Help with xtabond2 command

[Drishta Tengur]

Hello, I have a dynamic panel data of 26 countries and 21 years. I am struggling to transform the following into Stata commands. I want to include the following as a system-GMM command:

Instruments for Difference GMM regression: lagged levels (two periods) of the dependent variable and the variables A and B, and differences of the C dummies and the strictly exogenous variables D, E and F.
Instruments used in the system GMM regression are lagged levels (two periods) of the dependent variable and the variables A and B for the differenced equation, and lagged difference (one period) of these variables for the level equation. The C dummies and the strictly exogenous covariates variables D, E and F are instrumented by themselves in the differenced equation.

Can somebody please help me out?

[Sebastian Kripfganz]

It would be usually preferred if you could show us the Stata command line and corresponding output of what you have tried so far. The help file of xtabond2 will guide you. Nevertheless, try the following as a starting point.

Difference GMM:

xtabond2 Y L.Y A B C D E F, gmm(Y A B, lag(2 2) collapse eq(diff)) iv(C D E F, eq(diff)) twostep robust

System GMM:

xtabond2 Y L.Y A B C D E F, gmm(Y A B, lag(2 2) collapse eq(diff)) iv(C D E F, eq(diff)) gmm(Y A B, lag(1 1) collapse eq(level)) twostep robust

I was following your description with one modification: Because you have rather few countries relative to a not-so-small number of time periods, I highly recommend to use the collapse suboption of the gmm() options.

[Sebastian Kripfganz]

As it says: "Equation not identified. Regessors outnumber instruments." (sic!)

You have plenty of additional regressors specified but no instruments for them. How to choose these instruments is not primarily a Stata question. There should be some guideline in the related empirical literature, but that is something I cannot help with. As an aside: With your limited cross-sectional dimension, you might ask to much from your data by attempting to estimate such a large model.

[Drishta Tengur]

Sebastian:
Thank you for replying. I am planning to use GMM due to the dynamic panel model and the existing literature makes use of this estimator. Do you suggest that I decrease the number of regressors due to the limited cross-sectional dimension? Or how should I go ahead?

[Sebastian Kripfganz]

Not necessarily. Of course, you should ask yourself if you need all of these variables. There should be some guidance from economic theory. For all variables that you would like to include, you then need to ask if they are strictly exogenous, predetermined, or endogenous. Again, economic theory should help to tell you the answers. Given these answers, you need to choose the appropriate instrumental variables for each variable. In doing so, you need to make sure that the numbers of instruments does not become too large. Given your limited cross-sectional dimension, you can hardly have more than one instrument per variable.

That is why the system GMM estimator is probably not the ideal estimator for your situation. In fact, there is no ideal estimator for this kind of data set. 21 years might be just long enough to simply use the fixed-effects estimator and to ignore the dynamic panel data bias (although it is still too short for this to be a comfortable strategy). You could also look into simple instrumental variables estimators, as implemented in xtivreg.

[Sebastian Kripfganz]

The following code replicates the difference GMM estimates from my post #2 with ivreg2:

ivreg2 d.Y (ld.Y d.A d.B = l2.Y l2.A l2.B) d.C d.D d.E d.F, noconstant

That said, I did not intend to suggest in my earlier comment that you merely switch to a different Stata command. First of all, you really need to fundamentally think about your model, the assumptions you want to make, and the instrumental variables that are implied by these assumptions. That is the key step. Everything else then follows.

[Smara Ansar]

Hi,
I am new to statalist. I am currently working with xtabond2. I want to ask how we can calculate R square and what does F test means in case of Xtabond2 ?