Robust change point analysis

Abstract

The change point problem is usually treated by statistical procedures for detection of instabilities in statistical models. The problem is usually formulated in terms of hypothesis testing and estimation problem. Typically, we have observations X1, …, Xn obtained at ordered time points and the basic task is to decide whether the model remains stable during the whole observational period or whether the model changes at some unknown point(s) or become generally instable. In case of change(s) in the model being detected the further task is also to estimate the time of change and other parameters of the model in the periods where the model is stable. Various robust procedures for the above formulated problems are presented and discussed.