I'm trying to estimate a dynamic panel model using the two-step system GMM procedure; however, I'm not sure which lags to specify as instruments. I have reason to believe that the variables in my equation are predetermined, especially because the y variable is future firm profitability (not concurrent), but this is my first time using xtabond2.
I initially specified the model as F1.y y x1 x2 x3 x4 x5 _Year*, gmm(y x1 x2 x3 x4 x5, lag (1 2) collapse) iv(_Year*) small twostep robust
Out of curiosity, I also examined F1.y y x1 x2 x3 x4 x5 _Year*, gmm(y x1 x2 x3 x4 x5, lag (2 3) collapse) iv(_Year*) small twostep robust
where the only change was using the 2nd and 3rd lag of the independent variables as instruments; however, my results are very different in the second case. I'm not totally sure why, but I'd like to know if one makes more sense than the other.
In the first specification I get AR(1) p-value = 0.000, AR(2) p-value = 0.830, Hansen test of overid p-value = 0.656, and Difference-in-Hansen test p-value = 0.471
For the second, I get AR(1) p-value = 0.012, AR(2) p-value = 0.957, Hansen test of overid p-value = 0.179, and Difference-in-Hansen test p-value = 0.401
I should also note that one of my x variables for which the coefficient sign changes is not very persistent and unlikely to be strongly related to its own lags. I'm not sure if this would help explain the differences.
Any help regarding which specification and which lags to generally use as instruments would be much appreciated.
I initially specified the model as F1.y y x1 x2 x3 x4 x5 _Year*, gmm(y x1 x2 x3 x4 x5, lag (1 2) collapse) iv(_Year*) small twostep robust
Out of curiosity, I also examined F1.y y x1 x2 x3 x4 x5 _Year*, gmm(y x1 x2 x3 x4 x5, lag (2 3) collapse) iv(_Year*) small twostep robust
where the only change was using the 2nd and 3rd lag of the independent variables as instruments; however, my results are very different in the second case. I'm not totally sure why, but I'd like to know if one makes more sense than the other.
In the first specification I get AR(1) p-value = 0.000, AR(2) p-value = 0.830, Hansen test of overid p-value = 0.656, and Difference-in-Hansen test p-value = 0.471
For the second, I get AR(1) p-value = 0.012, AR(2) p-value = 0.957, Hansen test of overid p-value = 0.179, and Difference-in-Hansen test p-value = 0.401
I should also note that one of my x variables for which the coefficient sign changes is not very persistent and unlikely to be strongly related to its own lags. I'm not sure if this would help explain the differences.
Any help regarding which specification and which lags to generally use as instruments would be much appreciated.
