Robust Capital Asset Pricing Model Estimation through Cross-Validation

Abstract

Limitations of Capital Asset Pricing Model (CAPM) continue to present inconsistent empirical results despite its rm mathematical foundations provided in recent studies. In this thesis, we examine how estimation errors of the CAPM could be minimized using the cross-validation technique, a concept that is widely applied in machine learning (CV-CAPM). We apply our approach to test the assumption of CAPM as a well-diversified portfolio model with data from S&P500 and Dow Jones Industrial Average (DJIA). Our results from the CV-CAPM validate that both S&P500 and DJIA are well-diversified market indices with statistically insignificant variation in unsystematic risks during and after the 2007 financial crisis. Furthermore, the CV-CAPM provides the smallest root mean square errors and mean absolute deviations compared to the traditional CAPM.