Hi,
I am performing a two-step system GMM. My current code is the following:
xtabond2 rport1 l.rport1 age i.gender i.weight i.time i.year, gmm(l.rport1, lag(2 2)) iv(age i.gender i.weight i.time i.year,) nodiffsargan twostep robust orthogonal small cluster(firm)
However, I do face some problems.
1.
I find that I should reject the null-hypothesis of second-order serial correlation (AR(2)). That is a problem for the validity of my instruments. I attempt to add more lags as regressors and change the number of lags for the instruments. So, I am wondering:
- What if this does not work out? It is better to work with xtdpdgmm for non-linear moment conditions?
- How do you justify the number of lags for your instruments and regressor? You can not just write: it is trial and error. You can argue that you look for valid instruments and that you adapt your model based on the Arellano-Bond test. Is this enough?
2.
How to interpret the coefficient? Is this similar as a normal regression coefficient?
Thank you for your advice in advance.
I am performing a two-step system GMM. My current code is the following:
xtabond2 rport1 l.rport1 age i.gender i.weight i.time i.year, gmm(l.rport1, lag(2 2)) iv(age i.gender i.weight i.time i.year,) nodiffsargan twostep robust orthogonal small cluster(firm)
However, I do face some problems.
1.
I find that I should reject the null-hypothesis of second-order serial correlation (AR(2)). That is a problem for the validity of my instruments. I attempt to add more lags as regressors and change the number of lags for the instruments. So, I am wondering:
- What if this does not work out? It is better to work with xtdpdgmm for non-linear moment conditions?
- How do you justify the number of lags for your instruments and regressor? You can not just write: it is trial and error. You can argue that you look for valid instruments and that you adapt your model based on the Arellano-Bond test. Is this enough?
2.
How to interpret the coefficient? Is this similar as a normal regression coefficient?
Thank you for your advice in advance.
