Allais paradox

Overview

The Allais paradox is a classic finding in decision theory showing that many people’s choices under risk systematically violate expected utility theory (EUT), specifically the independence axiom. In carefully constructed pairs of lotteries, most respondents prefer a sure, moderate gain over a higher–expected-value gamble, but also prefer a riskier high-upside gamble when the same “common consequence” is removed. These reversals contradict the core EUT prediction that adding or removing a common outcome should not flip preferences. The paradox matters because it reveals robust “certainty effects” and nonlinear treatment of probabilities, shaping how consumers and managers respond to guarantees, warranties, promotions, and risky projects.

Origins and Credit

Maurice Allais introduced the paradox in 1953, critiquing the independence axiom of von Neumann–Morgenstern expected utility. Subsequent work showed the pattern is widespread across contexts and stakes. The paradox catalyzed alternative models of risky choice, including rank-dependent expected utility (Quiggin), prospect theory and cumulative prospect theory (Kahneman–Tversky), and regret/disappointment-based theories (Loomes–Sugden).

Core Idea and Mechanics

The paradox uses two choice problems with known probabilities (no ambiguity):

• Problem 1
  • A: 100% chance of $1 million
  • B: 89% chance of $1 million; 10% chance of $5 million; 1% chance of $0
  Most people choose A (the certainty effect), even though B has higher expected value.
• Problem 2
  • C: 11% chance of $1 million; 89% chance of $0
  • D: 10% chance of $5 million; 90% chance of $0
  Many people choose D.

Expected utility theory’s independence axiom says that if one prefers one lottery over another, mixing both with the same third lottery in the same proportions should not change the preference. Here, adding “an 89% chance of $1 million” to both C and D yields A and B, respectively. Therefore, a preference for D over C should imply a preference for B over A. Observing A over B and D over C is a direct violation. The behavioral intuition is the certainty effect: people put disproportionate weight on outcomes that are certain relative to outcomes that are merely very likely, and they treat probabilities nonlinearly.

Key Assumptions and Conditions

• Choices are over risky lotteries with known, objective probabilities (risk, not ambiguity).
• Expected utility theory imposes:
  • Completeness and transitivity (consistent rankings),
  • Continuity (no jumps in preferences), and
  • Independence: Adding the same “common consequence” to two lotteries should not reverse their ranking.
• One-shot, framed choices; no feedback or learning during the task.

Implications

• Certainty is unusually attractive: Guarantees, risk-free options, and sure discounts are valued more than their expected-utility equivalents would suggest.
• Probability weighting: People overweight small probabilities and underweight moderate-to-high ones; models like cumulative prospect theory capture this and fit Allais-type choices.
• Design of offers and policies: Per-unit warranties, money-back guarantees, and “sure thing” rebates can be more motivating than higher expected-value lotteries; conversely, lottery-like promotions can be effective because small probabilities of large prizes are overweighted.
• Risk governance: Managers may overinvest in eliminating small residual risks (to achieve certainty) while neglecting larger expected-value improvements—relevant for cybersecurity, quality assurance, and compliance.
• Limits of EUT for prediction: For consumer analytics and project appraisal, models that allow probability weighting and reference dependence can improve forecasts of take-up and risk-taking.

Example in Practice

Designing a consumer promotion versus a sure discount. A retailer considers two programs:

• Program A: a certain $20 instant discount at checkout.
• Program B: 89% chance of a $20 discount, 10% chance of a $100 discount, and 1% chance of $0 (revealed by a digital “spin” at checkout).

Although Program B has higher expected value for the customer, many shoppers prefer Program A because it guarantees savings (certainty effect). Now scale both down by removing a common $20 component (for instance, offer the discount only on certain SKUs so that, for those purchases, shoppers face):

• Program C: 11% chance of $20; 89% chance of $0.
• Program D: 10% chance of $100; 90% chance of $0.

Many will now prefer D over C (a higher-upside gamble). Preferring A over B and D over C is Allais-type behavior. For the retailer, the lesson is twofold: sure discounts are especially compelling when they eliminate uncertainty entirely; lottery designs are more attractive when outcomes are mostly zero with a chance at a big prize. Mixing the two within the same campaign can segment the market by risk attitudes and amplify overall engagement.

Limitations and Common Misunderstandings

• Not just “risk aversion”: The paradox cannot be explained by concave utility alone; it specifically violates independence. Probability weighting or certainty sensitivity is needed.
• Framing matters: How options are described (gains vs losses, graphical displays) affects the prevalence of Allais-type choices.
• Heterogeneity and stakes: Individuals differ; very large or very small stakes can attenuate or amplify the effect. Experience and feedback can reduce anomalies in some settings.
• Normative debate: The paradox is a descriptive challenge to EUT, not necessarily a normative refutation. Some economists retain EUT for prescription and adopt alternatives for prediction.
• Ambiguity is separate: Allais uses known probabilities; Ellsberg’s paradox concerns ambiguity (unknown probabilities).

Related Concepts

Expected utility theory; Independence axiom; Certainty effect and common consequence effect; Prospect theory and cumulative prospect theory; Rank-dependent expected utility; Ellsberg paradox.