[Gretl-devel] df question for modtest --autocorr after tsls
Dear all,
Gretl output makes a reference to Gogfrey(1994)
for Pseudo-LMF statistics.
It says (p. 533)
As a finite sample adjustment,
the tests are implemented in section IV
by comparing g-1 IVLR_j, g-1 IVLM_J and
g-1 IVWD_j to a right-hand-tail critical value of
the F(g, T - k - g) distribution, j = 0, g.
Here k is the number of rhs variables in
the original model, and g is the number
of lagged uhat's.
To reproduce:
open denmark.gdt
tsls LRM 0 LRY ; 0 IBO IDE
modtest --autocorr 4
We see
Test statistic: Pseudo-LMF = 13.186979,
with p-value = P(F(4,53) > 13.187) = 2.3e-07
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?
Oleh