Volatility Indicator Construction Via Deep Learning And The Application Of Multiple Volatility Indicators

The conventional approach to construct volatility indicator is theory-driven.Asset price is treated as a stochastic process,and an unbiased estimator of volatility is constructed accordingly.From the perspective of information extraction and volatility forecasting,the theory-driven approach is not necessarily optimal.Recent advances in the field of deep learning have made it possible to construct volatility indicators directly from high-frequency data.Based on this,we propose a data-driven and neural-network-based approach to construct volatility indicator.The indicator complements the existing volatility indicator system and provides additional volatility information.In-sample analysis shows that the indicator has additional predictive power for stock market volatility even after controlling lagged realized volatility and other volatility indicators.Out-of-sample analysis shows that the indicator can reduce the error of stock market volatility forecast,providing a 2%-37%relative improvement(based on out-of-sample R2 statistic)at several model settings.Any indicator alone is not enough to reflect volatility information in high-frequency data and it is more important to utilize all volatility indicators comprehensively.For the application of multi-volatility indicator system,this paper focuses on two aspects-volatility forecasting and investment strategy.For volatility forecasting,the main problem is the multicollinearity between the indicators.In contrast to the method by Engle and Gallo(2006),which puts all indicators into a unified model,we draw on Paye(2012)and Cai et al.(2017),adding each indicator as an exogenous variable to the model separately and then using model averaging to enhance prediction.Our approach is found to be better from the perspective of out-of-sample error,especially when there are lots of indicators.For investment strategies,we propose a low volatility strategy based on a combined volatility indicator.We found that the combined indicator can better explain the cross-sectional return of A shares.At the same time,the low volatility strategy based on the combined indicator achieves a higher annualized return and Sharpe ratio.