Volatility forecasting and volatility-timing strategies: A machine learning approach

Recent increases in stock price volatility have generated renewed interest in volatility-timing strategies. Based on high-dimensional models including machine learning, we predict stock market volatility and apply them to improve the performance of volatility-timing portfolios. Using various evaluation methods, we verify that those machine learning models have better prediction performances relative to the standard volatility models. Asset allocation results suggest that volatility-timing portfolios constructed using machine learning models tend to outperform the market, with higher average returns during the volatile market period. Our empirical evidence supports the application of machine learning in the construction of volatility-timing portfolios. • We improve volatility-timing strategies by forecasting the U.S. stock market volatility. • Machine learning and model averaging methods are applied to integrate predictive information. • High-dimensional models have higher accuracy compared to the standard volatility model. • Machine learning-based portfolios show outstanding investment performance.