Sep 17, 2026

Is Expected Utility Rational? Ecological Rationality and Ergodicity Economics

---

Emilie Rosenlund Soysal

Research Fellow, London Mathematical Laboratory.

Expected utility theory (EUT) is the standard benchmark for rational decision-making under uncertainty, but experimental evidence shows that people systematically violate its predictions. The violations are often interpreted as evidence that people are irrational, but it could instead be that the benchmark model for rational decisions is wrong. Both ergodicity economics and ecological rationality offer alternative benchmarks by including information about the environment in which a decision is made.

What is the standard economic definition of rationality?

Under EUT, a rational agent chooses the option that maximizes expected utility,

where $p_i$ is the probability of outcome $x_i$ and $u(x_i)$ is the utility derived from it.

The idea of transforming possible outcomes into their corresponding utility values before taking the expectation was first introduced by Daniel Bernoulli in 1738. Much later, in 1944, von Neumann and Morgenstern formalized the idea by deriving four decision axioms that specify a minimal set of conditions (completeness, transitivity, continuity and independence) under which an individual’s preferences can be represented by expected utility. Each of the conditions appears intuitively reasonable, suggesting that the preferences of a somewhat sane decision maker should satisfy them. This intuitive appeal is likely the reason why EUT has become the dominant normative model of rational choice under uncertainty.

Do people actually maximize expected utility?

However, experiments have repeatedly found systematic departures from EUT predictions, and it is now commonly accepted that EUT is a poor descriptive theory. For example, Risky Curves reviews decades of evidence and reports that expected-utility functions have not produced useful out-of-sample predictions.

There are two possible interpretations of the lack of descriptive power of EUT:

1. People are irrational: EUT is the correct normative benchmark, but human behavior systematically deviates from it.
2. The benchmark for rationality is misleading: some apparent violations of EUT may be appropriate responses to features of the environment that EUT does not represent.

Within the field of behavioural economics, the first proposition has received the most attention, with Daniel Kahneman and Amos Tversky among the most prominent scholars associated with the view (e.g. see their work on prospect theory). But evidence that people violate a model of rationality just establishes that the model is descriptively incomplete. It does not, by itself, establish that violators of the model are irrational. Perhaps we need a different normative model of rationality.

What is ecological rationality?

Ecological rationality, a framework developed by Gerd Gigerenzer and Peter Todd, evaluates a decision rule by how well it performs in the environment in which it operates. That is, the rationality of preferences should not be judged by their adherence to the set of decision axioms, but by whether they lead to good outcomes in the given context.

In many environments, expected utility maximisation isn’t appropriate. For expected utility maximisation to be an operational decision-making strategy, decision makers need not only preferences that satisfy the four decision axioms (and thereby well-defined utility functions), but also knowledge of possible outcomes and probabilities, and the capacity to actually do the maximisation. The full outcome space and corresponding probabilities are rarely known for real life problems, and even if they were, optimisation over all possible decisions may be computationally infeasible. Hence, simple heuristics can outperform the optimisation strategies in environments whose statistical structure favors those heuristics. Choosing the right heuristic requires information about the context in which the decision is made. Ecological rationality therefore treats context as part of the rationality problem, rather than something that can be abstracted away.

Ergodicity economics assumes rational choices are context-dependent

Like ecological rationality, ergodicity economics sees context as essential to the decision making. In particular, it identifies the temporal dynamics of an environment as one specific feature that can determine which decision rule performs well over time.

Ergodicity has to do with dynamics: A stochastic process can be evaluated in two different ways - by averaging across possible realisations (the expected value, also called the ensemble average) or by examining what happens to one realisation as it evolves over time (the time average).

For an ergodic observable, the expected value and the time average coincide in the relevant time and ensemble limits. For a non-ergodic observable, they need not. The distinction between the two averages matters because individuals experience outcomes through time, not across an ensemble of lives. A careful treatment of the difference can be found in The ergodicity problem in economics.

Why are dynamics important for decisions?

A decision is rational in the ecological sense when it helps the decision maker achieve their objective given the environment. A core insight from ergodicity economics is that if the objective of the decision maker is to maximise utility, then maximising the expected utility may do the exact opposite over time!

This problem is illustrated by Peters’ coin toss: Suppose a decision maker, who derives utility from wealth is offered the following gamble using a fair coin: After heads, wealth increases by 50 percent, but after tails, it decreases by 40 percent. This means wealth is multiplied by 1.5 in case of a gain and by 0.6 in case of a loss.

The expected wealth multiplier is the arithmetic average of outcomes:

so expected wealth increases by 5 percent per round. But the long-run growth factor is the geometric average:

so an individual wealth trajectory declines by roughly 5 percent per round in the long run. The gamble therefore has positive expected wealth growth but negative long-run wealth growth for an individual repeatedly taking it.

A risk-neutral expected-utility maximiser would accept the gamble and then experience utility decline over time almost surely. Maximising expected utility therefore does not, in general, maximise utility over time.

The mismatch between expected utility maximisation and experienced utility over time is developed in the detailed analysis of why expected-utility maximisers do not maximise utility and in The time interpretation of expected utility theory.

The example shows that the rationality of a decision cannot be judged without information about the dynamics of the environment in which the decision is made. Whether a decision is good or bad depends on context in the sense that the decision rule must be suited to the dynamics.

Last updated on Sep 17, 2026