Expected utility hypothesis

Expected utility theory explains risky choice by ranking lotteries according to the probability-weighted utility of their outcomes. If a gamble yields outcomes x₁, x₂, … with probabilities p₁, p₂, …, its expected utility is calculated as EU = Σpᵢu(xᵢ), and a person is predicted to select the available gamble with the greatest value. Daniel Bernoulli introduced an early version in 1738 to explain why people may value a fair monetary gamble at less than its mathematical expectation. The modern formulation came from von Neumann–Morgenstern utility, who showed that preferences over lotteries can be represented by expected utility when certain rationality conditions hold.¹

The theory distinguishes utility from money: the same financial gain can have different significance for different people. Utility is defined only up to positive affine transformation, so changing its scale or origin does not alter the predicted choices.

The hypothesis rests on preference axioms that make probabilistic choices coherent. The principal conditions are completeness, which requires that alternatives can be compared; transitivity, which prevents circular rankings; continuity, which rules out abrupt preference jumps; and independence, which says that mixing two lotteries with the same third lottery should preserve their ranking. Under these conditions, a utility function represents preferences over risky prospects, and its expected value represents the ranking of those prospects.¹

The independence axiom is especially distinctive: a person who prefers lottery A to lottery B should also prefer the same mixture of A and a third lottery to the corresponding mixture of B and that lottery. Concavity of the utility function captures risk aversion; a risk-averse person prefers a certain amount to a fair gamble with the same expected monetary value. The curvature also determines willingness to pay for insurance.

Expected utility provides a common language for insurance, finance, public policy, medical decisions, and Bayesian decision theory. It separates beliefs about probabilities from values assigned to outcomes, allowing analysts to study how information, wealth, and risk affect decisions. In economics, it also supports models of portfolio choice, consumption under uncertainty, and strategic behavior in games.

Actual choices often depart from the model, particularly when probabilities or outcomes are framed differently. The Allais paradox demonstrated that many people violate independence when choosing between a sure prize and a high-probability prize.³ The 1979 account of Prospect theory described additional regularities, including reference dependence, loss aversion, and overweighting of small probabilities.⁴ These findings challenge the hypothesis as a universal psychological description, though expected utility remains a powerful normative and analytical benchmark.²

The hypothesis does not require people to possess a single stable utility function for all circumstances; it concerns whether preferences over specified lotteries can be represented in that form. Its predictions can therefore be tested through revealed preference, rather than by directly measuring subjective happiness. The theory also allows subjective probabilities, as in Bayesian decision theory, rather than requiring objectively known frequencies.

A less visible implication is that the model can describe attitudes toward both gains and losses, provided the relevant outcomes and probabilities are specified. It does not by itself determine which outcomes matter, how probabilities are formed, or whether a decision-maker is morally justified in maximizing personal utility. Those questions belong to welfare economics, probability theory, and ethics. Modern decision theory consequently uses expected utility in several roles: as a benchmark for rational coherence, as an approximation in economic models, and as a standard against which behavioral deviations can be measured.¹

Expected utility

The probability-weighted average of the utilities assigned to a gamble’s possible outcomes.

Lottery

A prospect consisting of specified outcomes and associated probabilities.

Independence axiom

The condition that preference between two lotteries is preserved when both are mixed with the same third lottery.

Risk aversion

A preference for a certain outcome over a fair gamble with the same expected monetary value.

Utility function

A numerical representation of preferences, whose ranking is behaviorally meaningful even when its scale is not.