Multi-Dimensional Martingales from Mutual Information
Michael M. Kay · 2025
Read Multi-Dimensional Martingales from Mutual Information by Michael M. Kay (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-16544.txt.
Introduction
Michael M. Kay addresses a classic pricing task in finance. The task is to build a chance model for several assets at once. The model must match the market prices of plain European claims. Such claims pay once at a set date, based on asset values.
For one asset, the Dupire method gives a standard answer. For two or more, no natural unique build was known. Kay offers one based on martingale entropic optimal transport. That method moves chance mass at least cost while keeping fair-game rules. As a side gain, the build gives a hands-on proof of a classic result by Strassen. Kay also shows why a class of local link models can fail to match basket option prices.
Catalog record
| CREATOR | Michael M. Kay |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 106 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-16544
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Multi-Dimensional Martingales from Mutual Information Michael M. Kay Abstract 0 Introduction The aim of the present work is to introduce a concrete method to calibrate a Risk Neutral Markov process to an arbitrary set of European contingent claims 2 2 2 These are trades whose payoffs have a single cashflow at maturity that depends solely on the value of the tradables at maturity. that depend on multiple tradables. The archetypal example, where such a type of problem arises naturally is in the foreign exchange market. In particular, when pricing a trade that depends in some non-trivial manner, say on three currencies, which we shall refer to as d (domestic), f_{1} , f_{2} , a minimal set of vanilla products one needs to calibrate to are European options on the FX rates X_{d\leftarrow f_{1}},X_{d\leftarrow f_{2}},X_{f_{1}\leftarrow f_{2}} . Assuming no arbitrage, the latter is the ratio of the first two, hence its corresponding options can be recast as particular basket options on the first two. A popular class of models in this context is the one studied in [ 1 ] , however such a class fails in general (see Appendix A) to calibrate to arbitrage free prices. Other popular approaches for such a type of problems involve ideas from martingale Optimal Transport (see the comprehensive review [ 2 ] , and e.g., [ 3 ] , [ 4 ] , [ 5 ] , [ 6 ] ). Such an approach is more general and has a more general scope than what we propose, in particular the products it allows to calibrate to are in principle arbitrary. However a specialization thereof that could potentially be viewed as an extension of the work by Dupire ( [ 7 ] ) in the case of a single tradable, seems to be lacking. When dealing with a single tradable, the work of Dupire, aside from regularity considerations, allows one to construct a unique continuous local martingale given, say, the collection of all European call option prices C(t,K) for all maturities t and strikes K . In the setting of zero interest rates, in particular, the local volatility of such a process is obtained as \displaystyle\sigma^{2}_{t}(K)=2\frac{\partial_{t}C(t,K)}{\partial_{K}^{2}C(t,K)}. (0.1) In practice, one cleans (for no-arbitrage) and interpolates/extrapolates the function C(t,K) from a finite set of maturities and strikes. The above is nothing else than the result of solving for \sigma_{t}^{2}(x) in the forward Kolmogorov (Fokker-Planck) equation \displaystyle\partial_{t}\rho_{t}(x)=\frac{1}{2}\partial_{x}^{2}\left(\sigma^{2}_{t}(x)\rho_{t}(x)\right) (0.2) given the marginal probability densities \rho_{t} over \mathbb{R} . The appeal of Dupire’s result is that it offers a quick analytic solution (which however becomes more involved in the case of stochastic rates) and that when \sigma^{2}_{t} exists, it is unique. The existence of \sigma_{t}^{2} is essentially guaranteed in the absence of calendar arbitrage. More precisely, in the case of discrete time, when \rho_{s}\preceq\rho_{t} for s<t where \preceq denotes convex order , a classic result by Strassen (see [ 8 ] ) guarantees the existence of a Markov martingale with those marginals. Such a result is true not only in dimension one (over \mathbb{R} ), but for Borel measures over a separable Banach space. In the case of continuous time the analogous result is due to Kellerer (see [ 9 ] ) but only in dimension one, while a generalization to \mathbb{R}^{n} , requiring smoothing, was obtained very recently in [ 10 ] . Both in discrete and continuous time, however, the results are not constructive enough to make them useful in practice. Moreover, such Markov processes are no longer unique in \mathbb{R}^{n} for n>1 . The method we propose stems from a constructive proof of Strassen’s theorem for Radon probability measures over \mathbb{R}^{n} , whereby one obtains the unique Markov martingale that minimizes the mutual information of the joint distribution of the process 3 3 3 By mutual information we mean the Kullback-Leibler divergence of the joint distribution relative to the product of its marginals. . One way of interpreting such a martingale, is as the most unbiased martingale given marginals free of calendar arbitrage. We will refer to such a martingale as a Hessian martingale . Subsequently, we construct the unique Markov (Hessian) martingale that minimizes mutual information, where the constraints are not the knowledge of the full marginals, but rather the prices of a finite number of European Contingent claims. This latter result can be used as an effective calibration method in practice. The method we propose has the following properties in common with Dupire’s: • Up to numerical error, it succeeds if and only if there is no market arbitrage and can be also employed to detect/“clean away” arbitrage. • It is “model independent” in the sense that the form of the resulting Markov process is completely and solely determined by the form of the payoffs. In contrast to Dupire’s approach: • The form of the volatility matrix (or covariance matrix) is not determined from a simple algebraic formula, but requires instead the solution of a convex optimization problem. • The proposed approach only needs an arbitrary finite number of European contingent claims. In particular it does not require ad-hoc interpolation/extrapolation of prices or implied volatility (hyper-)surfaces. It is essential to mention that the idea of looking for a (Markov) process that minimizes the relative entropy between the joint distribution of the process and a given, reference one ( in our case the product of its marginals ) is nothing new, and in the physics literature goes back at least to Schrödinger (see [ 11 ] ), while in the mathematical finance literature such a line of research was pioneered in [ 12 , 13 , 14 ] . However, one added value we claim to provide, relative to that line of research, lies in unravelling the full structure of the resulting process when specializing to European contingent claims. Moreover, contrary to such approaches, here the martingale property is imposed a priori. The present note is subdivided into 7 sections plus 4 appendices and it has 3 main results called Theorem 1,2,3, the first of which is a version of the classic result by Strassen specialized to \mathbb{R}^{n} , for which, however, we offer a constructive proof 4 4 4 We note that for n=1 a novel proof in the same vein was also obtained recently in [ 15 ] . : 1. Revisitation of Strassen’s theorem: Theorem 1. 2. Intuitive explanation of the construction of the unique “most unbiased” Markov martingale incarnating Strassen’s result. 3. Proof of Theorem 1. 4. Intuitive understanding of the type of martingale Markov process constructed given the marginals. 5. Sketch of what the process looks like in continuous time. It shows that under mild assumptions, the process is a continuous local martingale, where the covariance matrix is the inverse of the Hessian of a time-dependent convex function on the space of tradables. 6. Statement and Proofs of Theorem 2,3: these generalize Theorem 1, by not requiring knowledge of the full marginals, but only of the prices of some European contingent claims. Theorem 2: shows exactly under what conditions the calibration procedure can be sequential. This condition incidentally is generically only satisfied in dimension 1. Theorem 3 shows how to construct the Hessian martingale to prices of an arbitrary finite