Portfolio selection based on shrinkage estimation of the covariance matrix and ADMM algorithm

Sun, Ruili, Wang, Kunqi, Ma, Tiefeng, Liu, Shuangzhe
Journal of Asset Management · Springer Nature · 2026-09-12 · DOI: 10.1057/s41260-026-00468-w

Abstract

Portfolio selection is difficult when the covariance structure of asset returns is unstable and only imperfectly predictable. In such settings, portfolios that rely heavily on the sample covariance matrix may become unstable and overly concentrated. This paper proposes a framework that uses transformed portfolio representations generated from shrinkage ideas to construct more stable portfolios with a minimum variance orientation. The method builds informative transformed dictionaries from shrinkage information and then selects portfolio exposures through sparse regularized optimization solved by an alternating direction method of multipliers algorithm. The approach is closely related to the Global Minimum Variance (GMV) portfolio, but avoids directly estimating a large number of shrinkage parameters. Rolling-window out-of-sample experiments on several asset universes show that the proposed portfolios often achieve stronger returns and better risk-adjusted performance than standard GMV, equal-weighted, and conventional shrinkage benchmarks in the considered sample. The results suggest that combining representations generated from shrinkage ideas with sparse coefficient regularization can improve portfolio stability under covariance uncertainty.

Read at publisher →

Similar articles