Em 7 de julho 2010 Allin <cottrell@wfu.edu> escreveu:

On Wed, 7 Apr 2010, Henrique Andrade wrote:

> Is it possible to save the residuals of the Augmented
> Dickey-Fuller (ADF) test regressions? I would like to save them
> to check if they follow a white noise process.

As matters stand, you have to formulate and run the ADF regression
yourself to get access to the residuals. This goes for all
auxiliary regressions run by gretl on behalf of diagnostic test
commands: they do not create "first-class" models from which you
could access $uhat, $ess and all the rest.

This seems OK to me: if you want to investigate "second-order"
aspects of the test that's fine, but you need to run the auxiliary
regression yourself. The built-in diagnostic commands are just
convenient short-cuts for procedures that are quite easily
scripted in gretl.

I've got your point and I think you are right. Thanks Allin!

Em 8 de abril de 2010 Ricardo <ricardogs@terra.com.br> escreveu:

If the test shows that you have a unit root, by definition, the residuals doesn't follow a random walk (white nose) and vice-versa.But if you really want to see the residuals, just run an AR(p) regression, by OLS, where p is the "augmented lag" (w/o truncation), then save the residuals.

Dear Ricardo, at first I would like to thank your for your answer!

I'm interested in the residuals of the ADF test regression:

d_y(t) = b0 + (a-1)*y(t-1) + d_y(t-1) + d_y(t-2) + ... + d_y(t-p) + e(t)

This is because in the ADF test we need to choose p-lags in order to obtain residuals, e(t), that follow a white-noise process.

Um abraço,
Henrique