Discussing differences of Firm vs. Industry fixed effects

[Luca Haseney]

Hello Stata community,

I currently test firm characteristics on the outcome of firm profitability.

To do this I have developed two models, once with time and firm fixed effect, clustered by firm:
1. reghdfe Profit x1 x2 Country x3 x 4 x5 x6, absorb(INDUSTRY Year) cluster(FIRM)

And once by time and industry fixed effects, clustered by firm:
2. reghdfe Profit x1 x2 Country x3 x 4 x5 x6, absorb(FIRM Year) cluster(FIRM)

As I learned in my other thread, since firms do not change the country, the variable country is collinear with the fixed effects and therefore omitted.

My question:
Is it fair/plausible to include both models and interpret the outcomes separately?
Because my first hypothesis is that the country of the firm has a siginificant influence on the Profit of the firm. I.e. interpret the results for my first research question, only focusing on the first equation?
The other variables, x1 to x6 then would be interpreted using the second equation.

Is this considered wrong?
The adjusted R squared for my firm fixed effect model is around 88 percent, while the industry fixed effect model has roughly 15 percent. At the same time, the coefficients themselves are not siginificant. For my understanding this is a sign of multicollinearity.
Should this influence my decision which model to prefer? And if yes, how?

[Joro Kolev]

Firms also rarely change industries -- it is not impossible, but still rare -- and therefore controlling for firm fixed effects is the more general way to estimate the relationship.

[Luca Haseney]

But then I would have to leave the country effect.
Would you suggest this is worth it to have more general effects?

[Joro Kolev]

But then I would have to leave the country effect.
Would you suggest this is worth it to have more general effects?

Yes, this is the point, that when you control for firm fixed effects, you are already controlling for everything that does no change by firm and across time, such as industry and country.

[Luca Haseney]

Thank you for the explanation.

I moved now to different specifications, but I still cant let go.
Would a suitable remedy be to split the sample by country and them compare differences?
Is there a sound statistical way to do this? Or is it again already included by firm fixed effects?