Semiparametric detection of changepoints in location, scale, and copula

Abstract

This paper proposes a new method to detect changepoints in the location and scale of univariate data sequences. The proposed method assumes that the data belong to the location‐scale family of distributions and estimate the associated densities nonparametrically. Specifically, the approach does not require knowledge of the functional form of the distribution of the data sequence. As such, the approach can detect changepoints in many distributions. We also propose a new method to detect changes in the location of multivariate sequences, using the marginals and a copula to capture the dependence between variables without the influence of marginal distributions. The performance of the proposed semiparametric approach is contrasted against both other competing nonparametric and Gaussian methods, via simulation studies, as well as applications arising from health and finance.