Information geometries and Microeconomic Theories

Richard Nock, Brice Magdalou, Nicolas Sanz, Eric Briys, Fred Celimene, Frank NielsenarXiv 2009 · arXiv

Introduction

In neoclassical economics, the Economic Production Function (EPF) serves as a fundamental tool for aggregating diverse inputs—such as labor and capital—into a single measure of output. Despite their widespread use, classic models like the Cobb-Douglas or Constant Elasticity of Substitution (CES) functions are frequently criticized for being ad hoc. They are often selected for their empirical fit or computational convenience rather than being derived from first principles.

This research provides a rigorous mathematical foundation for these theories by linking microeconomics to Information Geometry. The authors demonstrate that standard EPFs are not merely convenient heuristics but are the unique, optimal solutions for aggregating inputs when minimizing "transition costs" defined by Bregman divergences.

Theoretical Framework: Information Geometry and Aggregation

The paper utilizes the framework of Information Geometry to re-examine how economic inputs are combined. Two primary mathematical tools are central to this approach:

  1. Bregman Divergences: These are a class of functions that measure the "distortion" or distance between two points based on a strictly convex generator function. Unlike standard Euclidean distance, Bregman divergences can capture the specific curvatures inherent in different economic structures.
  2. Lowest Distortion Aggregators (LDAs): These are generalized means designed to minimize the total Bregman divergence from a set of input points.

By framing economic transitions—the shift from one set of inputs to another—as a movement through a geometric space of information, the authors provide a "weak axiomatization" that justifies the existence and specific forms of classical economic models.

Axiomatizing Economic Transition Costs

To establish this foundation, the authors propose three structural axioms for "global transition costs":

  • Non-negativity: The cost of transitioning between input states cannot be negative.
  • Identity of Indiscernibles: The cost is zero if, and only if, the inputs remain unchanged.
  • Optimal Averaging: The average input value must be the value that minimizes the global transition cost.

Through these axioms, the paper proves that any cost function satisfying these criteria must be a Bregman divergence, and the resulting optimal aggregation method must be an LDA. This mathematical necessity removes the ad hoc nature of production functions, placing them within a strict geometric requirement.

Key Insights: Economics as Geometry

The research yields several profound insights into the geometric nature of economic behavior:

Mainstream EPFs are LDAs

The authors prove that the most prominent functions in economic literature—including CES, Cobb-Douglas (under constant returns to scale), and Generalized Exponential Means (GEM)—are all specific instances of Lowest Distortion Aggregators. This suggests that the historical preference for these functions stems from an implicit (though previously unstated) requirement for geometric optimality.

Economic Duality and Legendre Duality

A cornerstone of microeconomics is the distinction between Marshallian demand (maximizing utility subject to a budget) and Hicksian demand (minimizing cost subject to a utility target). The paper reveals that this economic duality is a direct manifestation of Legendre duality within information geometry.

The Slutsky Identity as a Geometric Property

The "Slutsky identity," which decomposes price changes into substitution and income effects, is shown to map directly to the Bregman triangle equality. This finding implies that fundamental economic behaviors are not merely behavioral observations but are geometric properties of the transition path between different input states.

Formal Results

The paper consolidates these findings into three core theorems:

  • Theorem 1: Formally identifies the specific generator functions that produce the CES, Cobb-Douglas, and GEM models. Notably, it shows that other models, such as the "Translog" function, do not fit the LDA framework and thus lack the same geometric optimality.
  • Theorem 2 (Exhaustivity): Demonstrates that when standard economic properties (like homogeneity of degree 1) are assumed, the vast set of possible LDAs collapses into the specific mainstream EPFs used today.
  • Theorem 3: Establishes that Bregman divergences are the unique class of functions capable of satisfying the proposed axioms for economic transition costs.

Conclusion and Future Directions

This research suggests that information theory is the "Ariadne’s Thread" connecting static economic descriptions (production functions) with dynamic ones (transformation paths). While the current work focuses primarily on 1D scalar cases and separable transition costs (where the total cost is the sum of individual input contributions), it opens the door for significant future research. Exploring non-separable costs could provide a richer framework for studying complex interactions between inputs, while "minmax" cost minimizers might lead to the discovery of new aggregators that focus on a subset of critical inputs rather than a global average.

Summary updated

Abstract

More than thirty years ago, Charnes, Cooper and Schinnar (1976) established an enlightening contact between economic production functions (EPFs) -- a cornerstone of neoclassical economics -- and information theory, showing how a generalization of the Cobb-Douglas production function encodes homogeneous functions. As expected by Charnes \textit{et al.}, the contact turns out to be much broader: we show how information geometry as pioneered by Amari and others underpins static and dynamic descriptions of microeconomic cornerstones. We show that the most popular EPFs are fundamentally grounded in a very weak axiomatization of economic transition costs between inputs. The strength of this characterization is surprising, as it geometrically bonds altogether a wealth of collateral economic notions -- advocating for applications in various economic fields --: among all, it characterizes (i) Marshallian and Hicksian demands and their geometric duality, (ii) Slutsky-type properties for the transformation paths, (iii) Roy-type properties for their elementary variations.
BibTeX
@misc{nock2009information,
  author = {Richard Nock and Brice Magdalou and Nicolas Sanz and Eric Briys and Fred Celimene and Frank Nielsen},
  title = {{Information geometries and Microeconomic Theories}},
  year = {2009},
  eprint = {0901.2586},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/0901.2586},
}
Figure