The Application of Machine Learning in Value Investing
My research involves machine learning in asset pricing, value investing, earnings prediction and investor behaviour. 1. I achieved an AUC of 70% for predicting future company profit directions, which is the highest in the literature. 2. I propose a two-step machine learning approach to address the challenge left by previous studies, where momentum, liquidity, and volatility dominate the interpretability of algorithms in predicting future returns when using next month's returns to train the model. In a separate test set using post-2005 U.S. stock market data, the average return of a one-month holding buy-only portfolio can reach up to 3.92%, excluding transaction costs. 3. My research delves deep into the impact of price trend factors on value stock screening, including momentum, volatility, liquidity, liquidity volatility, and bid-ask spread. Using various neural networks, I identified trends of factors, providing a comprehensive understanding of their influence. 4. I propose the Investor Valuation Competition Theory to explain the disappearing value premium in the U.S. market and the difference in value premiums between developed and developing markets
The Application of Machine Learning in Value Investing