Entropic Dynamics of Exchange Rates and Options: A Novel Framework for FX Modeling

Table of Contents

1. Introduction

This paper presents an Entropic Dynamics framework for modeling foreign exchange (FX) rates and pricing European options. The core objective is to provide an alternative, information-theoretic foundation to financial dynamics, moving beyond traditional stochastic calculus. The authors, Mohammad Abedi and Daniel Bartolomeo, leverage the principles of entropic inference—a method for reasoning under incomplete information—to derive well-known financial models from first principles.

The work connects the abstract concepts of maximum entropy and information geometry to practical finance, culminating in the derivation of the Geometric Brownian Motion (GBM) for exchange rates and the Garman-Kohlhagen model for FX options. This approach highlights the scale invariance symmetry inherent in currency pairs, leading to the natural choice of modeling the logarithm of the exchange rate.

2. Theoretical Framework

2.1. Entropic Inference and Maximum Entropy

Entropic inference is an inductive framework for situations with incomplete information. Its first tool is probability theory to represent states of belief. The second is the relative entropy (or Kullback-Leibler divergence), used to update beliefs when new information arrives, guided by the Principle of Minimal Updating. Maximizing relative entropy yields the least biased posterior distribution that incorporates all available information.

The third tool is information geometry, which provides a metric on the space of probability distributions. While not deeply explored here, the authors note its potential significance for portfolio management and multi-asset dynamics.

2.2. Entropic Dynamics and Time

Entropic Dynamics applies entropic inference to model how systems change. A key innovation is the introduction of an entropic time parameter, which is emergent and tailored to the specific system rather than being a universal clock. This concept has been successfully applied in various physics contexts and is here adapted to finance.

2.3. Scale Invariance in FX

A fundamental symmetry in FX markets is scale invariance: the dynamics should not depend on whether we quote the exchange rate as USD/EUR or in its reciprocal form. This symmetry dictates that the model should be formulated in terms of the logarithm of the exchange rate, $x = \ln S$, where $S$ is the spot FX rate. Transformations like $S \to \lambda S$ (a simple scaling) leave the dynamics invariant when expressed in terms of $x$.

3. Model Derivation

3.1. From Entropic Principles to GBM

Starting with the prior information about an FX rate—specifically, its initial value and volatility—the authors use the entropic dynamics framework to derive its time evolution. By imposing constraints consistent with market observations (like finite variance) and maximizing entropy, the resulting probability distribution for the future log-exchange rate $x$ is shown to follow a drift-diffusion process.

Transforming back to the spot rate $S = e^x$, this process becomes the familiar Geometric Brownian Motion (GBM): $$ dS_t = \mu S_t dt + \sigma S_t dW_t $$ where $\mu$ is the drift, $\sigma$ is the volatility, and $W_t$ is a Wiener process. The derivation manifestly respects scale invariance.

3.2. Risk-Neutral Measure and Option Pricing

To price derivatives, the no-arbitrage principle is invoked. The authors demonstrate how to derive a risk-neutral measure $\mathbb{Q}$ within the entropic framework. This involves adjusting the drift of the GBM process to the risk-free rate differential between the two currencies, $(r_d - r_f)$.

Under $\mathbb{Q}$, the dynamics become: $$ dS_t = (r_d - r_f) S_t dt + \sigma S_t dW_t^{\mathbb{Q}} $$ Pricing a European call option on the FX rate with this dynamics leads directly to the Garman-Kohlhagen formula, the FX analogue of the Black-Scholes formula.

4. Results and Discussion

4.1. The Garman-Kohlhagen Model

The final output of the entropic derivation is the Garman-Kohlhagen model for the price of a European call option: $$ C = S_0 e^{-r_f T} \Phi(d_1) - K e^{-r_d T} \Phi(d_2) $$ where $$ d_1 = \frac{\ln(S_0/K) + (r_d - r_f + \sigma^2/2)T}{\sigma\sqrt{T}}, \quad d_2 = d_1 - \sigma\sqrt{T} $$ $S_0$ is the spot rate, $K$ is the strike, $T$ is time to maturity, $r_d$ and $r_f$ are domestic and foreign risk-free rates, $\sigma$ is volatility, and $\Phi$ is the standard normal CDF.

4.2. Comparison to Traditional Methods

The paper's primary contribution is methodological. It recovers established models (GBM, Garman-Kohlhagen) not through stochastic calculus and hedging arguments, but through an information-theoretic, first-principles approach based on entropy maximization and symmetry. This provides a deeper, more foundational justification for these models and opens the door to generalizing them by incorporating different or more complex information constraints.

5. Core Insight & Analyst's Perspective

Core Insight: This paper isn't about a new, better pricing formula; it's a philosophical power play. It argues that the entire edifice of continuous-time finance, from Bachelier to Black-Scholes, can be rebuilt from the ground up using information theory and the principle of maximum entropy. The authors are essentially saying, "Forget Ito's lemma for a second; the market's behavior is just the least surprising thing it could do, given what we know." This is a profound shift from modeling prices to modeling knowledge about prices.

Logical Flow: The argument is elegant and parsimonious. 1) We have incomplete information (a prior distribution). 2) We have symmetry (scale invariance). 3) We update our beliefs using the tool that changes them the least (maximum relative entropy). 4) This update, interpreted as dynamics, gives us GBM. 5) No-arbitrage pins down the drift, giving us the risk-neutral measure for pricing. It's a clean, axiom-driven derivation that makes the traditional PDE/hedging argument look almost clunky in comparison.

Strengths & Flaws: The strength is foundational elegance and the potential for generalization. As seen in physics with the work of E.T. Jaynes and later Caticha, entropic methods excel at deriving canonical results from simple principles. The flaw, as with many elegant theories, is the gap to messy reality. The framework elegantly derives GBM, but GBM itself is a flawed model for FX (it underestimates tail risk, ignores volatility clustering). The paper briefly mentions future work on jumps and information geometry, which is where the real test lies. Can this framework naturally incorporate the stylized facts of markets (e.g., fat tails) by simply adding the right constraints, or will it require ad-hoc adjustments that dilute its purity?

Actionable Insights: For quants and model validators, this paper is a mandatory read. It provides a new lens for model risk assessment. Instead of just testing a model's fit, ask: "What information is this model assuming? Is that information set complete or appropriate?" For innovators, the roadmap is clear. The next step is to use this framework to build new models. Constrain the entropy maximization with information about observed volatility smiles or jump frequencies, as hinted by the authors' reference to Bates and Heston models. The prize is a coherent, unified theory of derivative pricing that doesn't stitch together incompatible models. The work of Peters and Gell-Mann (2016) on ergodicity economics shows similar foundational rethinking is gaining traction. This paper is a solid step in that direction, but the market will be the ultimate judge of its utility beyond philosophical appeal.

6. Technical Details

The mathematical core involves maximizing the relative entropy $\mathcal{S}[P|Q]$ of a posterior distribution $P(x'|x)$ relative to a prior $Q(x'|x)$, subject to constraints. A key constraint is the expected squared displacement, which introduces the volatility $\sigma$: $$ \langle (\Delta x)^2 \rangle = \kappa dt $$ where $\kappa$ is related to the volatility $\sigma$. Maximization yields a Gaussian transition probability: $$ P(x'|x) \propto \exp\left(-\frac{(x' - x - \alpha dt)^2}{2\kappa dt}\right) $$ which in the continuum limit leads to the drift-diffusion SDE for $x_t$. The connection to the Black-Scholes-Merton PDE is made through the standard risk-neutral valuation argument applied to the derived GBM process.

7. Analysis Framework Example

Case: Incorporating Volatility Smile Information. The entropic framework allows for the integration of additional market data. Suppose, beyond the spot price and historical volatility, we also have information from the options market implying that the risk-neutral distribution of log-returns is not Gaussian but has negative skewness and excess kurtosis (a volatility smile).

Step 1: Define Constraints. In addition to the variance constraint $\langle (\Delta x)^2 \rangle = \sigma^2 dt$, we add moment constraints from the observed implied volatility surface: $$ \langle (\Delta x)^3 \rangle = \tilde{S} dt, \quad \langle (\Delta x)^4 \rangle - 3\langle (\Delta x)^2 \rangle^2 = \tilde{K} dt $$ where $\tilde{S}$ and $\tilde{K}$ capture skewness and kurtosis per unit time.

Step 2: Maximize Entropy. Maximizing relative entropy with these four constraints (mean, variance, skewness, kurtosis) leads to a transition probability $P(x'|x)$ described by a Gram-Charlier series or a more general exponential family distribution, not a simple Gaussian.

Step 3: Derive Dynamics. The resulting continuous-time limit would be a diffusion process with state-dependent drift and volatility, or potentially a jump-diffusion process, effectively deriving a model like those of Bates or Heston from informational first principles rather than pre-specifying a stochastic volatility process.

This example demonstrates the framework's power to systematically generalize models by explicitly incorporating more granular market information as constraints.

8. Future Applications & Directions

The entropic dynamics framework opens several promising avenues for future research in quantitative finance:

• Multi-Asset Portfolios & Information Geometry: The authors mention applying information geometry to portfolio selection. This could lead to novel asset allocation strategies based on the "distance" between the current market distribution and a target optimal distribution, going beyond mean-variance optimization.
• Modeling Stylized Facts: The framework is naturally suited to incorporate well-known empirical features like fat tails, volatility clustering, and leverage effects by adding appropriate dynamical constraints or making the constraints themselves time-dependent based on past information.
• Non-Stationary and Regime-Switching Markets: The prior distribution $Q$ in the relative entropy can be dynamically updated to reflect changing market regimes, potentially offering a principled way to build adaptive models that respond to structural breaks.
• Behavioral Finance Integration: The "information" constraints could be extended to include metrics of investor sentiment or attention, bridging the gap between traditional quantitative finance and behavioral models.
• Machine Learning Synergy: The principle of maximum entropy is a cornerstone of many machine learning methods. This framework could provide a rigorous information-theoretic foundation for hybrid ML-finance models, explaining why certain neural network architectures or regularization techniques work well for financial time series.

The ultimate goal is a unified, axiom-based theory of market dynamics that is both theoretically sound and empirically accurate, reducing the need for the ad-hoc model patching common in today's financial engineering.

9. 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 Proceedings of the MaxEnt 2012 conference.
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. Peters, O., & Gell-Mann, M. (2016). Evaluating gambles using dynamics. Chaos: An Interdisciplinary Journal of Nonlinear Science, 26(2), 023103. https://doi.org/10.1063/1.4940236
6. Amari, S. I. (2016). Information Geometry and Its Applications. Springer.
7. Bachelier, L. (1900). Théorie de la spéculation. Annales scientifiques de l'École Normale Supérieure, 3(17), 21–86.