Frequency Domain Bootstrap Methods For Time Series

Abstract

Two frequency domain bootstrap methods for weakly stationary time series will be proposed. The motivations for the proposed methods will be discussed, and the performance of the first method will be compared with that of a recently proposed method of Swanpoel and van Wyk, in a Monte Carol study. It is found that, when applied to the problem of estimating the variance of a log spectrum estimate, all methods under consideration can sometimes perform poorly. Overall, the frequency domain method used in conjunction with automatic spectrum estimate choice criterion developed by Hurvich, is found to perform best.