A1242
Title: Multivariate change-point detection via optimal transport and characteristic functions
Authors: Sarka Hudecova - Charles University, Prague (Czechia) [presenting]
Abstract: Detecting structural changes in a sequence of observations is a central problem in statistical inference, particularly when the observations are multivariate. In many applications, ranging from finance and environmental monitoring to biomedical signal analysis, data are collected as vectors, and potential changes may occur not only in the marginal distributions but also in the dependence structure of the joint distribution. It is therefore desirable to have a test that can identify a change point in the multivariate distribution while operating in a fully nonparametric setting, without imposing restrictive assumptions on the underlying distribution or on the nature of the change. A test for detecting distributional changes in a sequence of multivariate observations based on optimal measure transport is proposed. The test is constructed using empirical characteristic functions of the transported observations. The resulting procedure is distribution-free, which allows for straightforward implementation of an exact test. The asymptotic distribution of the test statistic is also derived. The finite-sample performance of the proposed method is illustrated through a Monte Carlo simulation study.
