How to run 2SLS using ivreg2 from the perspective of GMM xtabond2

[Emmy Pham]

Hello experts,

I would like to know how we can perform a dynamic model y = L.y + x1 + x2 + x3 +i.time +i.id using GMM (xtabond2) and 2SLS (ivreg2).

Is it correct to go xtabond2 like this:

xtabond2 y L.y x1 x2 x3 i.time, gmm(endogenous var) iv(exogenous var)
or
xtabond2 y L.y x1 x2 x3 i.time, gmm(L.y) iv(x1 x2 x3 i.time)

To be honest, I don't know how Stata runs the above model when L.y is endogenous because as I have read from theory, L.y should be instrumented using L2.y but in xtabond2, "x1 x2 x3 i.time" are all put in iv() rather than L2.y. This brings me to the question that, for 2SLS, which variable should we use to regress L.y on?

ivreg2 y x1 x2 x3 i.time (L.y=L2.y) ???

(If it is better to be more specific, you can use the example of Arellano and Bond (1991) model to give explanation.)
Thank you!

[Sebastian Kripfganz]

It is not entirely clear to me what your aim is. The following presentation might provide some helpful insights:

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

[Emmy Pham]

It is not entirely clear to me what your aim is. The following presentation might provide some insights:

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

Thank you, I want to have a robustness table with fe gmm and 2sls but I don't know what the command should be like for 2sls because I have no idea what variable I need to use to instrument L.y

[Sebastian Kripfganz]

The models estimated by xtabond2 and ivreg2 are usually not comparable. They typically impose very different assumptions. It is your responsibility as a researcher to specify those assumptions in the first place. Once you have done this, you can then look for an appropriate Stata command to implement the model. It is not informative to simply compare an "xtabond2" and an "ivreg2" model, because there are many models that can be implemented with these commands by changing syntax and options.

[David Roodman]

I agree with Sebastian.
Section 3 of the xtabond2 paper walks through the gradual elaboration of a model, starting with ivreg and ending with xtabond2.