On Sat, 29 Jun 2013, Allin Cottrell wrote: That said, an unrestricted arma(3,3) model has a very high chance of having near-common roots, which make the likelihood very flat at the maximum, which in turn is likely to cause horrendous numerical problems unless you have a very large sample. What happens if you decrease the AR and/or MA order by one? ------------------------------------------------------- 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 -------------------------------------------------------