Inability to apply Year Fixed Effects

[Shashank Srivastav]

Hello Members, I am facing one problem where I have my primary variable of such a nature that it remains same (constant) for all the firms in a year. For example, GDP or mean Temperature of a country. When we would have a panel data for multiple firm-year observations, GDP would remain same for all the firms in a particular year t. In such a case when I am applying Year FE (using reghdfe command) it drops my primary variable due to the multicollinearity issue with the primary variable. So, I am unable to apply Year FE. Can anyone suggest me a proper and appropriate resolution to this??

[Carlo Lazzaro]

Shashank:
unless you have access to more fine-grained data (say, country-specific GDP per month), there's non chance in heaven that the -fe- estimator will not wipe out time-invariant variables.
A possible option, if appropriate, is switching to the -re- estimator, with the caveat that -re- is inconsistent if -fe- is the way to go.

[Shashank Srivastav]

Dear Mr Carlo, Thanks for your reply. In the case I am unable to use year FE, as a substitute, can I add a few controls such as a Time-trend variable and macro-economic factors that may affect all the firms in a particular year. I think year fe is also used for the same purpose - to control for any such factors that may impact all the firms in a given year or to control any year-wise trend. Will I be correct ??

[Carlo Lazzaro]

Shashank:h
time fixed effect is basically included as a predictor to investigate, other things being equal, the contribution of time, taken as a categorical independent variable, to within panel variation of the regressand.