Static vs. dynamic panel FE

[Christian Danner]

I am estimating some panel FE model and I would like to know whether I can just rely on a static model or whether I need to go for a dynamic one.

My command is

areg y x1 x2 x3, absorb(panelid) vce(cluster panelid)

I tested for serial correlation - Ho of no autocorrelation is rejected

xtserial y x1 x2 x3

The clustering ensures robustness of SEs to autocorrelation as far as I understand.
1) Does this also ensure unbiased estimates?

Or do I need to go for a dynamic model then. I read that given my large N , large T dataset (N = 10 000, T= 800, obs = 1 000 000), I might just add a LDV:

areg y l1.y x1 x2 x3, absorb(panelid) vce(cluster panelid)

As my panels are unbalanced about 40% of the observation are excluded in the DFE model. The coefficients of x1-x3 are lowered in the DFE.
2) May I just go for the static model to get accurate results
3) If a dynamic model: Is this the right way do model it?

Any help is highly appreciated.

Thanks.

Christian

[Carlo Lazzaro]

Christian:
- -vce- options do not affect coefficients point estimates;
- large T, largel N panel dataset usually ask for -xtgls-;
- the choice between static vs dinamic panel data models should be theory-driven.

[Christian Danner]

Thank you so much Carlo. Your comments are very helpful !