Portfolio optimization via strategy-specific eigenvector shrinkage


We estimate covariance matrices for portfolio optimization that are tailored to given constraints. Our factor-based, data-driven construction relies on a generalized version of James-Stein for eigenvectors (JSE), which reduces estimation error in the leading sample eigenvector by shrinking toward a subspace of dimension ≥ 1. Unchecked, this error gives rise to excess volatility for optimized portfolios. Our results include a formula for the asymptotic improvement of JSE over the sample leading eigenvector as an estimate of ground truth, and provide improved optimal portfolio estimates when variance is to be minimized subject to finitely many linear constraints.

Publication date: 
September 8, 2023
Publication type: 
Other Research Areas