On Tue, 1 Sep 2015, oleg_komashko(a)ukr.net wrote:
1)Gretl have a nice option to output confidence intervals for ols,
hsk and the like Is there a possibility to do it via a script?
Since you have $coeff, $stderr, $df and critical(), yes.
2)It's very easy to add a function to my Waldtest package finding
confidence intervals for nonlinear functions of parameters Will it
be useful?
Could be.
3) Is it possible to modify heckit command so as it output full
covariance matrix. This is need for mfx
Here what's now in CVS. After heckit estimation you now have an
accessor named $model (this is also true for single-equation models of
all sorts). This gives you a bundle, which contains the regular
accessors such as coeff but also some extra stuff, and one of the
extra things for heckit is a matrix named full_vcv. The heckit
full_vcv has entries for each parameter in the main equation, each
parameter in the selection equation, then sigma and rho. Unlike the
regular $vcv it does not include an entry for lambda, which is a
derived magnitude, not subject to direct ML estimation. (I should say,
right now this is only for ML heckit).
Illustration:
<hansl>
heckit ...
bundle b = $model
# what's in the bundle?
print b
matrix V = b.full_vcv
k = rows(V)
scalar se_sigma = sqrt(V[k-1,k-1])
scalar se_rho = sqrt(V[k,k])
</hansl>
The standard errors produced in this way agree with Stata.
4) Is it possible to extract fixed effects stats from felogit.
Again, with mfx in mind
Maybe Jack can answer that.
Allin