Detecting Changes in Time Series Data using Volatility Filters

TL;DR

This paper introduces new sequential change detection algorithms based on windowed volatility filters, capable of identifying and locating transient changes in the volatility of univariate and multivariate time series data.

Contribution

It presents a novel adaptive filtering approach for change detection and a new method for change point localization, extending to multivariate data with distributed adaptive filters.

Findings

Effective detection of volatility changes demonstrated on synthetic data.

Successful application to real-world data showing practical utility.

Improved localization accuracy of change points.

Abstract

This work develops techniques for the sequential detection and location estimation of transient changes in the volatility (standard deviation) of time series data. In particular, we introduce a class of change detection algorithms based on the windowed volatility filter. The first method detects changes by employing a convex combination of two such filters with differing window sizes, such that the adaptively updated convex weight parameter is then used as an indicator for the detection of instantaneous power changes. Moreover, the proposed adaptive filtering based method is readily extended to the multivariate case by using recent advances in distributed adaptive filters, thereby using cooperation between the data channels for more effective detection of change points. Furthermore, this work also develops a novel change point location estimator based on the differenced output of the volatility filter. Finally, the performance of the proposed methods were evaluated on both synthetic and real world data.