The Allais Paradox
Your choices between two pairs of gambles are perfectly consistent with your own reasoning. And yet they violate the foundational axiom of rational choice theory. Are you irrational, or is the theory wrong?
Presented by French economist Maurice Allais at a 1952 conference in Paris, where he surveyed a room that included some of the era's leading decision theorists, the Allais Paradox showed that real human choices systematically violate expected utility theory. The responses from the audience, including Paul Samuelson and Milton Friedman, ranged from embarrassment to denial.
Allais, M. (1953). Le comportement de l'homme rationnel devant le risque. Econometrica, 21(4), 503–546.
The two choices
Allais constructed two pairs of gambles. Consider which you would choose from each pair.
Pair 1:
- Option A: $1,000,000 for certain
- Option B: 89% chance of $1,000,000, a 10% chance of $5,000,000, and a 1% chance of nothing
Pair 2:
- Option C: 11% chance of $1,000,000, otherwise nothing
- Option D: 10% chance of $5,000,000, otherwise nothing
Most people choose A over B, and D over C. The certain million feels better than the gamble in Pair 1. The better odds on the larger prize feel better in Pair 2.
The problem: these choices are mathematically inconsistent.
The independence axiom
Expected utility theory rests on the independence axiom: if you prefer A to B, you should still prefer A to B after adding the same element of risk to both options. Mixing in an identical "background lottery" shouldn't change your preference ordering.
Choosing A over B means you prefer certainty enough to give up the chance at $5M. Choosing D over C means you prefer the 10% shot at $5M over the 11% shot at $1M. But when you do the math, these preferences are only consistent if you treat the same probability weight completely differently depending on whether it eliminates uncertainty or merely reduces it.
The 1% chance of nothing in Option B disturbs people disproportionately because it breaks a guarantee. In Pair 2, there's no guarantee to break, so the 1% difference in probability feels smaller. People are not weighting probabilities linearly. The move from 0% chance of nothing to 1% chance of nothing gets far more psychological weight than the move from 89% to 90%.
What this means for rationality
Allais thought he was showing that expected utility theory was empirically inadequate as a description of human behavior. The theorists he surveyed mostly conceded the point about behavior but argued that their choices were mistakes, deviations from rationality rather than evidence against the theory.
That response has never fully satisfied anyone. If virtually everyone makes the same "mistake" in a predictable direction, the question becomes whether the theory is describing a real standard of rationality or a mathematical construct that happens to be logically coherent but disconnected from how any actual agent reasons.
Prospect theory, developed by Kahneman and Tversky in 1979, replaced expected utility with a model that explicitly builds in probability weighting: people overweight small probabilities and underweight large ones. It describes behavior much better than expected utility theory does, though it comes with its own normative problems.
The Allais Paradox sits at the center of a dispute that has not been resolved: whether rational choice theory is a description of how people should reason (and humans fail it), or whether "rational" should be derived from how careful, reflective humans actually do reason (and the theory fails them).
Discussion questions
- When was the last time you made a financial decision based on gut feeling rather than math, and do you regret it?
- Should businesses or governments be allowed to design choices that take advantage of the ways people are predictably irrational?
- Does consistency in your choices matter more to you than making the best decision in each situation?
Take it to the dinner table.
Get 3 thought experiments for memorable conversations, designed for dinner, with friends, at events, or anywhere small talk has gone on too long.