Research on Anomalies of China's Stock Market Based on Lasso and Ridge Regression Algorithm

Abstract

This study replicates 191 anomalies for six groups of anomalies using stock trading data and financial statement data of China’s A-share stock market from 2000-2021.With the single portfolio analysis of value-weighted and equally-weighted returns for a sample of all A-shares in China, and a comparison of anomalies replication results during the covid 19, confirms the existence of multiple significant anomalies in China A-share stock market.According to the Lasso and Ridge regression algorithms in machine learning, use effective anomaly factors form portfolios to predict stock returns. It also analyzes the performance differences between these two machine learning algorithms and the differences with the traditional OLS method under different scenarios and considering the real transaction costs.The machine learning algorithms provide higher predicted returns than traditional OLS, and in most cases the Ridge regression algorithm performs better than the Lasso regression algorithm for long-short portfolios.The analysis using the full anomaly factors is also considered in this paper.It provides reference for market investors to make stock return prediction based on anomalies.