B1017
Title: Subsampling-based change-point detection in LRD time series
Authors: Annika Betken - Ruhr-Universitat Bochum (Germany) [presenting]
Martin Wendler - Ernst Moritz Arndt Universitaet Greifswald (Germany)
Abstract: A robust change-point test is considered based on the Wilcoxon two-sample rank statistic to identify changes in the mean of dependent data. The scaling needed to ensure convergence of the corresponding test statistic to a non-degenerate limit usually depends on unknown parameters. Estimation of an unknown standardization may be avoided by an application of self-normalized test statistics. However, under long-range dependence the asymptotic distribution of self-normalized statistics still depends on parameters that characterize the intensity of dependence in the data. Approximating the limit distribution of self-normalized statistics by subsampling procedures bypasses estimation of these quantities. It is shown that the so-called sampling window method is valid for general statistics applied to long-range dependent subordinated Gaussian processes which satisfy mild regularity conditions. Furthermore, we investigate the finite sample performance of subsampling-based change point tests in a simulation study. We compare it to the performance of the usual testing procedure which generates test decisions on the basis of critical values that arise from the asymptotic distribution of the respective statistic.
Journal