Least-squares, Yule-Walker, and overdetermined Yule—Walker estimation of AR parameters: a Monte Carlo analysis of finite-sample properties

Abstract
A Monte Carlo analysis of the accuracy properties of least-squares (LS), Yule-Walker (YW), and overdetermined Yule-Walker (OYW) methods for estimating the parameters of autoregressive (AR) processes is presented. Comparisons of the estimated finite-sample accuracy to the theoretical asymptotic accuracy are included. It is shown that considerable differences may occur in some cases. Choice of the number of equations in the YW system of equations is discussed. Some remarks concerning the feasiblity and usefulness of an analytical study of the finite-sample accuracy properties are also included.