Change Point Detection in Financial Time Series for Market Regime Analysis

Abstract

Financial time series data are inherently non-stationary, and there are also structural breaks and regime changes. Thus, a static model cannot be applied to this problem. Change point detection (CPD) algorithms are used in this paper to find changes in the market regime, and these include cumulative sum (CUSUM), Bayesian online change point detection (BOCPD), and pruned exact linear time (PELT). The returns of the S&P 500 index during the 2008 financial crisis are applied as an empirical case study to show that the above methods can detect both changes in the level of returns and volatility regimes. This paper shows how the above methods detect changes in both return level and volatility regimes. Results show that PELT performs better than others in detecting multiple change points without knowing the number of shifts. Bayesian methods can also be applied to real-time risk monitoring. Based on the above results, CPD can be added to the risk management system and algorithmic trading strategies to react to changes in the market.