Nine steps to follow when performing a short-term event study
The blueprint of a standard event study is presented in the following. The underlying assumption of the event study methodology is that the capital market is semi-strong form efficient. This form of market efficiency assumes that asset prices comprise all publicly-available information relevant for price formation. You should follow the 9 steps provided below to accurately conduct an event study.
Exactly define the event
When conducting an event study, it is important to define as exactly as possible the corporate event you want to examine to make sure all identified events are comparable. For example, if you are interested in examining dividend cuts, first determine exactly which definition of a dividend cut you want to apply: all dividend cuts, only dividend cuts that have not been expected by analysts (in consensus), or only unexpected dividend cuts of more than X%. Always keep in mind that the event you want to examine should be new, unexpected information to the public. Else you should not expect a considerable reaction to asset prices. If you want to run an event study on, say, sudden deaths of CEOs, you may want to exclude cases of CEOs known to have heart diseases or cancer.
Define the sample and news sources
Before you start collecting your events of interest, define the sample period and the countries you want to examine. In general, you should choose a large sample period to be able to collect as many events as possible. Yet, make sure that over the years of your sample period the legal circumstances did not change in a way that influences your event identification and results. For example, make sure the event you are interested in was not subject to voluntary disclosure in the early years of your sample period and then became subject to new mandatory disclosure rules. If so, you must at least account for this change. A similar logic applies if you want to examine different countries within one event study. Further, be aware that event and event date identification tends to be more difficult for events that occurred earlier in the past, particularly for the time prior the (frequent) use of the internet. Including earlier periods in your study might lead to a bias towards larger firms for which information was more likely available (and collected) in the pre-internet period.
In addition, define a ranking of news sources to use for the identification of each event and event date. Often corporate events or actions of investors in the capital market must be announced in a prespecified format and database (e.g., filings for the U.S. Securities and Exchange Commission (SEC) in the EDGAR system or ad-hoc news in many European countries published on websites such as www.euroadhoc.com or www.dgap.de for Germany). Yet, standards differ dramatically between countries. If announcing information via a certain news source is not required, you need to search via news databases or general internet search (e.g., websites of newspaper companies).
News databases frequently used are Factiva® and LexisNexisTM. Many studies for the United States, particularly earlier ones, focus on the Wall Street Journal (WSJ) as their primary (or only) source of information. However, be aware that while the focus on a single, “top in class” news source reduces uncertainty with respect to the exact event date and adds to the comparability of your identified events (e.g., as they were all recognized by an important news source), you may have to (or want to) search all available news sources to make sure you really identify the earliest date for the announcement of a corporate event (especially if disclosure of the event is not mandatory).
Identify the exact event date
Make sure you identify the exact date on which the event of interest was announced. This is the first trading day on which the event became public information. If the event was announced on a non-trading day, the next trading day is the correct event day to choose. Before you run any event-study, use the Event Date Check feature provided in the Event Study Metrics software. This is an automatic check for each event in your list of events. Each event that falls on a non-trading day will be marked in your event list.
This feature can be activated as a standard feature in the software. If activated, the program first checks all events and, if necessary, adjusts each event date to a trading day each time before an event study is run. Furthermore, keep in mind that the asset prices you use in an event study usually are closing prices. In case you know the exact time when the news became public information, take this time into account. If news become public information after the trading system (e.g., the stock exchange) is closed they will be included in the closing price of the next trading day. If you are not able to identify the exact time of the news, you should choose a three-day event window (starting one day before and ending one day after the identified event date) to cope with event date uncertainty.
Drop confounding events
To get accurate and interpretable results, make sure you exclude all events that are announced together with other new, price relevant information. Some news may be released jointly on a systematic basis. You should identify such types of correlated news because even if your sample is very large, a systematic bias will not disappear. For example, if you want to examine the price effects of dividend cuts, you should drop events of dividend cuts announced together with other new information such as earnings warnings or announcements of share repurchases (or any other price relevant information).
Compose the event list and retrieve asset price data
If you follow steps 1-4 you will end up with a list of events, i.e., a list of all announcements for your event of interest over a specified period of time. This event list can consist of more events than firms given that most corporate events occur more than once (e.g., bond issues, dividend changes, M&As). The final event list includes the event date, the firm’s name and a firm identifier such as a CUSIP, CRSP® PERMNO, ISIN or Ticker. Via the firm identifier you can retrieve asset price data for your firms to run the event study and identify the average market reaction to your event of interest. You do not need asset prices for each event. You just need the asset price history for each firm in your event list. The program matches the events in your event list automatically with the asset price data via the firm identifier you choose. Therefore, when you retrieve asset price data, be aware that firm identifiers can change over time, for instance due to corporate name changes or mergers. Asset price databases frequently use current identifiers. In your news search, however, you may find old identifiers (in a few cases). Usually, changes of CUSIPs, Tickers, etc. can be identified via the internet.
Make sure each firm in your event list only has one unique firm identifier which is also the identifier used in your asset price data set. Stock price data is provided in several commercial databases such as Bloomberg©, CRSP® (only U.S. stock price data), or Datastream©, but can, for most stocks, also be retrieved freely from Yahoo Finance or Google Finance, among other sources. Bond price data, and other bond information, can be retrieved, for example, from Bloomberg© or TRACE (U.S. bond price data only). CDS data is available, for example, via Markit®.
Determine the estimation method for expected return calculation
Once you have retrieved the asset price data for your events, you already know the realized prices and returns for the event dates. In order to calculate the abnormal returns, i.e., the returns that can be attributed to the event of interest, you first need to estimate the expected return for the event date. This is the hypothetical return that would have occurred in the absence of the event you examine. The abnormal return is simply the realized return minus the expected return (on the same day). Several methods exist to estimate the expected return. They all have in common that they use asset price histories for expected return estimation. Thus, expected return estimation necessitates that you have a minimum period of asset price history available. The estimation methods differ in terms of sophistication and data requirements. The simplest method is the constant mean return model.
It uses the mean return of an asset over a period of time (the estimation window) as the estimate of the expected return of that asset. A similar approach is the market return model. Instead of the asset’s mean return it just uses a market’s mean return as the expected return for an asset. Frequently, industry-specific stock indexes (e.g., a telecommunication index in case you examine a telecommunications company like AT&T, Vodafone or Deutsche Telekom) are used as market proxies. If an industry index is used, one should exclude the respective firm of interest to get more accurate results. Standard & Poor’s (S&P), among others, constructs industry indexes and provides respective data. Apart from using the firm (or generally the asset) itself or a market index, one can also use comparable firms to estimate expected returns. This is done with the matched firms model. In this model, the expected return of a firm (or generally an asset) on a certain day is the return of its reference firm (i.e., the matched firm) on that day. Using this approach, it is important to make sure that the reference firm is really comparable to the firm of interest. Often firms are matched on their size (usually via market capitalization) and their book-to-market ratios (see, e.g., Lyon, Barber and Tsai, 1999). The general logic of the matched firms approach is similar to that of using peer groups in firm valuation. More sophisticated models for estimating expected asset returns are based on the capital asset pricing model (CAPM). The CAPM is the seminal asset pricing theory applied in both academia and practice. All other pricing models, both theoretical (APT) and empirical (multi-factor models), are modifications of the CAPM. The CAPM, market model (the most popular model in practice) and the three- and four-factor models are the predominant models for expected return calculation in event studies. The latter are extensions of the CAPM and the market model which simply include more risk factors in addition to the beta factor. Multi-factor models were introduced in the 1990s by Nobel laureate Eugene Fama and his colleague Kenneth French (see, e.g., Fama and French, 1993). If you apply one of these estimation methods, you need more data.
In fact, a broad stock index, which is the estimate for the market portfolio of the CAPM, is needed (for the return model) to calculate the beta factor which measures the sensitivity of an asset’s return to the return of the market portfolio. The majority of U.S. event studies use the CRSP index (equally or value weighted) or the S&P 500 or S&P 1500 to proxy for the market portfolio. The MSCI World and the MSCI Europe are regularly used in worldwide and European event studies, respectively. Generally, if an event study is run for a certain country, the country’s broadest stock index is used as the proxy for the market portfolio. For example, if you run an event study for Germany, you can use the German CDAX as the market proxy. In case you use three- or four-factor models, you need to include data for the other risk factors as well.
Be aware that these factors are not available for all countries. Finally, we note that the accurate choice of a model for expected return estimation is more important in long-term than in short-term event studies. Risk adjustment is less important in the latter because errors in abnormal return calculation caused by errors or inaccuracy in the adjustment of risk tend to be small (see Kothari and Warner, 2008). Simple risk adjustment approaches usually perform quite well in short-term event studies as shown by Brown and Warner (1985).
Before you run any event study, make sure that the asset price and market index data have the same format, i.e., either prices or returns. If you have price data for your assets and return data for the market index (or vice versa), use the Price-Return Converter provided in the software to adjust the data format accordingly. Whether your data is price or return data must be specified within the program (in the left corner at the top). You can also specify how the program treats missing data in an asset’s price or return history. Furthermore, make sure you use total return data, i.e., asset data which accounts for capital gains, dividends, interests, etc. When stock price data is used, make sure it is adjusted for stock splits.
Determine the estimation and event window
Once you have chosen a model and gathered all necessary data for expected return estimation, you need to determine the estimation window and the event window. The two windows do not overlap, i.e., the trading days prior to the event day which are part of the event window are not part of the estimation window. The estimation window is the period of trading days (before the event date) that is used to estimate the expected return for each asset and each event (using the constant mean return model, the market model, or any other of the aforementioned models). The event window is the period of trading days over which you want to calculate abnormal returns. An event window of one day, abbreviated , only includes the event day itself. In many studies the maximum event window includes 41 trading days symmetrically surrounding the identified event day, abbreviated [-20,+20].
The program allows you to determine up to eight different event windows at once, i.e., you can calculate abnormal returns over eight different periods of time. There is no consensus with respect to the length of the event window and particularly with respect to the length of the estimation window. Most studies either use an estimation window of 180 or 200 trading days ending either 10 or 20 days prior to the event. The end date of the estimation window should depend on the likelihood of information leakage of the event of interest. Event studies on mergers and acquisitions (M&A), for example, often use 40 trading days prior to the M&A announcement as the end date for the estimation window. The choice of the estimation window also determines the amount of asset price history needed for the event study.
For example, if you choose an estimation window of 200 trading days ending 20 trading days prior to the event day, you should have asset price histories for each firm in your event list that reach back 220 trading days from each respective event day. Too much missing data for asset price histories either lead to the exclusion of some events or necessitate the choice of a shorter estimation window. If only some data is missing, i.e., price information for single days is not available (an issue that occurs at times), you do not have to drop the respective events. The software can account for single days of missing data in the (estimation window) calculations and adjust test statistics accordingly.
Calculate cumulative (average) abnormal returns or buy-and-hold abnormal returns
When the estimation model and the event and estimation windows are chosen, the program calculates a firm’s abnormal return (AR) for each respective event and each day in the prespecified event window. For the event day itself, this is the realized return on that day minus the expected return on that day. For the day before or the day after the event day, for example, the calculation is the same, but the realized return is usually different (as another trading day is taken into account). The cumulative abnormal return (CAR) is just the sum of a firm’s abnormal returns over a certain period around, prior to or after an event. As an example, consider the three days symmetrically surrounding an event, abbreviated [-1,+1]. The respective CAR is just the sum of the firm’s abnormal returns on the day before the event, the event day itself, and the day after the event. The average (usually non-weighted) of each firm’s AR and CAR over a certain period of trading days in the event window is called average abnormal return (AAR) (for the event day) and cumulative average abnormal return (CAAR) (for several days in the event window), respectively.
Some scholars have argued that cumulative abnormal returns are not appealing on economic grounds as investors usually invest in assets and hold them for a certain period (i.e., they do not earn abnormal returns on each day) and have proposed the use of buy-and-hold abnormal returns, abbreviated BHARs, (see, e.g., Ritter, 1991; Barber and Lyon, 1997). BHARs and CARs are simply two different ways of aggregating abnormal returns (arithmetic versus geometric). As the name indicates, buy-and-hold returns consider the return from buying and holding an asset over a certain period of time (in trading days or months). Therefore, they are closer to the actual investment experience of the investor. The BHAR is defined as the difference between the realized buy-and-hold return and the expected (or normal) buy-and-hold return over the same period. The buy-and-hold return of an asset is calculated as the product of 1 plus the asset’s return on day t for all days t over a defined period of time. Calculation on a monthly basis is possible as well (in case of long-term event studies). To approximate the expected (or normal) return of an asset, a characteristic-based matching approach is applied. The matching is usually based on the industry, on firm size and on growth perspectives (using the market-to-book ratio). However, more sophisticated matching, for example via firm age, past performance, or governance attributes, is possible and can increase the accuracy of the results. In most cases, either a market or an industry index or the aforementioned matched firms approach (see step 6) is used to calculate the normal return of an asset over a defined period of time.
Thus, the counterfactual is what an investor would have earned had he invested the same amount of cash over the same period of time in a comparable stock index or a comparable firm (or portfolio of firms). The average buy-and-hold abnormal returns (ABHARs) can be calculated on an equal- or value-weighted basis. The value-weighted average is usually calculated as the BHAR of each firm weighted with its market capitalization. The use of weights can be activated in the Event Study Metrics software (by selecting ‘Use Weights’ in the left corner at the top).
Although CARs and BHARs are often very similar for short time horizons, CARs are the method predominantly applied in short-run event studies. The contrary is true for long-run event studies for which BHARs seem conceptually better. We refer to Fama (1998) for a comparison and discussion of CARs and BHARs.
ARs, CARs, BHARs and ABHARs can easily be inspected (and exported) in Event Study Metrics by clicking on the respective tabs ‘ARs’, ‘CARs’, ‘BHARs’ and ‘ABHARs’ (in the right corner at the bottom). For each event window, the main event window and the sub-windows, the software calculates CARs and CAARs as well as BHAR and ABHARs. It also calculates the ratio of positive to negative CAARs and BHARs (denoted ‘Pos : Neg’) to provide an indication of how heterogeneous the results are. Furthermore, CAARs and ABHARs are graphically illustrated in the software for the main event window when clicking on the tab ‘Graph’. The software calculates both CARs and CAARs (BHAR and ABHARs) because usually one is interested in the overall average effect (i.e., the CAAR) and in explaining firm-specific effects (i.e., the CARs) in a subsequent cross-sectional analysis. Event Study Metrics Plus allows you to immediately perform such a cross-sectional analysis of the calculated CARs subsequent to the event study.
Test for statistical significance
As a last step, you need to test whether the abnormal returns (AARs and CAARs) and buy-and-hold returns (BHARs) are significantly different from zero on a statistical basis. To do so, several test statistics can be applied. The software includes all state of the art test statistics: the time-series t-test and the cross-sectional t-test (Brown and Warner, 1980 and 1985), Patell’s Z-test (Patell,1976), the standardized cross-sectional test (or BMP-test) developed by Boehmer, Musumeci, and Poulsen (1991) which accounts for event-induced volatility, the rank test (Corrado, 1989; Corrado and Zivney, 1992), the general sign test (Cowan, 1992) as well as the new, more sophisticated version of the BMP test-statistic with cross-correlation adjustment proposed by Kolari and Pynnönen (2010). The latter can be activated in the settings form. It is not run as a basic test because of its large computational effort. The rank test and the general sign test are non-parametric tests, i.e., they do not assume that the data have a particular probability distribution. All other tests are parametric tests.When you perform long-run event studies, be aware that buy-and-hold abnormal returns (BHARs) tend to be right-skewed, as shown by Barber and Lyon (1997) and Kothari and Warner (1997). As a result, the test statistics of a basic t-test likely suffer from a skewness bias causing them to be less reliable. To overcome this problem, a skewness-adjusted t-test, initially developed by Johnson (1978), is used. It is a transformation of the basic t-test which eliminates the skewness bias and allows for more reliable conclusions about the significance of your long-run event study results. An advanced version of the skewness-adjusted t-test was developed later by Lyon, Barber and Tsai (1999). The authors recommend the use of a bootstrapped version of the skewness-adjusted t-test and show that it yields well-specified test statistics. Kothari and Warner (1997) also recommend the use of bootstrap procedures. Besides the basic t-test, Event Study Metrics provides the skewness-adjusted t-test and its bootstrapped version to test BHARs for statistical significance. The bootstrapped version can be activated in the settings form.
Return calculation models (step 6), abnormal return calculation (step 8) and significance tests (step 9) are further motivated and explained in detail in our methodology section. The section includes formulas for all return models and all test statistics as well as null hypotheses for the latter. If you wish to get more information apart from this event study blueprint and the methodology section, you can read additional published academic papers providing excellent overviews of the event study methodology such as Kothari and Warner (2008), MacKinlay (1997) and McWilliams and Siegel (1997). As a text book, we recommend Campbell, Lo and MacKinlay (1997). The seminal paper on event studies, i.e., the first that applies the event study methodology as we know it today, is Fama et al. (1969). An overview and discussion of efficient capital markets is provided in Fama (1998, 1991 and 1970).
Barber, B.M. and Lyon, J.D. (1997): Detecting Long-Run Abnormal Stock Returns: Empirical Power and Specification of Test-Statistics, Journal of Financial Economics 43, 341-372.
Boehmer, E., Musumeci, J. and Poulsen, A.B. (1991): Event-Study Methodology under Conditions of Event-Induced Variance, Journal of Financial Economics 30, 253-272.
Brown, S. and Warner, J. (1980): Measuring Security Price Performance, Journal of Financial Economics, 205-258.
Brown, S. and Warner, J. (1985): Using Daily Stock Returns – The Case of Event Studies, Journal of Financial Economics 14, 3-31.
Corrado, C.J. (1989): A Non Parametric Test for Abnormal Security Price Performance in Event Studies, Journal of Financial Economics 23, 385-395.
Corrado, C.J. and Zivney, T.L. (1992): The Specification and Power of the Sign Test in Event Study Hypothesis Tests Using Daily Stock Returns, Journal of Financial and Quantitative Analysis 27, 465-478.
Cowan, A.R. (1992): Nonparametric Event Study Tests, Review of Quantitative Finance and Accounting 2, 353-371.
Fama, E.F. (1998): Market Efficiency, Long-Term Returns, and Behavioral Finance, Journal of Financial Economics 49, 283-306.
Fama, E.F. (1991): Efficient Capital Markets: II, Journal of Finance 46, 1575-1617.
Fama, E.F. (1970): Efficient Capital Markets: A Review of Theory and Empirical Work, Journal of Finance 25, 383-417.
Fama, E., Fisher, L., Jensen, M. and Roll, R. (1969): The Adjustment of Stock Prices to New Information, International Economic Review 10, 1-21.
Fama, E.F. and French, K.R. (1993): Common Risk Factors in the Returns on Stocks and Bonds, Journal of Financial Economics 33, 3-56.
Johnson, N. J. (1978): Modified t Tests and Confidence Intervals for Asymmetrical Populations, Journal of the American Statistical Association 73, 536-547.
Kolari, J. and Pynnönen, S. (2010): Event Study Testing with Cross-Sectional Correlation of Abnormal Returns, Review of Financial Studies 23, 3996-4025.
Kothari, S.P. and Warner, J.B. (1997): Measuring Long-Horizon Security Price Performance, Journal of Financial Economics 43, 301-340.
Kothari, S.P. and Warner, J.B. (2008): Econometrics of Event Studies, in: Eckbo, B.E. (ed.), Handbook of Corporate Finance: Empirical Corporate Finance, Vol. 1, Elsevier/North-Holland, 3-36.
Lyon, J. D., Barber, B.D. and Tsai, C. (1999): Improved Methods for Tests of Long-Run Abnormal Stock Returns, Journal of Finance 54, 165-201.
MacKinlay, C.A. (1997): Event Studies in Economics and Finance, Journal of Economic Literature 35, 13-39.
McWilliams, A. and Siegel D. (1997): Event Studies in Management Research: Theoretical and Empirical Issues, The Academy of Management Journal 40, 626-657.
Patell, J. (1976): Corporate Forecasts of Earnings per Share and Stock Price Behavior: Empirical Tests, Journal of Accounting Research 14, 246-276.
Ritter, J. (1991): The Long-Run Performance of Initial Public Offerings, Journal of Finance 46, 3-27.