Hello, everyone.
I am trying to estimate a static and a dynamic model for a macropanel unabalanced dataset, of provinces in a single country over years, with N=27 and T=32. My model is nonlinear, and I understand that my dataset falls in the category of big T and small N. I have reviewed previous similar posts, but they seem to have T that is too big and N that is too small, unlike my borderline dataset.
I started with a fixed effects model with time dummies for both the static and dynamic models using -xtreg- as follows:
xtreg y x x_sq i.year, fe robust
xtreg y L.y x x_sq i.year, fe robust
I got the expected signs for the x and x_squared, and both were significant. But from the macropanel literature, I am not sure the fixed effects model would be suitable for my case, due to issues including the Nickell bias, cross-sectional dependence, heterogenous slopes, possible endogeneity, non-linearity, and others. I have seen suggestions for using -xtmg- or -xtscc- or -xtlsdvc- or even GMM, and others, to address some of these issues.
Which of these estimators/commands should I use for my case, for the static and the dynamic models, or are there other alternatives that fit this dataset and specifications best?
Thank you very much.
I am trying to estimate a static and a dynamic model for a macropanel unabalanced dataset, of provinces in a single country over years, with N=27 and T=32. My model is nonlinear, and I understand that my dataset falls in the category of big T and small N. I have reviewed previous similar posts, but they seem to have T that is too big and N that is too small, unlike my borderline dataset.
I started with a fixed effects model with time dummies for both the static and dynamic models using -xtreg- as follows:
xtreg y x x_sq i.year, fe robust
xtreg y L.y x x_sq i.year, fe robust
I got the expected signs for the x and x_squared, and both were significant. But from the macropanel literature, I am not sure the fixed effects model would be suitable for my case, due to issues including the Nickell bias, cross-sectional dependence, heterogenous slopes, possible endogeneity, non-linearity, and others. I have seen suggestions for using -xtmg- or -xtscc- or -xtlsdvc- or even GMM, and others, to address some of these issues.
Which of these estimators/commands should I use for my case, for the static and the dynamic models, or are there other alternatives that fit this dataset and specifications best?
Thank you very much.
