Am 13.07.2015 um 10:50 schrieb yinung at Gmail:
The feasible GLS comes from Wooldridge's Introductory Econometrics: A
Modern Approach, 4th ed., Chapter 8, example 8.7. Actually, Wooldridge
two ways of correcting for heteroskedasticity.
The fgls.gfn package was intended to implement those two ways (but the
other is not finished yet).
Ok thanks, it seems to be the same in the 3rd ed which I have here. The
second way (I'm referring to eq. 8.34) would seem to be very similar to
what gretl's hsk does. (Would have to check/think more about it, whether
it's really identical.)
I wrote fgls.gfn because I can't use gretl's default "wls" or
replicate the example in Wooldridge's example. That is why.
It would have been helpful to understand that already from the help text
of the package. Implementing an estimation which is not possible in
gretl is certainly a good reason for a package.
Personally I think it would be worthwhile to think about adding options
to the built-in 'hsk' command for these estimation variants. But since I
wouldn't be doing the actual implementation work of course I cannot
decide that. You or I could add it on the feature request tracker at
I did not attached Wooldridge's smoke.gdt since I am not sure the data
is proprietary or not. Though the data was obtained from gretl official
Will fgls.gfn be back on gretl's server? Thanks.
No, not in its current form -- as I said, the example script fails, and
that's a show-stopper. You could change the example script easily to
open one of the standard example files available in gretl. Then you
could tell the user in the help text that if she installs the Wooldridge
data she will be able to replicate the book example.
Adding one or two sentences to the help text on what the package does
would also be nice. Feasible GLS could be many things, for example
initially I thought it had to do with panel data.