Re: [Gretl-users] Weak Instrument test with TSLS
On Sat, 8 Oct 2011, Anutechia Asongu wrote:
I'm applying the TSLS estimation method. In a regression
with just one endogenous regressor, the output shows an
F-test(Weak Instrument test) for the First Stage regression.
Meanwhile, when I perform a similar regression with more
than two endogenous regressors, the output reveals an F-test
statistics without any information on whether it is that for
the Weak Instrument test at First stage regression. I hope
someone must have come accross this in the past. The concern
is to know how to interprete this F-stats in the TSLS with
more than two endogenous regressors. Do I inteprete it as
the weak instrument test or the second stage F-stats?
The F-test that is reported in the first block of information
under the table of coefficients is always simply the test for
the null that all the slopes are zero in the model printed
above. So in the case of TSLS it's a second-stage test.
As for the weak instruments test, the first-stage F-test is
not applicable when there are two or more endogenous
regressors. Gretl shows the Cragg-Donald minimum eigenvalue
and additional information based on the Stock-Yogo approach,
where these things can be computed.
www.nber.org/~myogo/papers/published/RothenbergCh5.pdf
Allin Cottrell