High-Dimensional Mean–Variance Optimization with Nuclear Hedging Portfolios
-
Series
-
SpeakerRasmus Lönn (Erasmus University Rotterdam)
-
FieldEconometrics, Data Science and Econometrics
-
LocationUniversity of Amsterdam, Roeterseilandcampus, E5.07
Amsterdam -
Date and time
October 31, 2025
12:30 - 13:30
We introduce a novel framework for constructing mean–variance efficient portfolios when the number of assets is large. By formulating the estimation problem as a system of hedging regressions, we jointly estimate the expected excess returns and the precision matrix. We show that, under general factor structures, hedging returns exhibit a near–low-rank structure. We therefore reduce estimation risk by adapting high-dimensional penalized reduced rank regression techniques, which regularize the nuclear complexity (i.e., sum of the singular values) of hedging portfolio returns. This provides an intuitive estimation framework that delivers mean–variance optimal portfolio weights without requiring explicit high-dimensional matrix inversion.