
Behavioral Economics — Prospect Theory I: Reference Points and Loss Aversion
Behavioral Economics — People Respond to Changes, Not to Levels of Wealth
In expected utility theory, utility is a function of final wealth. A person with 100 million won who gains 1 million won and a person with 102 million won who loses 1 million won both end up with 101 million won, so the two are the same. Kahneman and Tversky’s prospect theory abandons this assumption. People evaluate outcomes as gains and losses relative to a reference point, and they feel losses much more strongly than gains of the same size.
1. Three properties of the value function
- Reference dependence: value is defined over changes . The reference point is usually the status quo, but it can be an expectation or a goal.
- Loss aversion: because , the loss curve is steeper than the gain curve. It kinks at the origin.
- Diminishing sensitivity: because , the function is concave for gains and convex for losses. Going from 0 to 100,000 won feels large; going from 1 million to 1.1 million won feels small.
The same change of 50 is worth about 31 as a gain and about −70 as a loss. The kink in the slope at the origin is loss aversion.
2. How large is loss aversion?
Consider a coin toss in which heads wins won and tails loses 100,000 won. If the expected value is positive ( > 100,000 won), a risk-neutral person accepts. In practice most people accept only when approaches 200,000 won. If the value function is close to linear, the condition for accepting is , that is, won. That is why estimates of across many experiments and field studies are mostly around 2.
Explaining this behaviour with expected utility theory requires extreme risk aversion even over small stakes. Rabin showed that a utility function concave enough to reject small bets must make absurd choices over large stakes that no one would make. Risk aversion over small stakes is more naturally explained by the kink at the reference point than by the curvature of utility over wealth.
3. The reflection effect: safety in gains, gambling in losses
| Problem | Option A | Option B | Majority choice |
|---|---|---|---|
| Gains | Gain 3,000 for sure | Gain 4,000 with 80% probability | A (80%) |
| Losses | Lose 3,000 for sure | Lose 4,000 with 80% probability | B (92%) |
B’s expected value, 3,200, exceeds A’s. In the gain problem people chose the sure thing with lower expected value; in the loss problem they chose the gamble with the worse expected value. The concavity in the gain domain and the convexity in the loss domain show through directly. This asymmetry is called the reflection effect.
A real-world example is holding on to shares that are showing a loss rather than selling. Once in the loss domain, a gamble with a chance of recovery is more attractive than locking in a certain loss. This leads to the disposition effect of Chapter 8.
4. Framing: change the reference point and the choice changes
If the reference point depends on how a problem is described, the way it is expressed alone can change choices even when the outcomes are identical. In Tversky and Kahneman’s (1981) “Asian disease” problem, with 600 people at risk:
- “200 people will be saved” vs “a one-third chance that 600 people will be saved” → 72% chose the sure option
- “400 people will die” vs “a two-thirds chance that 600 people will die” → 78% chose the gamble
The two pairs of options have exactly the same outcomes. The first description makes “everyone dies” the reference point and turns the outcomes into gains; the second makes “no one dies” the reference point and turns them into losses. The standard model’s assumption of description invariance (the same problem yields the same choice however it is described) breaks down.
5. Narrow framing
People tend to evaluate decisions one at a time rather than as a bundle. Offered the 50:50 bet above once (gain 150,000 won, loss 100,000 won), a person with rejects it. Viewed as a bundle of 100 such bets, the probability of a loss is tiny and the expected gain is 2.5 million won. The person rejects a bet they would accept as a bundle because they look at it on its own. This is the logic of the explanation of the equity premium puzzle in Chapter 8 (myopic loss aversion).
Check your understanding
If and the value function is linear, what is the largest X for which someone accepts a bet that wins 100,000 won with 60% probability and loses X won with 40% probability? From , won. A risk-neutral person would accept up to 150,000 won. Loss aversion halved the acceptable size of the loss.
References
- Daniel Kahneman and Amos Tversky, “Prospect Theory: An Analysis of Decision under Risk,” Econometrica (1979)
- Amos Tversky and Daniel Kahneman, “The Framing of Decisions and the Psychology of Choice,” Science (1981)
- Amos Tversky and Daniel Kahneman, “Advances in Prospect Theory: Cumulative Representation of Uncertainty,” Journal of Risk and Uncertainty (1992)
- Matthew Rabin, “Risk Aversion and Expected-Utility Theory: A Calibration Theorem,” Econometrica (2000)
OIYO Editorial
Editorial DeskThe OIYO editorial desk researches money, law, lifestyle, and self-understanding topics against primary sources and public statistics. Every piece carries source notes and is reviewed on a regular cycle for accuracy and usefulness.