To apply the event study methodology, scholars work through a fairly structured workflow of methodological choices and analytical steps. Further, the method imposes a conceptual structure on the analyzed data that is inherently linked to important methodolical choices. Together, this research workflow and data structuring make up an event study blueprint that underpins all event studies. The characteristics workflow for performing an event study is described by the steps listed beneath and illustrated in Figure 1.
- Choose an expected/normal return model that matches your research question and data availabitity
- To set the event window length and the position of the event window, answer the following questions: On what date did the market receive the information about the event you want to study? Has there been potential information leakage prior and a certain market pricing/digestion period subsequent to the date?
- Calculate expected/normal returns (i.e., returns for the firms' stocks assuming the event had not taken place)
- Calculate abnormal returns (ARs) as the difference between the actual observed returns and the 'normal/ no-news' returns for each firm and day in the event window
- These single-day 'abnormal returns' can then be further aggregated across time to 'cumulative abnormal returns' (CARs) or cross-sectionally to 'average abnormal returns' (AARs). Aggregating the abnormal returns across both time and firms yields the 'cumulative average abnormal returns' (CAARs).
- In a final step, calculate significance tests to establishe whether the abnormal returns found at any of the AR-, AAR-, CAR- or CAAR-levels are statistically valid/significant
Figure 1: Event Study Blueprint / Workflow
Depending on the return model chosen, event studies either imply the use of an event window only (e.g., the market-adjusted model) or an event and an estimation window (e.g., the market model). Most common, the 'market model' is used. It predicts normal returns with a regression analysis that regresses stock returns on market returns over the 'estimation window'. Through this analysis, the typical relationship between the stock and its reference index is captured in two parameters (i.e., alpha, beta). Figure 2 sketches the data structure used by event studies and provides further information on how this data structure is used by the 'market model'.
Figure 2: Data Structure of an Event Study
Adapted from Benninga (2008: 372)
While the above described event study blueprint offers considerable room for methodological variety, many scholars have convered to a similar set of choices. Specifically, most scholars opt for the 'market model' to predict normal returns, choose a symmetric three-day event window (i.e., [-1, +1]), and locate an estimation window of at least 120-day length right before the event window.
Event studies can be conducted with different tools, such as Excel, Stata, or this website's 'abnormal return calculators'. We recommend to use our websites free calculators for several reasons. First, conducting event studies in either Excel or Stata will require substantial programming from you. Second, even if you succeed in this programming, you will likely find yourself missing critical features such as 'parameter-setting as per event' or 'event grouping'. And finally, you will have to verify the correctness of your results and code by benchmarking it with prior results. We went through this tedious excercise and recommend you to invest this time into other areas of your research project.
References and additional links
Benninga, S. 2008. Financial modeling (3 ed.). Boston, MA: MIT Press.