[Gretl-devel] Re: Regression with many dummies (RFC)
Am 18.10.2025 um 14:28 schrieb Cottrell, Allin:
On Fri, Oct 17, 2025 at 10:27 AM Riccardo (Jack) Lucchetti
<p002264(a)staff.univpm.it> wrote:
> Hi all,
>
> I've begun to explore the issue of the numerical performance of OLS regression,
where you want to condition on a qualitative variable with many different values [and have
come up with an efficient solution].
This is nice. I'd say it's a bit too specialized to be an option to
"ols", and a function package would be a good way to go. But since it
seems that aggregate() does the heavy lifting, we could revisit that
to see if there's a tweak that could speed things up in this case.
However, I wouldn't be opposed to a new "fols" command (fols y xlist ;
faclist) if that has an additional speed advantage.
Jack mentioned that the approach is a generalization of fixed-effects
regression, and that the foundation is through the Frisch-Waugh-Lovell
theorem. So I wonder whether one should go one step further and also
attempt to cover non-qualitative control variables. Basically, I'd like
to point out the apparent connection to double machine learning. I'm
sure that in terms of the algorithm it's non-trivial to go from discrete
to continuous variables. However, before a new estimation command is
introduced, I think it would be good to think about what else it might
cover in the future, if only to make the name of the command general enough.
cheers
sven
Attachments:
- attachment.htm (text/html — 2.1 KB)