Cumulative prospect theory
The refined version of prospect theory that weights cumulative probabilities and never breaks dominance.
What it means
Cumulative prospect theory is the 1992 revision of prospect theory that replaced separate weighting of each outcome's probability with rank-dependent weighting of cumulative probabilities, applied separately to the gain and loss branches. This change cured the original theory's embarrassing flaw — that it could prefer a dominated prospect — while preserving its core features of reference dependence, loss aversion, and an inverse-S weighting function. It also extended the theory to prospects with many or continuous outcomes and to uncertainty as well as risk. By borrowing Quiggin's rank-dependent machinery, it became the version used in most quantitative applications and parameter estimates, including the canonical loss-aversion coefficient near 2.25. It stands today as the leading descriptive model of decision under risk.
How the weighting actually works
Cumulative prospect theory does not weight each outcome's probability on its own. It ranks the outcomes within a branch from worst to best, transforms the cumulative probability of doing at least that well, and sets each outcome's decision weight to the marginal change in that transformed cumulative probability. Because every weight is carved from one monotone transformation of the whole distribution, the model cannot rank a dominated prospect above a dominating one — the flaw that had forced the 1979 theory into an ad hoc editing phase. The inverse-S shape of the transformation means the extreme outcomes, the best and the worst, carry more weight than their raw probabilities warrant, while the middling outcomes carry less. That single structural feature is what generates the fourfold pattern of risk attitudes.
What the evidence shows
Estimated weighting functions are reliably inverse-S: small probabilities are overweighted and moderate-to-large ones underweighted, with the fitted curve crossing objective probability near one-third (Wu and Gonzalez 1996; Fehr-Duda and Epper 2012). The qualitative choice patterns travel well — a 19-country, roughly 4,100-person study reproduced about 94 percent of the original items and twelve of thirteen theoretical contrasts (Ruggeri et al. 2020). The famous loss-aversion coefficient is shakier. Tversky and Kahneman's median estimate of 2.25 hardened into a fixed constant in applied work, but a 2024 meta-analysis of 607 estimates put the mean nearer 1.95 and showed it swinging with elicitation method and stakes. The lesson is that cumulative prospect theory fits aggregate data well while its parameters remain population summaries, not universal constants.
Where it breaks down
The model is descriptive, not mechanistic. It curve-fits choices without saying what cognitive process produces them, so its parameters can absorb almost any in-sample pattern yet predict individual choices poorly out of sample. Estimates are also heterogeneous: hierarchical Bayesian fits reveal wide person-to-person spread and correlations among parameters that make loss aversion and probability weighting hard to disentangle (Nilsson et al. 2011). And the inverse-S holds only for decisions from description. When people learn probabilities from repeated experience rather than a stated number, rare events tend to be underweighted rather than overweighted — the description-experience gap — which the standard weighting function gets backwards. The theory is best read as a compact account of how people respond to explicitly described risks, not a general law of choice.
Related but distinct
Cumulative prospect theory is routinely conflated with the 1979 original, but the machinery differs. The original weighted each probability separately and leaned on an editing phase to strike out dominated options by hand; the cumulative version builds dominance-respect into the weighting itself and discards most of that editing. It is also, formally, Quiggin's rank-dependent utility made sign-dependent: the same cumulative-weighting device is applied twice, once to the gain branch above the reference point and once, with its own separate function, to the loss branch below it. Remove the reference point and collapse the two branches into one, and you are back to plain rank-dependent expected utility over a single distribution. Keeping these three layers apart — reference dependence, loss aversion, and rank-dependent weighting — clarifies which assumption any given piece of evidence actually tests.
Examples
Valuing a complex gamble with several possible payoffs, the theory ranks the outcomes and weights their cumulative odds, so a clearly inferior bet can never come out ahead.
Insurance and lottery tickets sell to the same people because the inverse-S weighting inflates tiny probabilities — a one-in-a-million jackpot and a one-in-a-million house fire both loom larger than they are.
Almost nobody accepts a coin flip paying 100 for heads and losing 100 for tails; with the theory's loss-aversion coefficient near 2.25, the win has to approach 225 before it tempts.
A cumulative prospect theory model of investors is commonly invoked to explain the disposition effect: holding losing stocks too long because the loss branch is risk-seeking, so people gamble to break even rather than book a certain loss.
Facing surgery with a small but catastrophic complication rate, patients decline more often than expected value predicts; the theory's overweighting of the worst-ranked outcome captures that refusal directly.
First described in Tversky & Kahneman (1992).
Key references
- Brown, A. L., Imai, T., Vieider, F. M., & Camerer, C. F. (2024). Meta-analysis of empirical estimates of loss aversion. Journal of Economic Literature, 62(2), 485-516. doi.org/10.1257/jel.20221698
- Ruggeri, K., Ali, S., Berge, M. L., et al. (2020). Replicating patterns of prospect theory for decision under risk. Nature Human Behaviour, 4(6), 622-633. doi.org/10.1038/s41562-020-0886-x
- Fehr-Duda, H., & Epper, T. (2012). Probability and risk: Foundations and economic implications of probability-dependent risk preferences. Annual Review of Economics, 4, 567-593. doi.org/10.1146/annurev-economics-080511-110950
- Nilsson, H., Rieskamp, J., & Wagenmakers, E.-J. (2011). Hierarchical Bayesian parameter estimation for cumulative prospect theory. Journal of Mathematical Psychology, 55(1), 84-93. doi.org/10.1016/j.jmp.2010.08.006
- Wu, G., & Gonzalez, R. (1996). Curvature of the probability weighting function. Management Science, 42(12), 1676-1690. doi.org/10.1287/mnsc.42.12.1676
- Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5(4), 297-323. doi.org/10.1007/BF00122574