B0838
Title: Multiple autocovariance change-points problems in high-dimensional time series
Authors: Yuan Ke - University of Georgia (United States) [presenting]
Abstract: A framework is established to study multiple autocovariance change-points problems in high-dimensional, piecewise stationary, and heavy-tailed time series. First, we propose an element-wise truncated autocovariance estimator for high dimensional and nonstationary time series. We prove the estimator enjoys nice nonasymptptic and asymptotic properties when the time series data exhibits nonlinear temporal dependency and heavy-tailedness. Next, we introduce a moving sum statistic and a recursive segmentation algorithm to consistently detect the number and locations of autocovariance change-points in high-dimensional time series. The detection threshold in the algorithm is selected by a block-wise Gaussian multiplier bootstrap method. Further, we study the inference for the existence of a change-point around a pre-specified location and false discovery rate control for multiple autocovariance change-points detection. The superior empirical performance of the proposed methods is evaluated by various simulated and real data examples.
