Do industries predict stock market volatility? Evidence from machine learning models

Abstract

In a novel take on the gradual information diffusion hypothesis of Hong et al. (2007), we examine the predictive role of industries over aggregate stock market volatility. Using high frequency data for U.S. industry indexes and various heterogeneous autoregressive (HAR) type and machine learning models, we show that most industries are informative for future aggregate market volatility in out-of-sample tests. While the oil and gas industry plays a more dominant role for the component of aggregate market volatility that is associated with discount rate fluctuations, consumer services are most informative over market volatility that is attributable to cash flow fluctuations. More importantly, we find that the predictive information captured by industries not only helps improve the volatility forecasts for the stock market, but can also be used to generate significant economic benefits for investors who use these volatility forecasts in their asset allocation strategies.