Re: [Gretl-users] Two-Step dynamic panel system GMM
On Wed, 13 Jun 2012, Anutechia Asongu wrote:
I have a few qualms with the output of 'Two-Step dynamic
panel system GMM'. Normally I need three information
criteria upon regression: Test for AR(2) errors; Sargan
over-identification(OIR) test and Wald (joint) test. However
quite often, just two pairs of the information criteria
appear in the output: i.e, either the first, second or third
of the information criteria is absent.
If any of these tests are absent, it means it was not
possible to compute the test in question. With the AR test it
can happen that there are no relevant data on which to base
the test; with the Wald test it can happen that the relevant
portion of the covariance matrix is singular to machine
precision; I'm not sure why the Sargan test might fail.
However, none of these tests fail in any of my dynamic panel
test cases, so it's anomalous that they're failing on your
data. Perhaps you could send me a dataset and a model
specification and I'll take a look.
Allin Cottrell