Generalized Black-Litterman with Decision Fusion

At the heart of the portfolio selection problem lies the challenge of accurately estimating asset expected returns and covariance matrices. The classical Black-Litterman model addresses this challenge by combining market equilibrium and investors' views within the Markowitz mean-variance framework. This paper identifies three potential weaknesses of this allocation framework and proposes solutions to address them. First, we characterize the equilibrium model as a flexible, data-driven framework and present a new portfolio combination strategy for estimating equilibrium expected returns. Second, we leverage information from multiple sources to understand asset return dynamics comprehensively. Then, we establish a connection between investors' knowledge and the decision-making process through the use of an entropy-based decision-fusion system. Finally, we propose a GLASSO-Wishart model for analyzing realized covariance matrices of asset returns and show how estimation risk can be reduced by restricting the matrix of a Wishart distribution to sparse parameterization. In the presence of transaction costs, the generalized Black-Litterman (GBL) strategy can consistently and significantly increase the Sharpe ratio over the 1/N strategy. Regarding risk-reward maximization, we observe our GBL strategy to outperform eight well-studied benchmark strategies proposed in the literature to manage estimation risk. Extensive robustness experiments show that our approach withstand various choices of parameters