Prospect Theory: Why Losses and Gains Are Not Symmetric

Scope. This article explains what prospect theory says, what the famous number attached to it actually measured, and how much of it survives testing outside a laboratory. The structure of the theory is very well established. One widely quoted parameter is not, and this article separates the two. It supplies no technique, no corrective and no rule. Findings verified 17 August 2026. Figures are small non-canonical illustrations.

That sounds like a technicality and it is the whole thing. A portfolio worth $95,000 is a different experience depending on whether it was $80,000 last year or $120,000 — and under expected utility theory, the classical alternative, it should not be.

The three features of the value function

Concave for gains. Each additional unit of gain adds less than the one before, which produces risk aversion when someone is ahead.

Convex for losses. Each additional unit of loss hurts less than the one before, which produces risk seeking when someone is behind. This is the feature people find hardest to believe and it is central.

Steeper for losses than for gains. This is loss aversion — the asymmetry the theory is best known for.

The theory also replaces probabilities with decision weights, capturing the finding that people underweight merely probable outcomes relative to outcomes obtained with certainty.

The famous number, and what it actually measured

"Losses hurt about twice as much as equivalent gains" is the best-known quantitative claim in behavioural economics. Its provenance is worth knowing precisely.

The 1979 paper did not state a ratio. It established the qualitative asymmetry.

The figure of about 2.25 comes from a later paper — Tversky and Kahneman (1992), in the Journal of Risk and Uncertainty — which fitted a power-function value model to fresh experimental choices. That estimation used 25 graduate students at Berkeley and Stanford, making hypothetical choices about modest monetary sums.

What the size of the coefficient implies

Worked example — the coefficient's size is not a detail. The win probability required to accept an even-money bet is the coefficient divided by one plus itself. At the textbook figure a person would decline unless the odds of winning reached 69.2%; at the upper end of field estimates, 60.0%. That is a 9.2 percentage point difference in described behaviour, produced by a parameter most accounts quote as though it were a constant of nature. Figures are illustrative and describe no actual decision.

What survives testing outside the laboratory

The direction is robust. The magnitude is not, and the honest summary is that the textbook figure overstates it.

Meta-analytic work — notably from Yechiam's group and from Gal and Rucker's 2018 review in the Journal of Consumer Psychology — finds loss-aversion coefficients in field settings often closer to 1.0 to 1.5 than to the 2.0 to 2.5 range. Some field studies find no loss aversion at all in particular domains, especially for small-stakes everyday decisions. A 2024 meta-analysis of the published parametric estimates put the average coefficient at about 1.9 to 2.0, with a wide spread across studies and contexts — above the field-only figures and below the textbook one, which is the honest picture of a parameter that is real and unstable.

The structure of the theory has fared better than the parameter. A large multinational replication of the 1979 experiments produced positive results overall, with mixed findings on the reflection effect — the specific prediction that preferences flip between the gain and loss domains.

This is the case study the pillar's reporting standard exists for, so it is worth stating plainly. The most famous number in behavioural finance was estimated from 25 graduate students making hypothetical choices, and field measurement puts it materially lower. None of that makes prospect theory wrong. Reference dependence, the concave-convex shape and the direction of the asymmetry are among the better-established findings in the social sciences. What it makes wrong is the confident recitation of "twice as much" as though it were measured in the world rather than fitted in a room. A reader who takes away the structure and treats the coefficient as approximate and context-dependent has understood the state of the evidence. A reader who takes away 2.25 has learned a number that field studies do not support.

Where the reference point comes from

The theory says outcomes are judged against a reference point and does not say what sets it. That is a genuine gap rather than an omission this article can fill.

Candidates observed in practice include the purchase price, the highest value the holding reached, the amount originally invested, and a round number. They frequently disagree, and the same position can be a gain against one and a loss against another. Which one a person is using is not observable from outside, and may not be stable over time.