A1839
Title: Distribution-free change-point detection in high dimensions: A generalized homogeneity approach
Authors: Xianyang Zhang - Texas A\&M University (United States) [presenting]
Runmin Wang - Texas A&M University (United States)
Abstract: A distance-based method is proposed to detect and localize general structural breaks in high-dimensional data beyond the first two moments, a problem largely unexplored compared to standard mean and covariance shifts. The approach uses generalized homogeneity tests and constructs a cumulative sum process in an embedded Hilbert space, for which the limiting null distribution and asymptotic consistency are rigorously established under the high-dimensional medium sample size (HDMSS) framework. To estimate multiple changepoints, the statistic is integrated with the Narrowest-Over-Threshold (NOT) strategy and componentwise monotone transformations are introduced for enhanced robustness. Empirical studies on simulated and U.S. stock return data demonstrate the method's superiority. The methodology is implemented in the R package KDist, available at https://github.com/zhangxiany-tamu/KDist.
