A Machine-Learning Approach in Assessing the Fama French Three and Fama French Five Factor Model

The purpose of this paper is to find out the disparity in prediction accuracy between the Fama French Three Factor Model and the Fama French Five Factor Model in portfolio return. The data used in this experiment are the 49 U.S. Industry Portfolio Dataset, a panel dataset from 1926 to 2023 across 49 industries. The data are assessed through a machine-learning approach, XGBoost regression, and the prediction accuracy is measured by root-mean-square error. Knowing the non-Gaussian characteristic of the root-mean-square error data collected, a Mann–Whitney U Test is used to compare the differences in prediction accuracy. Furthermore, a robustness analysis is conducted by adding time variants into the models to test the result. The result reveals no statistically substantial disparity between the Fama French Three and Fama French Five Factor Model under XGBoost Regression. However, this result might be altered by various pre-specifications of the XGBoost regression or under a different machine learning. This paper creates value for portfolio managers and investment professionals using various models to estimate expected returns and assess risk. Understanding the comparative performance of different factor models can guide investment decisions.