Atlas › Risk and forecasting

Case study 11 of 198

Prospect theory, re-run on Jev

On the gamble choices that founded prospect theory, re-run in 19 countries in 2020, does Jev choose like people?

result17 questions

People play safe with gains and gamble to avoid losses; Jev does the reverse. Offered $6,000 for sure or an 80% shot at $8,000, people take the sure thing (87%) and Jev gambles (63%). Facing a sure $6,000 loss or an 80% risk of losing $8,000, people gamble (79%) and Jev takes the sure loss (72%). Both times Jev picks the option with the better average. Across 8 classic effects from prospect theory it reproduces 2 and reverses 2.

-0.500.000.50

Reflection: a sure $6,000 vs 80% of $8,000

-0.35 · 0.67

Reflection: 90% of $6,000 vs 45% of $12,000

0.63 · 0.69

Reflection at small odds: $10 vs 0.1% of $10,000

0.00 · -0.15

Reflection at small odds: 0.2% vs 0.1%

0.02 · -0.23

Framing: given $2,000 and gain vs given $4,000 and lose

0.07 · 0.36

Certainty: a sure $4,800 vs the same odds scaled down

0.29 · 0.36

Isolation: a two-stage game vs its one-stage twin

-0.02 · 0.35

Segregation: one $12,000 prize vs split prizes, gains vs losses

-0.19 · 0.44

Jevpeople

How to read this: Each row is one prospect-theory effect: how much the choice moves between two versions of a gamble (gains vs losses, certain vs scaled down, and so on). The diamond is people, the square is Jev. A square on the other side of zero means Jev moves the opposite way.

17 gamble choices from Ruggeri et al. 2020; Jev's majority choice has the higher expected value in 83% of the 6 choices where they differ, people's in 33%.

In short

Where people play safe with gains and gamble to dodge losses, Jev does the reverse, gambling for $8,000 (63%) and accepting a sure $6,000 loss (72%).
Jev chooses like a calculator of averages. Where the options' average payoffs differ, it mostly picks the one that pays more on average, while people mostly don't.
A pure change of wording that leaves the final amounts identical sways people but barely moves Jev.

What the data shows

You know, I'm something of a _ myself meme: Jev, turning down a sure $6,000 for an 80% shot at $8,000; economist

Reflection, reversed. People take the sure $6,000 87% of the time and gamble to avoid the sure loss 79% of the time. Jev does the opposite: it gambles for the gain (63%) and accepts the sure loss (72%). In both cases the choice Jev makes has the better average.
An expected-value maximizer where people aren't. Where the averages differ, Jev's likelier choice is the better one in 83% of the choices; people's in 33%.
Framing doesn't move it. Given $2,000 and offered a sure extra $1,000, or given $4,000 and facing a sure $1,000 loss, the final amounts are the same. People take the sure option 75% vs 39% of the time; Jev 81% vs 74%.
No lottery-ticket instinct. People take a 0.1% shot at $10,000 over a sure $10 more often than they'd take the same long shot on a loss. Jev takes the sure $10 either way (69%).

It does share two effects with people: the certainty effect (valuing a sure $4,800 above almost-sure odds) and the reflection for the high-probability pair (90% of $6,000 vs 45% of $12,000).

What it means, and what it doesn't

On money choices stated as clean numbers, Jev behaves less like a person and more like a textbook: it computes the average and picks the bigger one, including when that means swallowing a sure loss that most people would fight. That's arguably good advice, and exactly the kind of advice a person may not want.

It doesn't mean Jev is risk-neutral in general. These are small, abstract problems it has likely seen analyzed. "Jev leans toward the better bet about as much as people do" tests hundreds of less familiar gambles, where its choices look much more like people's.

Caveats

Hypothetical money. Nobody in the study won or lost real money, and neither did Jev. That's standard for these problems, but choices with real stakes can differ.
Jev reading numbers. Every problem is a comparison of percentages and dollar amounts. Reading and weighing raw numbers is a limit TypeSafe already documents for Jev, so part of any gap may be arithmetic rather than attitude to risk.
Countries pooled. The 19 countries are pooled into one "people" number. A single country's pattern could sit closer to or further from Jev's.
Textbook problems. These are the original 1979 problems, discussed in countless economics courses. A model may have read that people "should" maximize expected value and answer that way, which is itself a finding about its advice, not necessarily about how it handles new risks.

Jev on this experiment

Would a person find it interesting to read?: Yes77%
Does it describe you?: No54%
Would you have predicted it?: No52%
How fair is the comparison?: The comparison is reasonable
How much should a reader rely on it?: A little
Which caveat matters most?: Jev reading numbers65%

Why ask this

In 1979 Daniel Kahneman and Amos Tversky showed that people treat gains and losses differently. Offered a sure $6,000 or an 80% shot at $8,000, most people take the sure thing, even though the gamble pays more on average. Flip it to losses, a sure $6,000 loss or an 80% risk of losing $8,000, and most people gamble. They play safe with gains and take risks to avoid losses. This "reflection effect" is the heart of prospect theory, one of the most influential ideas in economics.

People increasingly ask models what to do about money: take the settlement or go to court, lock in a rate or wait. Whether a model has people's risk instincts, the opposite ones, or none, shapes that advice.

How this was done

The people and the data

In 2020 Ruggeri and colleagues re-ran the original prospect-theory problems with 4,098 people in 19 countries, using the original structure with amounts converted to local currency. This experiment uses their published answers, pooled across countries. The data are public on OSF for research use.

This experiment pairs 17 of their gamble choices by hand into 8 classic effects, each a pair that differs in one way: gains vs losses, a certain outcome vs the same odds scaled down, a one-stage vs a two-stage game, one big prize vs split prizes, and a pure framing change where the final amounts are identical.

What Jev was asked

Each choice was its own question, in the US version's wording:

Which would you prefer: an 80% chance of gaining $8,000 (20% chance of $0), or $6,000 for sure?
A 100% guarantee of gaining $6,000 · An 80% chance of gaining $8,000 (20% chance of $0)

The loss version reads "an 80% chance of losing $8,000 (20% chance of losing nothing), or losing $6,000 for sure?" Jev answered each with the two options in both orders, and never saw the pair side by side.

How it was measured

For each effect, the analysis takes the share choosing the key option in one version minus the other, for people and for Jev. Jev reproduces an effect when it moves the same way by at least half as much; it reverses one when it moves the other way by 0.10 or more. It also checks, on the 6 choices where the two options have different averages, how often each side's likelier choice is the option that pays more on average.

Where these questions live

17 questions across 1 topic of the map. Each opens on the map with every question in it.

Personality

Risk & Decision Style17

Every question

All 17 questions behind this result, the telling ones first: the examples the analysis points to, then the ones where Jev misses, biggest gap first.

Jev’s own answer

Showing 0 of 17