[Gretl-users] Cointegration Engle-Granger ?

[Allin Cottrell]

On Mon, 24 Jul 2006, john w wrote:

> under both ADF tests stays that: P-values based on MacKinnon 
> (JAE, 1996) I think(not sure) that if this two tests have 
> exactly the same estimated values and test statistics, and are 
> based on the same p-value calculation ie. MacKinnon then 
> p-values for both must be the seme.

Please take a look at MacKinnon's 1996 article.  The 
distributions, and hence the p-values, are *not* the same in the 
two cases of (a) a "stand-alone" ADF test on a single 
time-series and (b) an EG test based on the residuals from a 
cointegrating regression.  In the latter case the distribution 
of what MacKinnon calls the tau statistic depends on the number 
of (possibly) cointegrated variables.

Allin Cottrell