Re: [Gretl-devel] df question for modtest --autocorr after tsls
Dear Sven, Dear all
This is the script to clarify the question
(stupid model, just to illustrate quickly)
set echo off
set messages off
open denmark.gdt -q
tsls LRM 0 LRY; 0 IBO IDE -q
modtest --autocorr 4
# F(4,53) = 13.186979
uhat = $uhat
list luhat = lags(4,uhat)
loop foreach i luhat -q
z$i=misszero($i)
endloop
list zhat = zuhat_*
tsls LRM 0 LRY zhat; 0 IBO IDE zhat -q
# by hand
omit zhat --test-only # 2-nd df differs from 'modtest' results
# by hand, asymptotic version
omit zhat --test-only --chi-square
# F(4, 49) = 13.186979
## there are two related but different questions:
## 1) whether to include a new option to modtest: --chi-square
## 2) what is correct: F(4,53), or F(4, 49)
Oleh
18 вересня 2016, 19:19:30, від "Sven Schreiber" <svetosch(a)gmx.net>:
Am 18.09.2016 um 13:00 schrieb oleg_komashko(a)ukr.net:
> Dear all,
> Gretl output makes a reference to Gogfrey(1994)
> for Pseudo-LMF statistics.
> I.e. gretl uses F(g, T - k ) instead of F(g, T - k - g)
>
> Is this a bug, or there is a newer corrected
> reference?
I haven't looked at the cited paper, but whether small-sample
adjustments are used or not is open to debate I guess, especially for
asymptotically justified estimators. But I agree that it should be
transparent at least.
cheers,
sven
_______________________________________________
Gretl-devel mailing list
Gretl-devel(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-devel