Dynamic Panel Models

[Sladjana Bodor]

Dear professor Sebastian Kripfganz,



Would you be so kind to help me determine which command in Stata is most suitable for evaluating the model I am considering. The idea is to estimate a regression with at most 21 explanatory variables where the model also contains a lagged dependent variable as regressor. My sample contains almost the same number of countries and years (N=28 and T=27). It is clear to me that with lagged dependent variable as regressor, dynamic specification should be estimated and I have looked into Stata options that involve the xtdpdgmm, xtdpdbc, xtdpdqml and xtabond2 commands. I also know that GMM estimators for dynamic panel models are designed for large N and small T, and I am aware that the N dimension in my sample is rather small and time dimension is long to apply these commands. Taking into account all this, I am not sure if it is correct to apply some of these commands. I would be very grateful if you recommend me another estimation methods for my dynamic specification.


Thank you very much in advance.



Best regards,


Sladjana Bodor

[Sebastian Kripfganz]

Estimating a dynamic panel model with your data set is challenging, as you already noticed. You could try xtdpdgmm with the onestep option, but make sure that you are strongly limiting the number of instruments. (You might want to try a just-identified instrumental variables estimator in the spirit of Andersen and Hsiao; see the first example in the xtdpdgmm help file.) You could also try xtdpdbc, but be aware that it makes the strong assumption that all of your 21 explanatory variables are strictly exogenous with respect to the idiosyncratic error component. In any case, standard errors and confidence intervals may not be very reliable given that N and T are of the same order, and relatively small.

Also, ask yourself whether you really need a lagged dependent variable.