Chow test with fixed effects panel estimation

[Noemi Seng]

Dear community,

I estimated a fixed effects model with panel data and added also time fixed effects. Now, I would like to test whether two subsets of my data follow the same regression function. I therefore added the group dummy and interacted it with each other variable to then conduct a Chow test. Does someone know, whether this is possible like this or this procedure is not possible with a fixed effects estimation?

I would appreciate your help!

[Clyde Schechter]

I'm not sure what your question is really about. With the interaction model you can separately test equality of regression coefficient for each variable across the two subsets in your model. You can also test joint hypotheses of equality of coefficients for any subset of the variables (including the full set of variables). Depending on how strictly you want to use language, it could be said that all or none of these are Chow tests.

The actual Chow test was a different way of testing equality of regression coefficients across two subpopulations. It was invented by Frank Chow back in the 1960's when computer resources were scarce and expensive--it had the advantage that the regressions did not have to be re-run and the calculations could be done by hand just from the regression outputs. Today, it is hard to imagine that anybody does the actual Chow test any more. The interaction approach, or, for supported estimation commands, -suest- provides the same answers that the Chow test would. And that is what you are proposing to do. By abuse of language, many people now refer to any test of equality of coefficients in subpopulations a Chow test.