关键词: 横截面收益预测;资产配置策略, 我国股票市场;组合LASSO-logistic方法

Abstract: This paper analyzes the out-of-sample predictability for the Chinese stock market returns with a large panel of 41 individual firm characteristics. We propose a machine-learning-driven approach-combination LASSO logistic regression to solve 3 problems of cross-sectional forecasts for stock return: 1)Incomplete construction of anomalous factors because of the insufficiency of firm characteristics; 2 )" Curse of Dimensionality" problem with too many feature variables; 3)Prediction instability problem caused by the high correlation between feature variables. Our empirical evidence indicates that the combination LASSO-logistic model can effectively identify the complex patterns hidden in the characteristic variables and expected returns better than the traditional multiple logistic models, support vector machine (SVM) models, and random forest models. We further find the firm characteristics variables affecting expected stock returns are not constant, and their significant dynamics suggest to some extent the weak stability of the Chinese stock market

Key words: Cross-sectional Expected Stock Returns, Portfolio Strategy, Chinese Stock Market, Combination LASSO-logistic Model