In classical mathematical finance, pricing a derivative typically begins by assuming a specific probability model that governs asset prices over time (such as the geometric Brownian motion in the Black-Scholes model). Under this assumed probability measure, the fair price of a derivative is the expected value of its future payoff under a risk-neutral measure, which perfectly matches the cost of a replicating trading strategy.
However, in reality, the true probability distribution of asset prices—and particularly their future volatility—is unknown. This is known as model uncertainty or Knightian uncertainty. If a financial institution mis-specifies the model, its hedging strategy can fail, leading to unexpected losses.
To protect against this, one can look for a superhedging strategy: a trading strategy that requires some initial capital and guarantees a portfolio value greater than or equal to the derivative's payoff along every possible path or under all plausible models. The minimum initial capital required to achieve this is the superhedging price.
The robust pricing-hedging duality is a mathematical equivalence that bridges this worst-case risk management perspective with a probabilistic valuation perspective. It states that:
$$\text{Minimum Superhedging Cost} = \sup_{\mathbb{P} \in \mathcal{M}} \mathbb{E}_{\mathbb{P}}[\text{Payoff}]$$
where $\mathcal{M}$ is the set of all risk-neutral (martingale) probability measures consistent with observed market prices or physical constraints.
In other words, the cheapest way to completely insulate yourself from risk across all possible models is exactly equal to the highest possible price assigned to the derivative by any single valid model.
The Primal and Dual Perspectives
This duality transforms a difficult pathwise or multi-model optimization problem (the primal problem of finding an optimal trading strategy) into a functional optimization problem over a space of probability measures (the dual problem).
- The Primal Problem (Hedging): Find a predictable trading strategy and the minimum initial capital such that the portfolio value at maturity covers the claim, regardless of which model or path is realized.
- The Dual Problem (Pricing): Find the worst-case scenario by maximizing the expected payoff of the claim over all martingale probability measures that satisfy the market's constraints.
As established by Biagini et al., in continuous-time markets subject to model uncertainty, market viability is fundamentally tied to the existence of a local martingale measure for every possible physical measure in a given family. Under these conditions, the maximum expected value of a financial claim across all risk-neutral measures is equal to the minimum initial capital required to hedge that claim across all possible physical models. This ensures that the cost of protecting against the worst-case scenario is mathematically consistent with the highest possible valuation under risk-neutrality.
Martingale Optimal Transport
A powerful way to understand why this duality works is through its connection to optimal transport theory. In classical optimal transport (the Monge-Kantorovich problem), the goal is to find the most efficient way to transform one probability distribution into another given a certain transportation cost.
In financial contexts, we observe the market prices of liquid options, which effectively reveal the marginal distributions of the asset price at different times. However, the joint distribution—the path the asset takes between those times—is unknown. To find the upper and lower bounds for an exotic, path-dependent option, we want to find the joint distribution that maximizes or minimizes the expected payoff.
As Cheridito et al. explain, extending optimal transport duality to include martingale or supermartingale constraints provides a functional method for determining superhedging prices. This framework, known as Martingale Optimal Transport (MOT), adds a crucial constraint to the transport problem: the price process must be a martingale (i.e., it must be fair, preventing risk-free arbitrage). The dual variables in these transport problems directly represent the robust hedging strategies that guarantee payoffs regardless of market paths, as noted by Beiglböck et al..
For example, Guyon et al. apply this constrained martingale optimal transport framework to determine sharp, model-independent bounds for VIX futures given S&P 500 option smiles. They demonstrate a zero duality gap, meaning the cost of an optimal superreplicating portfolio is mathematically identical to the extreme expected value of the volatility index under all market-consistent distributions.
Quasi-Sure vs. Pathwise Formulations
How do we mathematically define the set of possible scenarios? Researchers have developed two main paradigms, which have been shown to be deeply connected:
- The Quasi-Sure Approach: Model uncertainty is represented by a large, non-dominated family of probability measures. A property holds "quasi-surely" if it holds almost surely under every measure in this family. Nutz establishes a robust version of the no-arbitrage principle by requiring the condition to hold simultaneously across all measures in such a non-dominated set.
- The Pathwise Approach: Instead of probability measures, we specify a "prediction set" of physically possible trajectories that the asset price can take. Hou et al. demonstrate that the minimum cost required to superhedge an exotic option across a set of possible asset paths is equivalent to the maximum expected payoff calculated over all calibrated martingale models.
Obloj et al. show that these two frameworks are mathematically equivalent under specific structural conditions. The Fundamental Theorem of Asset Pricing and the Superhedging Theorem maintain a consistent duality across both perspectives, proving that the relationship between no-arbitrage and hedging costs is a fundamental property of robust markets.
Resolving Technical Challenges: Measurability and Attainment
While the intuition behind the duality is elegant, proving that it holds rigorously involves overcoming significant mathematical hurdles, particularly when moving from discrete-time to continuous-time settings or dealing with irregular payoffs.
- Borel Measurable Claims: Early robust finance theories required payoffs to be continuous or upper semicontinuous. However, real-world options like digital options have discontinuous, "rough" payoffs. Neufeld et al. extended the duality to all Borel measurable claims under volatility uncertainty, leveraging analytic sets to ensure that dynamic programming and universal hedging strategies remain valid without assuming a single reference probability measure.
- Existence of Strategies (Closedness): To ensure that the minimum hedging cost is an attained minimum rather than a theoretical limit, the set of reachable claims must be closed. Nutz used the Hahn-Banach theorem in discrete time to show that the set of financial claims hedgeable from zero capital is sequentially closed under a norm defined by the maximum expectation, guaranteeing that an optimal trading strategy always exists.
- Dual Attainment: In Martingale Optimal Transport, ensuring that a dual optimizer (the actual hedging portfolio) exists requires specific conditions on the cost function. Beiglböck et al. proved that dual attainment is guaranteed when the cost function exhibits semiconvexity in the target variable, providing a stable theoretical foundation for risk management.
By establishing that the worst-case expectation across a family of models matches the exact cost of an executable, pathwise hedging strategy, robust pricing-hedging duality provides a mathematically rigorous and practically viable framework for managing financial risk under model ambiguity.
Key Foundational Work
A Harmonic Analysis Solution to the Static Basket Arbitrage Problem (2003) — contributed 3 concept elements to this direction.