Consolidating Joint Predictability of Firm Characteristics in Chinese Stock Market : A Subset Combination Approach

We adopt a portfolio perspective and apply a subset combination approach to consolidate the joint predictability of a large number of firm characteristics in the Chinese A-share market. Our approach incorporates higher moments of stock return distribution and imposes shrinkage on the predictive power of each individual characteristic. The resultant optimized portfolio can achieve higher Sharpe ratio than factors from conventional asset pricing models. It also outperforms alternative optimized portfolios that use machine learning methods, such as Lasso and Neural Network, to consolidate characteristics. Augmenting our optimal portfolio with the Liu, Stambaugh, and Yuan (2019) factors, the new factor model can explain 20 out of 23 significant anomalies in the Chinese stock market