Thanks, Egon
I would think this would be very difficult to test for co-integration.
If my memory serves me correctly a Dickey Fuller unit root test can be
conducted via OLS and then the t-values compared to a DF table. But Co-
integration (and the subsequent residuals for the error correction
model) I know in Stata are estimated via some maximum likelihood
method. Stata also does other desirable things like transforms lost
lags via the Prais-Winsten method I believe (so all time series models
are estimated via iteration in Stata). I would suggest you work with a
program that has commands to do this directly, and R is always an
option.
Good luck
Andy W