Static Model with xtabond2

[John Don]

Dear Users,

I have the following equation for panel data

xtreg zer esg cap turn tdrdta prbv betal i.year, fe vce(cluster id)

However, I wonder about endogeneity problem and would like to use GMM equation for my model. Note that I do not have any instruments.

However as far as I see, xtabond2 command is being used for dynamic models rather than static ones.

Is there a way that I can convert my code above to a xtabond2 one without using the lag of the dependent variable?

[Sebastian Kripfganz]

On slide 89 of my 2019 London Stata Conference presentation, you can find an example about the use of xtdpdgmm (which is an alternative for xtabond2) to replicate the fixed-effects estimator for a static model with a GMM command. You can use this as a starting point and then modify the instruments etc. as desired.

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

[John Don]

Dear Sebastian,

Thank you very much for your prompt response and kind information. Your slides are very informative and helpful.

However, there is something I do not understand: how come estimating my model above with GMM does provide identical results to "xtreg"? Is there something I am missing?

[Sebastian Kripfganz]

The generalized method of moments, as the name says, is a general framework that encompasses other approaches (OLS, IV, 2SLS) as special cases. A fixed-effects estimation is equivalent to a just-identified instrumental variables estimation where the instruments are the regressors in deviations from their within-group means (which is equivalent to using untransformed regressors for the model transformed in mean deviations). This requires that all regressors are strictly exogenous. If you have endogenous regressors, you need to replace the respective instruments. Valid instruments could be lags of those variables for a model transformed in first differences, as in the standard Arellano-Bond approach for dynamic panel data models (just without the lagged dependent variables). For the strictly exogenous regressors, you could keep the same instruments for the model in mean deviations as in the fixed-effects approach, or you could also replace them with instruments for the model in first differences. The GMM approach can flexibly combine instruments for different model transformations.

[John Don]

Thank you very much for your kind response.

It is very much appreciated.