Title: Change point detection in multivariate two-sample setup
Authors: Zdenek Hlavka - Charles University (Czech Republic) [presenting]
Marie Huskova - Charles University (Czech Republic)
Simos Meintanis - University of Athens (Greece)
Abstract: New methods are discussed for detecting structural breaks in a series of multivariate observations but, instead of considering general alternatives, we concentrate on a two-sample setup. In other words, we assume that two random subvectors have identical distribution until the unknown change-point. Most often, these random subvectors will consist of the same variables observed for two different populations and the goal will be to estimate the unknown change-point. The procedures are based on $L_2$-type criteria utilizing multivariate empirical characteristic functions leading to computationally attractive closed-form expressions. We present asymptotic and Monte-Carlo results for both on-line and off-line type procedures.