[Paper Review] Wishart conditional tail risk measures: An analytic approach

Jose Da Fonseca, Patrick Wong|arXiv (Cornell University)|Feb 6, 2026

Risk and Portfolio Optimization0 citations

TL;DR

The paper develops an analytical framework using the Wishart process to compute multivariate tail risk measures and intertemporal risk quantities via explicit MGF-based formulas, with a Fourier-transform approach to conditional tail expectations.

ABSTRACT

This study introduces a new analytical framework for quantifying multivariate risk measures. Using the Wishart process, which is a stochastic process with values in the space of positive definite matrices, we derive several conditional tail risk measures which, thanks to the remarkable analytical properties of the Wishart process, can be explicitly computed up to a one- or two-dimensional integration. These quantities can also be used to solve analytically a capital allocation problem based on conditional moments. Exploiting the stochastic differential equation property of the Wishart process, we show how an intertemporal (i.e., time-lagged) view of these risk measures can be embedded in the proposed framework. Several numerical examples show that the framework is versatile and operational, thus providing a useful tool for risk management.

Motivation & Objective

• Motivate and quantify multivariate risk measures with dependence captured through positive definite matrices.
• Extend tail conditional expectation computation to multivariate and higher-order moments using the Wishart process.
• Derive closed-form MGF representations and their derivatives to enable analytical risk measure calculations.
• Introduce a Fourier-transform framework to express conditional tail moments via MGFs and joint MGFs at multiple dates.
• Show intertemporal (time-lagged) risk measures within a dynamic Wishart-process setting.

Proposed method

• Model losses and dependencies in the space of positive definite matrices using the Wishart process (matrix SDE).
• Utilize the affine property to obtain exponentially affine MGFs of x_t with explicit Riccati-type solutions (a(t,θ), b(t,θ)).
• Compute derivatives of MGFs with respect to scalar multipliers to obtain conditional moments and tail measures (Proposition 2.3, Corollaries 2.4–2.7).
• Represent conditional tail expectations and higher moments via multidimensional and one-dimensional Fourier transform formulas (Propositions 3.1, 3.2; Remarks on Plancherel-Parseval).
• Extend to intertemporal settings by deriving joint MGFs for (x_t0, x_t1) and their derivatives (Propositions 2.5–2.6).
• Provide a numerical implementation illustrating speed, accuracy, and dependence effects (Section 4).

Experimental results

Research questions

• RQ1 How can one compute tail conditional moments (including higher orders) for multivariate risks with dependence structure captured by positive definite matrices?
• RQ2 Can the Wishart process yield closed-form, computable MGFs and their derivatives to enable analytic TCE and related risk measures?
• RQ3 How can intertemporal (time-lagged) tail risk measures be embedded in a Wishart-process framework using joint MGFs at multiple dates?
• RQ4 What is the computational cost and accuracy of the Fourier-transform representation for multivariate conditional tail measures compared to density-based methods?
• RQ5 How does dependence influence multivariate tail risk measures within this analytic Wishart framework?

Key findings

• The Wishart process provides an analytically tractable framework with MGFs that are exponentially affine, enabling explicit computation of conditional tail measures.
• Derivatives of MGFs with respect to scalar multipliers yield higher-order conditional moments and tail risk measures (Proposition 2.3 and Corollaries 2.4–2.7).
• Joint MGFs at two dates allow intertemporal tail risk measures, with closed-form structure for the two-time case (Propositions 2.5–2.6).
• Conditional tail measures can be expressed via Fourier transform representations (Propositions 3.1 and 3.2), enabling one-dimensional integrations under known MGFs.
• The framework accommodates general dependence among risks through matrix-valued losses, avoiding strict copula constraints and allowing richer dependency modeling.
• Numerical implementations demonstrate accuracy and computational efficiency, with explicit attention to dependence effects (Section 4).