Minimax theorems for American options in incomplete markets without time-consistency

Denis Belomestny, Volker KraetschmerarXiv 2017 · DOI · arXiv

Abstract

In this paper, we give sufficient conditions guaranteeing the validity of the well-known minimax theorem for the lower Snell envelope with respect to a family of absolutely continuous probability measures. Such minimax results play an important role in the characterization of arbitrage-free prices of American contingent claims in incomplete markets. Our conditions do not rely on the notions of stability under pasting or time-consistency and reveal some unexpected connections between the minimax result and the path properties of the corresponding density process.

Introduction

In financial mathematics, the pricing of American options—which can be exercised at any time up to maturity—presents a significant challenge in incomplete markets. Unlike complete markets, where a unique martingale measure exists, incomplete markets offer a range of arbitrage-free prices. The lower bound of this range is characterized by the lower Snell envelope.

Mathematically, determining this value requires solving a minimax problem. We seek to establish the identity: $$\sup_{\tau \in \mathcal{T}} \inf_{Q \in \mathcal{Q}} E_Q[Y_\tau] = \inf_{Q \in \mathcal{Q}} \sup_{\tau \in \mathcal{T}} E_Q[Y_\tau]$$ where $\mathcal{T}$ is the set of stopping times and $\mathcal{Q}$ is a family of probability measures. While this identity is well-established under certain conditions, the standard literature heavily relies on the assumption of time-consistency, also known as "stability under pasting." This paper provides a new framework that bypasses this restrictive requirement, focusing instead on topological properties and path behavior.

The Limitation of Time-Consistency

Time-consistency implies that if two pricing models (measures) are valid, a third model created by "pasting" them together at an arbitrary stopping time must also be included in the set $\mathcal{Q}$. While this algebraic structure simplifies the recursive calculation of Snell envelopes, it is often too restrictive for practical applications. Traders or researchers may wish to consider a specific, "small" family of models—such as those defined by a narrow range of volatility parameters—that do not satisfy the pasting property.

This research demonstrates that time-consistency is a sufficient but not a necessary condition for the minimax identity to hold. By relaxing this requirement, the authors allow for the pricing of American options across a broader and more realistic class of model sets.

A Topological Approach to Minimax Results

Instead of the structural "pasting" property, the authors employ a functional analytic approach. The core innovation lies in analyzing the vector measure associated with the set of probability measures $\mathcal{Q}$.

1. Vector Measure Compactness

The authors define a mapping $\mu_{\mathcal{Q}}$ that maps events to a bounded function on $\mathcal{Q}$. The primary condition for their minimax theorem (Theorem 2.1) is that the range of this vector measure must be relatively compact in the supremum norm. This shift from algebraic structure to topological compactness allows for the inclusion of sets of measures that are fundamentally not time-consistent.

2. Path Properties of Density Processes

To provide a more applicable criterion for stochastic calculus, the authors translate the abstract compactness condition into the language of density processes. They show that if the Radon-Nikodym derivatives ($Z_t^Q = dQ/dP|_t$) exhibit certain path properties—specifically, if they are $P$-almost surely uniformly continuous across the set $\mathcal{Q}$—the minimax identity is guaranteed.

3. Nearly sub-Gaussian Fields

The paper extends these results to "nearly sub-Gaussian" families of local martingales. This provides a concrete framework for models driven by Brownian motion where parameters may vary across a set, yet the pricing identity remains valid due to the entropy conditions of the underlying fields.

Key Theoretical Results

  • Theorem 2.1: Establishes that the minimax identity for the lower Snell envelope holds if the range of the vector measure associated with $\mathcal{Q}$ is relatively compact. This is the foundational result that replaces the need for stability under pasting.
  • Theorem 2.3: Provides a practical criterion: if the density processes $dQ/dP$ are dominated by a $P$-integrable variable and possess $P$-almost surely uniformly continuous paths, the minimax result holds.
  • Proposition 3.1: Demonstrates the application for parameterized families of measures. It shows that if the underlying martingales are nearly sub-Gaussian and satisfy specific entropy conditions, the pricing identity remains valid even in the absence of time-consistency.

Implications and Conclusion

The "aha moment" of this research is the realization that the minimax property in American option pricing is a topological feature rather than a structural one. By focusing on the path properties of density processes, the authors prove that non-consistency is not an obstacle to rigorous pricing. This allows for the use of "smaller," more targeted sets of measures that better reflect specific market views or parameter uncertainties.

Limitations and Future Work

While this work advances the theory of lower Snell envelopes, it primarily focuses on the greatest lower bound of arbitrage-free prices. Additionally, the proofs assume an atomless probability space, a standard but necessary constraint in continuous-time finance. Future research could explore the numerical implementation of these non-consistent models, providing tools for applied practitioners to compute these bounds in complex market scenarios.

Summary updated

Abstract

In this paper we give sufficient conditions guaranteeing the validity of the well-known minimax theorem for the lower Snell envelope with respect to a family of absolutely continuous probability measures. Such minimax results play an important role in the characterisation of arbitrage-free prices of American contingent claims in incomplete markets. Our conditions do not rely on the notions of stability under pasting or time-consistency and reveal some unexpected connection between the minimax result and the path properties of the corresponding density process.
BibTeX
@misc{belomestny2017minimax,
  author = {Denis Belomestny and Volker Kraetschmer},
  title = {{Minimax theorems for American options in incomplete markets without time-consistency}},
  year = {2017},
  eprint = {1708.08904},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1708.08904},
}
Figure