8 presents nonparametric approaches to event studies which eliminate the need for parametric structure. In some cases theory provides hypotheses concerning the relation between the magnitude of the event abnormal return and firm char- acteristics. Section 9 presents a cross- sectional regression approach that is use- ful to investigate such hypotheses. Section 10 considers some further issues relating event study design and the pa- per closes with the concluding discussion in Section 11. 2. Procedure for an Event Study At the outset it is useful to briefly dis- cuss the structure of an event study. This will provide a basis for the discussion of details later. While there is no unique structure, there is a general flow of analysis. This flow is discussed in this section. The initial task of conducting an event study is to define the event of interest and identify the period over which the security prices of the firms involved in this event will be examined-the event window. For example, if one is looking at the information content of an earnings with daily data, the event will be the earnings announcement and the event window will include the one day of the announcement. It is customary to define the event window to be larger than the specific period of interest. This permits examination of periods surrounding the MacKinlay: Event Studies in Economics and Finance 15 event. In practice, the period of interest is often expanded to multiple days, in- cluding at least the day of the an- nouncement and the day after the an- nouncement. This captures the price effects of announcements which occur after the stock market closes on the an- nouncement day. The periods prior to and after the event may also be of inter- est. For example, in the earnings an- nouncement case, the market may ac- quire information about the earnings prior to the actual announcement and one can investigate this possibility by ex- amining pre-event returns. After identifying the event, it is neces- sary to determine the selection criteria for the inclusion of a given firm in the study. The criteria may involve restric- tions imposed by data availability such as listing on the New York Stock Exchange or the American Stock Exchange or may involve restrictions such as membership in a specific industry. At this stage it is useful to summarize some sample char- acteristics (e.g., firm market capitaliza- tion, industry representation, distri- bution of events through time) and note any potential biases which may have been introduced through the sample se- lection. Appraisal of the event's impact re- quires a measure of the abnormal return. The abnormal return is the actual ex post return of the security over the event win- dow minus the normal return of the firm over the event window. The normal re- turn is defined as the expected return without conditioning on the event taking place. For firm i and event date t the abnormal return is AR1t = R1, - E(RjrjXr) (1) where AR,,, Ri,, and E(Ri,IXt) are the ab- normal, actual, and normal returns re- spectively for time period t. Xl is the conditioning information for the normal return model. There are two common choices for modeling the normal re- turn-the constant mean return model where Xl is a constant, and the market model where Xl is the market return. The constant mean return model, as the name implies, assumes that the mean return of a given security is constant through time. The market model as- sumes a stable linear relation between the market return and the security re- turn. Given the selection of a normal perfor- mance model, the estimation window needs to be defined. The most common choice, when feasible, is using the period prior to the event window for the estima- tion window. For example, in an event study using daily data and the market model, the market model parameters could be estimated over the 120 days prior to the event. Generally the event period itself is not included in the esti- mation period to prevent the event from influencing the normal performance model parameter estimates. With the parameter estimates for the normal performance model, the abnor- mal returns can be calculated. Next comes the design of the testing frame- work for the abnormal returns. Impor- tant considerations are defining the null hypothesis and determining the tech- niques for aggregating the individual firm abnormal returns. The presentation of the empirical re- sults follows the formulation of the econometric design. In addition to pre- senting the basic empirical results, the presentation of diagnostics can be fruit- ful. Occasionally, especially in studies with a limited number of event observa- tions, the empirical results can be heav- ily influenced by one or two firms. Knowledge of this is important for gaug- ing the importance of the results. Ideally the empirical results will lead to insights relating to understanding the sources and causes of the effects (or lack 16 Journal of Economic Literature, Vol. XXXV (March 1997) of effects) of the event under study. Ad- ditional analysis may be included to dis- tinguish between competing explana- tions. Concluding coiimments complete the study. 3. An Example of an Event Study The Financial Accounting Standards Board (FASB) and the Securities Ex- change Commission strive to set report- ing regulations so that financial state- ments and related information releases are informative about the value of the firm. In setting standards, the informa- tion content of the financial disclosures is of interest. Event studies provide an ideal tool for examiining the informnation content of the disclosures. In this section the description of an example selected to illustrate event study methodology is presented. One particular type of disclosure-quarterly earnings announcements-is considered. The objective is to investigate the infor- mation content of these announce- ments. In other words, the goal is to see if the release of accounting informnation provides information to the marketplace. If so there should be a correlation be- tween the observed change of the mar- ket value of the company and the infor- mation. The example will focus on the quar- terly earnings announcements for the 30 firms in the Dow Jones Industrial Index over the five-year period from January 1989 to December 1993. These an- nouncements correspond to the quar- terly earnings for the last quarter of 1988 through the third quarter of 1993. The five years of data for 30 firms provide a total sample of 600 announcemnents. For each firm and quarter, three pieces of in- formation are compiled: the date of the announcement, the actual earnings, and a measure of the expected earnings. The source of the date of the announcement is Datastream, and the source of the ac- tual earnings is Compustat. If earnings announcements convey in- formation to investors, one would expect the announcement impact on the mar- ket's valuation of the firm's equity to de- pend on the magnitude of the unex- pected component of the announcement. Thus a measure of the deviation of the actual announced earnings from the mar- ket's prior expectation is required. For constructing such a measure, the mean quarterly earnings forecast reported by the Institutional Brokers Estimate Sys- tem (I/B/E/S) is used to proxy for the market's expectation of earnings. I/B/E/S compiles forecasts from analysts for a large number of companies and reports summary statistics each month. The mean forecast is taken from the last month of the quarter. For example, the mean third quarter forecast from Sep- tember 1990 is used as the measure of expected earnings for the third quarter of 1990. To facilitate the examination of the impact of the earnings announcement on the value of the firm's equity, it is essen- tial to posit the relation between the in- formation release and the change in value of the equity. In this example the task is straightforward. If the earnings disclosures have information content, higher than expected earnings should be associated with increases in value of the equity and lower than expected earnings with decreases. To capture this associa- tion, each announcement is assigned to one of three categories: good news, no news, or bad news. Each announcement is categorized using the deviation of the actual earnings from the expected earn- ings. If the actual exceeds expected by more than 2.5 percent the announce- ment is designated as good news, and if the actual is more than 2.5 percent less than expected the announcement is des- ignated as bad news. Those announce- MacKinlay: Event Studies in Economics and Finance 17 ments where the actual earnings is in the 5 percent range centered about the ex- pected earnings are designated as no news. Of the 600 announcements, 189 are good news, 173 are no news, and the remaining 238 are bad news. With the announcements categorized, the next step is to specify the parameters of the empirical design to analyze the eq- uity return, i.e., the percent change in value of the equity. It is necessary to specify a length of observation interval, an event window, and an estimation win- dow. For this example the interval is set to one day, thus daily stock returns are used. A 41-day event window is em- ployed, comprised of 20 pre-event days, the event day, and 20 post-event days. For each announcement the 250 trading day period prior to the event window is used as the estimation window. After presenting the methodology of an event study, this example will be drawn upon to illustrate the execution of a study. 4. Models for Measuring Normal Performance A number of approaches are available to calculate the normal return of a given security. The approaches can be loosely grouped into two categories-statistical and economic. Models in the first cate- gory follow from statistical assumptions concerning the behavior of asset returns and do not depend on any economic ar- guments. In contrast, models in the sec- ond category rely on assumptions con- cerning investors' behavior and are not based solely on statistical assumptions. It should, however, be noted that to use economic models in practice it is neces- sary to add statistical assumptions. Thus the potential advantage of economic models is not the absence of statistical assumptions, but the opportunity to cal- culate more precise measures of the nor- mal return using economic restrictions. For the statistical models, the assump- tion that asset returns are jointly multi- variate normal and independently and identically distributed through time is imposed. This distributional assumption is sufficient for the constant mean return model and the market model to be cor- rectly specified. While this assumption is strong, in practice it generally does not lead to problems because the assumption is empirically reasonable and inferences using the normal return models tend to be robust to deviations from the assump- tion. Also one can easily modify the sta- tistical framework so that the analysis of the abnormal returns is autocorrelation and heteroskedasticity consistent by us- ing a generalized method-of-moments approach. A. Constant Mean Return Model Let [,u be the mean return for asset i. Then the constant mean return model is Rit= - i + Git (2) (2 E(4st) = 0 var (4i) = G7; . where Rit is the period-t return on secu- rity i and Cit is the time period t distur- bance term for security i with an expec- tation of zero and variance cy . Although the constant mean return model is perhaps the simplest inodel, Brown and Warner (1980, 1985) find it often yields results similar to those of more sophisticated mnodels. This lack of sensitivity to the model can be attributed to the fact that the variance of the abnor- mal return is frequently not reduced much by choosing a more sophisticated model. When using daily data the model is typically applied to nominal returns. With monthly data the model can be ap- plied to real returns or excess returns (the return in excess of the nominal risk free return generally measured using the U.S. Treasury Bill with one month to inaturity) as well as nominal returns. 18 Journal of Economic Literature, Vol. XXXV (March 1997) B. Market Model The market model is a statistical model which relates the return of any given security to the return of the mar- ket portfolio. The model's linear specifi- cation follows from the assumed joint normality of asset returns. For any secu- rity i the market model is Rit = (xi + PiR,,t + ?it (3) E(eit = 0) var(eyt) = (T where Rit and Riiit are the period-t re- turns on security i and the market port- folio, respectively, and Lit is the zero mean disturbance term. oci, fPi, and G2 are the parameters of the market model. In applications a broad based stock in- dex is used for the market portfolio, with the S&P 500 Index, the CRSP Value Weighted Index, and the CRSP Equal Weighted Index being popular choices. The market model represents a poten- tial improvement over the constant mean return model. By removing the portion of the return that is related to variation in the market's return, the variance of the abnormal return is reduced. This in turn can lead to increased ability to de- tect event effects. The benefit from us- ing the market model will depend upon the R2 of the market model regression. The higher the R2 the greater is the vari- ance reduction of the abnormal return, and the larger is the gain.