Loading field map
How should a person choose when the consequences of their actions depend on events they cannot predict with certainty? This question sits at the heart of microeconomics, and the frameworks built to answer it have shaped fields from finance to public policy. The central tension running through the history of choice under uncertainty is between normative ideals—what a perfectly rational agent would do—and descriptive accuracy—what real people actually do. Each major framework has had to navigate this tension, and the sequence of frameworks tells a story of axioms challenged, paradoxes uncovered, and psychological realism gradually absorbed into formal theory.
Expected Utility Theory (EUT), formalized in the mid-twentieth century, provided the first rigorous axiomatic treatment of choice under risk—situations where probabilities are known. Its core idea is that a decision-maker assigns a numerical utility to each possible outcome and then chooses the option that maximizes the expected value of that utility. The theory's normative appeal rests on a set of axioms, most notably the independence axiom: if a person prefers lottery A to lottery B, they should also prefer a compound lottery that mixes A with any third lottery C over the same mixture with B. This axiom gives EUT its clean mathematical structure and makes it the benchmark for rational choice under risk. For decades, EUT served as the dominant framework for modeling behavior in economics and finance, and it remains the standard normative model taught in textbooks today.
EUT assumed that probabilities are objectively known, but many real-world decisions involve uncertainty where probabilities are not given. Subjective Expected Utility Theory (SEU), developed by Leonard Savage in the 1950s, extended the expected utility logic to cover these situations. SEU's central innovation was to treat probabilities as subjective beliefs that can be inferred from a person's choices, provided those choices satisfy a set of consistency axioms. The key axiom in SEU is the sure-thing principle: if two acts yield the same consequence in some state of the world, the decision-maker's preference between them should depend only on what happens in the other states. This principle is a direct analogue of EUT's independence axiom, adapted to the setting of unknown probabilities. SEU thus absorbed EUT as a special case (where subjective probabilities coincide with objective ones) and provided a unified framework for all decisions under uncertainty. Together, EUT and SEU defined the normative ideal of rational choice for nearly three decades.
By the 1950s and 1960s, experimental evidence began to accumulate that real people systematically violate the axioms of EUT and SEU. The Allais paradox showed that people's choices between risky lotteries often violate the independence axiom: they prefer a certain gain over a gamble with a slightly higher expected value, but then reverse that preference when both options are shifted into a domain where the certain outcome is no longer available. The Ellsberg paradox targeted the sure-thing principle of SEU: people prefer to bet on an urn with known proportions of colored balls over an urn with unknown proportions, even when the known urn offers no objective advantage. These paradoxes were not mere curiosities; they revealed deep patterns in human decision-making—ambiguity aversion, a nonlinear response to probabilities, and a sensitivity to the framing of outcomes as gains or losses—that the normative frameworks could not accommodate. The pressure to explain these patterns gave rise to a new generation of descriptive models.
Prospect Theory, introduced in 1979, was the first major framework to break decisively with the expected utility tradition. Instead of assuming that people evaluate final wealth levels, Prospect Theory proposed that people evaluate outcomes relative to a reference point (typically the status quo), with gains and losses treated differently. The value function is concave for gains (risk aversion) but convex for losses (risk seeking), and it is steeper for losses than for gains—a feature called loss aversion. Equally important, Prospect Theory replaced the linear probability weighting of EUT with a nonlinear weighting function that overweights small probabilities and underweights moderate and large ones. This weighting function explained why people buy both lottery tickets (overweighting the small chance of a big gain) and insurance (overweighting the small chance of a big loss). Prospect Theory was explicitly descriptive: it aimed to capture how people actually choose, not how they should choose. It coexisted with EUT and SEU by occupying a different role—the normative frameworks remained the standard for welfare analysis and optimal policy, while Prospect Theory became the leading model for predicting behavior.
Prospect Theory's original formulation had a technical flaw: its probability weighting could violate stochastic dominance, meaning a person could prefer a lottery that is clearly worse than another in every possible outcome. Rank-Dependent Expected Utility Theory (RDEU), developed in the 1980s, addressed this problem by introducing a cumulative weighting scheme. Instead of weighting each probability individually, RDEU weights the entire cumulative distribution function, transforming the ranking of outcomes rather than the probabilities themselves. This approach ensures that stochastic dominance is preserved while still allowing the nonlinear probability sensitivity that Prospect Theory had shown to be empirically important. RDEU narrowed the gap between descriptive realism and formal consistency: it kept the psychological insight of probability weighting but embedded it in a structure that satisfied the same dominance principle that EUT had respected. RDEU thus transformed Prospect Theory's insight into a more rigorous framework, though it remained a single-domain model that did not incorporate reference dependence or loss aversion.
Cumulative Prospect Theory (CPT), introduced in 1992, combined the psychological richness of the original Prospect Theory with the formal rigor of RDEU. CPT applies RDEU's cumulative weighting method separately to gains and losses, each with its own weighting function. This dual application preserves stochastic dominance while allowing the reference-dependent value function and loss aversion that were central to the original Prospect Theory. The result is a framework that can accommodate the Allais paradox, the Ellsberg paradox, and a wide range of other empirical regularities within a single, axiomatically grounded model. CPT has become the leading descriptive framework for choice under uncertainty, used extensively in behavioral finance, insurance economics, and public policy analysis. It did not replace EUT and SEU, which continue to serve as normative benchmarks, but it transformed the landscape by showing that a descriptively accurate model could still be formally coherent.
Today, the five frameworks coexist in a clear division of labor. EUT remains the default normative model: it is used for welfare analysis, optimal contract design, and any setting where the analyst wants to prescribe what a rational agent should do. SEU extends this normative logic to situations of ambiguity, and it remains the foundation for models of learning and belief formation. Prospect Theory and CPT are the workhorses of behavioral economics, applied wherever empirical predictions about actual choice are needed—in marketing, consumer finance, and the study of risk-taking behavior. RDEU occupies a middle ground: it is used when a researcher wants nonlinear probability weighting without the full psychological apparatus of reference dependence, often in theoretical work on insurance and portfolio choice.
The leading frameworks today—EUT as normative benchmark and CPT as descriptive workhorse—agree on the importance of formal axiomatic foundations, but they disagree on the role of the independence axiom and the nature of the utility function. EUT insists on linearity in probabilities as a requirement of rationality; CPT treats nonlinear probability weighting as a stable feature of human cognition. This disagreement is not a sign of weakness but a productive tension: normative models provide clear benchmarks, while descriptive models reveal the systematic patterns that any complete theory of choice must explain. The ongoing research frontier includes efforts to integrate ambiguity attitudes into CPT-style models, to understand the neural foundations of reference dependence, and to extend these frameworks to dynamic choice settings where commitment and time inconsistency matter.