I have a system of two equations, which I have been trying to estimate
both
through equation-by-equation "ivreg" and also 3-SLS using "reg3".
However,
the point-estmates in the two procedures seem to differ a lot.
Crucially,
the co-efficients of the instruments used in ivreg becomes all
insignificant
in reg3. Can anybody shade some light on why is it happening this way.
For
IV procedure, I used both ivreg and ivreg2 with GMM option
Are you sure that in your ivreg(ivreg2) single equation estimates you
are using the same instrument list for both equations? In reg3, you
give only one instrument list, which applies to all equations.
The use of a system (full information) estimator theoretically contains
a potential gain in efficiency which would make coefficient estimates
more precise. But a limited-information estimator (ivreg on each
equation) has one clear advantage: if you totally misspecify eqn 1,
your estimates of eqn 2 are not affected. When you estimate them
jointly, the specification error in one equation a.s. pollutes the
entire system. So one explanation for what you are getting from ivreg
vs. reg3 would be that you have some specification problems in one of
the equations.
