Prospect Theory

What it is

Economics had an account of risky choice before 1979, and it was normative. A rational agent evaluates each option by the final state of wealth it leaves them in, weights those states by their probabilities, and picks the highest expected utility. Kahneman and Tversky's paper, "Prospect Theory: An Analysis of Decision under Risk", published in Econometrica in 1979, is a descriptive account instead. It asks what people do, finds that it is not that, and proposes a model of what they do instead.

Nobody evaluates final states of wealth. People evaluate changes, and a change has to be a change from something.

Reference dependence is the first component and the one everything else hangs from. Whether an outcome counts as a gain or a loss depends on where the person is standing when they meet it, and where they are standing is often an accident of description or of last year. A salary of £50,000 is a raise or a pay cut depending on what preceded it. The theory does not predict the reference point. It only insists that there is one and that the answer changes when it moves.

Diminishing sensitivity is the second. The value function is concave over gains and convex over losses, so the gap between £10 and £20 registers more than the gap between £110 and £120, and this happens in both directions. That is what produces the S-shape the theory is drawn with.

Probability weighting is the third, and it is the one most often left out of popular summaries. People do not use stated probabilities as given. Small probabilities are overweighted and moderate to large ones are underweighted. This is why the same person, on the same afternoon, buys a lottery ticket and an insurance policy without noticing any tension between the two.

Loss aversion is the fourth: the value function is steeper below the reference point than above it. This is the part that escaped into general circulation, and for a while the vault had it filed as the parent concept with prospect theory as a footnote, which is backwards. Loss aversion is one property of one limb of one function in a larger model, and it is the property currently under the most pressure.

Because value is measured from a reference point, and the reference point is set by how a thing is described, the same choice described two ways produces two different decisions. That is framing, and it is the part of the theory with the widest reach outside the laboratory, running through health messaging, pricing and choice architecture. It is also the part that has held up best.

Separating what is solid from what is not matters here, because the theory tends to be defended or attacked as a single object and it is not one. Reference dependence, diminishing sensitivity, framing effects, probability weighting and the fourfold pattern are extensively replicated, and they are the work the 2002 Nobel Memorial Prize in Economic Sciences recognised. Kahneman received that prize alone. Tversky had died in 1996, and the prize is not awarded posthumously, which is the origin of a great deal of subsequent miscrediting.

The contested part is narrow and specific. The claim that losses loom roughly twice as large as gains, and in particular the fixed coefficient λ = 2.25 taken from the 1992 cumulative prospect theory paper, does not hold as a universal constant. David Gal and Derek Rucker argued in 2018 that the asymmetry is highly sensitive to how preferences are elicited, and the full disagreement, with both cases stated at length, sits at loss aversion.

Our editorial position is the one the vault records, and it is a split rather than a verdict. Prospect theory as a descriptive framework is robust. The 2:1 constant inside it is mixed. An article may lean on the first without inheriting the second, and should say which one it is using. A robust theory carrying a contested parameter is unusual enough to be worth stating in those terms rather than smoothing into a single grade.

In effect

The fourfold pattern

The theory's most testable output is a table of four cells, and the reason to reach for it is that it predicts behaviour that looks like inconsistency and is not.

The same person does all four, and expected utility theory cannot generate that pattern from any single risk attitude. Prospect theory produces it directly, from the S-shaped value function crossed with probability weighting. It is the strongest thing the framework has, and it is the reason the contested coefficient does not take the whole model down with it.

The failing business that doubles its bet

The bottom-left cell has a practical form that anyone who has watched a project die will recognise. People facing a near-certain loss will gamble rather than crystallise it, because in the loss domain the curve makes the gamble look better than the sure thing. That is the shape of the manager committing more money to the failing launch, the trader adding to a losing position, and the person staying another year in the job they should have left. Escalation of commitment and the sunk cost fallacy both have their foothold here.

Berger's pricing chapter, and whose work it is

Jonah Berger's Practical Value chapter in Contagious is applied prospect theory throughout. Reference points and diminishing sensitivity drive every pricing result in it, which is why the same discount can be worth sharing on a cheap item and not worth mentioning on an expensive one. Berger's own endnotes concede the scenarios are adapted from Richard Thaler's work on mental accounting rather than taken from Kahneman and Tversky directly. The book also presents Kahneman's 2002 Nobel without naming Tversky as the co-author of the work it was awarded for, mentioning him only in passing across pages 160 to 173. Mental accounting has no page on this site yet. The vault's audit of Contagious grades the book mixed, on the strength of a different chapter, so treat the pricing material as Berger's exposition of a literature rather than as evidence he has produced.

What it does not say

It does not say people are irrational. The model is descriptive. It predicts the fourfold pattern rather than deploring it, and reference dependence is a sensible way to run a perceptual system that has to work with limited range. What it says is that the standard economic model was a poor description, not that its subjects were defective.

It does not tell you what the reference point is. This is the largest hole in applying it, and it is structural rather than an oversight. Everything interesting about a real pricing or policy decision turns on what the person is comparing the outcome to, and prospect theory takes the reference point as given. A price rise judged against last year's price and the same rise judged against a competitor's are different decisions in the same room.

It does not carry the 2:1 ratio as a law. The framework survives without it. Gal and Rucker's 2018 paper is the counter-case, and the coefficient λ = 2.25 belongs to the 1992 paper rather than to 1979 or 1991, an attribution error common enough to be worth naming.

It does not belong to Kahneman alone. The Nobel went to one living author of a two-author body of work, and popular writing has been quietly dropping Tversky ever since. The 1979 paper is Kahneman and Tversky; the 1991 and 1992 papers are Tversky and Kahneman.

Who developed it

Daniel Kahneman and Amos Tversky, jointly, in Econometrica in 1979. The pair extended the model to multiple outcomes as cumulative prospect theory in 1992, and gave loss aversion its own paper, with a coefficient, in 1991. Tversky died in 1996. Kahneman received the Nobel Memorial Prize in Economic Sciences in 2002 for work that was never his alone, and the resulting habit of crediting him singly is the most reliable citation error in this field. Neither has a page on this site yet.

---