Re: [Gretl-users] weights in system estimation?
On Wed, 29 Apr 2015, Simona De Lauretis wrote:
Dear Gretl users,
I have a question about system estimation and a (possible) bug report.
The question is: is it possible to include weights for observations when
estimating a system of equations in gretl (with sur method for example)?
I've read in the command reference that the supported estimators
are ols, tsls, sur, 3sls, fiml and liml. However, if I estimate the
system using the
"simultaneous equations" dialog box, there is also wls as one of the
proposed estimators
(but it's not possible to specify a weight variable, so maybe it does
not perform a
weighted regression after all).
Thanks for the report. The "wls" option has somehow slipped through
undocumented, we need to fix that. But it's not intended that the user
can supply weights, rather the idea is that weights are derived from
the relative variances of the error terms in the equations of the
system. It's the same use of "wls" as in the systemfit package for R,
in case you've come across that. See
http://cran.r-project.org/web/packages/systemfit/systemfit.pdf
for brief documentation.
The possible bug occurs after the estimation of a system using
method SUR. In the results window, the command "Tests -> Normality
of residuals" is active, but nothing happens when I click on it
(gretl does not crash either).
I see what you mean. For some reason the normality test via the GUI
was not hooked up to its "callback" to actually do the job (or
actually, was hooked up only for models estimated on time-series
data). That's now fixed in CVS and gretl snapshots.
Allin Cottrell