Abstract

This study proposes a novel Black–Litterman portfolio model that leverages machine learning predictions based on size, book-to-market, momentum, and volatility. Our model integrates insights from the four-factor model and the low volatility anomaly, paving the way for a more systematic and automated process in view construction. The proposed methodology significantly augments the out-of-sample portfolio performance, outpacing benchmark strategies across various metrics, such as alpha and the Sharpe ratio. This improvement underscores the substantial economic gains offered by our model, with its Sharpe ratio being approximately 2.4 times that of the market index. Furthermore, the alpha of our portfolio exhibits an impressive annual rate of 18.7%. Additionally, forward-looking views based on machine learning prove superior to naive, backward-looking views based on historical means by more accurately aligning portfolio returns more closely with the actual distribution of future returns, further optimizing risk-adjusted returns. This study bridges the gap between asset pricing theory and portfolio management literature, demonstrating the potential of machine learning in enhancing portfolio management efficiency.