Re: [Gretl-devel] Proposal for function package cleanup
On Mon, 13 Jul 2015, yinung at Gmail wrote (regarding the package
"fgls", in response to a query from Sven):
Dear Sven
In the Help text, it says that "This function package is for the feasible
GLS in Wooldridge's textbook."
It means that this package is to replicate the result from the example of
the 'deasible GLS' in Wooldridge's textbook.
I've taken a look at this, and the estimator offered by Yi-Nung's
package is very close to gretl's "hsk" command (heteroskedasticity
correction via automatically weighted least squares). The only
difference is that in the auxiliary (variance) regression "hsk"
includes squares as well as levels of the original regressors, while
Wooldridge's variant includes only levels.
In CVS and snapshots I've added a --no-squares option to the "hsk"
command: with this option you get exactly the Wooldridge estimator.
The following script illustrates:
<hansl>
open http://ricardo.ecn.wfu.edu/pub/gretldata/smoke.gdt
list X = lincome lcigpric educ age agesq restaurn
ols cigs 0 X
# gretl's original hsk behaviour
hsk cigs 0 X
# the Wooldridge variant: see Example 8.7 in
# Introductory Econometrics 4e
hsk cigs 0 X --no-squares
</hansl>
Allin