The VAR in system is estimated using the GMM Arellano-Bond

[zen]

I have to deal with running VAR by GMM Arellano-Bond approach. My panel data has 9 cross sections and 32 time series, it is unbalanced
By testing lag length, i get lag 3 and this is my model: dfx dfx(-1) dfx(-2) dfx(-3) ddirect.invest(-1) ddirect.invest(-2) ddirect.invest(-3) d1 d2 d3 c (with d1 d2 d3 are dummy). After that, i have to run GMM but i have no ideal to choose suitable intrumental variables and run GMM