Changepoint detection in non-exchangeable data

TL;DR

This paper introduces a Bayesian changepoint model that accounts for local correlation patterns within segments, improving detection accuracy in non-exchangeable data such as count data and financial time series.

Contribution

It proposes a novel Bayesian model relaxing exchangeability assumptions by incorporating segment-specific dependence, with a reversible jump MCMC algorithm for inference.

Findings

Enhanced changepoint detection in correlated data

Effective segmentation of financial time series

Improved monitoring of network count data

Abstract

Changepoint models typically assume the data within each segment are independent and identically distributed conditional on some parameters which change across segments. This construction may be inadequate when data are subject to local correlation patterns, often resulting in many more changepoints fitted than preferable. This article proposes a Bayesian changepoint model which relaxes the assumption of exchangeability within segments. The proposed model supposes data within a segment are $m$-dependent for some unkown $m \geqslant0$ which may vary between segments, resulting in a model suitable for detecting clear discontinuities in data which are subject to different local temporal correlations. The approach is suited to both continuous and discrete data. A novel reversible jump MCMC algorithm is proposed to sample from the model; in particular, a detailed analysis of the parameter space is exploited to build proposals for the orders of dependence. Two applications demonstrate the benefits of the proposed model: computer network monitoring via change detection in count data, and segmentation of financial time series.