I an making something similar to R::forecast:ets It uses NMmin to find set of alpha, beta, gamma and starting values which maximize likelyhood for e-smoothing model (for additive models it is the same as minimize sum of squared errors Empirically, I detected that setting maxiter to 100 gives minimal sum of squares comparable to that of Hyndman. I think , empirical setting of the tolerance could bring a better time to exactness tradeoff Oleh P.S. Your unwilling to accept Rforecast made me code a couple non-rubbish things, e.g. lagreg   16 лютого 2018, 09:58:08, від "Riccardo (Jack) Lucchetti" <r.lucchetti(a)univpm.it&gt;: On Fri, 16 Feb 2018, oleg_komashko(a)ukr.net wrote: As of now, you can't. You can set the maximum number of iterations, though. Can you provide an example of what you need to set the tolerance for? ------------------------------------------------------- Riccardo (Jack) Lucchetti Dipartimento di Scienze Economiche e Sociali (DiSES) Università Politecnica delle Marche (formerly known as Università di Ancona) r.lucchetti(a)univpm.it http://www2.econ.univpm.it/servizi/hpp/lucchetti ------------------------------------------------------- _______________________________________________ Gretl-users mailing list Gretl-users(a)lists.wfu.edu http://lists.wfu.edu/mailman/listinfo/gretl-users