A0518
Title: Change-point tests in locally stationary time series via sparse subsampling and pooling
Authors: Kin Wai Chan - The Chinese University of Hong Kong (Hong Kong) [presenting]
Cheuk Hin Cheng - The Chinese University of Hong Kong (Hong Kong)
Abstract: Standard cumulative sum (CUSUM)-type tests for detecting mean change points are biased when the stationarity assumption does not hold. By considering locally stationary time series with, for example, complex time-varying variance-covariance structures, an intuitive sparse-CUSUM process is proposed as a general principle for handling time-dependent autocovariance structures. It allows the automatic construction of various types of pivotal CUSUM-type tests that do not require change-point-sensitive bootstrapping or computationally intensive simulation. Testing and estimating multiple change points in mean and general parameters are investigated both theoretically and empirically. The simulation shows a significant improvement in power and size accuracy, even under complicated non-stationary time series. The method is applied to an environmental time series for illustration.
