Generalizing the Skorokhod Embedding Problem

21 concepts10 papers1 proposal

Imagine you are observing a standard Brownian motion $B_t$ starting at zero. If you stop the process at a fixed time, say $t = 1$, the resulting value $B_1$ follows a standard normal distribution. But what if you want the stopped value to follow a completely different distribution—perhaps a uniform distribution, a discrete distribution, or a distribution with heavy tails?

The Skorokhod Embedding Problem (SEP) asks whether we can define a stopping time $\tau$—a rule that decides when to stop based only on the history of the process up to the current moment—such that the stopped process $B_\tau$ exactly matches a pre-specified target distribution $\mu$.

As Jan Obłój unifies in the comprehensive survey The Skorokhod embedding problem and its offspring, solving the SEP allows us to represent independent random variables as components of a continuous-time martingale. This connection is incredibly powerful: it lets mathematicians use the rich analytical toolkit of continuous stochastic calculus to prove bounds, derive inequalities, and solve complex problems in quantitative finance.

Beyond Brownian Motion: General Markov and Lévy Processes

While the classical SEP is framed around Brownian motion, modern applications often require modeling systems with jumps or non-linear dynamics. Generalizing the SEP to arbitrary Markov processes, however, introduces a major challenge: we lose the convenient properties of Brownian motion, such as continuous paths and explicit local times.

How do we find a stopping rule when the underlying process is a general Markov chain or a Lévy process? The key lies in shifting our perspective from path geometry to operator theory.

In On Skorokhod embeddings and Poisson equations, Gonon et al. establish that finding a finite-mean Skorokhod embedding for a general Markov process is mathematically equivalent to solving a specific Poisson equation. This equation directly relates the infinitesimal generator of the process (the operator describing its expected local movement) to the difference between the target and initial distributions. If this operator equation is solvable, a valid embedding exists.

For processes that exhibit jumps, John Schoenmakers develops a statistical framework in Statistical Skorohod embedding problem and its generalizations that extends the SEP to arbitrary Lévy processes. By leveraging Mellin and Laplace transforms, the author demonstrates how to infer the distribution of the random stopping time from observations of a time-changed process. To further solidify this mathematical foundation, Gonon et al. prove existence and uniqueness results for irregular Fokker-Planck equations in Existence and Uniqueness Results for Time-Inhomogeneous Time-Change Equations and Fokker–Planck Equations, enabling the construction of martingale diffusions that precisely match target marginal distributions.

The Discrete Frontier: Simple Symmetric Random Walks

Intuitively, one might expect that solving the SEP on a discrete grid—like a simple symmetric random walk—would be simpler than in continuous space. In reality, the discrete nature of the walk introduces severe rigidities. Because a random walk can only move up or down by a fixed step at discrete time intervals, it cannot smoothly adjust to hit a target distribution.

In Classes of measures which can be embedded in the Simple Symmetric Random Walk, Cox et al. reveal that attempting to embed arbitrary probability distributions into a simple random walk gives rise to highly complex, fractal-like mathematical structures. Characterizing exactly which distributions are embeddable remains an open and difficult question.

To bridge this gap and make discrete embeddings practical for decision-making and behavioral finance, He et al. introduce two explicit, algorithmic constructions in Two explicit Skorokhod embeddings for simple symmetric random walk. These methods provide concrete stopping rules that guarantee the random walk terminates with the exact desired target distribution, despite the rigid discrete environment.

Multiple Marginals and Path Constraints

In practical scenarios like financial option pricing, we rarely care about just a single final distribution. Instead, we often need to understand the state of a process at multiple intermediate times (multi-maturity options) or the behavior of its running extremes (lookback and barrier options).

To accommodate multiple time horizons, Peter Spoida introduces an elegant extension of classical embedding techniques in An iterated Azéma-Yor type embedding for finitely many marginals. By iteratively applying an embedding rule across a sequence of distinct target distributions, the author constructs a single martingale that satisfies multiple intermediate marginal constraints while optimizing properties tied to the process's global maximum.

Furthermore, the SEP serves as a tool for finding fundamental probabilistic limits. In On joint distributions of the maximum, minimum and terminal value of a continuous uniformly integrable martingale, Cox et al. derive tight, explicit bounds on the joint exit probabilities of a martingale. The authors prove the optimality of these bounds by constructing novel, extremal solutions to the SEP. These specific embeddings act as proof that the derived bounds are "tight"—meaning there exists a real martingale that achieves those exact worst-case or best-case joint probabilities.

Higher Dimensions and Random Clocks

When moving from a one-dimensional line to higher-dimensional space ($d \geq 2$), the geometry of the Skorokhod Embedding Problem changes completely. In one dimension, a process is easily trapped or stopped by hitting specific points or intervals. In higher dimensions, a process can flow around obstacles, making the design of non-randomized stopping regions highly complex.

In Optimal Brownian Stopping When the Source and Target Are Radially Symmetric Distributions, Ghoussoub et al. study this problem for radially symmetric distributions in dimensions $d \geq 3$. The authors prove that the optimal strategy to minimize or maximize a power-law cost function is determined by a unique geometric barrier. When the Brownian motion hits this barrier, it stops. This insight establishes a deep mathematical duality between higher-dimensional stochastic stopping and subharmonic martingale optimal transport.

Finally, complexity can also be introduced by changing how time itself is measured. Instead of looking for a stopping rule based on the process's history, what if the process is stopped by an external, random clock? In Time-homogeneous diffusions with a given marginal at a random time, Cox et al. invert the problem: given a fixed target distribution and an independent, exponentially distributed random stopping time, they show how to construct a unique time-homogeneous diffusion process that matches the target. By adjusting the "speed measure" of the diffusion rather than the stopping rule, the authors provide a fresh perspective on how stochastic processes can be tailored to fit target distributions.

Key Foundational Work

Classes of measures which can be embedded in the Simple Symmetric Random Walk (2008) — contributed 2 concept elements to this direction.
The Skorokhod embedding problem and its offspring (2004) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_markov_process_embedding_dee57d2a,
  title = {Markov process embedding},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/markov-process-embedding/rdr_dee57d2a21894ec892d30cd695a09db1},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_markov_process_embedding_dee57d2a_v1,
  title = {Markov process embedding},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/markov-process-embedding/rdr_dee57d2a21894ec892d30cd695a09db1/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake