Hedging Maturity-Specific Risk in Forward Curve Derivatives under Stochastic Volatility

2026

Read Hedging Maturity-Specific Risk in Forward Curve Derivatives under Stochastic Volatility (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-28891.txt.

Introduction

Alberti and Karbach study how to hedge financial claims written on forward curves, the schedules of prices for delivery at future dates, when volatility itself moves randomly. Their setting uses a Heath-Jarrow-Morton-Musiela equation, a standard framework for whole forward curves, driven by a covariance structure of infinite rank. They seek the variance-optimal hedge, the strategy whose remaining error varies least.

The optimal hedge, they show, is a projection of the claim onto the traded instruments, measured in a norm built from the forward curve's own fluctuations. Simple strategies using finitely many maturities are dense among all strategies, so they can approximate the ideal hedge. Finite-rank projections converge as the rank grows. The total hedging error splits exactly into bucket, rank and residual parts. In enlarged information settings, the residual becomes a floor set by volatility risk from noise that cannot be traded.

Catalog record

YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE112 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.28891

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Hedging Maturity-Specific Risk in Forward Curve Derivatives under Stochastic Volatility Abstract . We study the variance-optimal hedging of European contingent claims written on forwards. We assume that the dynamics of the underlying forward curves follow a Heath–Jarrow–Morton–Musiela stochastic partial differential equation modulated by an infinite-rank stochastic covariance component. The variance-optimal hedge is then given by the Galtchouk–Kunita–Watanabe projection with respect to some covariance-norm quotient generated by the forward curve martingale. We show density of finite-maturity and delivery-window strategies, convergence of spectral finite-rank hedge projections and an exact decomposition of the quadratic hedging error into bucket, rank and residual risk components. In enlarged filtrations, the residual risk is a stochastic-volatility floor for claims loading on non-traded covariance noise. We illustrate the hedging framework in affine stochastic covariance and multiplicative HJMM models, and give a concrete example of the decomposition in a CIR stochastic covariance model. Keywords: Quadratic hedging, variance-optimal hedging, forward-curve derivatives, maturity-specific risk, stochastic volatility, Heath–Jarrow–Morton framework, finite-rank approximation, energy markets. Riccardo Alberti Sven Karbach [ [email protected] ] [ [email protected] ] 1. Introduction Derivatives whose underlying is a forward rate or forward price arise naturally in both fixed–income and commodity markets. Under some pricing measure, collateralised forward or futures prices are martingales, and European payoffs may depend either on one maturity or on a delivery window. For example, a caplet with strike \kappa and payment date T is a single-maturity option written on a forward rate with payoff \big(F_{T}(\tau^{\ast})-\kappa\big)^{+},\qquad\tau^{\ast}\geq T\ \text{fixed}. By contrast, a typical power option is a European option written on an electricity base-load future delivering over [\tau_{1},\tau_{2}] that pays \left(\frac{1}{\tau_{2}-\tau_{1}}\int_{\tau_{1}}^{\tau_{2}}F_{T}(u)\,\mathrm{d}u-\kappa\right)^{+}, so the payoff depends on a whole segment of the curve. Both cases fit the same infinite-dimensional Heath–Jarrow–Morton framework, in which the field (t,\tau)\mapsto F_{t}(\tau) is treated as a single forward-curve state variable. More precisely, we denote by x\coloneqq\tau-t the time-to-maturity of the forward, and set f_{t}(x)\coloneqq F_{t}(t+x) . In these coordinates, the risk-neutral evolution of the forward curve is described by the Heath–Jarrow–Morton–Musiela (HJMM) equation [ 12 , §2.4] : (1) \displaystyle\begin{cases}\mathrm{d}f_{t}(x)=\partial_{x}f_{t}(x)\,\mathrm{d}t+\sum_{i=1}^{d}\sigma_{t}^{(i)}(x)\,\mathrm{d}W_{t}^{(i)},&t>0,\ x>0,\\ f_{0}(x)=F_{0}(x),&x>0.\end{cases} Here \{W^{(i)}\}_{i=1,\ldots,d} denotes a family of independent Brownian motions with d\in\mathbb{N}\cup\{+\infty\} . The parameter d is the rank of the noise: finite d gives a finite-rank HJM model, while d=\infty gives an infinite-rank model. In this framework, European claims written on forwards are interpreted as nonlinear functionals of the state variable x\mapsto f_{T}(x) . In this paper, we develop a quadratic hedging theory for such claims in HJMM models with both stochastic volatility and infinite-rank noise, i.e., the family \left\{\sigma^{(i)}\colon i=1,\ldots,d\right\} may vary randomly in time and we let d=\infty . Note that empirical evidence from energy and commodity forward markets supports high-dimensional, maturity-dependent risk structures with time-varying correlations [ 1 ] . Principal component analyses of NordPool power futures require more than ten factors to explain 95\% of total variance [ 30 , 24 , 7 ] , in contrast with many interest-rate term structures where a few factors explain most variation [ 12 , §1.7.3] . Explained variance, however, is not the appropriate criterion for hedging. A direction with low unconditional variance may still materially affect option values, hedge ratios, quantiles, or value-at-risk whenever the payoff is sufficiently exposed to that direction [ 14 ] . Consequently, an a priori rank reduction based solely on principal-component analysis may lead to substantial hedging errors. The correct and relevant notion of approximation error is instead determined by the covariance-based hedging norm introduced in this paper. More precisely, we shall formulate the hedging problem in a full infinite-rank setting and assess any subsequent rank reduction directly in the metric relevant for hedging. 1.1. Stochastic Covariance and Maturity-Specific Risk Throughout the paper, we work on a filtered probability space supporting a cylindrical Brownian motion W on a separable Hilbert space G . The instantaneous volatility is represented by a predictable process \displaystyle\sigma_{t}\in\mathcal{L}_{2}(G,H), where H denotes the forward-curve state space. Accordingly, the noise term in ( 1 ) is understood as the Hilbert-space stochastic integral \sigma_{t}\,\mathrm{d}W_{t} . The corresponding instantaneous covariance operator is \displaystyle\Sigma_{t}\coloneqq\sigma_{t}\sigma_{t}^{\ast}\in\mathcal{L}(H),\qquad t\geq 0. For every t\geq 0 , the operator \Sigma_{t} is self-adjoint, non-negative, and trace class, but it may have infinite rank. We refer to the joint process (f_{t},\Sigma_{t})_{t\geq 0} as a stochastic-volatility-modulated HJMM model . Infinite-rank covariance provides a natural representation of maturity-specific risk. In an idealised market with a continuum of maturity dates, distinct regions of the forward curve may be exposed to distinct sources of risk. Therefore, capturing such effects generally requires infinitely many factors ( d=\infty ), see also the discussion [ 12 , Section 6.5.3] . 1.2. Hedging European Contingent Claims in Forward Markets Let K=h\bigl(F_{T}(\cdot)\bigr) be a European contingent claim in L^{2}(\Omega,\mathcal{F}_{T},\mathbb{P}) written on the forward curve at time T . This class includes both point-maturity payoffs and payoffs depending on averages over delivery windows. We assume that trading is possible in a strip of forward or futures contracts, with admissible self-financing gains defined in Section 2.3 below. Among all admissible strategies and initial endowments, we minimise the mean-square hedging error. We refer to the resulting optimisation problem as the variance-optimal hedging problem . Let M_{t}\coloneqq\int_{0}^{t}\sigma_{s}\,\mathrm{d}W_{s},\qquad t\geq 0, denote the H -valued martingale component of the forward-curve dynamics in ( 1 ). Since the forward curve is a martingale up to the Musiela shift under the pricing measure, the variance-optimal hedge is characterised by the Galtchouk–Kunita–Watanabe (GKW) decomposition (Theorem 3.3 ): K=\mathbb{E}[K]+\int_{0}^{T}\phi_{s}^{\star}\,\mathrm{d}M_{s}+N_{T},\qquad N\perp M. Here \phi^{\star} is the predictable GKW integrand and N a square-integrable martingale strongly orthogonal to M ; the initial capital \mathbb{E}[K] together with \phi^{\star} minimises the mean-square hedging error, with minimal residual risk \mathbb{E}\bigl[N_{T}^{2}\bigr] . Maturity-specific risk has two conceptually distinct consequences for this problem. First , even when the infinite-rank volatility is deterministic, exact replication by strictly viable portfolios involving only finitely many maturities may fail. Writing Q_{M}(t) for the martingale covariance density introduced below, its range is dense in its closed support, \overline{\operatorname{Ran}Q_{M}(t)}=\bigl(\ker Q_{M}(t)\bigr)^{\perp} , but need not itself be closed. The pointwise optimal hedge may then involve an unbounded pseudo-inverse and exist only as a limit of admissible hedges in the covariance norm, so the abstract GKW integrand \phi^{\star} need not be implementable by finitely many traded maturities or delivery periods. This is a failure of strict viability, not of market completeness: under deterministic volatility, a Brownian filtration, and the usual covariance-support condition the market remains approximately complete (Corollary 3.17 ; Example 3.16 works out an explicit hedge of this kind). Second , a strictly positive irreducible hedging error arises only when the covariance is stochastic and its additional source of randomness affects the claim without being spanned by trading in the forward curve. This unspanned covariance risk generates the stochastic-volatility floor \mathbb{E}[N_{T}^{2}] of the quadratic hedging problem. Along a nested family of closed approximation spaces, the mean-square hedging error then admits an exact decomposition into (i) a finite-bucket implementation error; (ii) a finite-rank spectral truncation error; and (iii) the irreducible residual risk \mathbb{E}[N_{T}^{2}] , separating the cost of the traded contracts from that of covariance-rank reduction and from the intrinsic incompleteness generated by unspanned risk (Proposition 3.18 ). The floor is made fully explicit for an affine covariance model driven by a Cox–Ingersoll–Ross process in Section 4.1.1

Read the complete text