Generalized Black-Litterman Portfolio Optimization with Decision Fusion

In the realm of investment decision-making, it is widely recognized that effective decision-making necessitates the integration of various knowledge domains. This paper presents a novel approach to estimating investors’ views in the Black-Litterman model by accumulating evidence from various sources. We provide explicit expressions to represent investors’ uncertainty and lack of knowledge, connecting their knowledge to their decisions using Dempster-Shafer Theory of Evidence. We obtain equilibrium expected return through a portfolio combination strategy and model covariances using Wishart stochastic volatility specification with a sparse initial conditional covariance matrix. The sparsity is introduced using Graphical Lasso, with regularization based on asset diversification distribution. Post-transaction costs, the proposed strategy consistently and significantly improves the Sharpe ratio over 1/N and outperforms eight well-established strategies for managing estimation risk. Extensive robustness experiments show that our approach withstands various choices of parameters