New with econometrics - Changes in Beta event study question

[Quatze Quatze]

Hello,

I am trying to do an event study regarding beta and check if beta changes after stock splits dates.

I am planning on checking if the beta computed within the time frame [-30, -1] is statistically different from the beta calculated within the time frame [+1, +30]. You can find my data file attached to this post [ 1mb ]. Any idea how I can check if the 'after beta' is statistically different from the 'before beta'?

Kind Regards,
Quatze

[Clyde Schechter]

Well, I gather this is finance data. I know that "beta" has a specific meaning in finance, but I don't know what that is. In statistics generally, beta is often used to refer to a regression coefficient. So I'm going to illustrate the approach using a fixed-effects (for permno) regression of your variable ret on vwretd. If that's not right, modify this code accordingly:

xtset permno

gen byte before = inrange(date, event_date-30, event_date-1)
replace before = . if !inrange(date, event_date-30, event_date+30) | date == event_date
label define before 1 "Before Event" 0 "After Event"
label values before before

xtreg ret i.before##c.vwretd, fe // OR SUBSTITUTE APPROPRIATE REGRESSION COMMAND

The difference between the before and after beta's (before minus after) will be estimated by the coefficient of the interaction term, and the statistical significance by the corresponding p-value.

Side point: I initially tried -xtset permno date- and learned that you have multiple observations with the same permno and date (specifically there are 23 distinct combinations of permno date which occur twice each). If that's really how the data are supposed to be, that's fine. But usually there would be just one observation per date on any given firm in this kind of data. So you might want to re-check your data to make sure it isn't corrupted, or something other than you think it is.

[Quatze Quatze]

Dear Mr Clyde,

Thank you very much you are of tremendous help.

There is a small problem with the code that you have provided. It seems to take into account actual calendar days, while my file contains trading days (exactly -30 before day 0 and +30 after). I would like to test if beta changes not based on calendar days, but actual trading days. Furthermore, you are right and I am sorry I did not specify this before, in my finance data the beta is represented by the regression coefficient.

Small questions:
1. Why did you use a fixed-effect for permno?
2. What does this line do?
xtreg ret i.before##c.vwretd, fe Kind Regards, Quatze

[Clyde Schechter]

With regard to the calendar vs trading days, this implies that you have to create a Stata business calendar and convert your dates accordingly. Read -help datetime_business_calendars- and the corresponding manual sections for how to do that. With that done, the code will capture the correct dates. I can't offer you any specifics on doing this: the issue never comes up in my line of work and I have never used them. I just know about them, and I know that it is what you need here.

My -xtreg, fe- command, which does a fixed-effects regression of ret on vwretd with an additional main effect for the before event condition and the interaction between that and vwretd, was just for purposes of illustration. You may have reasons for preferring a random-effects model, or for pooling the data into a flat model for that matter. And perhaps you want to do separate regressions for each firm and do some kind of summary statistics on the firm-specific results. You did not say what kind of regression you wanted to perform, and I do not know the substance of what you are working on: I'm an epidemiologist who uses Stata extensively and have only a minimal knowledge of finance. I've noticed on the Forum that fixed-effects regression models are very popular in finance and economics, so I used that as a placeholder in my code.

Since you could not decipher that -xtreg, fe- command on your own, I'm inferring that you're not familiar with factor variable notation. This is one of Stata's most useful features, and you should invest time in mastering it. It isn't hard, and it's all documented in -help fvvarlist- and the corresponding manual section. The short amount of time it takes to learn them will be paid back very rapidly.

[Quatze Quatze]

Dear Mr Clyde,

Thank you very much for your help. I've researched into it and also I've looked for other methods online. It seems simpler and more intuitive if I use a Chow test on my data based on the values before the event date (day 0) and the values after the event date.

Can you please provide some help in how I would be able to run a Chow test in stata?

Kind Regards,
Quatze

[Clyde Schechter]

I don't use the Chow test myself. Suffice it to say, if you run -findit chow-, you will get links to several sources of information about this, including some user-written programs you could download for this purpose.

[Quatze Quatze]

Quick question,

I've installed the module called chowreg , apparently it will help me run a chow test. However, how do I run a module? Is there a special button that I can press in order to run this particular module?

Sorry if the question is too simple.

I really appreciate the answer.

[Clyde Schechter]

Type -help chowreg- in the Stata command window. The viewer will open up and will give you information on how to use the command. This is true of any official Stata command, and nearly all user-written ones as well.