Expert Dialog

Question #58: Event Study Calculator Instructions (ARC)

I am writing to you concerning the Event Study Calculator. Indeed, I would like to use this helpful tool but I am confused with two elements:
- In what code should the firm ID be? (ISIN, SEDOL, NAME...)
- For the firm and market data, how many days should we use for the returns (the example shows around 800 days which is much more than the event and estimation period)?

Answer by Dr. Markus Schimmer:
The ID does not have to follow any specific convention. It must simply be unique as per each line. Hence, I recommend using simple/plain numbering. Regarding the amount of return data needed: You need as much as your event and estimation windows require. So if these windows are next to each other and stretch across 200 trading days, you'll need 202 closing prices as input data to calculate 200 returns. There is no harm when you include more data in your upload files - the tool will pick the required data sequence. You're correct in that our own sample files include more data than necessary... the reason is that we started off with a larger request file, trimmed it down, but did not adjust the firm and market data files.
Question #57: Scope of Service & Data Inputs

I would like to use your server app to conduct an event study and trying to find out if the app can completely replace an analysis with a statistics software like Stata or R? Also, I was wondering if you offer further advice or forums to better understand the preparations of the datasets before one is able to upload the request files? Is it also possible to use own (rather specific) indices as benchmark returns for CAARs and to use non-Yahoo or Google Finance sources but data from commercial data providers like Datastream for the share price data to be analyzed with the app?

Answer by Dr. Markus Schimmer:
EST's server-side research apps replace the use of other statistical software for the defined scopes of each individual research app. For the case of an event study, this means the following: As long as want to perform an event study only and are interested only in ARs, AARs, CARs, CAARs, and the respective test statistics, you are all set. Instead, if you want to explain the abnormal returns using multiple regression analysis, you will need another statistical software package for the regression. EST only provides analytics and no data. Hence, you are free to choose whatever financial data you can organize and put into the apps. Simply make sure the data lives up to the requirements of the respective research app - as described on the introductions to the individual apps.
Question #56: AR vs. AAR / CAR vs. CAAR

I have calculated the AR and CAR, as well as the AAR and CAAR. However, while understanding the different calculations behind these i am not sure when it is better to use AR or AAR, CAR or CAAR respectively. Could you explain what the differences are in terms of interpretation? When should which version be used for analysis? Are there different kind of 'significance' tests to be done to check these two different calculations?

Answer by Dr. Markus Schimmer:
The extra A in AAR and CAAR stands for average, meaning, it is the average of the ARs and CARs as per each sub-sample of your overall set of events. You define the sub-samples you want to apply in your analysis by using the "grouping variable" in the request file. In terms of interpretation, take the following example: You ran an analysis about differnt M&A types, let's say type 1 where the buyer paid in cash and type 2 where the buyer paid in stock. Consequently, you used as grouping variables "type 1" and "type 2". The ARs and CARs will give you insights on each and every single M&A transaction studied, whereas the AARs and CAARs will tell you about your two merger types in an aggregated manner. Yes, the test statistics differ. Please visit our test statistics page for details on how they are different.
Question #55: Importance of currency

I am conducting event study with different companies across Europe and all of them has different currency. Is it important in event study calculations to have the same currency for all companies in your sample? Or, as in the end we still get Returns, so currency is not significant?

Answer by Dr. Markus Schimmer:
I tend to agree with your assessment. Currencies shouldn't matter much since the analyses are performed based on returns. What matters, however, is market efficiency and if you're doing a multi-country study, market efficiencies may differ between countries - particularly if some are developed and others are emerging market countries. In your case, however, you should be fine.
Question #54: Large scale event study

I am currently conducting a large-scale event study. I am using R and the EventStudy package. It turns out that I am only able to upload around 2000 stock prices otherwise I receive the error code:

Request Status Code: 500
Error: Argument 'txt' must be a JSON string, URL or file

Is there any way to use the package for a large amount of data e.g. 200k stock prices in the overall analysis?

Answer by Dr. Markus Schimmer:
We strongly recommend using the API directly for large-scale studies. This will give you also some additional flexibility in structuring the analysis - if needed.
Question #53: can not calculate event window before event day

I can not calculate event window before event day like -7, -1. System force me to write the last day of event window bigger than 0.

Answer by Dr. Markus Schimmer:
Yes, for technical reasons, we need to enforce that the event date needs to be in your event window. You can conceptually circumvent this requirement by changing the calendar date of your event accordingly.
Question #52: What's your current API key?
Answer by Dr. Markus Schimmer:
You find the API key on this page:
Question #51: length of event window

I am working on a study of biopharma stock prices's responses to receiving "breakthrough therapy designation" (BTD) from the FDA. Our hypothesis is that pre-commercial biotechs will experience a transient stock bump, but excess returns will dissipate by ~90 days. My question relates to the appropriate / maximum length of the event window. What's the maximum or optimal length of the event window, and what factors might dictate the appropriateness of a longer vs. shorter one? We have seen papers with 5-30 day windows, but have not seen a good discussion of how to select the length.

Answer by Dr. Markus Schimmer:
The appropriate length of the event window is driven by your research assumptions on how the capital market is processing the information that is released by the event studied. From what you wrote I assume you have the following assumptions: (1) Prior to the BTD release, there is no information leakage (2) You are studying companies in the US, which means that we should assume semi-strong market efficiency (3) You have the hypothesis that during the 90 days subsequent to the BTD release, abnormal returns are negative and undo the positive effect surrounding the BTD release This will give you the following type of parameters to your event study: + Short event window around the BTD date to capture the positive effect: [0,2] if you are certain that there is no information leakage or [-2,2] if you assume some leakage. You choose the second number based on how opaque the news might be to the capital markets. BTD releases sound quite clear, so you wouldn't assume the US capital market to take more than 2 days to price it in. Note: the smaller your companies, the less coverage they have, the long it may take... + Then, for your 90 day hypothesis you obviously will take a 90 day event window. Most likely, you would ground the calculation on the same estimation window as your [-2,2] window, meaning on the period prior to the BTD event - this will ensure twofold: one, your second analysis uses the same presumed relationship of your firms to the stock markets, and two, you overrule the effect of the BTD event on the stock market-company relationship, which you factually negate given your hypothesis (meaning, this choice is in line with your hypothesis)
Question #50: Event Study about rating changes

Dear Event study tools,

I am studying the rating changes effect on stock price, at the moment I use the market model to calculate the abnormal returns in the event window for each rating change for each company in the sample. I have this data but I don't understand how I can aggregate them into the Average abnormal returns and later into the CAAR, also I am a concerned about how to determine the standard deviation for T- test.

Many thanks in advance for your kind attention

Answer by Dr. Markus Schimmer:
You can use the grouping variable in our abnormal return calculator to get the AARs and CAARs as well as all associated test statistics calculated.
Question #49: t-test fpr CAAR

Dear Dr. Müller,

for the t-test (mentioned on eventstudytools - parametric test statistics number 1) it is possible to test H0:ARi,t=0 and H0:CARi=0.

For my study I am using the paper of MacKinlay (1997). On page 24, formula (20), he used a test of which I thougt that it is a t-test for CAAR. Dymke (2010) used this formula of MacKinlay on page 77 too. But on page 79 he talks about the parametric test of Brown/Warner (1985).

I am confused and need some help.

Thanks in advance.

Answer by Dr. Simon Müller:
Hi. You are correct formula (20) on page 24 is a CAAR t-test. This is the same test as in point 2 on our test statistic page. Brown and Warner uses this test in their publication (1985), so you may name this test "parametric test of Brown/Warner", but this is not common. More common is Cross-sectional test. Best, Simon