On Fri, 23 Dec 2011, clarodina(a)lycos.com wrote: My script can be thought of as giving you a d1-hat that satisfies your "maximum ADF t" criterion. Of course, it may bear little relationship to the actual d1 data. If in addition you want d1-hat to "resemble" d1 in some way, you'll have to constrain the maximization of the ADF t -- e.g. provide a criterion for maximization that's a weighted sum of SSR for d1 and the ADF t-ratio for d1-hat (with a negative sign on the SSR). [SSR = sum of squared residuals] But this seems very weird, and there's no telling what statistical properties the resulting estimator would possess. Allin Cottrell