Help regarding Event Study Analysis

[IvoBrit]

Hello,

Im a master student, currently developing my thesis on Post-Earnings Announcement Drift (An Euro Stoxx 50 Analysis) between 2012 and 2017. I've defined the event window (-20,0,20) and computed the normal returns and market model parameters for each firm in my study(51 companies). However, the Beta parameters of all aren't statistically significant (p-value > 5%), which makes my study irrelevant.

Also, how can one then estimate AAR and CAAR for all companies in Excel when you have multiple events per company throughout your analysis period?

I am sorry, I had a course on Excel but they only teached us the basics and I'm currently stuck in my thesis since i don't know how to aggregate the abnormal returns for all companies/events on Excel. Any help or suggestions would be welcome.

Thank you!

I can't upload my excel file because it is 5MB, but basically i computed all the returns for market and the companies in excel and computed the intercepts and slopes of the various companies, i think what led to the very small betas and alphas might be the period i used when estimating since i used almost 1530 days.

[scottiex]

Sounds to me like you need to review some literature and find some papers doing similar analysis and copy their methodology as this is about your supervisors (and any destination journals etc) having confidence in your method of calculating the Cumulative Average Abnormal Return.

For example the first thing I found was googling was
https://www.researchgate.net/publica...sted_Companies
I haven't read it in detail but women on boards might be a scenario where multiple events occur per company for a period. Presumably there must be cases of that you can find in the literature.

I guess you have access to the academic journals and can find far more than me. I'd skim read as many of these as I can find and extract out whatever methods look good.

If you can use those to lay out the exact rules and dummy up something in excel and explain exactly what you want to happen maybe we can try to help you with the technical excel issues?

[IvoBrit]

Thank you for the reply! I have studied the methodology a lot but I can't figure out the technical aspects of it. In a sense, how to apply it on Excel. I have computed the returns for all 51 companies and the market index in my study on Excel. I've also computed the intercept and slope of every regression for each company (Returns of the company = intercept + slope*Market Returns
My current concern is how to apply the next step of the methodology for a high number of companies with a high number of events on Excel. Sorry if im not being clear.

[scottiex]

In that case its a communication issue and you can just move to that last step.
Explain to us what that next step (the one you have difficulty with) is in simple terms and maybe give us some example numbers in a spread sheet that we can connect together and use to test that our understanding is exactly right?
I presume you have worked out some sample results you just don't know how to put those results into excel so you can process all results?
or you know the formula but you don't know what excels functions are for them?

[IvoBrit]

Scottiex thank you so much for the replies, even if I can't figure it out I want to say thank you for taking the time out to answer my topic. I have an excel with returns, abnormal returns calculated. However its 5MB and i can't post it here due to size restrictions on the forum.

Basically my next step is to include over the event window i chose -20 until 20 in terms of days so basically 41 cells in excel and then for each one of this days compute the Average Abnormal return, average cumulative abnormal return. However I only know how to do this for a single event, but i have 1405 events and i have to aggregate them all for each day of the event window from -20 to 20.

[scottiex]

As to the file size. maybe cut out half the data then upload it?

well you can add them together but you want to do some statistical analysis on them don't you... so maybe you can get a statistically significant result for some sort of hypothesized effect. I presume you want to do statistical analysis on 1405 * 40 data points to try to show.... maybe earnings are bit depressed before and rebound after? (obviously considering the 40 data points are not independent).

I presume other papers will explain the tests they applied and will say the names like Bartlett's test, or Levene's test Mann-Withney Wilcoxon, Wilcoxon range etc. which will say various things about your data. And quite often would have those formula written out that we might try to emulate.

Althoguh TBH my gut feel is you should get at best a pretty hard to detect result for major shares on major markets because if you did get a result then traders would have detected it and arbitraged out most of the effect. So finding your result is p > .5% is quite likely a valid result...

[IvoBrit]

I removed a lot of data from the excel file but you can see some of the companies i left what i did so far. Basically the idea now is to use all the event dates i have which are 1045 and instead of doing a event study analysis 1 by 1 aggregate them under 1 single analysis with average AR and average CAR for each day of the event window from -20 to 20. 0 is the day the event takes place.

[scottiex]

http://www.eajournals.org/wp-content...k-Exchange.pdf
seems like you have your abnormal returns as % but I think standard is to have them divided by standard deviation

i attached a file to see if it gets you started.
I don't really know what I am doing with the stats (anymore, its been many a year) and I am not entirely sure what you want so I haven't gone too deep into this here but I created a little table of Formula to do this from T1 on here.
It is to get the AAR and the SAAR(standard AAR) and then sum them and based on a table of companies and date ranges which I presume you have.

that said I guess your non significant result was pretty similar to this and you may not get any result in which case maybe you are suggesting you want to amend the hypothesis.
Beating data until it gives a result is a little dubious but people do do it.

[IvoBrit]

Thank you scottiex, any help is valuable to me! Thank you for taking some time to help me out with this, I appreciate it!