A factor pricing model based on machine learning algorithm

Abstract

We adopt Discrete Wavelet Transform (DWT) and Support Vector Regression (SVR) algorithm to predict stock returns and form a time-varying Machine Learning (ML) factor based on the predicted returns for improving the classical asset pricing in the Chinese stock market from 2000 to 2020. The results show that the sorted portfolios formed by the predicted return rates can obtain significant excess returns in the Chinese market. Furthermore, our research shows that incorporating the ML factor into the CH4 and FF5 model can improve the pricing power significantly. It indicates that the ML factor can complement the traditional pricing models. We also find the performance of factor models depend on macroeconomy and market sentiment, that is, the better the macroeconomic and stock market performance, the stronger the pricing power of the factor models in the Chinese market. Additionally, we explore whether ML factor falls under short-term, median-term, and long-term momentum and reversal factors, and our analysis demonstrates that the ML factor is more effective than momentum and reversal factors.