Data analytics enhanced component volatility model

Yao, Y; Zhai, J; Cao, Y; Ding, X; Liu, J; Luo, Y

Abstract

Volatility modelling and forecasting have attracted many attentions in both finance and computation areas. Recent advances in machine learning allow us to construct complex models on volatility forecasting. However, the machine learning algorithms have been used merely as additional tools to the existing econometrics models. The hybrid models that specifically capture the characteristics of the volatility data have not been developed yet. We propose a new hybrid model, which is constructed by a low-pass filter, the autoregressive neural network and an autoregressive model. The volatility data is decomposed by the low-pass filter into long and short term components, which are then modelled by the autoregressive neural network and an autoregressive model respectively. The total forecasting result is aggregated by the outputs of two models. The experimental evaluations using one-hour and one-day realized volatility across four major foreign exchanges showed that the proposed model significantly outperforms the component GARCH, EGARCH and neural network only models in all forecasting horizons.

Citation

Yao, Y., Zhai, J., Cao, Y., Ding, X., Liu, J., & Luo, Y. (2017). Data analytics enhanced component volatility model. Expert systems with applications, 84, 232-241. https://doi.org/10.1016/j.eswa.2017.05.025

• Journal Article Type: Article
• Acceptance Date: May 9, 2017
• Online Publication Date: May 10, 2017
• Publication Date: Oct 30, 2017
• Deposit Date: May 10, 2017
• Publicly Available Date: May 10, 2018
• Journal: Expert Systems with Applications
• Print ISSN: 0957-4174
• Publisher: Elsevier
• Volume: 84
• Pages: 232-241
• DOI: https://doi.org/10.1016/j.eswa.2017.05.025
• Publisher URL: https://doi.org/10.1016/j.eswa.2017.05.025
• Related Public URLs: http://www.sciencedirect.com/science/journal/09574174/84/supp/C