In this paper we present a multiscale approach for change point detection. The algorithm estimates likelihood-ratio (LR) test in several scrolling windows simultaneously. This makes the method adaptive to structural breaks of different scales. Critical values are calibrated in a data-driven way using multiplier bootstrap, which estimates nonasymptotic distribution of the test statistics.
- Likelihood ratio test
- Multiplier bootstrap
- Multiscale change point detection
- Scrolling window