Introduction
For over a century, economics has relied on specific mathematical functions to aggregate heterogeneous data—such as labor, capital, or prices—into single representative indices. The Cobb-Douglas and Constant Elasticity of Substitution (CES) functions are the primary workhorses of macroeconomic modeling. However, these functions were largely developed inductively: economists observed empirical patterns and selected mathematical forms that fit the data.
This research bridges the gap between economics and information theory by demonstrating that these classic aggregators are not merely empirical "best fits." Instead, they are mathematically optimal Low-distortion Aggregators (LDAs) that minimize information loss. By viewing economic models through "information lenses," the authors provide a rigorous theoretical foundation for why these specific models have dominated economic thought for decades.
The Information Lens: Bregman Divergences and LDAs
The core methodology of this study is rooted in Bregman Divergences, a broad class of distortion measures that generalize concepts like squared Euclidean distance and Kullback-Leibler (KL) divergence.
The authors define a Low-distortion Aggregator (LDA) as a summary statistic (an average) that minimizes the total Bregman distortion relative to the original data points. In this framework, an LDA is the "most informative" summary possible for a given type of data. By applying the properties of Legendre duality—a concept from convex analysis—the authors prove that the set of mathematically optimal information aggregators collapses specifically into the well-known economic functions used today.
Key Insights: The Convergence of Two Worlds
The paper reveals a striking convergence: when an aggregator is required to be optimal from an information-theory standpoint while satisfying basic economic axioms (such as unit elasticity), the mathematical result is inevitably the Cobb-Douglas form.
Furthermore, the research highlights that Legendre Duality in mathematics perfectly mirrors economic dualities. For instance, the relationship between a function and its convex conjugate is shown to be identical to the relationship between a consumption index and its corresponding price index.
Interestingly, the authors also examined older, 19th-century models (such as the Mitscherlich-Spillman-von Thünen aggregators) and found they are not LDAs. This suggests that as economic modeling evolved during the 20th century, it unconsciously gravitated toward functions that are more information-theoretically sound.
Theoretical Results and Optimality
The paper presents several key theorems that establish the unique status of these aggregators:
- Theorem 5 (Completeness): If two dual aggregators (e.g., prices and consumption) satisfy the standard identity $\sum x_i z_i = x_* z_*$ and at least one is an LDA, then both must be CES functions.
- Theorem 7: If an LDA has a substitution elasticity of exactly one, it is guaranteed to be a Cobb-Douglas function.
- Theorem 8: The only LDAs that are "homogeneous of degree 1" (representing constant returns to scale) are the Cobb-Douglas and CES families.
These results demonstrate that popular aggregators aren't just members of the "optimal set" of information summaries—under common economic constraints, they are often the only members.
Conclusion and Future Directions
While the current analysis assumes "separable" generators—meaning the influence of one variable does not depend on others in complex, non-additive ways—it successfully covers the vast majority of standard economic models. Future work may extend this "information lens" to more complex areas like international trade and growth theory to determine if other specialized aggregators also possess hidden information-theoretic optimality. By grounding economic theory in the principles of information preservation, this research provides a new, deductive justification for the mathematical foundations of the field.
Summary updated