Robust monitoring conditional volatility change for time series based on support vector regression

Abstract

This study considers a robust monitoring procedure aimed at detecting an anomaly of conditional volatility from sequentially observed time series following a (nonlinear) generalized autoregressive conditional heteroscedastic (GARCH) model. We employ a specifically designed cumulative sum (CUSUM) method hybridized with the asymmetric Huber support vector regression, named AHSVR. AHSVR-GARCH model provides an effective way to model nonlinear GARCH time series while significantly augmenting the performance of the proposed monitoring process in terms of stability and detection ability. Monte Carlo simulations illustrate the functionality of AHSVR-GARCH monitoring scheme, demonstrating AHSVR-GARCH model’s superiority over the standard SVR-GARCH model when monitoring GARCH-type time series. Data analysis using the S&P 500, NASDAQ index, and Korea Composite Stock Price Index (KOSPI) also affirms our method’s efficacy in real-world applications.