Dynamic Panel Regression - T=15, N is around 14-18 (unbalanced)

[Lizelle Cruz]

Hi

I am currently doing my undergraduate thesis and our dataset includes an unbalanced panel with 14 to 18 banks with the years 2003-2017 (15 years). We are asked to find a dynamic panel data model that accounts for dynamic endogeneity, but we cannot use GMM since our N is very small. I'd like to ask what other models or stata commands can we use that will fit our data?

Thanks

[Sebastian Kripfganz]

I am afraid to say that - with the small dimensions of your data set - there is no estimator that would yield precise estimates for a dynamic panel model with unobserved heterogeneity (which causes the "dynamic endogeneity"). Ideally, you would be able to gather data on more banks to increase your cross-sectional dimension. Otherwise, you could only hope that the persistence of the process is relatively low such that the bias of the conventional fixed-effects estimator will be small, or you would need to assume that there is no unobserved heterogeneity in the first place such that you can apply the pooled OLS estimator.

Another idea is to estimate only a static model without a lagged dependent variable. There can still arise a bias in this case due to the omitted dynamics, but given your data there is simply no way to estimate a sophisticated model without bias.

I would thus recommend to admit the fact that all estimators will suffer from some sort of bias, and to report the estimates from different estimators rather than arbitrarily picking one of them.