Hi everyone,

I am running a system GMM model with a N=31 T=10 dataset. I included a time trend (equal to 1 in year=1, equal to 2 in year=2...) as independent variable and I am using it in the ivstyle specification of xtabond2 (for the level equation only). I then realised that in Roodman (2009) "How to xtabond2" it is clearly suggested to include time dummies "to remove universal time-related shocks from the errors" and ensure that errors are correlated only within individuals and not across them.
Can a time trend ensure that the same condition is satisfied or is it not a sufficient specification, compared to time dummies?
Also, I was wondering what are the implications of including time dummies, or any other instrument, as external instruments in ivstyle but not in the initial model? I thought that was possible, since by definition external instruments are expected to be "externally determined".
Thank you very much!

Flora