A1143
Title: Inference for quantile change points in high-dimensional time series
Authors: Jiaqi Li - University of Chicago (United States) [presenting]
Likai Chen - Washington University in Saint Louis (United States)
Mengyu Xu - University of Central Florida (United States)
Abstract: Change-point detection methods for quantiles are able to effectively detect structural breaks in extreme values. A novel change-point detection scheme is proposed that utilizes fixed quantiles of moving sums (MOSUM) from high-dimensional time series. The approach employs a MOSUM test statistic that aggregates the component series by the max norm. The asymptotic properties of the proposed test statistic are investigated in the context of high-dimensional time series, allowing for strong or weak cross-sectional dependence by establishing a powerful uniform Bahadur representation. Specifically, the existing uniform Bahadur representation is extended to the high-dimensional setting for dependent data. To the best of knowledge, this is the first proposal for a change-point detection approach that leverages quantiles from high-dimensional time series. Simulation studies demonstrate the effectiveness of the approach. An application is also presented on a real-world dataset for the value-at-risk in S\&P500, which showcases the validity of the method in practical settings.
