On Fri, 23 Dec 2011, clarodina(a)lycos.com wrote: If d2 and d3 are actual data, while d1 is just "some linear combination of d2 and d3, the coefficients to be determined somehow", and the object is to have that linear combination as close to stationary as possible, then it sounds as if you want a cointegration analysis. In that case you don't need BFGS; see the "coint2" and "vecm" commands. However, if you really just want to maximize the ADF tau for d1 with respect to alpha and beta, taking d2 and d3 as given, then here's an example: <hansl> function scalar adf_t (matrix *b, series d2, series d3) series d1 = b[1]*d2 + b[2]*d3 adf 0 d1 --c --quiet scalar tau = $test return tau end function open data3-6 matrix b = {1,1} set max_verbose on tau_max = BFGSmax(b, adf_t(&b, Ct, Yt)) print b series d1 = b[1]*Ct + b[2]*Yt adf 0 d1 --c </hansl> Allin Cottrell