Distributionally Robust Martingale Optimal Transport

Zhengqing Zhou, José Blanchet, Peter W. GlynnarXiv 2021 · DOI · arXiv

Abstract

We study the problem of bounding path-dependent expectations (within any finite time horizon $d$) over the class of discrete-time martingales whose marginal distributions lie within a prescribed tolerance of a given collection of benchmark marginal distributions. This problem is a relaxation of the martingale optimal transport (MOT) problem and is motivated by applications to super-hedging in financial markets. We show that the empirical version of our relaxed MOT problem can be approximated within $O(n^{-1/2})$ error where $n$ is the number of samples of each of the individual marginal distributions (generated independently) and using a suitably constructed finite-dimensional linear programming problem.

Introduction

In mathematical finance, Martingale Optimal Transport (MOT) serves as a cornerstone for "model-free" pricing and the hedging of exotic options. Unlike classical optimal transport, which seeks a coupling between distributions to minimize cost, MOT imposes an additional martingale constraint. This constraint ensures that the underlying stochastic process is consistent with risk-neutral pricing principles, where the current price of an asset is the discounted expected value of its future price.

Despite its theoretical elegance, MOT faces significant practical hurdles. First, it assumes that marginal distributions of asset prices are known with absolute certainty. In practice, we only have access to discrete market data—such as option prices at specific strikes—or empirical samples. If these empirical samples do not perfectly satisfy the "convex ordering" property required for a martingale to exist, the standard MOT problem becomes infeasible. Second, multi-period MOT is computationally intensive, and the field has lacked rigorous convergence rate analysis for numerical approximations. This paper introduces the $\epsilon$-Distributionally Robust Martingale Optimal Transport ($\epsilon$-DRMOT) framework to address these limitations.

The $\epsilon$-DRMOT Framework

The core innovation of this work is the relaxation of strict marginal constraints. Instead of requiring the joint measure to match target marginals exactly, the authors allow the marginals to reside within an $\epsilon$-neighborhood of the benchmark distributions, measured using the Wasserstein distance.

This relaxation transforms MOT from a "brittle" problem into a robust one. By allowing for a small tolerance, the framework can absorb the noise inherent in empirical data, such as bid-ask spreads or sampling errors. This ensures that the optimization problem remains feasible even when the data does not strictly adhere to theoretical martingale properties.

Theoretical Contributions and Duality

A significant contribution of the paper is the proof of Strong Duality (Theorem 1). The authors demonstrate that the primal problem—maximizing the expected payoff of an option—is dual to a robust super-hedging problem. In this dual perspective, a trader seeks the cheapest portfolio of vanilla options and dynamic trading strategies that stays above the exotic option's payoff across all possible models within the $\epsilon$-uncertainty set.

This duality provides a clear financial interpretation: the "adversary" in the distributionally robust optimization (DRO) framework selects the worst-case marginal distributions within the Wasserstein ball to maximize the expected payoff, forcing the hedger to account for model misspecification.

Sample Complexity and Convergence

One of the most impactful results of the study is the analysis of sample complexity. The authors prove that the empirical version of the relaxed MOT problem converges to the true value at a rate of $O(n^{-1/2})$, where $n$ is the number of independent samples from the marginal distributions. This $O(n^{-1/2})$ rate is considered the "canonical" or gold-standard rate in statistical estimation, providing high confidence in the framework's reliability as more data becomes available.

Furthermore, the authors address the issue of feasibility. They provide a lemma showing that for a sufficiently large sample size $n$, the empirical version of the $\epsilon$-DRMOT problem is guaranteed to be feasible with high probability. This solves a major practical headache where standard MOT often fails due to the lack of convex ordering in raw data.

Computational Tractability via Linear Programming

To bridge the gap between theory and application, the authors propose a discretization scheme using a uniform grid. They demonstrate that the infinite-dimensional optimization problem can be reduced to a finite-dimensional Linear Programming (LP) problem.

Key features of this reduction include:

  • Error Preservation: The discretization maintains the $O(n^{-1/2})$ error bound.
  • Complexity: The resulting LP involves $O(n^{3d/2})$ variables, where $d$ is the number of time steps.
  • Generalization: For cases where asset prices are not restricted to a compact interval, the authors utilize Skorokhod’s embedding to reduce the non-compact problem into a compact one, ensuring the results apply to a wide range of financial models.

Discussion and Future Directions

While the $\epsilon$-DRMOT framework represents a significant leap forward, it is not without challenges. The number of variables in the LP grows exponentially with the number of time periods ($d$), a manifestation of the "curse of dimensionality." While tractable for a few observation dates, very long-dated exotic options may still require further computational optimizations.

Additionally, while the paper establishes theoretical bounds for the tolerance parameter $\epsilon$, practical implementation on a trading desk will require methods for selecting an "optimal" $\epsilon$ that balances the trade-off between robustness (safety) and conservativeness (cost). Future research may also look toward extending these discrete-time results into continuous-time martingale settings.

Conclusion

By integrating Wasserstein-based distributional robustness into the Martingale Optimal Transport problem, this research provides a mathematically rigorous and computationally feasible path for robust financial modeling. The $O(n^{-1/2})$ convergence rate and guaranteed feasibility make $\epsilon$-DRMOT a powerful tool for practitioners dealing with noisy, real-world market data.

Summary updated

Abstract

We study the problem of bounding path-dependent expectations (within any finite time horizon $d$) over the class of discrete-time martingales whose marginal distributions lie within a prescribed tolerance of a given collection of benchmark marginal distributions. This problem is a relaxation of the martingale optimal transport (MOT) problem and is motivated by applications to super-hedging in financial markets. We show that the empirical version of our relaxed MOT problem can be approximated within $O\left( n^{-1/2}\right)$ error where $n$ is the number of samples of each of the individual marginal distributions (generated independently) and using a suitably constructed finite-dimensional linear programming problem.
BibTeX
@misc{zhou2021distributionally,
  author = {Zhengqing Zhou and José Blanchet and Peter W. Glynn},
  title = {{Distributionally Robust Martingale Optimal Transport}},
  year = {2021},
  eprint = {2106.07191},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2106.07191},
}
Figure