The Fama–French Five-Factor Model with Hurst Exponents Compared with Machine Learning Methods

Scholars and investors have been interested in factor models for a long time. This paper builds using the monthly data of A-share market. We construct seven-factor model by adding Hurst exponent momentum to Fama–French five-factor find that there is 7% improvement average R–squared. Then, we compare five machine learning algorithms with ordinary least squares (OLS) one representative stock all A-Share stocks. regularization algorithms, such as lasso ridge, worse performance than OLS. SVM random forests good fitting power, while neural network not always better OLS, depending on data, frequency, period, etc.