Re: [Gretl-users] ADF lag length selection and normality of residuals?
On Mon, 6 Oct 2008, Olle Olsson wrote:
Hello,
I'm doing some work with unit roots and cointegration applied to energy
prices. Currently I've run into a bit of a dilemma regarding the choice
of lag length for the Augmented Dickey Fuller tests for stationarity.
The problem is, that for one of the tested variables, I don't get normal
residuals regardless of how many lags I include in the ADF regression.
Does anyone have any suggestions on how i should tackle this?
As far as I know, normality is not needed at all for the validity of the
asymptotics of the ADF test. Ignacio, can you confirm this?
A personal opinion: I think that the importance of normality tests as
linear regression diagnostics is _highly_ overrated among practitioners.
None of the asymptotic properties of OLS depend in any way on normality.
In small samples, it's another matter, but then, all normality tests I can
think of are asymptotic anyway. The normality assumption is crucial (and
can be tested) in models like probit or heckit, but definitely not OLS or
ADF.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti