Pricing Options on Forwards in Function-Valued Affine Stochastic Volatility Models

This paper derives semi-closed Fourier-based pricing formulas for European options on forward contracts within function-valued infinite-dimensional affine stochastic volatility models, specifically analyzing Gaussian models driven by finite-rank Wishart processes and pure-jump extensions of the Barndorff-Nielsen-Shephard framework to capture term structure risks.

The Gist

Imagine you are a farmer trying to sell your wheat next year, or an airline trying to lock in the price of jet fuel for next winter. You don't want to guess the price; you want a forward contract—a promise to buy or sell at a specific price on a specific future date.

But here's the problem: the future is messy. Prices don't just move up or down; they jump, they spike, and they behave differently depending on how far away the delivery date is. A price change for wheat delivered in one month might look totally different from a price change for wheat delivered in ten years.

This paper is about building a super-smart calculator to figure out the fair price of options (insurance policies) on these forward contracts, even when the market is chaotic and the "volatility" (the speed and size of price swings) is constantly changing.

Here is the breakdown of their approach using simple analogies:

1. The Problem: The "Infinite" Curve

Usually, financial models look at a single price point. But in reality, you have a whole curve of prices: one for next week, one for next month, one for next year, and so on. It's like a snake with infinite segments.

• The Paper's Solution: Instead of trying to count every single segment of the snake, they treat the whole curve as a single, living object in a high-dimensional space. They use a framework called HJMM (think of it as a GPS for the forward curve) that tracks how the entire shape of the price curve evolves over time.

2. The Engine: "Stochastic Volatility"

In old models, the "roughness" of the market (volatility) was like a car driving at a constant speed on a highway. But in real life, the market is like a car driving through a storm: sometimes it's smooth, sometimes it's bumpy, and sometimes it gets hit by a hailstorm (a sudden jump).

• The Paper's Solution: They model volatility not as a fixed speed, but as a living, breathing creature that changes its own shape and behavior over time. They look at two specific types of "creatures":
  1. The Gaussian (Wishart) Model: Think of this as a smooth, rolling wave. It's predictable and continuous, like the ocean swells. It's great for normal market days.
  2. The Pure-Jump Model: Think of this as a drunkard stumbling. It walks normally, but occasionally takes a giant, random leap. This is crucial for modeling sudden market shocks (like a war starting or a supply chain breaking).

3. The Magic Trick: "Affine" Math

Calculating the price of these options with such complex, moving parts is usually impossible without a supercomputer running for days.

• The Paper's Solution: They use a special type of math called Affine Processes. Imagine the market is a complex machine with thousands of gears. Usually, you'd have to track every single gear. But "Affine" math is like finding a master key. It proves that even though the machine is complex, you can describe its future behavior using a simple set of rules (called Riccati equations).
• The Analogy: It's like predicting the weather. Instead of simulating every single air molecule, you use a few key variables (pressure, humidity, temperature) to get a very accurate forecast. The authors found the "key variables" for these infinite-dimensional markets.

4. The Two Approaches (The "Recipes")

The paper tests two different recipes for their calculator:

• Recipe A (The Jump Model): This is for markets that get shocked. They derived a formula that uses Fourier transforms (a mathematical way of breaking a complex wave into simple sine waves). It's like taking a complex song and breaking it down into individual notes so you can analyze it easily. They proved this works even when the "jumps" depend on the current state of the market (e.g., the market is more likely to jump if it's already nervous).
• Recipe B (The Wishart Model): This is for smoother, more continuous markets. Since the math here is too heavy to solve perfectly, they used a finite-rank approximation.
  • Analogy: Imagine trying to draw a perfect circle. It's hard. But if you draw a 100-sided polygon, it looks like a circle. If you draw a 1,000-sided polygon, it looks even better. They proved that using a "polygon" with just a few sides (a low number of factors) is enough to get a price that is almost identical to the perfect circle, saving massive amounts of computing power.

5. The Result: Fast and Accurate

The authors tested their formulas against Monte Carlo simulations.

• Analogy: Monte Carlo is like running a video game 100,000 times to see what happens on average. It's accurate but incredibly slow.
• The Win: Their new formulas are like having a crystal ball. They give the answer almost instantly (in milliseconds) and match the slow, heavy simulations almost perfectly.

Why Does This Matter?

In the real world, energy traders (oil, gas, electricity) and commodity traders (wheat, corn, metals) need to price options every second.

• Before: They might have had to use slow, inaccurate models or wait hours for a computer to crunch the numbers.
• Now: With this paper, they have a toolkit that handles the entire curve of prices (not just one date) and accounts for sudden market shocks, all while running fast enough to be used in real-time trading.

In a nutshell: The authors built a high-speed, high-precision navigation system for the chaotic, infinite-dimensional ocean of commodity prices, allowing traders to price their insurance policies (options) instantly and accurately, even when the market is storming.

Technical Summary

1. Problem Statement

The paper addresses the challenge of pricing European-style options written on forward contracts within commodity and energy markets. These markets are characterized by:

• Function-valued dynamics: Forward prices are not single values but entire curves evolving over time-to-maturity ($x = T - t$).
• Stochastic Volatility (SV): Volatility is time-varying, exhibits clustering, and is prone to sudden spikes (jumps), which constant volatility models fail to capture.
• Infinite-dimensional complexity: Modeling the full forward curve requires an infinite-dimensional state space, making standard finite-dimensional affine models insufficient.
• Computational intractability: Traditional Monte Carlo simulations for infinite-dimensional stochastic partial differential equations (SPDEs) are computationally expensive and slow.

The authors aim to derive semi-closed pricing formulas for these options that capture the term structure risk and maturity-specific dynamics while remaining computationally tractable.

2. Methodology

The authors adopt the Heath-Jarrow-Morton-Musiela (HJMM) framework, where forward prices are modeled directly as functions of time-to-maturity. The dynamics are governed by a stochastic partial differential equation (SPDE) driven by a cylindrical Brownian motion, with the covariance structure modulated by an infinite-dimensional affine stochastic volatility process.

The paper investigates two specific classes of affine models for the instantaneous covariance process:

A. The Pure-Jump Affine Model

• Structure: Based on an extension of the operator-valued Ornstein-Uhlenbeck process. The covariance process $Y_t$ takes values in the cone of positive self-adjoint trace-class operators.
• Jumps: The model incorporates state-dependent jumps (compound Poisson processes) where the jump intensity depends affinely on the current covariance state. This captures sudden market shocks (e.g., supply disruptions).
• Mathematical Tool: The authors derive generalized Riccati equations in Banach spaces. They rigorously establish conditions for the existence of exponential moments in complex domains, allowing for the use of Fourier-Laplace transforms.
• Pricing: They derive semi-explicit pricing formulas for call and put options using Fourier inversion, relying on the solution to these complex Riccati equations.

B. The Infinite-Dimensional Wishart Model

• Structure: A Gaussian model where the covariance process is driven by a Wishart process on the cone of positive trace-class operators.
• Finite-Rank Approximation: Since the Wishart process remains finite-rank if initialized as such, the authors propose a finite-rank Riccati approximation. They project the infinite-dimensional problem onto a finite-dimensional subspace (matrix-valued Riccati equations).
• Pricing: This yields a Fourier-based pricing approximation that preserves the infinite-dimensional maturity structure of the forward curve while making the covariance dynamics computationally manageable.

C. Numerical Implementation

• Basis Functions: The forward curve space is discretized using a specific Hilbert space $H_w$ with a weight function $w(x) = e^{\alpha x}$. The basis functions are constructed from Laguerre polynomials, allowing for efficient evaluation of the shift semigroup and integrals.
• Validation: The authors validate their analytical formulas against conditional Monte Carlo simulations. For the Wishart model, they use a conditional Gaussian scheme; for the jump models, they simulate the jump paths directly.

3. Key Contributions

1. Rigorous Existence Theory: The paper provides a rigorous mathematical proof for the existence of solutions to the complex generalized Riccati equations in infinite-dimensional Hilbert spaces for the pure-jump model. This establishes the validity of the exponential moments required for Fourier pricing.
2. Semi-Closed Pricing Formulas:
   • Derivation of exact semi-closed formulas for the pure-jump model.
   • Development of finite-rank pricing approximations for the Wishart model, bridging the gap between infinite-dimensional theory and numerical implementation.
3. State-Dependent Jumps: The framework successfully handles state-dependent jump intensities, a feature crucial for modeling realistic commodity market dynamics where volatility spikes often correlate with the current volatility level.
4. Computational Efficiency: The paper demonstrates that the affine approach is orders of magnitude faster than Monte Carlo simulation for explicit Lévy and Barndorff-Nielsen-Shephard (BNS) specifications, while maintaining high accuracy.

4. Results

• Accuracy: Numerical experiments show that the affine pricing formulas align closely with Monte Carlo benchmarks.
  • Wishart Model: The finite-rank approximation converges rapidly. With just 5 basis functions ($N=5$), the relative error in option prices is below 0.02% compared to a benchmark with $N=10$.
  • Pure-Jump Models: Both the Lévy-driven and state-dependent jump models show fast convergence with respect to the number of basis functions.
• Convergence: Tables 1–4 demonstrate that increasing the number of basis functions ($N$) or the simulation paths ($N_{sim}$) leads to negligible differences, confirming the robustness of the method.
• Runtime:
  • For explicit Lévy and BNS models, the affine formula takes < 0.2 seconds, whereas Monte Carlo takes ~88–103 seconds (a speedup of ~500x).
  • For the state-dependent jump model (where Riccati equations must be solved numerically), the affine method is still competitive (55s vs 71s), though the gap narrows due to the complexity of solving the ODEs.
• Term Structure Dynamics: The models successfully capture the non-monotonic behavior of option prices with respect to delivery lag, a feature observed in real commodity markets but often missed by simpler models.

5. Significance

This work is significant for both theoretical finance and practical quantitative trading:

• Theoretical Advancement: It extends the affine transform formula to infinite-dimensional settings with state-dependent jumps, a non-trivial mathematical feat that opens the door to more realistic modeling of term structure markets.
• Practical Applicability: It provides a tractable alternative to Monte Carlo simulation for pricing complex derivatives on forward curves. The ability to price options in seconds rather than minutes/hours is critical for risk management and real-time trading in energy and commodity markets.
• Modeling Flexibility: By accommodating both Gaussian (Wishart) and Jump (Pure-Jump) dynamics within a unified function-valued framework, the paper offers a versatile toolkit for capturing the full spectrum of market behaviors, from smooth volatility clustering to abrupt shocks.

In summary, the paper successfully bridges the gap between high-dimensional stochastic modeling and computational feasibility, offering a robust framework for pricing options on forward curves in modern commodity markets.