Hello everyone,

This is my first post here.

I have a question about the xtabond2 command. I would like to incorporate the interaction of an endogenous and an exogenous variable. My question is whether I should treat the interaction term as endogenous, exogenous or not at all.

So which of the following options is preferable? Could anyone please explain what happens if I do not list it under gmm nor under iv (first option)? I thought it would be treated such that it interacts the instrument of the endogenous variable and the actual value of the exogenous variable. To me it seems the most reasonable to do it like that, but I cannot find any information confirming it.

1) xtabond2 y l.y x1 x2 c.x1##c.x2 i.year, gmm(l.y x1) iv(x2 i.year), noleveleq twostep robust

2) xtabond2 y l.y x1 x2 c.x1##c.x2 i.year, gmm(l.y x1 c.x1##c.x2) iv(x2 i.year), noleveleq twostep robust

3) xtabond2 y l.y x1 x2 c.x1##c.x2 i.year, gmm(l.y x1) iv(x2 c.x1##c.x2 i.year), noleveleq twostep robust


Also, I was wondering how to deal with an interaction of two exogenous regressors. Should I include the interaction to the iv option or should I leave it out, since x1 and x2 are already defined as exogenous? Why do I not get the same results for these two options?

4) xtabond2 y l.y x1 x2 c.x1##c.x2 i.year, gmm(l.y) iv(x1 x2 i.year), noleveleq twostep robust

5) xtabond2 y l.y x1 x2 c.x1##c.x2 i.year, gmm(l.y) iv(x1 x2 c.x1##c.x2 i.year), noleveleq twostep robust


Thank you already.

Sofie