Image: Oniff, CC BY-SA 4.0Why a loss hurts twice as much as a gain
Here is a bet. A fair coin is tossed; heads you receive 150 euros, tails you pay 100 euros. The expected value is 25 euros in your favour, and if you could take the bet a thousand times you would be almost certain to end up richer. Offered once, most people refuse it. Raise the prize to 200 euros and about half accept; many need 250 before they say yes. Daniel Kahneman, who spent decades asking people questions like this, summarized the result in one sentence: losses loom larger than gains, by a factor of about two.
The standard model and its problem
Economics since Daniel Bernoulli in 1738 has explained caution with a concave utility of wealth: each extra euro adds a little less satisfaction than the one before, so a gamble that is fair in money is unfavourable in utility. The trouble, pointed out most sharply by Matthew Rabin in a 2000 paper that is the research spotlight of Unit 14 of the masterclass, is that concavity over small sums implies absurd caution over large ones. A person who turns down the 150-or-100 coin toss at every level of wealth would, if the standard model were right, also turn down a coin toss in which the loss is 1,000 euros and the gain is any amount of money whatsoever. Nobody is like that. Something other than the curvature of utility over lifetime wealth must be at work when people face small gambles.
The value function
Kahneman and Amos Tversky proposed the answer in 1979 in a paper in Econometrica called Prospect Theory, which became one of the most cited papers in economics and earned Kahneman the 2002 Nobel prize in economic sciences; Tversky had died in 1996. Their value function differs from the standard utility of wealth in three ways. First, it is defined over gains and losses relative to a reference point, usually the status quo, rather than over total wealth. Second, it is concave for gains and convex for losses, so that the difference between gaining 100 and gaining 200 feels smaller than the difference between gaining nothing and gaining 100, and the same on the loss side. Third, and this is loss aversion, it is steeper for losses than for gains, with a kink at zero. The function looks like a flattened S that has been bent downwards at the origin.
In their 1992 paper, which added the treatment of probabilities that made the theory complete, Tversky and Kahneman estimated the parameters from the choices of 25 graduate students facing many gambles. The curvature parameter came out at 0.88 for both gains and losses, and the loss aversion coefficient, the ratio of the slope on the loss side to the slope on the gain side near zero, came out at 2.25. That is where the factor of about two comes from. Later studies on larger and more varied samples have found values from about 1.5 to 2.5, with a few lower, and the masterclass uses 2.25 as the agreed figure with that range stated. The Nobel lecture Kahneman gave in 2002 is the best short account of the theory by one of its authors.
The worked number
Apply the 1992 function to the coin toss. The value of gaining 150 is 150 to the power 0.88, which is about 82. The value of losing 100 is minus 2.25 times 100 to the power 0.88, which is minus 2.25 times 57.5, about minus 129. With equal weights on the two outcomes, the gamble is worth 0.5 × 82 minus 0.5 × 129, which is negative, so the person refuses. Now ask what gain makes the gamble just acceptable against a loss of 100. The gain must have a value of 129, and 129 to the power 1/0.88 is about 250 euros. The theory therefore predicts that people will need a gain of roughly two to two and a half times the possible loss before they accept a fifty-fifty bet, which is what the laboratory finds and what most readers will recognize in themselves.
The mug
The most famous demonstration of loss aversion involved no gamble at all. In 1990 Kahneman, Jack Knetsch, and Richard Thaler gave coffee mugs with the university crest, worth about 6 dollars in the bookshop, to half the students in a classroom at Cornell. They then asked the students who had a mug the lowest price at which they would sell it, and the students without one the highest price they would pay. If a mug is worth the same to a person whether or not they happen to hold it, the two prices should be about equal on average. Instead the median owner asked for about 7 dollars and the median buyer offered about 3, a ratio a little above two, and very few trades took place. The students had owned the mugs for a few minutes. Giving one up was a loss, and the loss weighed about twice as much as the gain of acquiring one. The result is known as the endowment effect, it has been replicated many times with many goods, and it is one of the reasons housing markets freeze when prices fall: sellers who would lose relative to their purchase price refuse to sell at the market price.
Where it turns up
Once you know the shape of the value function you see it everywhere. Taxi drivers who set themselves a daily earnings target work fewer hours on busy days, when they reach the target early, and longer on slow days, which is the opposite of what a profit-maximizing driver would do and exactly what a driver who hates falling short of a reference point would do. Professional golfers, in a study of more than two million putts by Devin Pope and Maurice Schweitzer published in the American Economic Review in 2011, hole putts for par about two percentage points more often than otherwise identical putts for birdie, because missing a par putt is a loss relative to the reference point of par and missing a birdie putt is only a forgone gain. Investors sell shares that have risen and hold shares that have fallen, the disposition effect, because realizing a loss hurts more than the equivalent gain pleases. Shops advertise a cash discount rather than a card surcharge, since the second is a loss. A supermarket that charges for plastic bags cuts their use far more than a supermarket that offers the same few cents as a reward for bringing your own.
What it does not say
Loss aversion is a regularity, with a measured size and a measured range, and a student should treat it as one. It is not a claim that people are irrational, since a person who refuses small gambles and keeps a mug is making a choice that is consistent with a stable preference, merely a preference with a kink. It is not a universal constant; the coefficient is smaller for experienced traders, for small sums that people treat as part of a stream, and in some studies for decisions made on behalf of others. And the reference point, which decides what counts as a loss, is itself something the theory has had to work out, which Unit 14 discusses with the expectations-based account of Botond Kőszegi and Matthew Rabin. The unit builds the value function with the 1992 parameters, works through the Allais paradox and the mug experiment, and then hands the idea on to Unit 16, where loss aversion helps explain why shares have paid so much more than bonds for a century.


