Parameter-Covariance Maximum Likelihood Estimation

Abstract

Linear time series modelling is dominated by the use of purely autoregressive models even though incorporating moving average components can greatly improve parsimony. We present a convex formulation for vector-ARMA system identification which respects this fundamental property, thus granting access to the nice properties afforded by convex programming. The identification procedure is done purely in the time domain which can accommodate non-stationarity through regime switching. As a proof of concept, we present experimental results demonstrating this convex program in action. Next, we show how to adapt the expectation-maximization algorithm to support regime switching behavior.