Change Point Detection with Multivariate Observations Based on Characteristic Functions

Abstract

We propose change-point detectors for multivariate independent observations, as well as corresponding methods involving observations which are driven by vector autoregressive models. The methods make use of characteristic functions (CFs).

Apart from other favorable features, an extra reason for using CFs is that with CFs vector observations are linearly projected onto the real line and the resulting statistics may be written in convenient closed-form expressions.

Keywords

• change-point
• characteristic function
• vector autoregressive model