The pricing ability of factor model based on machine learning: Evidence from high-frequency data in China

Abstract

The existing literature mainly documents the asset pricing models estimated on low-frequency data, lacking the empirical evidence for exploring the “right” systematic factors based on high-frequency (HF) level. This study develops a revised HF factor model and evaluates the asset pricing performance. Using machine learning algorithms, we find that HF factor model includes three very persistent systematic factors, well-approximated by a portfolio of market, finance, and information. Sharpe ratios and out-of-sample tests prove that the HF revised factor model has the best explanatory power compared to the CAPM, Fama-French three-factor and five-factor models. The findings contribute to an in-depth understanding of the characteristics and mechanisms of risk and return from an HF perspective in the Chinese stock market.