Semi-Static Variance-Optimal Hedging of Covariance Risk in Multi-Asset Derivatives

TL;DR

This paper introduces a semi-static hedging framework for multi-asset derivatives that combines dynamic trading with static positions in auxiliary claims, effectively reducing hedging errors in complex covariance risk scenarios.

Contribution

It develops a novel decomposition approach and systematic selection method for static hedging instruments, enabling explicit semi-static replication formulas under broad asset models.

Findings

Static hedging significantly reduces mean-squared error.

Explicit formulas for covariance swaps and dispersion trades.

Framework applies to diverse asset dynamics including jumps and stochastic volatility.

Abstract

We develop a semi-static framework for the variance-optimal hedging of multi-asset derivatives exposed to correlation and covariance risk. The approach combines continuous-time dynamic trading in the underlying assets with a static portfolio of auxiliary contingent claims. Using a multivariate Galtchouk--Kunita--Watanabe decomposition, we show that the resulting global mean-variance problem decouples naturally into an inner continuous-time projection onto the space spanned by the underlying assets and an outer finite-dimensional quadratic optimization over the static hedging instruments. To systematically select suitable auxiliary claims, we leverage multidimensional functional spanning theory, establishing that otherwise unhedgeable cross-gamma exposures can be structurally mitigated through static strips of vanilla, product, and spread options. As a central application, we derive explicit semi-static replication formulas for covariance swaps and geometric dispersion trades. Our framework accommodates a broad class of asset dynamics, including quadratic and stochastic Volterra covariance models, as well as affine stochastic covariance models with jumps, yielding tractable semi-closed-form solutions via Fourier transform techniques. Extensive numerical experiments demonstrate that incorporating optimally weighted static strips of cross-asset instruments substantially reduces the mean-squared hedging error relative to purely dynamic benchmark strategies across various model classes.