Growing Semantic Vines for Robust Asset Allocation

The vine structure has been widely studied as a graphical representation for high-dimensional dependence modeling, depiction of complicated probability density functions, and robust correlation estimation. However, the number of candidate vine structures grows exponentially as the number of elements increases, making the specification of the best vine structure a challenging issue. In this article, we propose to leverage semantic prior knowledge of assets extracted from their descriptive documents to find a suitable vine structure for financial portfolio optimization. A vine growing algorithm is provided and the robust covariance matrix estimation process is performed on this vine structure. The experiments show that our construction of a semantic vine is superior to the state-of-the-art arbitrary vine-growing method. The effectiveness of using semantic vines for robust correlation estimation for the classic asset allocation model on a large scale is also demonstrated