Title: On robust and nonparametric change-point detection in multiple time series
Authors: Roland Fried - TU Dortmund University (Germany) [presenting]
Alexander Duerre - TU Dortmund University (Germany)
Herold Dehling - Ruhr-University Bochum (Germany)
Daniel Vogel - University of Aberdeen (United Kingdom)
Martin Wendler - Ernst Moritz Arndt Universitaet Greifswald (Germany)
Abstract: A basic issue in the analysis of time series data is the question of stability of the data generating process. Especially in multivariate time series, there are many different aspects which can change over time. Methods for the detection of abrupt changes at unknown time points have been developed during several decades and offer some power also against monotonic drifts. Most of the existing work is based on linear statistics and/or for the situation of a single time series. Starting from previous work on robust and nonparametric change-point detection in univariate time series based on the usage of U-statistics and U-quantiles, a general class of tests for the detection of changes of possibly several types in multivariate time series is developed and investigated in different scenarios concerning the type of time series model and the type of structural change.