Abstract
Change-point analysis is often used to detect changes in distribution of a single random sequence. The power of the sequential test can be improved by looking at differences with respect to a positively correlated reference sequence, i.e., by using the so-called paired change-point test.
In this contribution, we investigate the possibility of detecting changes with respect to two (or more) reference sequences. Our approach is based on a measure of differences between empirical characteristic functions leading to computationally attractive algorithms suitable for high-dimensional observations.