Events

Date: 26 March 2026   Time: 14:00 - 15:00     

Location: Hybrid: MB503, SMS, QMUL, or via the Teams link (see seminar webpage)

In time series analysis, many data sets of practical interest contain abrupt changes in structure, such as the canonical setting of change points in the mean. We propose new methodology based on estimating functions in a general framework for detection of change points in the model parameters of a multivariate time series, including scenarios such as change in mean, regression, and potentially nonlinear autoregression. We propose a two-stage method; in the first stage, a multiscale scanning step produces initial change point estimators, which are used to define search intervals that are as large as possible and each contain only one true change point. In the second step, a cumulative sum (CUSUM)-based approach refines the initial change point estimators using the intervals generated in step one. The proposed approach is computationally efficient, suitable for heavy-tailed and dependent time series, and is robust to model misspecification. The theoretical properties of the procedure are examined, and the flexibility of the method is illustrated by applying it to a data example from the environmental sciences.