Panel with Fixed Effects - variable omitted because of collinearity

[Sebastian Griebling]

Hello everyone,

I am running a regression on panel data including a bunch of companies. Hence I control for the effects of time and the individual company.
Within my variables I have, 4 which are varying over time and company (no problem).
But I also have one variable which is only varying over time (inflation). When I include it with the other variables, it (not surprisingly) gets omitted because of collinearity.
I am asking myself where can I declare it as what it is - so Stata knows that it is a variable that only varies over time ...

Thank you very much in advance for any help

[Carlo Lazzaro]

Sebastian:
welcome to this forum.
Some comments about your post:
1) from your description, I can only guess that you went -regress- instead of the (much better) -xtreg,fe-;
2) I fail to get the rest of your description. However, the -fe- estimator wipes out all the time-invariant variable.
As an aside, in your future posts please post what you typed and what Stata gave you back (as per FAQ). Thanks.

[Sebastian Griebling]

Thanks for your remarks Carlo.

I entered:

reghdfe rpt beta turnover dummy_neg_ret dummy_vola_change sentimentindex, absorb(date name)

--> rpt as dependent variable (volatility proxy)
--> fixed effects to account for are date (daily) and (Company-)name
--> Beta, Turnover, 2 Dummys and Sentimentindex as independent variables --> the first four are varying over time and company; the sentimentindex however only varies over time and is for each company the same (similar as if someone would analyse inflation as explanatory variable)

Stata returned:

note: sentimentindex is probably collinear with the fixed effects (all partialled-out values are close t
> o zero; tol = 1.0e-09)
(MWFE estimator converged in 5 iterations)
note: sentimentindex omitted because of collinearity

As mentioned, this is not surprising as the sentimentindex is the same for all companies, logically Stata omits it for collinearity.
The problem is, that I entered the sentimentindex with all the other (time and company varying) variables, I would like to declare this variable however as only time-varying and I am wondering if this is possible.

To make it hopefully more clear. All the other variables would have index X_it with i for company and t for time but the sentimentindex is only X_t and this is something I could not put into the software, yet.

Best regards
Sebastian

[Carlo Lazzaro]

Sebastian:
as per your description of -sentimentindex- I do not think you can obtain a coefficient under the -fe- framework (which is the first choice in general and in situation like this one is also the only reasonable way to go).
Just out of curiosity, I'd try a -re- panel regression to see whether a coefficient for -sentimentidex- can be estimated or not.