[Gretl-users] GPH and Whittle

[Riccardo (Jack) Lucchetti]

On Tue, December 12, 2006 14:15, javier garcia enriquez wrote:

>   On the other hand, I've found one possible bug. When periodogram gives us
> the estimator of  'd', there is a big difference between GPH and Whittle. In
> our opinion that difference can't be so big, since in some cases one is
> positive and the other estimation negative. I add you one example, where
> m=24 and a=0,5.

...

> GPH test for fractional integration (T = 615, m = 24)
>   Orden de integración estimado = 0,257321 (0,222388)
>   Estadístico de contraste: t(22) = 1,15708, con valor p  0,2596
>   Estimador local de Whittle (T = 615, m = 24)
>   Orden de integración estimado = -0,0621984 (0,102062)
>   Estadístico de contraste: z = -0,609417, con valor p  0,5422

It is possible that the code is buggy, but let me point out that:

1) neither estimate is significantly different from 0
2) you've got a huge spike at frequency ~25. Are these bi-weekly data? Or
perhaps daily (6 days per week?) I suspect that if you set m a little lower
than 24, the difference should be less apparent. With seasonal data, it is
well known that semiparametric estimators of d may perform poorly in finite
samples.

Riccardo (Jack) Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona