Faking Brownian motion with continuous Markov martingales

Mathias Beiglböck, George Lowther, Gudmund Pammer, Walter Schachermayer

TL;DR

This paper constructs a "very fake" Brownian motion—a stochastic process that has the exact same marginal distributions and continuous paths as Brownian motion but is fundamentally a different process. The authors prove that while their construction is Markovian, it fails the strong Markov property, highlighting a subtle but critical boundary in the uniqueness of stochastic processes.

What Problem Does This Paper Address?

In mathematical finance and probability theory, a common task is the "fitting problem": constructing a martingale (a "fair game" process) that matches specific marginal constraints—essentially, it must look like a specific distribution (like a Bell curve) at any given time $t$.

Brownian motion is the most famous example of such a process. For a long time, researchers have tried to create "fake" Brownian motions—processes that look like Brownian motion at any single point in time but behave differently over time. Previous attempts resulted in processes that either had jumps (non-continuous) or weren't Markovian (the future depended on the entire past, not just the present).

Earlier theorems by Gyöngy, Dupire, and Lowther established that Brownian motion is the unique continuous strong Markov martingale with these marginals. This left one final gap: Could a process be continuous and Markovian (but not strongly Markovian) and still "fake" Brownian motion? This paper proves the answer is yes.

Methods

The authors use two distinct approaches to build this "very fake" Brownian motion:

  1. The "Bare Hands" Construction: This is the core of the paper. They divide the world into two types of particles: "lazy" and "busy."
    • Lazy Territory: A set $C$ (specifically a "fat Cantor set") that has a positive area (Lebesgue measure) but contains no intervals (empty interior). Particles here stay still for a random amount of time.
  2. Busy Territory: The complement $G$. Here, particles move according to a time-changed Brownian motion.
  3. The Transition: Using the heat equation, the authors calculate exactly when a "lazy" particle should become "busy" to ensure the overall distribution always matches a Gaussian Bell curve. Because the set $C$ has an empty interior, the jumps required to move from lazy to busy states become "invisible" in the limit, resulting in a process with continuous paths.
  4. The Abstract Construction: They utilize a known existence theorem (Lowther’s Theorem 6.1) to show that one can split the starting distribution of a process, evolve two different strong Markov martingales, and then randomly choose between them to create a single Markovian (but not strongly Markovian) process.

Datasets

Not discussed in the paper. This is a theoretical mathematics paper; it relies on formal proofs and stochastic analysis rather than empirical data.

Key Insights and Ideas

  • The "Strong" Markov Distinction: Most students learn the Markov property (the future depends only on the current state). The strong Markov property extends this to "stopping times" (random times like "the first time the stock hits $100"). This paper shows that this distinction is not just a technicality—it is the specific property that makes Brownian motion unique.
  • Pathological Sets as a Bridge: By using a fat Cantor set, the authors exploit a mathematical loophole. The set is "large" enough to hold particles still (preserving the Markov property) but "thin" enough that moving out of it doesn't require a visible jump (preserving path continuity).
  • Mimicking via the Heat Equation: The construction relies on the fact that the flow of probability in Brownian motion is governed by the spatial derivative of the heat kernel. By matching their transition rates to this derivative, they ensure the "fake" process is indistinguishable from the real one if you only look at a snapshot in time.

Key Results

  • Theorem 1.1: The authors successfully prove the existence of a 1-dimensional Markovian martingale $X$ that has continuous paths and Brownian marginals but is not strongly Markovian.
  • Continuity vs. Jumps: They demonstrate that while previous fake Brownian motions required jumps to maintain the Markov property, their use of empty-interior sets allows for a continuous version.
  • Failure of Strong Markovianity: Using a "coupling" argument, they show that if you know a particle is at a certain point at a stopping time, you still don't know if it was previously "lazy" or "busy." This "hidden memory" is what breaks the strong Markov property.

Limitations and Future Work

  • Complexity of the Set: The construction relies on highly specific, "pathological" sets like fat Cantor sets. While theoretically brilliant, these are difficult to work with in practical financial engineering or simulation.
  • Dimensionality: The proof is centered on 1-dimensional processes. Extending this "very fake" behavior to multi-dimensional manifolds or more complex volatility surfaces remains an open challenge.
  • Financial Application: While the paper mentions the "fitting problem" in finance, it does not provide a framework for how these non-strongly Markovian processes might be used to price derivatives or manage risk in a way that Brownian motion cannot.

Summary updated

Abstract

Hamza-Klebaner posed the problem of constructing martingales with Brownian marginals that differ from Brownian motion, so called fake Brownian motions. Besides its theoretical appeal, the problem represents the quintessential version of the ubiquitous fitting problem in mathematical finance where the task is to construct martingales that satisfy marginal constraints imposed by market data. Non-continuous solutions to this challenge were given by Madan-Yor, Hamza-Klebaner, Hobson, and Fan-Hamza-Klebaner whereas continuous (but non-Markovian) fake Brownian motions were constructed by Oleszkiewicz, Albin, Baker-Donati-Yor, Hobson, Jourdain-Zhou. In contrast it is known from Gy\"ongy, Dupire, and ultimately Lowther that Brownian motion is the unique continuous strong Markov martingale with Brownian marginals. We took this as a challenge to construct examples of a "very fake'' Brownian motion, that is, continuous Markov martingales with Brownian marginals that miss out only on the strong Markov property.
BibTeX
@inproceedings{beiglbock2023faking,
  author = {Mathias Beiglböck and George Lowther and Gudmund Pammer and Walter Schachermayer},
  title = {{Faking Brownian motion with continuous Markov martingales}},
  booktitle = {Finance},
  year = {2023},
  doi = {10.1007/s00780-023-00526-w},
}
Figure