ivreg2h : (1) for panel data? (2) the same command for single equation and set of fully simultaneous equations?

[Susanne Elsas]

Dear Stata list members,

I want to analyze a model using Lewbel's (2012) method of internal instruments for identification, because I am not aware of valid external instruments.

Let Y1 and Y2 be the endogenous variables and X a set of strictly exogenous variables.

Y1_it = \alpha1 + X_it'\beta1 + Y2_it\gamma1 + u1_it
Y2_it = \alpha2 + X_it'\beta2 + Y1_it\gamma2 + u2_it

I roughly understand the intuition behind Lewbels method (where heteroskedasticity in the errors is used to construct internal instruments) but I am not sure

(1) if I can apply it in a context of repeated measures : Assume that I have repeated measures of Y1, Y2 and X, with some of the X not varying over time.
Can I ignore the panel structure of the data for identification when the elements of X are strictly exogeneous? Is it sufficient if I only invoke the cluster(id) option to account for the panel structure of the data when standard errors are computed?

(2) how I instruct Stata to estimate parameters of a set of fully simultaneous equations, instead of a single equation model with an endogenous RHS variable, or is the method independent of the (true) value of \gamma2?

Will the following commandline produce the estimate of \gamma1 irrespective of \gamma2 equalling zero or not?

ivreg2h Y1 X (Y2=), cluster(id)

And the following commandline will produce the estimate of \gamma2 in the the system of simultaneous equations?

ivreg2h Y2 X (Y1=), cluster(id)

Thank you very much for your kind help!

Susanne

[Phil Bromiley]

You didn't get a quick answer. You'll increase your likelihood of a useful answer by following the FAQ on asking questions. We don't even know exactly what you plan to estimate.

If you want to estimate multiple equations at once, you're probably going to want reg3 or SEM/GSEM. ivreg2h is user written so getting detailed help on it depends on whether someone active on the listserve uses it. I must admit I would be concerned about using a novel technique without a detailed understanding of its technical aspects. You may need to contact the authors if you want detailed information on the routine.

I doubt that you can ignore the panel structure of your data - I don't know any panel estimation problem that doesn't need to test for panel effects. Exogenous x's doesn't mean you can ignore panel structure - xtreg is strictly for exogenous x's.