CONDITIONS FOR LOCAL CONVERGENCE OF MAXIMUM LIKELIHOOD ESTIMATION FOR ARMAX MODELS

摘要

This paper analyzes the conditions for local convergence of Maximum Likelihood estimation for ARMAX models. We do this by examining the region in which the steepest descent direction leads to a reduction in the Euclidean norm of the parameter error. Inter-alia this gives new insights into the question of existence of local maxima of the likelihood function for various commonly used model structures.