eventstudy2: how to read carfile for BHAR model

[Lynn Lyu]

Hello everyone

I have a quick question about significance test for BHAR model, using eventstudy2

for example, how to read this carfile generated by eventstudy2?

Which significance test this command uses to generate significance level for BHAR model? Is it possible for me to have the p-value?

Many thanks and best wishes,
Lynn

[George Ford]

I believe to be significant, the BHAR value would need to exceed (in this case) the Sig*Up values (or be below them at the other end), which it does not. BHAR is within all the confidence intervals.

[Lynn Lyu]

Thank you very much George! I checked the results and theory - your explanation make sense.

I just wonder is this super large BHAR value common? It is not significant and we should not believe it. But it is really beyond my expectation.

[George Ford]

Are you interested in long-run or short-run abnormal returns?

[Dalal AlQadeeri]

Dear Thomas,

Could you please clarify/confirm the interpretation of the “Sig95Low” and “Sig95Up” values that result from the BHAR model? (just like the post above).

My colleague and I understood these terms differently and thus wanted to confirm. Thank you!

[Thomas Kaspereit]

Dear Sir,

I confirm that Sig* are bootstrapped confidence intervals, however, it is not BHAR which needs to be compared to them (as Gorge Ford wrote; nevertheless: thanks a lot, George, for intervening here and having prodivded guidance as best as possible given the limited documentation in my package), it is the SkewAdjT.

Here some lines from a study of mine which make it very clear:

foreach i in "90" "95" "99"{
gen sig_`i' = 1 if Skew < Sig`i'Low | Skew > Sig`i'Up
}


sig_* are indicator vars which capture the "asterisks-levels" ... *, ** , *** :


gen sig_level = ""
replace sig_level = "*" if sig_90_`v' == 1
replace sig_level = "**" if sig_95_`v' == 1
replace sig_level = "***" if sig_99_`v' == 1

Regarding the concrete BHAR values in the example above: They look way too high; I suspect that Lynn Lyu had an issue with the decimals, e.g. she used 10 instead of 0.10 for a ten percent daily return.

Best
Thomas

[Saima Nazir]

Dear Thomas,

I have a sample of 89 firms and have to calculate BHAR using monthly stock returns for event windows of 6, 12, 24 and 36 months. Can I use eventstudy2 for this as it seem that eventstudy2 calculates BHAR using daily returns.

Regards,
Saima

[Thomas Kaspereit]

Dear Saima,

I have not tried yet because even if one looks at 6, 12 ... etc months of BHAR, one could use daily returns as a basis of calculations.

But, yes, I think it should work even with monthly data if you feed a date variable to eventstudy2 that has monthly format, i.e., one integer notch representing one month plus or minus. It will not work if one simply keeps daily dates and just one per month end -- this will certainly not work.

By this way: this applies to any other frequency, too. One could even imagine to use eventstudy2 on intraday tick data, as long as each intervall is represented by one integer notch of the date variable.

Best Thomas

[Saima Nazir]

Dear Thomas,

Thank you so much for your instant response. I will try to apply this on my data and see whether it runs or not.

Regards,
Saima

[Saima Nazir]

Dear Thomas,

I tried to run the BHAR model on monthly stock returns using the following command and it showed some error:

eventstudy2 Security_id Date using Security_Returns_BHAR, returns(Security_Returns) model(BHAR) marketfile(Mkt_Returns_BHAR) marketreturns(Market_Returns) idmarket(Market_Reference)

Generating dateline ...
...succeeded
Preparation of event list ...
...succeeded
Preparation of security return data...
...succeeded
Preparation of market and/or factor return data...
...succeeded
Merging event dates and stock market data...
...succeeded
Calculating abnormal returns...
variable __00002H not found
r(111);


Could you please guide me what it is referring to?