1
$\begingroup$

I have a question about a panel data regression model in general. I have firm-level data in a single country. Is it okay to control for both time fixed effects and entity-invariant variables, such as GDP growth and interest rate (which are the same across firms but vary across years)? Ideally, I want my model to be as parsimonious as possible, i.e., without controlling for the entity-invariant variables.

However, my regression result without including them is quite bad (none are statistically significant), while the results are significant by including them. I am not sure whether the latter result is valid.

$\endgroup$
2
  • $\begingroup$ Do you also adjust for “entity” fixed effects? $\endgroup$ Sep 26, 2022 at 18:22
  • $\begingroup$ Hi Thomas. Yes, I adjust for both entity and time fixed effect. I am not sure whether it is okay to include entity-invariant variables given that I have included the time fixed effect. Thanks $\endgroup$
    – rprass
    Sep 26, 2022 at 18:25

1 Answer 1

0
$\begingroup$

Is it okay to control for both time fixed effect and entity-invariant variables, such as GDP growth and interest rate (which are the same across firms but vary across years)?

No.

But let's get our terminology straight. The term "entity-invariant" means: common across entities but changes over time. In short, the year fixed effects will adjust for the unobserved factors that evolve over the years but are constant across entities.

In practice, including a full set of year dummies adjusts for both observed and unobserved factors, which means most software packages would drop your entity-invariant covariates. It appears you're sampling firms/industries within the same country, so annual GDP growth for the units within those countries should be the same. In short, all units (i.e., firms, industries, etc.) experience the same rate of change over time, assuming they're all nested within the same country. Remember, GDP is a core measure of economic prosperity—for the whole country.

However, my regression result without including them is quite bad (none are statistically significant), while the results are significant by including them.

I wouldn't assume it's a "bad" result simply because you're missing a few asterisks next to the coefficients. Maybe a statistically insignificant result has a meaningful interpretation. It's hard to offer further guidance without actually seeing your results, but I would recommend either dropping the entity-invariant variables, or modeling "time" in a different way.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.