A1627
Title: Sequential monitoring for distributional changepoints in time series
Authors: B Cooper Boniece - Drexel University (United States) [presenting]
Lajos Horvath - University of Utah (USA)
Lorenzo Trapani - University of Pavia (United Kingdom)
Abstract: A nonparametric sequential monitoring procedure for detecting distributional breaks in time series is developed based on a class of degenerate U-statistics associated with possibly indefinite kernels, encompassing energy distances and kernel-based maximum mean discrepancy (MMD). The framework accommodates a general class of monitoring schemes and is developed in a Hilbert space setting. The asymptotic null behavior is characterized under weak dependence, and tractable approximation of the limit distribution is developed via spectral methods. Simulations across a range of monitoring schemes, along with real data examples, illustrate the method's performance.
