Arbitrage and duality in nondominated discrete-time models

Bruno Bouchard, Marcel Nutz

TL;DR

This paper tackles the challenge of modeling financial markets with "model uncertainty," where the true probability distribution of asset prices is unknown and can only be described by a set of possible models. They establish a novel framework for discrete-time financial markets that allows for this uncertainty, proving that the absence of arbitrage is equivalent to the existence of a family of "martingale measures." This leads to a robust theory for pricing and hedging financial derivatives, even when the market is nondominated.

What Problem Does This Paper Address?

Traditional financial modeling often assumes a single, known probability measure governing asset price dynamics. However, in reality, market participants face significant uncertainty about the true underlying model. This "model uncertainty" means that a set of possible probability measures, rather than a single one, better describes the market. A key challenge arises when this set of measures is "nondominated," meaning there's no single reference measure under which all other measures are absolutely continuous. This nondomination breaks many standard tools in financial mathematics, like the Dominated Convergence Theorem, and makes it difficult to define concepts like arbitrage and hedging in a robust way. This paper aims to develop a rigorous mathematical framework for such nondominated discrete-time financial markets, extending fundamental results like the First and Second Fundamental Theorems of Asset Pricing and the Superhedging Theorem.

Methods

The authors develop a general measure-theoretic framework for discrete-time financial markets. They introduce the concept of "quasi-sure" absence of arbitrage, which accounts for events that are polar (negligible) under at least one model in the set of possible measures. A central result is the equivalence between this quasi-sure no-arbitrage condition and the existence of a suitable family of "martingale measures" that are "P-equivalent" (share the same polar sets). This family of martingale measures acts as a generalization of the unique risk-neutral measure in classical finance.

For superhedging, they establish a duality result: the minimal superhedging price of a contingent claim is the supremum of its expected value over this family of martingale measures. Crucially, they prove the existence of optimal superhedging strategies. The proofs often involve a "local" approach, first solving problems in a one-period setting and then "gluing" the solutions together for the multi-period case. They also leverage tools from the theory of analytic sets and measurable selections to handle the complexities introduced by nondomination.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Nondominated Markets are Tractable: The paper demonstrates that even in the presence of significant model uncertainty (nondominated probability measures), fundamental financial concepts like arbitrage and hedging can be rigorously defined and analyzed.
  • Martingale Measures as a Generalization: The concept of a family of "P-equivalent" martingale measures provides a powerful generalization of the unique risk-neutral measure in classical finance, allowing for robust pricing and hedging under model uncertainty.
  • Duality and Optimal Strategies: The paper establishes a clear duality between the superhedging price and the supremum of expected payoffs over the set of martingale measures. Importantly, it proves that optimal superhedging strategies always exist in this setting, which is a significant result for practical applications.
  • Optional Decomposition in Nondominated Settings: A nondominated version of the Optional Decomposition Theorem is derived, extending a key tool from martingale theory to markets with model uncertainty.

Key Results

  • First Fundamental Theorem: Absence of arbitrage in a quasi-sure sense is equivalent to the existence of a suitable family of P-equivalent martingale measures.
  • Superhedging Theorem: In an arbitrage-free nondominated market, the minimal superhedging price of a contingent claim is the supremum of its expected value over the family of martingale measures, and optimal superhedging strategies exist.
  • Second Fundamental Theorem: Market completeness (all claims are replicable) is equivalent to the uniqueness of the martingale measure in the family.
  • Nondominated Optional Decomposition Theorem: A process that is a supermartingale under all martingale measures can be decomposed into a martingale transform and an increasing process.

Limitations and Future Work

The paper focuses on discrete-time models. While the authors mention that their approach can be extended to continuous-time settings, this is not the primary focus of the current work. The proofs often rely on sophisticated measure-theoretic tools, which might present a barrier to readers less familiar with these areas. Future work could explore the implications of these results for specific financial instruments and markets, and potentially extend the framework to more complex market structures or trading constraints. The paper also hints at connections to nonlinear expectations and robust control, suggesting avenues for further research.

Summary updated

Abstract

We consider a nondominated model of a discrete-time financial market where stocks are traded dynamically, and options are available for static hedging. In a general measure-theoretic setting, we show that absence of arbitrage in a quasi-sure sense is equivalent to the existence of a suitable family of martingale measures. In the arbitrage-free case, we show that optimal superhedging strategies exist for general contingent claims, and that the minimal superhedging price is given by the supremum over the martingale measures. Moreover, we obtain a nondominated version of the Optional Decomposition Theorem.
BibTeX
@article{bouchard2015arbitrage,
  author = {Bruno Bouchard and Marcel Nutz},
  title = {{Arbitrage and duality in nondominated discrete-time models}},
  journal = {The Annals of Applied Probability},
  year = {2015},
  doi = {10.1214/14-aap1011},
}
Figure