The Allais Paradox: The Decision Theory Result That Forced Economics To Update (Anti-Example)

Most behavioral-economics findings in this hub did not survive scrutiny. The Allais paradox did. In 1953, a French economist with a habit of disagreeing with American Nobel laureates handed people two pairs of gambles, asked them to choose, and showed that their answers were mathematically inconsistent with the dominant theory of rational choice. Seventy years and a hundred replications later, the result still holds. It motivated prospect theory. It won Allais his own Nobel Prize in 1988. And it is the cleanest historical example of what robust behavioral economics actually looks like when it is done right.

If you have been reading through this hub, you have watched a long parade of canonical behavioral findings get dismantled. Ego depletion collapsed. Power posing was recanted by one of its own authors. Money priming evaporated. The marshmallow test shrank dramatically once socioeconomic status was controlled for. The bystander effect’s Kitty Genovese mythology turned out to be a journalist’s invention. The grit construct shrank to near-conscientiousness redundancy. The replication crisis has done real, justified damage to the credibility of social science as a whole, and behavioral economics has not been immune.

A rational reader by now might conclude that behavioral economics as a field is simply unreliable. That conclusion would be wrong, and the Allais paradox is the cleanest case for why. Unlike most of the findings dismantled elsewhere in this hub, the Allais paradox was not generated by a clever lab paradigm that fails to generalize. It was generated by Maurice Allais writing down two gambles on a piece of paper at a 1952 economics conference in Paris and asking Leonard Savage --- one of the architects of expected utility theory --- which he preferred. Savage gave the “wrong” answer in the sense that his choices violated his own theory’s independence axiom. Allais had him.

That same demonstration has been run on undergraduates, MBAs, professional economists, Russian and Chinese subjects, and high-stakes experimental subjects in laboratories from Pittsburgh to Tel Aviv to Shanghai, with real money on the line. The qualitative pattern --- the modal subject picking A over B and D over C, which together imply a contradiction under expected utility theory --- has held up everywhere. Camerer 1989, Conlisk 1989, Starmer 2000, the comprehensive review evidence summarized in Wakker 2010: the same pattern, year after year, paradigm after paradigm.

This article is the story of why this single counterexample, presented in 1953 in a paper most Anglophone economists ignored for two decades, eventually forced the entire field of decision theory to update. It is also the calibration this hub means to deliver: when you are evaluating a behavioral-economics claim, the Allais paradox is the benchmark for what real evidence looks like. Not “a single underpowered lab study found a marginal effect at p = 0.048.” But: “the same choice pattern shows up in every laboratory, in every decade, across every methodology, and forces a formal theoretical framework to change.”

Here is what holds up, why it survived, and what makes it the cleanest anti-example in this hub.

The Two Choice Pairs

The Allais paradox is most often stated as a pair of binary choices over monetary gambles. The exact numbers Allais used in his 1953 paper involved francs and large nominal stakes, but the version that became canonical --- used in essentially every replication --- uses dollars and is structured as follows.

Choice 1. Choose between:

• A: $1 million for certain.
• B: A gamble with a 10% chance of $5 million, an 89% chance of $1 million, and a 1% chance of nothing.

Choice 2. Choose between:

• C: A gamble with an 11% chance of $1 million and an 89% chance of nothing.
• D: A gamble with a 10% chance of $5 million and a 90% chance of nothing.

Most people --- and this is the empirical finding that has survived 70 years of replication --- choose A in the first pair and D in the second. The choice of A over B reflects a preference for certainty: getting $1 million for sure feels better than a gamble that, with 1% probability, leaves you with nothing despite an upside of $5 million. The choice of D over C reflects a preference for the larger upside: when both gambles already involve a high probability of getting nothing, the difference between 11% and 10% feels small relative to the difference between $1 million and $5 million.

Each individual choice, in isolation, seems perfectly sensible. The problem is that the two choices, taken together, are mathematically inconsistent with expected utility theory.

The proof is two lines of algebra. Under expected utility theory, the agent has some utility function $u(\cdot)$ over outcomes, and the agent prefers prospect $X$ to prospect $Y$ if and only if the expected utility of $X$ exceeds the expected utility of $Y$. So:

Choosing A over B means: $u($1M) > 0.10 \cdot u($5M) + 0.89 \cdot u($1M) + 0.01 \cdot u($0)$.

Subtract $0.89 \cdot u($1M)$ from both sides: $0.11 \cdot u($1M) > 0.10 \cdot u($5M) + 0.01 \cdot u($0)$.

Choosing D over C means: $0.10 \cdot u($5M) + 0.90 \cdot u($0) > 0.11 \cdot u($1M) + 0.89 \cdot u($0)$.

Subtract $0.89 \cdot u($0)$ from both sides: $0.10 \cdot u($5M) + 0.01 \cdot u($0) > 0.11 \cdot u($1M)$.

The two inequalities, after the algebra, are direct contradictions. There is no utility function $u(\cdot)$ that can satisfy both at the same time. The modal preferences --- A in pair one, D in pair two --- are therefore inconsistent with expected utility theory at the level of mathematical structure, not at the level of empirical fit or statistical power.

This is the cleanest possible kind of behavioral-economics result. It does not require a p-value. It does not require an effect-size estimate. It does not require a meta-analysis. It requires only that you ask a person two questions, observe their answers, and notice that no consistent utility function could have generated both. Allais understood this immediately in 1952. The American economists at the conference, including Savage, took considerably longer to be convinced.

Why The Choices Violate The Independence Axiom

The specific axiom of expected utility theory that the Allais paradox violates is the independence axiom, sometimes called the substitution axiom or the sure-thing principle.

The independence axiom says: if you prefer prospect $X$ to prospect $Y$, then for any third prospect $Z$ and any probability $p$ in $(0, 1)$, you should also prefer the compound prospect “$X$ with probability $p$, $Z$ with probability $1-p$” to “$Y$ with probability $p$, $Z$ with probability $1-p$.” The intuition is that the common component $Z$ should cancel out of the comparison. If you would rather have apples than oranges, you should also rather have “apples with 50% chance, bananas with 50% chance” than “oranges with 50% chance, bananas with 50% chance,” because the banana branch is identical in both compound prospects and contributes the same expected utility either way.

The Allais paradox is constructed so that each pair of choices can be decomposed into the same underlying “core” gamble combined with different “common consequence” branches. Specifically:

• Choice 1 can be rewritten as a comparison between two prospects that both have an 89% common branch of $1M, and differ only in what happens in the remaining 11% of cases.
• Choice 2 can be rewritten as a comparison between two prospects that both have an 89% common branch of $0, and differ only in what happens in the remaining 11% of cases.
• And the 11% non-common branch in each pair is identical across the two choices. In Choice 1’s A versus B, the non-common branch is “$1M with certainty” versus ”($5M with 10/11 probability, $0 with 1/11 probability)”. In Choice 2’s C versus D, the non-common branch is exactly the same comparison.

So under the independence axiom, an agent who prefers A over B must prefer C over D, and an agent who prefers B over A must prefer D over C. The pairings (A, C) and (B, D) are the only ones consistent with the independence axiom. The modal pairing (A, D) is the one that violates it.

What Allais argued, and what prospect theory would later formalize, is that the independence axiom fails because the “common consequence” branches do not cancel out psychologically. In Choice 1, the 89% common branch is $1M --- a sure thing if you choose A, and a near-sure thing if you choose B. The 1% probability of getting $0 in B feels enormously costly because it disrupts what would otherwise be certainty. The certainty itself is what is being valued, and switching from A to B costs you that certainty in exchange for an upside that might not even materialize.

In Choice 2, the 89% common branch is $0 --- a likely outcome regardless of whether you choose C or D. Neither gamble offers anything close to certainty. The question becomes purely “11% chance of $1M versus 10% chance of $5M,” and the larger upside wins. The certainty premium that pulled people toward A in Choice 1 is simply not available in Choice 2, because nothing in Choice 2 is certain.

Prospect theory’s probability-weighting function captures this asymmetry directly: small probabilities at the extremes (the 1% chance of $0 in prospect B, the 10% versus 11% comparison in prospects C and D) are weighted differently from medium probabilities, and the discontinuity at certainty creates a “kink” in the way probability-weighted utility is computed. The Allais paradox is the empirical fingerprint of that kink.

The Replication Record

The Allais paradox is one of the most-replicated findings in all of behavioral economics. Unlike the long list of replication failures cataloged elsewhere in this hub, the Allais paradox has been replicated, in some form, in essentially every study that has tried to replicate it. The key references:

Allais, M. (1953). “Le comportement de l’homme rationnel devant le risque: critique des postulats et axiomes de l’école américaine.” Econometrica, 21(4), 503—546. DOI: 10.2307/1907921. The founding paper. Allais presents the original choice pairs (with francs and somewhat different numbers than the canonical English version) and argues at length, with mathematical rigor, that the modal pattern of choices is inconsistent with expected utility theory and reflects a systematic preference for certainty that the von Neumann—Morgenstern axioms cannot accommodate. The paper was written in French, in a journal whose American readership was largely uninterested in challenges to the new expected utility framework, and was largely ignored in the United States for nearly two decades.

Kahneman, D., & Tversky, A. (1979). “Prospect theory: An analysis of decision under risk.” Econometrica, 47(2), 263—291. DOI: 10.2307/1914185. This is the paper that formalized the alternative to expected utility theory and that took the Allais paradox seriously as a foundational anomaly rather than a curiosity. Kahneman and Tversky present their own replication of the Allais choice pairs with new samples and confirm the modal A/D pattern. They use the paradox as one of the central motivating examples for prospect theory’s reference-dependent value function and non-linear probability weighting function.

Camerer, C. F. (1989). “An experimental test of several generalized utility theories.” Journal of Risk and Uncertainty, 2(1), 61—104. DOI: 10.1007/BF00055711. A systematic experimental comparison of expected utility theory against several alternative models, including prospect theory and various weighted-utility frameworks. Camerer documents that Allais-paradox-type violations of the independence axiom show up reliably across multiple choice problems and that they cannot be dismissed as the artifact of any one specific gamble structure. The paper is one of the cleanest demonstrations that the violations are systematic and predictable, not random noise.

Conlisk, J. (1989). “Three variants on the Allais example.” American Economic Review, 79(3), 392—407. Conlisk runs three modified versions of the Allais choice pairs designed to test whether the paradox depends on specific features of the original gambles --- the magnitudes, the certainty branch, the framing of probabilities. He shows that the modal violation pattern survives across all three variants, including a version with smaller stakes and a version with no certainty option. The paradox is not an artifact of the specific numbers Allais chose. It is a general property of how people weight low-probability outcomes against high-probability outcomes near the boundaries of the probability simplex.

Tversky, A., & Kahneman, D. (1992). “Advances in prospect theory: Cumulative representation of uncertainty.” Journal of Risk and Uncertainty, 5(4), 297—323. DOI: 10.1007/BF00122574. The 1992 follow-up to the original prospect theory paper, which fixed the mathematical limitations of the 1979 specification and presented “cumulative prospect theory” --- a version that satisfies stochastic dominance and applies to arbitrary numbers of outcomes. Tversky and Kahneman re-derive the Allais paradox under the new specification and show that cumulative prospect theory’s rank-dependent probability weighting function explains the modal A/D choice pattern as a structural prediction of the model, not as a stylized fact bolted on after the fact.

The pattern across these and dozens of subsequent studies is consistent: when you give people the Allais choice pairs (or any of the standard variants), the modal response is the one that violates the independence axiom. The exact percentage varies with the sample, the stakes, and the framing --- ranges of 40 to 80% of subjects exhibiting the violation are typical --- but the qualitative pattern is robust. It shows up in undergraduate samples, MBA samples, and samples of professional economists who explicitly know what the independence axiom is and what the violation looks like. It shows up with hypothetical stakes and with real stakes. It shows up in studies run in the United States, Germany, France, Israel, China, Singapore, and India. The cross-cultural replication record is, frankly, better than most “WEIRD” psychology findings have any right to claim.

Two caveats are worth noting. First, the violation rate does decline somewhat when subjects are given more time to deliberate, when the choices are presented in a “matrix” format that makes the common-consequence structure explicit, or when subjects are given the opportunity to revise their answers after being shown the implied contradiction. This is exactly what prospect theory predicts: the editing operations in prospect theory’s first stage include “cancelling common components,” and explicitly highlighting those components in the presentation makes the cancellation more likely to occur. Second, there is some evidence that highly mathematically sophisticated subjects --- professional decision theorists, certain kinds of trained economists --- exhibit the violation at lower rates than the general population, presumably because they recognize the structural form of the gambles and apply the independence axiom self-consciously. But neither of these caveats undermines the core finding. The Allais paradox is a real, robust, cross-cultural empirical regularity that any descriptive theory of decision under risk must explain.

The 1988 Nobel Prize and What It Meant

Maurice Allais was awarded the Nobel Prize in Economic Sciences in 1988, 35 years after his original paper. The official citation was “for his pioneering contributions to the theory of markets and efficient utilization of resources” --- language broad enough to encompass his work on monetary theory, his theory of intertemporal allocation, and his earlier work on general equilibrium. But the Allais paradox was prominently cited in the Nobel committee’s background documents and in essentially every retrospective on his career.

The 1988 Nobel mattered for two reasons.

First, it was the formal acknowledgment, from the most institutionally weighty body in economics, that Allais’s challenge to expected utility theory had been correct on the merits. Allais had spent 35 years being mostly ignored by the Anglophone economics profession. His original 1953 paper was in French. His subsequent defenses of the position were often polemical and uncompromising in a style that did not endear him to American economists committed to the von Neumann—Morgenstern framework. The 1988 Nobel was, in part, a formal recognition that the position Allais had been defending was right, and that the profession’s neglect of it had been a failure of intellectual seriousness, not a failure of the work.

Second, the timing of the Nobel mattered. By 1988, prospect theory had been published for nine years. Kahneman, Tversky, and a growing community of behavioral economists were doing serious mathematical work on alternatives to expected utility theory. The field of decision theory had begun to internalize that the von Neumann—Morgenstern axioms were not a complete description of how rational agents do or should make decisions under risk, and that systematic violations like the Allais paradox required formal accommodation rather than dismissal. Awarding the Nobel to Allais in 1988 was, in effect, the profession’s announcement that the Allais paradox was no longer a curiosity. It was a foundational result that any serious theory of decision under risk had to engage with.

Kahneman would win his own Nobel in 2002, in part for prospect theory itself. The line from Allais 1953 to Kahneman—Tversky 1979 to the 1992 cumulative prospect theory paper to the 2002 Nobel is one of the cleanest examples in the history of economics of a foundational empirical anomaly being rigorously documented, formally accommodated by a new theoretical framework, and eventually recognized by the highest institutional authority in the field. It is what successful paradigm transition looks like when the underlying empirical evidence is real.

The contrast with the failed behavioral findings cataloged elsewhere in this hub is sharp. The Allais paradox was not generated by a single clever lab paradigm that failed to generalize. It was not buoyed by publication bias that collapsed under preregistered replication. It was not the work of a single charismatic researcher whose methods turned out to be sloppy or fraudulent. It was a robust empirical regularity that survived every test the field could throw at it, motivated a complete reconstruction of decision theory, and was eventually awarded a Nobel Prize. That is the standard. When a behavioral-economics finding does not look like this --- when it is a single underpowered study with a clever paradigm and a press release --- you are right to be skeptical. When it does look like this --- robust, cross-cultural, theoretically generative, eventually institutionalized --- you are right to take it seriously.

The Connection To Prospect Theory

The Allais paradox is one of the founding empirical anomalies that prospect theory was designed to accommodate, and it remains one of the cleanest illustrations of prospect theory’s two core innovations: reference-dependent valuation and non-linear probability weighting.

Under prospect theory’s value function, outcomes are evaluated relative to a reference point rather than as absolute levels of final wealth. In the Allais choice pairs, the reference point in Choice 1 is plausibly the certainty branch of $1M --- the outcome you are guaranteed if you choose A. From that reference point, the 1% probability of $0 in prospect B is a loss, and losses are weighted more heavily than equivalent gains. The $4M upside in prospect B’s 10% branch is a gain, and gains are weighted less heavily than equivalent losses. So even though the expected monetary value of B exceeds the expected monetary value of A by a substantial margin, the loss-aversion structure of the value function can rationalize preferring A.

But the more important component of prospect theory’s explanation of the Allais paradox is the probability weighting function. The probability weighting function in prospect theory is a non-linear transformation of objective probabilities into “decision weights” that reflect how much each probability is actually weighted in the decision. The function has a characteristic shape: small probabilities are systematically overweighted, while medium-to-large probabilities are systematically underweighted, and the function has discontinuities at the endpoints (probability 0 and probability 1) that capture the special status of impossibility and certainty.

This last feature --- the discontinuity at certainty --- is the part of prospect theory that explains the Allais paradox most directly. In Choice 1, prospect A offers certainty and prospect B offers a near-certain but technically uncertain outcome. The jump from probability 1.0 to probability 0.99 is treated as much larger, in prospect theory’s probability weighting function, than the equivalent jump from probability 0.10 to probability 0.11. This is the “certainty effect,” and it is the formal mechanism in prospect theory that generates Allais-paradox-style violations of the independence axiom.

The 1992 cumulative prospect theory paper extended this framework using rank-dependent probability weighting --- a more mathematically sophisticated approach where decision weights depend not just on individual probabilities but on the cumulative distribution of outcomes. Under cumulative prospect theory, the Allais paradox is no longer an awkward anomaly that the model can rationalize on a case-by-case basis. It is a structural prediction of the model’s specification. Any agent with the canonical inverse-S probability weighting function will exhibit Allais-paradox-style violations of the independence axiom whenever the choice pairs are structured as the canonical examples are.

This is the gold standard for what behavioral economics should look like as a research program. Start with a clean empirical anomaly (Allais 1953). Document its robustness (the 1989 Camerer and Conlisk papers, plus dozens of subsequent replications). Formalize an alternative theoretical framework that explains the anomaly as a structural prediction rather than an ad hoc patch (Kahneman—Tversky 1979, 1992). Get the resulting framework adopted by the broader field. Watch the original anomaly turn from a paradox into a textbook case. The Allais paradox is no longer “a curious violation of expected utility theory.” It is “one of the standard predictions of cumulative prospect theory, exhibited by the modal subject in essentially every laboratory setting where the relevant gambles have been administered.” The conceptual reframing from paradox to prediction is what successful theory development looks like.

For the deeper account of prospect theory’s value function, four-fold pattern of risk attitudes, and probability weighting function, see the dedicated prospect-theory article in this hub.

Real-World Implications

The Allais paradox is not just a laboratory curiosity. The same psychological mechanisms that generate the canonical A/D choice pattern show up in three important real-world domains: insurance markets, lottery markets, and regulatory decision-making about low-probability hazards.

Insurance markets. The puzzle of why people buy insurance against rare events --- earthquake insurance, extended warranties on consumer electronics, flight insurance --- is structurally identical to the puzzle of why people choose A over B in the canonical Allais pair. In each case, the consumer is paying a premium that exceeds the actuarially fair expected value of the coverage in exchange for the elimination (or near-elimination) of a low-probability bad outcome. Under expected utility theory with reasonable utility function curvature, this behavior is hard to rationalize at the observed scale. Under prospect theory’s probability weighting function, the same low probabilities of large losses get overweighted in the consumer’s decision, and the willingness to pay premiums above actuarial fair value falls out as a structural prediction. The insurance industry’s pricing models implicitly rely on the same probability-weighting asymmetries that generate the Allais paradox, even when the underwriters who set the premiums have never read Allais.

Lottery markets. State lotteries are economically puzzling for the same reason insurance is, but in mirror image. Lottery tickets have negative expected value --- the state keeps a substantial cut --- and would be unattractive to any expected-utility-maximizing agent with reasonable risk aversion. They are nevertheless purchased in enormous volumes, predominantly by people for whom even small monetary losses are economically significant. Under prospect theory, the same probability weighting function that overweights low-probability losses (motivating insurance purchases) also overweights low-probability gains (motivating lottery ticket purchases). The Allais paradox’s identification of asymmetric probability weighting at the extremes of the probability simplex is the same mechanism that explains why the same individual can rationally (in prospect-theory terms) hold both an insurance policy and a Powerball ticket.

Regulatory decision-making. Government regulators routinely make decisions about how much to invest in mitigating low-probability hazards --- nuclear-plant safety upgrades, earthquake-resistant building codes, pandemic-preparedness stockpiles. Standard cost-benefit analysis under expected utility theory says these investments should be made when the expected reduction in harm exceeds the expected cost. But the Allais paradox shows that real human decision-makers do not weight low-probability outcomes the way expected utility theory says they should. Regulators who follow strict expected-value cost-benefit analysis will systematically underweight low-probability catastrophic events relative to public preference; regulators who follow public preference will systematically overinvest in mitigating low-probability hazards relative to what expected-value analysis would recommend. This tension --- between expected-value rationality and prospect-theory-consistent public preference --- shows up in essentially every regulatory debate about catastrophic risk, and the resolution requires acknowledging that the Allais paradox is not just a laboratory curiosity but a real feature of how humans evaluate the risks they care about. Frameworks like the precautionary principle in environmental regulation can be read as implicit attempts to formalize the certainty premium that drives the Allais paradox.

These three domains are not exhaustive. The same probability-weighting asymmetries show up in financial portfolio choice, in legal decisions under uncertainty, in medical-treatment decisions involving small probabilities of severe side effects, and in the design of incentive systems where employees evaluate low-probability bonuses or penalties. The Allais paradox is the cleanest laboratory demonstration of a psychological mechanism that operates throughout economic life.

The Strategist’s Takeaway

If you are evaluating a behavioral-economics claim --- a vendor pitching a “behavioral nudge” intervention, a consultant citing a “cognitive bias” that should reshape your product strategy, an academic paper claiming a novel decision-making effect --- the Allais paradox is your benchmark for what real evidence looks like. Specifically:

Robust replication across paradigms, decades, and cultures. The Allais paradox has been replicated in dozens of countries, using gamble structures that vary widely in their specific numerical parameters, with hypothetical and real stakes, on undergraduates, MBAs, professional economists, and general-population samples. If a behavioral claim has only been demonstrated once, in a single sample, with a single paradigm, by a single research group, you are looking at the kind of finding that the replication crisis has repeatedly shown does not generalize. The Allais paradox does not look like that. It looks like the opposite of that.

Theoretical generativity. The Allais paradox did not just sit there as an isolated empirical curiosity. It motivated an entire formal theoretical framework --- prospect theory --- that organized it and a dozen other anomalies under a coherent mathematical structure with predictive content. If a behavioral claim is just “people sometimes do this surprising thing in this specific context,” and there is no formal framework that explains why it happens or predicts when it should generalize, you are looking at folk psychology, not behavioral economics. The Allais paradox is generative. It produced new theory, new predictions, and new experimental paradigms.

Eventual institutional recognition. Allais won the Nobel Prize in 1988. Kahneman won the Nobel Prize in 2002. The line from the original anomaly to formal theoretical accommodation to institutional acknowledgment took 35 years, but it happened, because the underlying empirical regularity was real and the theoretical framework that explained it was rigorous. If a behavioral claim has been around for 20 years and has not been integrated into any serious theoretical framework, has not generated a community of researchers building on it, and has not produced any new empirical predictions that have themselves been confirmed, you are probably looking at a finding that has quietly failed to replicate even if no one has done the formal preregistered replication study to bury it.

The Allais paradox is the calibration. When someone tells you about a behavioral-economics result, ask: how many times has this been replicated? In how many countries? Using how many different paradigms? With what stakes? Is there a formal theoretical framework that organizes this result with others? Has the framework been institutionally adopted? If the answers are “many,” “many,” “many,” “various,” “yes,” and “yes,” you are looking at evidence that meets the Allais standard. If the answers are “once,” “one,” “one,” “hypothetical,” “no,” and “no,” you are looking at the kind of finding that this hub has spent dozens of articles documenting as not having survived scrutiny.

The Allais paradox is not the only behavioral finding that meets this standard. Prospect theory itself meets it. The default effect meets it. A small number of other findings in this hub meet it. But the bar is genuinely high, and the canonical behavioral-science findings that meet the bar are a small fraction of the canonical behavioral-science findings that get cited in popular-press books. The Allais paradox is the cleanest historical example of a behavioral finding that genuinely cleared the bar, and it is the standard against which every other behavioral claim should be measured.

Sources

• Allais, M. (1953). “Le comportement de l’homme rationnel devant le risque: critique des postulats et axiomes de l’école américaine.” Econometrica, 21(4), 503—546. DOI: 10.2307/1907921.
• Kahneman, D., & Tversky, A. (1979). “Prospect theory: An analysis of decision under risk.” Econometrica, 47(2), 263—291. DOI: 10.2307/1914185.
• Camerer, C. F. (1989). “An experimental test of several generalized utility theories.” Journal of Risk and Uncertainty, 2(1), 61—104. DOI: 10.1007/BF00055711.
• Conlisk, J. (1989). “Three variants on the Allais example.” American Economic Review, 79(3), 392—407.
• Tversky, A., & Kahneman, D. (1992). “Advances in prospect theory: Cumulative representation of uncertainty.” Journal of Risk and Uncertainty, 5(4), 297—323. DOI: 10.1007/BF00122574.
• Starmer, C. (2000). “Developments in non-expected utility theory: The hunt for a descriptive theory of choice under risk.” Journal of Economic Literature, 38(2), 332—382. DOI: 10.1257/jel.38.2.332.
• Wakker, P. P. (2010). Prospect Theory: For Risk and Ambiguity. Cambridge University Press. ISBN: 978-0521765015.

Related

• Prospect Theory: The Behavioral-Economics Framework That Actually Replicates (Anti-Example) --- The 1979 framework that incorporated the Allais paradox as a structural prediction.
• Ellsberg Paradox --- The parallel demonstration of ambiguity aversion, which similarly violates expected utility theory and motivated separate theoretical extensions.
• Loss Aversion --- The reference-dependent valuation mechanism that does part of the work in explaining the Allais paradox.
• Hyperbolic Discounting --- The intertemporal analog: systematic violations of the discounted-utility model that have motivated alternative formal frameworks in the same way the Allais paradox motivated prospect theory.
• Framing Effect --- The broader family of decision-context dependencies of which the Allais paradox is the canonical formal example.

FAQ

Why did Maurice Allais’s 1953 paper get ignored for so long? It was written in French in a journal whose Anglophone readers were largely committed to defending the new von Neumann—Morgenstern expected utility framework. Allais’s polemical style and uncompromising defense of his position did not help. The American economics profession did not seriously engage with the result until the 1970s, when Kahneman and Tversky’s work on prospect theory made the anomaly impossible to ignore. The 35-year delay between the 1953 paper and the 1988 Nobel is, in retrospect, a documented case of the field’s failure to take a foundational empirical result seriously until decades after it should have.

Does the Allais paradox replicate with real money? Yes. Most early demonstrations used hypothetical stakes because of the impracticality of running gambles with $1M payouts, but subsequent work has shown that the modal violation pattern persists with real-money stakes scaled down to laboratory-feasible amounts. Camerer 1989 and many subsequent studies have used real-money versions and found the violations at substantively similar rates to hypothetical-stakes versions. The Allais paradox is not an artifact of subjects “not really caring” because the stakes are hypothetical. It is a robust feature of how people evaluate gambles regardless of stake size.

What is the difference between the Allais paradox and the Ellsberg paradox? The Allais paradox involves choices over gambles where the probabilities are known and well-defined --- the violations arise from how subjects weight those known probabilities. The Ellsberg paradox involves choices over gambles where some probabilities are known and others are ambiguous (not well-defined), and the violations arise from subjects’ systematic preference for known probabilities over ambiguous ones. Both paradoxes violate expected utility theory, but they violate it in different ways and motivate different theoretical extensions. Allais motivates prospect theory’s probability weighting function; Ellsberg motivates ambiguity-aversion frameworks like Choquet expected utility and maxmin expected utility.

Has anyone shown that the Allais paradox does not replicate? No serious attempt to replicate the Allais paradox has failed to find the basic pattern. Some studies have shown that the violation rate declines when subjects are explicitly walked through the common-consequence structure of the gambles, or when highly mathematically trained subjects are tested. But these qualifications do not constitute a replication failure. The qualitative pattern --- modal A/D choice rather than the consistent A/C or B/D pattern --- has held up in essentially every population, paradigm, and decade in which it has been tested. This is what distinguishes the Allais paradox from most of the findings dismantled elsewhere in this hub.

What does the Allais paradox say about whether people are “rational”? This is a contested philosophical question. Allais himself argued that the choices people make in his paradox are not irrational --- they reflect a legitimate preference for certainty that the von Neumann—Morgenstern axioms incorrectly rule out as inadmissible. Other economists argue that the choices are irrational in the sense that they cannot be rationalized by any consistent utility function. Prospect theory takes a third position: the choices are predictable consequences of a coherent psychological architecture that is different from but not necessarily worse than the von Neumann—Morgenstern framework. For most practical purposes, the question of “rationality” is less important than the question of “predictability” --- and on the predictability dimension, the Allais paradox is one of the most reliable findings in all of behavioral economics.

Does the Allais paradox have implications for real-world financial decision-making? Yes. The same probability-weighting mechanisms that generate the Allais paradox in the laboratory show up in real-world financial behavior: in insurance markets (overpaying premiums for protection against low-probability losses), in lottery markets (overpaying for tickets with low-probability gains), in portfolio choice (the equity-premium puzzle), and in the pricing of out-of-the-money options. Financial professionals who incorporate prospect-theoretic adjustments into their models routinely outperform those who use strict expected-utility frameworks, particularly for products that involve low-probability extreme outcomes.