Robust Pricing-Hedging Dualities in Continuous Time
Abstract
This paper establishes a robust mathematical framework for pricing and hedging financial options in continuous time, bridging the gap between model-free and model-specific finance. By introducing "prediction sets"—specific beliefs about possible asset paths—the authors prove a general duality: the minimum cost to perfectly hedge an exotic option equals the maximum expected payoff across all market-calibrated models. This framework allows for the quantification of the impact of modeling assumptions and extends martingale optimal transport duality to multi-dimensional, multi-maturity settings.
Introduction
In quantitative finance, practitioners often face a trade-off between model-specific approaches, like the Black-Scholes model, and model-independent approaches. While model-specific methods provide precise prices, they are vulnerable to "Knightian uncertainty"—the risk that the underlying model is fundamentally incorrect. Conversely, model-independent methods offer robustness but often yield price bounds that are too wide for practical application. This paper, authored by Zhaoxu Hou and Jan Obłój, introduces a framework that interpolates between these two extremes by incorporating "prediction sets."
The Problem: Bridging the Gap
The core challenge addressed is the interpolation between a completely model-free view and a specific model. A trader might have certain beliefs—for instance, that volatility will remain within a specific range or that asset prices will not experience jumps exceeding a certain percentage. The authors aim to provide a mathematical structure where these beliefs can be formally integrated to tighten price bounds without committing to a single, potentially flawed probability measure.
Methodology
The authors employ several advanced mathematical techniques to establish their pricing-hedging duality:
- Path Space Restrictions (Prediction Sets): Instead of a single probability measure, the authors define a set $\mathfrak{P}$ of possible continuous paths. Superhedging is required only for paths within this set.
- Lebesgue Discretization: To manage continuous-time complexities, paths are discretized based on hitting specific price levels rather than fixed time intervals. This allows the approximation of continuous problems with discrete ones where duality is more tractable.
- Variational Formulation: Using Lagrange multipliers, the authors incorporate constraints from statically traded European options, transforming the problem into a minimax-solvable unconstrained version.
- Asymptotic Duality: To address the lack of compactness in calibrated martingale measures in continuous time, the authors introduce "approximate" problems that allow for slight mis-calibration, ensuring the duality holds.
Key Insights
- The Robust Spectrum: The framework enables a smooth transition from total uncertainty to model-specific certainty. As the prediction set $\mathfrak{P}$ shrinks, robust price bounds converge toward a single model price.
- Time-Change Invariance: Many practical beliefs, such as drawdown constraints, are time-invariant. The authors demonstrate that for such beliefs, the duality is particularly robust.
- Handling Mis-calibration: By defining an $\eta$-market model, the theory accounts for real-world market noise and bid-ask spreads, making it more applicable than models requiring perfect calibration.
- Multi-Dimensional Extension: The paper extends Martingale Optimal Transport duality to include multiple assets and maturities, a significant advancement over previous one-dimensional, single-maturity results.
Key Results
- Theorem 3.2 (General Duality): For bounded and uniformly continuous payoffs, the approximate superhedging price is shown to equal the approximate model-based price.
- Theorem 3.6 (Finite Options): In scenarios with a finite number of traded put options, the duality holds exactly, extending existing literature.
- Theorems 3.12 & 3.16: These provide the Martingale Optimal Transport duality for complex scenarios involving multiple maturities and time-invariant belief sets.
- Pathwise Superreplication: The authors prove that hedging strategies are "pathwise," meaning they depend only on the observed asset path, not on a specific probability measure.
Conclusion and Future Work
The framework provided by Hou and Obłój offers a powerful tool for robust finance, allowing for the integration of modeling beliefs into a rigorous duality framework. While the current results primarily focus on uniformly continuous payoffs and frictionless markets, they lay the groundwork for future research. Potential extensions include handling "digital" options with discontinuous payoffs, integrating transaction costs, and developing methods for practitioners to specify prediction sets $\mathfrak{P}$ using historical data or machine learning.
Summary updated