Fine Properties of the Optimal Skorokhod Embedding Problem

Mathias Beiglböck, Marcel Nutz, Florian Stebegg

Abstract

We study the problem of stopping a Brownian motion at a given distribution $\nu$ while optimizing a reward function that depends on the (possibly randomized) stopping time and the Brownian motion. Our first result establishes that the set $\mathcal{T}(\nu)$ of stopping times embedding $\nu$ is weakly dense in the set $\mathcal{R}(\nu)$ of randomized embeddings. In particular, the optimal Skorokhod embedding problem over $\mathcal{T}(\nu)$ has the same value as the relaxed one over $\mathcal{R}(\nu)$ when the reward function is semicontinuous, which parallels a fundamental result about Monge maps and Kantorovich couplings in optimal transport. A second part studies the dual optimization in the sense of linear programming. While existence of a dual solution failed in previous formulations, we introduce a relaxation of the dual problem that exploits a novel compactness property and yields existence of solutions as well as absence of a duality gap, even for irregular reward functions. This leads to a monotonicity principle which complements the key theorem of Beiglböck, Cox and Huesmann [Optimal transport and Skorokhod embedding, Invent. Math., 208:327-400, 2017]. We show that these results can be applied to characterize the geometry of optimal embeddings through a variational condition.

Introduction

In stochastic calculus, the Skorokhod Embedding Problem (SEP) is a foundational challenge: given a Brownian motion $B$, find a stopping time $\tau$ such that the stopped process $B_\tau$ follows a target probability distribution $\nu$. While the existence of such stopping times has been established for decades, the "Optimal" SEP—which seeks to maximize a reward function $G(\tau, B)$—presents significant mathematical hurdles.

A recent study by Mathias Beiglböck, Marcel Nutz, and Florian Stebegg provides a major breakthrough in this field. By establishing a robust framework for duality and proving the density of pure stopping times, the authors resolve long-standing questions regarding the relationship between randomized and non-randomized strategies, as well as the existence of dual solutions for irregular reward functions.

The Density of Pure Embeddings

A central question in optimization is whether "randomized" strategies (where an external source of randomness helps decide when to stop) provide a higher reward than "pure" strategies. In the context of SEP, this is the distinction between the set of stopping times $\mathcal{T}(\nu)$ and the set of randomized embeddings $\mathcal{R}(\nu)$.

The authors prove a Density Theorem, showing that $\mathcal{T}(\nu)$ is weakly dense in $\mathcal{R}(\nu)$. This result is the SEP analogue of the relationship between Monge maps and Kantorovich couplings in Optimal Transport theory. The implication is profound: for any lower semicontinuous reward function, the maximum reward achievable via randomization is no better than what can be achieved (or closely approximated) by a pure stopping time.

The technical innovation behind this proof is the concept of internal randomness. The authors demonstrate that Brownian motion is sufficiently "wiggly" at its inception that the germ $\sigma$-field (the behavior of the path in the first infinitesimal moment) can be used to simulate any external coin flip. By using a tiny initial segment of the path as a randomization device, any randomized strategy can be converted into a pure stopping time that is nearly indistinguishable in its distribution and reward.

Robust Duality and Relaxation

In linear programming and optimization, solving the "dual" problem often provides the solution to the "primal" one. However, previous attempts to formulate a dual for the SEP often encountered a "duality gap" or failed to guarantee the existence of a dual optimizer, particularly when the reward function $G$ was not continuous.

To address this, the authors introduced several key methodological shifts:

  1. Relaxed Dual Space: Instead of restricting the dual problem to standard martingales, they expanded the search space to include local martingales and functions within $L^1(\nu)$. This expansion provides the necessary compactness to ensure that a dual solution actually exists.
  2. Capacity Theory: By utilizing Choquet’s capacity theory, the researchers extended the zero-duality-gap result ($S(G) = I(G)$) to all upper semianalytic reward functions, moving far beyond the limitations of simple continuous functions.

These adjustments ensure that the primal and dual values match, even for "messy" or irregular rewards, providing a complete theoretical framework for the optimization problem.

Geometric Characterization: Cave Embeddings

The theoretical results have immediate applications in characterizing the geometry of optimal stopping strategies. One such application is the study of "cave" embeddings—strategies where the stopping rule is defined by two boundaries (a "floor" and a "ceiling").

By applying their duality theory, the authors derived a variational condition that uniquely identifies the optimal boundaries for a given reward. This complements the monotonicity principle, which describes the support of optimal embeddings. The study notes, however, that this principle requires the reward to be of "class D" (an integrability condition); without this, the geometric support of the optimal strategy can become significantly more complex.

Conclusion and Future Directions

This work provides a definitive treatment of the fine properties of the Optimal SEP in one dimension. By bridging the gap between pure and randomized embeddings and establishing a robust duality theory, it provides the tools necessary to solve complex optimization problems in stochastic control.

Looking forward, the techniques developed here—particularly the relaxation to local martingales and the use of internal randomness—hold promise for extension into higher dimensions and more general Markov processes. Furthermore, these insights are expected to influence the field of Martingale Optimal Transport, which is critical for robust finance and the pricing of exotic options in volatile markets.

Summary updated

Abstract

We study the problem of stopping a Brownian motion at a given distribution $\nu$ while optimizing a reward function that depends on the (possibly randomized) stopping time and the Brownian motion. Our first result establishes that the set $\mathcal{T}(\nu)$ of stopping times embedding $\nu$ is weakly dense in the set $\mathcal{R}(\nu)$ of randomized embeddings. In particular, the optimal Skorokhod embedding problem over $\mathcal{T}(\nu)$ has the same value as the relaxed one over $\mathcal{R}(\nu)$ when the reward function is semicontinuous, which parallels a fundamental result about Monge maps and Kantorovich couplings in optimal transport. A second part studies the dual optimization in the sense of linear programming. While existence of a dual solution failed in previous formulations, we introduce a relaxation of the dual problem that exploits a novel compactness property and yields existence of solutions as well as absence of a duality gap, even for irregular reward functions. This leads to a monotonicity principle which complements the key theorem of Beiglb\"ock, Cox and Huesmann [Optimal transport and Skorokhod embedding, Invent. Math., 208:327-400, 2017]. We show that these results can be applied to characterize the geometry of optimal embeddings through a variational condition.
BibTeX
@article{beiglbock2021fine,
  author = {Mathias Beiglböck and Marcel Nutz and Florian Stebegg},
  title = {{Fine Properties of the Optimal Skorokhod Embedding Problem}},
  journal = {Journal of the European Mathematical Society},
  year = {2021},
  doi = {10.4171/jems/1122},
}

Loading related content...

Figure