Choosing between Stata commands for system GMM

[Mohit Economics]

Hello,

I am running a panel regression for 51 US states and 10 years (2010-2019) with state and year fixed effects. I have a lagged dependent variable in my model. Even though the number of time periods is not too small, I am worried about the Nickell bias. Hence, I want to run a system GMM for robustness. I am inclined to use the command xtdpdsys for this purpose. I have some questions in this respect --

1. Is this better than xtabond and xtabond2? I know that xtabond does not do system GMM but not sure about xtabond2.

2. If I use the command xtdpdsys, how many lags of the dependent variable to be used as instruments should be specified inside maxldep()? I think 2 is a good number to go with to avoid proliferation of instruments and overidentification but not sure.

3. In the results I obtain from xtdpdsys, the 2010 year dummy gets omitted which I understand is because of the first year being a benchmark for the year dummies. However, I was expecting that the estimate for one more year would not be reported (2011) because of the generation of lags for the dependent variable. That did not happen though. I wonder why.

I will appreciate any help in this regard.

[George Ford]

I think xtabond2 is the most flexible, which you may prefer as you'll want to control the number of instruments given the panel dimensions.

[Sebastian Kripfganz]

1. Naturally, as the author of the xtdpdgmm package, I recommend my own package. The official packages xtdpd, xtabond, and xtdpdsys have considerably shortcomings and lack of flexibility (e.g., lack of collapsing), and I would not recommend them.
2. There is no general answer here, as it depends on type of data and the number of time periods you have. It is always good checking for robustness to different choices, and to be transparent about the choices and the process of making these choices when reporting your results.
3. It is not possible to answer the last question without knowing details about your data and seeing the Stata output.

More on dynamic panel data GMM estimation in Stata:

• Kripfganz, S. (2019). Generalized method of moments estimation of linear dynamic panel data models. Proceedings of the 2019 London Stata Conference.

[George Ford]

Do what SK recommends. He's the pro.