A Fast Algorithm for Estimating Parameters of a Multivariate Autoregressive Moving Average Processes

Abstract

We propose in this paper a fast and iterative algorithm for estimating the parameters of a Gaussian vector autoregressive-moving average (VARMA) model. This algorithm is a multivariate generalization of that suggested by Sabiti (1996) for estimating the parameters of a univariate ARMA(p,q) process. It is proposed, mainly for providing initial estimators for the iterative maximization of a log-likelihood function. Comparisons about the number of computations in terms of multiplication operations are made with a method that uses gradients to locate a maximum of the likelihood function and the fast method suggested by Spliid (1983).