Enhancing Market Predictability and Investment Decision‐Making: Machine Learning Models for Predicting Stock Market Crashes in China

ABSTRACT This study aims to enhance the predictive accuracy of China's stock market crashes and optimise the investment decision‐making process by leveraging machine learning techniques and a diverse array of constructed aggregate factors. The empirical analysis validates the superior market‐timing capabilities of machine learning tools, particularly the feedforward neural network, in comparison to the conventional logistic regression model. This advancement, in turn, translates into nontrivial economic benefits for mean–variance investors across various portfolio types, even after accounting for trading costs. Moreover, our findings illuminate that stocks with small capitalisations, traded on the Shenzhen Stock Exchange, and within aggressive industry sectors exhibit heightened sensitivity to market crashes. Additionally, we identify substantial supplementary effects of aggregate market‐level characteristics in conjunction with well‐established macro factors. Notably, momentum, valuation and liquidity emerge as the most influential market‐level feature groups in predicting market crashes.