Re: [Gretl-users] Cointegration Engle-Granger ?
On Wed, 26 Jul 2006, John C Frain wrote:
Can I make a brief comment on this debate. A residual based
test for cointegration is base on the residuals from a
cointegrating regression. Depending on whether the variables
in the cointegrating regression have non zero drifts on should
include a constants or a trend in the cointegration. If this
is done properly there is no need to include a constant or a
trend in the cointegrating regression.
I take it you mean "in the final DF test regression" at the end
there?
What we do currently is consistent with your comment, as I
understand it. Based on the selection of case "nc", "c",
"ct"
or "ctt" we include or omit the constant, trend and
trend-squared in the cointegrating regression. When we do
the DF test on the residuals from this regression, we do not
include any deterministic terms, but we select the "tau
statistic case" (that is, the distribution to which the "t"
ratio should be referred) based on the setup of the
cointegrating regression.
Can I recommend the account in Hayashi , F (2000),
Econometrics, Princeton University Press.
Thanks for the reference.
Allin.