The mathematical framework connects utility functions with the long-run growth patterns of wealth, not only with average outcomes.
Economic models often choose actions by maximizing an average—or expected—wealth or utility. In settings where outcomes change unpredictably over time, that average can differ from what an individual experiences in the long run, and maximizing it can reduce long-run wealth or utility.
The authors show that expected-utility maximization and long-run growth optimization coincide more generally when the utility function matches a mathematical transformation that makes wealth dynamics suitable for long-run analysis. The framework includes the known cases of linear utility with additive wealth changes and logarithmic utility with multiplicative changes.
How wealth and utility connect
The paper generalizes the mapping between expected utility theory and growth optimization. It finds that the two decision approaches agree when the utility function is the same as the transformation that converts a wealth process into one whose additive changes have the required long-run properties.
For the class examined in detail, those changes must be stationary and independent—meaning their statistical behavior does not change over time and one increment does not determine another. When the paths are also continuous, the transformed utility follows Brownian motion with drift, a continuous random process with a steady component and random fluctuations.
The analysis also shows how the shape of a utility function corresponds to wealth dynamics. With a concave utility function, wealth fluctuations become larger at higher wealth levels; with a convex function, they become smaller as wealth rises. The authors present this as a possible dynamical explanation for risk-averse and risk-seeking preferences, respectively.
Why wealth dynamics matter
The framework provides a theoretical basis for choosing utility functions from the way wealth actually changes over time, rather than treating preferences as separate from those dynamics. It also suggests a testable hypothesis: the dynamics of wealth may help predict how people value risk.
The result matters because an expected outcome can be a poor guide to an individual’s long-run experience when the underlying process is not ergodic—that is, when an average across many possible outcomes differs from the average outcome over time for one individual.
// Source
Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences · 2026 · DOI: 10.1098/rsta.2025.0166
Authors: Ole Peters, Alexander Adamou
Institutions: Santa Fe Institute, London Mathematical Laboratory