Entropic Dynamics of Exchange Rates and Options: A Maximum Entropy Framework

1. Introduction

This paper presents an Entropic Dynamics framework for modeling foreign exchange (FX) rate dynamics and pricing European options. The core objective is to provide an alternative, information-theoretic foundation to traditional stochastic calculus approaches. The authors, Mohammad Abedi and Daniel Bartolomeo from the University at Albany-SUNY, leverage the principles of entropic inference and maximum entropy to handle situations of incomplete information—a common reality in financial markets. The framework systematically incorporates known symmetries, such as scale invariance, leading to the derivation of established models like Geometric Brownian Motion (GBM) and the Garman-Kohlhagen model from first principles.

2. Theoretical Framework

The methodology is built upon three pillars of entropic inference.

2.1. Foundations of Entropic Inference

Entropic inference is an inductive framework designed for reasoning under uncertainty. It extends classical logic to handle partial information. Probability distributions represent the state of knowledge about a system.

2.2. Principle of Minimal Updating

When new information becomes available, the prior probability distribution is updated using the relative entropy (Kullback-Leibler divergence). The update is governed by the Principle of Minimal Updating, which ensures changes are made only as necessitated by the new data, yielding the least biased posterior distribution.

2.3. Information Geometry

The space of probability distributions forms a Riemannian manifold with a unique metric derived from the Fisher information. This information geometry provides a notion of distance between distributions, which is crucial for defining dynamics. The authors note its potential significance for portfolio optimization, to be explored in future work.

3. Entropic Dynamics for FX Rates

Entropic Dynamics applies the inference framework to model how systems change, introducing an entropic time specific to the system.

3.1. Scale Invariance and Variable Selection

A key symmetry in FX markets is scale invariance: the dynamics should be invariant under transformations like $S \rightarrow \lambda S$, where $S$ is the exchange rate. To make this symmetry manifest, the authors identify $x = \log S$ as the natural variable to model, as the transformation becomes a translation $x \rightarrow x + \log \lambda$.

3.2. Derivation of Geometric Brownian Motion

By imposing constraints based on available information about the FX rate (e.g., its expected drift and volatility) and maximizing the relative entropy subject to these constraints, the framework naturally leads to a dynamics for $x$. Translating back to $S$ yields the Geometric Brownian Motion (GBM) equation: $$ dS_t = \mu S_t dt + \sigma S_t dW_t $$ where $\mu$ is the drift, $\sigma$ is the volatility, and $W_t$ is a Wiener process. This derivation shows GBM emerges as the least biased model consistent with the given moment constraints and scale symmetry.

4. Option Pricing Framework

To price derivatives, a risk-neutral valuation framework is essential to avoid arbitrage.

4.1. Risk-Neutral Measure Derivation

Within the entropic framework, changing from the real-world measure $\mathbb{P}$ to a risk-neutral measure $\mathbb{Q}$ is interpreted as an inference problem. It involves updating the prior (real-world dynamics) with the new information that the discounted asset price must be a martingale (no arbitrage). Applying the Principle of Minimal Updating under this constraint leads to the Girsanov theorem transformation, defining $\mathbb{Q}$.

4.2. Garman-Kohlhagen Model

Applying the risk-neutral measure to the GBM dynamics for an FX rate (which involves two interest rates, domestic $r_d$ and foreign $r_f$) and solving the Black-Scholes-Merton PDE for a European option yields the Garman-Kohlhagen formula: $$ C = S_0 e^{-r_f T} \Phi(d_1) - K e^{-r_d T} \Phi(d_2) $$ where $$ d_1 = \frac{\ln(S_0/K) + (r_d - r_f + \sigma^2/2)T}{\sigma\sqrt{T}}, \quad d_2 = d_1 - \sigma\sqrt{T}. $$ This result aligns the entropic dynamics approach with the standard FX options pricing model.

6. Original Analysis: A Critical Perspective

The paper by Abedi and Bartolomeo presents a compelling intellectual exercise in reframing classical financial mathematics through the lens of information theory. Its primary contribution is not a new model, but a novel derivation and justification for existing ones—Geometric Brownian Motion (GBM) and the Garman-Kohlhagen model. This aligns with a broader trend in quantitative finance seeking more fundamental principles, reminiscent of the axiomatic approach in economics or the search for first principles in physics.

Technically, the application of maximum entropy principles to derive dynamics is elegant. The identification of $\log S$ as the correct variable due to scale invariance is a crucial and well-justified step. It echoes the use of log-prices in virtually all stochastic volatility and jump-diffusion models that succeeded GBM. However, the framework's output—the standard GBM—is its greatest limitation. The financial literature since the 1987 crash and the 2008 crisis has overwhelmingly demonstrated the empirical shortcomings of GBM: it fails to capture volatility clustering (as seen in GARCH models), fat-tailed returns, and the volatility smile/skew pervasive in options markets. Models like Heston (1993) or the infinite-activity Lévy processes reviewed by Cont and Tankov (2004) were developed precisely to address these gaps.

Therefore, the paper's significance lies not in its final equations but in its methodological promise. The entropic inference framework is inherently flexible. The constraints used to derive GBM (mean and variance of returns) are simplistic. The true test would be to impose more realistic constraints—such as the observed volatility of volatility or certain moments of the return distribution—and see what dynamics emerge. Could it derive a Heston-type model? This would be a far more impactful contribution. The reference to future work on information geometry for portfolio optimization is particularly tantalizing. The Fisher information metric could provide a rigorous way to measure the stability or sensitivity of a portfolio to parameter estimation errors, a topic of great practical concern often addressed heuristically.

In conclusion, this work is a sophisticated proof of concept. It successfully transplants the entropic dynamics framework from physics to finance and shows it can replicate foundational results. Its value will be determined by whether subsequent research can leverage this framework's machinery to tackle the known deficiencies of those very foundations, moving from elegant justification to genuine innovation.

7. Mathematical Framework & Technical Details

The core mathematical engine is the maximization of relative entropy (Kullback-Leibler divergence) subject to constraints. Given a prior distribution $q(x)$ and new information in the form of expected values $\mathbb{E}_p[f_i(x)] = F_i$ for several functions $f_i$, the posterior $p(x)$ is found by minimizing: $$ D_{KL}[p||q] = \int p(x) \ln \frac{p(x)}{q(x)} dx $$ subject to $\int p(x) f_i(x) dx = F_i$ and normalization $\int p(x) dx = 1$. Using Lagrange multipliers $\lambda_i$, the solution is: $$ p(x) = \frac{1}{Z} q(x) \exp\left(-\sum_i \lambda_i f_i(x)\right) $$ where $Z$ is the partition function. In the context of dynamics, $q(x)$ represents the probability of a transition from an initial state, and the constraints encode the system's expected drift and fluctuation. For the FX application, with $x = \log S$, a constraint on the expected change $\mathbb{E}[\Delta x]$ and its variance $\mathbb{E}[(\Delta x)^2]$ leads to a Gaussian transition probability, which in the continuous limit produces the diffusion equation underlying GBM.

The shift to the risk-neutral measure $\mathbb{Q}$ involves adding a new constraint: the expected return of the discounted asset must equal the risk-free rate. This modifies the Lagrange multipliers, effectively introducing a drift adjustment term $\theta$ such that $dW^{\mathbb{Q}}_t = dW^{\mathbb{P}}_t + \theta dt$, which is the essence of the Girsanov theorem.

8. Analytical Framework & Case Example

Case: Justifying Model Choice for a Currency Pair (EUR/USD)

Scenario: A quantitative analyst at a bank is tasked with developing a model for pricing vanilla EUR/USD options. They must justify their model choice to the model validation committee.

Application of Entropic Framework:

1. State Prior Information: The analyst lists known facts: EUR/USD is positive, its percentage changes are more relevant than absolute changes (scale invariance), and historical data provides estimates for average drift and volatility.
2. Apply Principle of Minimal Updating: Starting from a state of maximum ignorance (a flat prior for $\log S$), the analyst updates beliefs by incorporating the drift and volatility constraints via maximum entropy.
3. Derive Dynamics: The framework outputs GBM as the least biased model consistent with the two moment constraints. The analyst presents this derivation to the committee, arguing that using any model with more parameters (e.g., stochastic volatility) would require corresponding additional, statistically robust information to justify the more complex update.
4. Pricing: To price options, the analyst adds the no-arbitrage constraint, deriving the risk-neutral measure and the Garman-Kohlhagen formula.

Outcome: The committee accepts GBM/Garman-Kohlhagen as the baseline model due to its principled derivation from limited information. They may approve a more complex model (like SABR) for specific tenors/moneyness only if the analyst can demonstrate, perhaps using the same entropic logic, that additional market data (e.g., the volatility smile) provides sufficient information to warrant the more complex update from the GBM prior.

9. Future Applications & Research Directions

The entropic dynamics framework opens several promising avenues beyond replicating classical results:

• Beyond GBM: Incorporating constraints on higher moments (skewness, kurtosis) or the volatility process itself could lead to entropy-based derivations of local/stochastic volatility or jump-diffusion models.
• Information Geometry in Portfolio Construction: As hinted by the authors, the Fisher metric can quantify the "statistical distance" between different market environments. This could be used to: 1) Develop robust portfolio strategies that minimize sensitivity to errors in estimated parameters. 2) Create early warning signals for regime shifts by monitoring the information distance between recent returns and the current model.
• Modeling Illiquid Assets: For assets with sparse data, the maximum entropy approach provides a rigorous method to specify a prior distribution based on economic principles or similar assets, and update it minimally as new trades occur.
• Multi-Asset Dynamics: Extending the framework to multiple correlated assets. The constraints would include correlations, and the resulting dynamics would naturally respect the geometry of the covariance structure, potentially offering insights into systemic risk.
• Integration with Machine Learning: The "prior updating" paradigm aligns with Bayesian machine learning. The framework could guide the design of neural networks that incorporate financial constraints (like no-arbitrage) directly into their architecture or loss functions, improving interpretability and robustness.

10. References

1. Jaynes, E. T. (1957). Information Theory and Statistical Mechanics. Physical Review, 106(4), 620–630.
2. Caticha, A. (2012). Entropic Inference and the Foundations of Physics. In 11th Brazilian Meeting on Bayesian Statistics.
3. Garman, M. B., & Kohlhagen, S. W. (1983). Foreign currency option values. Journal of International Money and Finance, 2(3), 231–237.
4. Black, F., & Scholes, M. (1973). The pricing of options and corporate liabilities. Journal of Political Economy, 81(3), 637–654.
5. Cont, R. (2001). Empirical properties of asset returns: stylized facts and statistical issues. Quantitative Finance, 1(2), 223–236.
6. Heston, S. L. (1993). A closed-form solution for options with stochastic volatility with applications to bond and currency options. The Review of Financial Studies, 6(2), 327–343.
7. Cont, R., & Tankov, P. (2004). Financial Modelling with Jump Processes. Chapman and Hall/CRC.
8. Amari, S. I., & Nagaoka, H. (2000). Methods of Information Geometry. American Mathematical Society.