A Machine Learning Approach to Stock Returns Prediction in China
Log in to collectAcademic paper
Are Stock Returns Predictable in China? A Machine Learning Approach
huihang wu; xingkong wei; Xiaoyan Zhang
- Tsinghua University
- National Postdoctoral Association
- ?Postdoctoral Fellow
- ?Tsinghua University - PBC School of Finance
- China Institute of Finance and Capital Markets
- ?China Securities Co., Ltd
Strategy in a nutshell
This is a monthly machine learning stock selection strategy for A-shares in the Shanghai and Shenzhen markets using 108 US-market anomalies as input features. Multiple models—including OLS with penalties, PCA, PLS, regression trees, random forests, GBDT, and neural networks with 1–5 hidden layers—predict next-month stock returns. Stocks are sorted by predicted returns, and portfolios are formed using long-only, long-short, or short-only strategies, rebalanced monthly. Out-of-sample results (2010–2019) show that the long-short portfolio achieves an average monthly return of 1.99% and an annualized Sharpe ratio of 1.13%, demonstrating robust predictive performance.
Economic rationale
The strategy works because machine learning models capture complex patterns in stock anomalies—momentum, liquidity, and volatility—that traditional econometric models often miss. In China’s developing stock market, ML methods significantly improve out-of-sample return predictions, demonstrating strong forecasting power beyond classic Efficient Markets assumptions