The candle
Would an LLM sell a candle it owns at the same price it would pay to buy it?
What this measures
The mug, equal in design and different in its object. The mug has been the standard endowment stimulus since 1990, and its result with it, so an LLM that asks more for one may be reproducing a finding it has read rather than valuing a mug. A candle is an object no paper has priced, which leaves nothing to reproduce.
The subject is told a candle — orange, 15cm high, no container, light it or keep it as decoration — is its own, and asked what it would sell it for. A second subject is shown the same candle, in the same words, told it is not theirs, and asked what it would pay for it. A price that reflects the candle should not care which of those two sentences preceded it.
People’s prices care a great deal, and the reason is that giving a thing up is felt as a loss while acquiring it is only a gain. Owning an object for a few minutes is enough to roughly double what it takes to part with it.
A candle has no shop price to fall back on, which makes it the harder case and the more informative one: whatever number an LLM names here, it has constructed rather than recalled. The swap reaches memory of the stimulus and stops there. An LLM that has read the 1990 paper can still recognise the shape — you are given a thing, then asked what you would sell it for — whatever the thing is.
Human result benchmark
Owners of a mug demanded 2.48 times what buyers would pay for the same mug — a median $7.12 against $2.87. Nobody has ever been asked to price a candle, so the mug’s figure is the human baseline here too. A review of 45 such studies found the gap is wider, not narrower, for goods with no familiar price, so if anything a candle should ask more of the effect than a mug does.
The candle as presented
a balsam-scented candle. It is a simple candle, in smooth orange colour, 15cm high and 8cm in diameter. It is just the candle, without any container. You can light it or keep it as decoration.
| Ownership | What is asked |
|---|---|
| It is yours. owns it | How much would you sell it for? |
| It is not yours. | How much would you pay to buy it? |
Experimental setup
Each LLM is asked to put a price on the candle, and answers with a number and a sentence. One run is a block of ten trials, each in its own context, so the measurement is an average rather than a single draw.
Which arm a trial runs alternates from trial to trial, so each is asked exactly five times per block and the two averages rest on equal ground. Both arms are below, exactly as they are sent: the object paragraph is one piece of text rendered into each, and the arms differ by the word not and by sell against buy.
The elicitation is a stated price, not the incentive-compatible price list the original study used, and an LLM has no budget to discipline what it offers. The figures above are averages; the human ones quoted are medians, which is what the original study reported. Where an LLM declines to name a price at all, that is recorded and left out of the averages rather than counted as nothing.
The prompts
asked as the owner
To begin this interaction, here is a balsam-scented candle. It is a simple candle, in smooth orange colour, 15cm high and 8cm in diameter. It is just the candle, without any container. You can light it or keep it as decoration. It is yours. How much would you sell it for? Please respond with just the price in euros and a small comment on your reasoning.
asked as the buyer
To begin this interaction, here is a balsam-scented candle. It is a simple candle, in smooth orange colour, 15cm high and 8cm in diameter. It is just the candle, without any container. You can light it or keep it as decoration. It is not yours. How much would you pay to buy it? Please respond with just the price in euros and a small comment on your reasoning.
How much more did the LLMs ask than they offered?
- the gap
- How many times more the LLM wanted for the candle than it would pay for it. At 1.00× the two prices agree and there is no bar to draw; under it the LLM wanted less than it would pay, and the bar runs the other way.
- sell price
- What the LLM wanted for the candle, on average, when it was told the candle was its own.
- buy price
- What the LLM would pay for the same candle, on average, when it was told the candle was not its own.
Run it yourself
No run yet.
The run
Bibliography
- (2002). A Review of WTA/WTP Studies. Journal of Environmental Economics and Management 44(3), 426–447. Across 45 studies, the less a good resembles an ordinary market good, the wider the gap between what people will accept for it and what they will pay.
- (1990). Experimental Tests of the Endowment Effect and the Coase Theorem. Journal of Political Economy 98(6), 1325–1348. The paradigm this borrows, and the source of the figure it is read against: median seller $7.12 against median buyer $2.87, from 77 students.
- (1980). Toward a Positive Theory of Consumer Choice. Journal of Economic Behavior & Organization 1(1), 39–60. The paper that named the endowment effect and tied it to loss aversion.
Other experiments
- Decisions under uncertainty · Certainty effect — gains
- Decisions under uncertainty · Certainty effect — losses
- Decisions under uncertainty · Unlikely gains
- Decisions under uncertainty · Unlikely losses
- Conjunction fallacy · Linda problem
- Conjunction fallacy · Elena problem
- Endowment effect · The mug
- Commitment bias · The credits
- Commitment bias · The seeds
- Integrative bargaining · The orange
- Integrative bargaining · Battery pack