[Gretl-users] Re: F statistic not showing with HAC errors
On Fri, 17 Feb 2023, maisiedarby7(a)gmail.com wrote:
I am doing a manual 2SLS, and so i need an F statistic to check the
independent variable in the first stage is above 10 and a suitable IV.
When i do classic errors i get an F stat of 82 (although it's unclear if
this is for the whole regression or just my independent variable). As my
data has autocorrelation I would like to use HAC errors, however when I
do an F statistic is not computed and so I cannot see if my IV is
strong. Can anyone help?
The tsls command takes care of applying robust inference whenever
necessary: the following script exemplifies what happens (although the
data are not time series here, but it should be irrelevant). Note that the
F statistic you get with tsls as the "Weak instrument test" is identical
to what you get with the "omit" command below.
<hansl>
open mroz87.gdt
logs WW
list X = const WE WA AX
list Z = const WMED WFED WA AX
smpl WW > 0 --restrict
tsls l_WW X ; Z --robust
# 1st stage "by hand"
list Excluded_Inst = Z - X
ols WE Z --robust
omit Excluded_Inst --quiet
</hansl>
-------------------------------------------------------
Riccardo (Jack) Lucchetti
Dipartimento di Scienze Economiche e Sociali (DiSES)
Università Politecnica delle Marche
(formerly known as Università di Ancona)
r.lucchetti(a)univpm.it
http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------