Is xtreg command suitable for a dynamic panel with dimensions T=21 and N=110?

[Laura FR]

I would like to estimate a dynamic model using a sample of N=110 (countries) and T=21 (years).
I have previously tried with xtabond2 command but, even using the option collapse, the null hypothesis of the Sargan test (Overidentifying restrictions are valid) could not be accepted.
My question is: being T= 21, would it be correct to estimate a dynamic panel using xtreg (, fe)? I have also read about xtregar but I understand this command is only suitable when T>N, which is not the case.
Thank you!

[Joao Santos Silva]

Dear Laura FR

Yes, the FE estimator should be OK in this case because with T = 21 the Nickell bias should be small.

Best wishes,

Joao

[Sebastian Kripfganz]

It depends: If the data-generating process is very persistent, the bias can still be substantial when T = 21. In a typical macroeconomic setting, I would not feel confident using the FE estimator. In microeconomic applications, the persistence is often much lower.

On top of that, keep in mind that the FE estimator requires that also all other regressors are strictly exogenous.

The FE estimator requires stronger assumptions than the usual GMM estimators. If you believe in the overidentification test from the GMM estimation, then the FE estimator is even less justified.

[Laura FR]

Thank you, @Joao Santos Silva and @Sebastian Kripfganz, for your replies.