Re: [Gretl-users] Breusch-Pagan test for Heteroskedasticity (Gretl-users Digest, Vol 100, Issue 3)
Can I make my first contribution to this list, regarding Breusch-Pagan's het test. My
understanding is that the original statistic requires disturbance term normality for the
chi-square limiting distribution to apply and the (also implemented by Gretl) reworking by
Koenker does not (hence the description of Koenker as a `robust' statistic I assume.
It's just a matter of the consistent estimation of an unknown parameter and BP use a
relationship between fourth and second moments which requires normality in doing this). I
always teach my students to use Koenker. IMHO Gretl should present Koenker only and
describe is as (something like) the Breush-Pagan-Koenker test for heteroscedasticity.
D Holden