[Andrea Zaugg]

I just found a paper that did a similar analysis as I am doing at the moment.

They however did not only cluster standard errors by firm, but they also included dummy variables for time.

Does it make sense to do both? I understand it, that with the clustering of the standard errors per firm, we control for factors that influence one specific firm over the whole observation period. On the other hand, including time-dummies controls for influences on all the firms in specific years. Is this quite simple interpretation correct?

I fitted my model with and without such time-dummies. The results are quite similar but a little more significant when time-dummies are included. Would you recommend including time dummies or not? I could the the respective other option as a robustness test.

best,

Andrea

[Andrea Zaugg]

I just found a paper that did a similar analysis as I am doing at the moment.

They however did not only cluster standard errors by firm, but they also included dummy variables for time.

Does it make sense to do both? I understand it, that with the clustering of the standard errors per firm, we control for factors that influence one specific firm over the whole observation period. On the other hand, including time-dummies controls for influences on all the firms in specific years. Is this quite simple interpretation correct?

I fitted my model with and without such time-dummies. The results are quite similar but a little more significant when time-dummies are included. Would you recommend including time dummies or not? I could the the respective other option as a robustness test.

best,

Andrea

[Carlo Lazzaro]

Andrea:
you probably missposted the same query twice.
You can actually test -i.year- via -testparm-.

[Andrea Zaugg]

Hi Carlo

yes, I realized that. I however tried to delete one of the posts and couldn't find how to do that.

Just first to not mix up different topics: Is my interpretation in above post correct?

Ok I in fact didn't know about -testparm-, thank you. I thus assume that you would recommend including time-dummies and correct the standard error per company at the same time if the -testparm- function for all time-dummies is significant?

[Carlo Lazzaro]

Andrea:
the two things are in fact independent.
-if there's a theoretical reason to cluster your standard errors, it remains valid with or without -i.time- included in the set of your predictors (set aside the instance when -i.year- can cause the problem that leads you to cluster your standard errors);
-yes: if -testparm- reaches statistical significance, keep -i.year- among predictors.