Hello,
I am running a one step dynamic gmm model of this kind. My data consists of a N=185 and T=27.
I have a lagged depent variable (l.y), 3 endogenous regressors (x1 x2 x3) and a series of exogenous controls ($exogen_controls).
To deal with endogeneity I would like to use a dynamic GMM with also time fixed effects.
In practice I am estimating the following model:
Code:
 xtabond2 y l.y x1 x2 x3 $exogen_controls yeardum*, ///
gmm(l.y x1 x2 x3, lag(1 15) collapse ) iv(yeardum* $exogen_controls) noleveleq small noconstant robust
While I am aware that the number of instruments should be less than the number of groups, I am not sure if there is a way to understand the appropriate number of lags to include as instruments.

My concern is that depending on the number of lags that I include in my gmm style instrument my endogenous variables are significant (or not).
For instance, while in the model specified above my results are significant, If I run the alternative below (lag 1-7 or 1-20) my coefficients on the endogenous variables are no longer significant.

Is there a way to understand what is the appropriate lag length for my model?

I have checked xtabond2 helpfile and the slides from prof. Kripfganz (https://www.stata.com/meeting/uk19/s..._kripfganz.pdf) but I did not understand how to go about this issue

thanks a lot in advance for your help


Best

Code:
 xtabond2 y l.y x1 x2 x3 $exogen_controls yeardum*, ///
gmm(l.y x1 x2 x3, lag(1 7) collapse ) iv(yeardum* $exogen_controls) noleveleq small noconstant robust

 xtabond2 y l.y x1 x2 x3 $exogen_controls yeardum*, ///
gmm(l.y x1 x2 x3, lag(1 20) collapse ) iv(yeardum* $exogen_controls) noleveleq small noconstant robust