R-squared for Panel Data Fixed Effects

[Deepika Deshpande]

Hi,

I am doing a fixed effects regression (impact of geographic diversification on firm performance) on panel data for a set of firms over 20 years. The R-squared for my model is 13%. It is comparable to other research studies in this field. However, my own understanding is that R-squared for fixed effects regression would normally be lower than for OLS regression since it excludes the impact of firm-specific time-invariant factors.

I wanted to double-check my understanding of this. Is it fair to assume that fixed effects models would have a lower R-squared given the way they are constructed?

Thank you.

[Carlo Lazzaro]

Deepika:
it is not clear the R-sq you're referring to (R-sq within for -xtreg,fe-).

[Deepika Deshpande]

Thanks Carlo. You are right - the R-squared I am referring to is the (R-sq within for -xtreg,fe-).

[Joro Kolev]

The R-squared has the usual decomposition of variance meaning only in OLS regressions.

You can compare your R-squared to those of other authors computed by the same formula but for other data. Comparing R-squares across different estimation methods do not make sense.