xtabond2 syntax

[Elliott Blain]

Dear Statalists,

I am currently doing research into the link between economic growth and inequality in a panel of countries, using system GMM. I am getting slightly confused with the xtabond2 terminology and specifically the difference between gmm() style instuments and iv() instruments.

My current code for the system GMM estimation is:

xtabond2 growth lnGDP_L1 Primenroll_L1 gini_disp_L1 i.Year, ///
gmm(lnGDP_L1 Primenroll_L1 gini_disp_L1 ) twostep robust small

As you can see here, I have specified all regressors as gmm style instruments, which I am assuming is incorrect.

As it is a dynamic panel, GDP is endogenous and should the other regressors be specified as IV instruments? What is the criteria for gmm style and iv style instruments?

Thanks

[Sebastian Kripfganz]

Because these are fundamental questions about the functioning of this kind of estimators, I recommend that you start by doing some more background reading, e.g. the "how to do xtabond2" Stata Journal article by David Roodman. A good starting point might also be the following presentation I gave in 2019 at the London Stata Conference, which might already answer some of your questions:

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

You can also search this forum for other threads on the same topic.

[Elliott Blain]

Hi Sebastian,

Thank you for your response and recommendations. I have read both resources and they have helped my understanding of the STATA implementation greatly.

I just have one question. If the only instrument I would like to use for the levels equation is the first difference of each regressor in period t-1, i.e. only delta Xi,t−1, how would I specify this in the gmm() syntax?

My current code is:
xtabond2 (growth) lnGDPpc_L1 Primenroll_L1 gini_disp_L1 i.Year, ///
gmm(lnGDPpc_L1, lag(2 5) equation(diff)) gmm(L1.gini_disp_L1 L1.Primenroll_L1, collapse) iv(Year) twostep robust small

Here, I lnGDPpc_L1 is only being used as an instrument for the differenced equation and not for the levels.

Thanks