Coefficients/results instability - Eventstudy2

[Gabriele More]

Dear stata community,

In using the writing the following commands to stata

eventstudy2 event date using market.dta, returns(pcreturn) model(MA) marketfile(index.dta) marketreturns(ri) idmarket(ref) car1LB(-1) car1UB(0) car2LB(-2) car2UB(-2) car3LB(-1) car3UB(-1) car4LB(0) car4UB(0) car5LB(+1) car5UB(+1) car6LB(+2) car6UB(+2) car7LB(-20) car7UB(+20) car8LB(-2) car8UB(+2) car9LB(-1) car9UB(+1) car10LB(0) car10UB(+1)

I get very unstable AAR and CAR values, statistics and significance tests. I tried to specify the estimation period window option, but the results are still unstable, and can change significantly between themselves.

Yet, the data is exactly the same from one estimation to the next, the commands are the same and I don't know at all what could drive this instability.

Any advice is immensely appreciated!

Gabriele

[Tobias Bornemann]

Hi Gabriele,

I experienced the same issues. The problem is with the ado nearmrg. See here for a solution which did the trick for me:

https://www.stata.com/statalist/arch.../msg00459.html

Hope I could help.

[Gabriele More]

Hi Gabriele,

I experienced the same issues. The problem is with the ado nearmrg. See here for a solution which did the trick for me:

https://www.stata.com/statalist/arch.../msg00459.html

Hope I could help.

Hi

I am not too sure i understand the answer. The link talks about a nearmrd command (which I believe is in eventstudy2 source code). In reading the command instructions, we cannot theorically add this option to the main stata command. Am I wrong?

So only the first iteration is the good one as the porblem is in the sort function which is ambiguously defined, if I understand correctly.

Thank you for the reply, you saved my heart
Gabriele