AI-generated research, checked but not peer-reviewed. It may contain errors.

E816　 KILL　 UnderpoweredTested · October 2026

Film-score long shots showed no sign of overpricing

Buying NO on film-score contracts priced 3 to 40 cents for YES, a day before settlement, lost 11.3 cents per contract after costs on 14 films; the interval (-29.8 to +6.1 cents) cannot rule out either sign.

prediction-markets film-scores longshot-bias fans event-contracts

Readable summary Full research record JSON recordFeedback on E816

A US event-contract exchange lists contracts on a film's critic score, such as "score above 80", on the Monday after release. The idea tested: fans overpay for the high-score side. We bought the opposite side (NO) 24 hours before settlement whenever the YES price was between 3 and 40 cents. Only 28 films had settled contracts available, and 14 films produced a qualifying trade. The 15 trades lost 11.3 cents per contract after spread and fee. NO won 80% of the time but cost 91 cents on average. The interval, -29.8 to +6.1 cents, is far too wide to call the effect zero.

Net per contract-11.3 c95% interval -29.8 to +6.1 c, after spread and fee

Films with a qualifying trade1415 contracts; the pass rule needed 40 films

NO hit rate80%average NO price 91 cents

Films with settled contracts28all that were available

01 · The idea being tested

The idea tested: fans overpay for the high-score side of film review score contracts, so the YES side of high thresholds is overpriced.

Why people believe it

The belief is that fans of a film expect a high critic score and buy YES on the high thresholds without checking how rarely a score clears them. If so, high-threshold YES contracts would trade above their true chance, and the opposite side would collect the difference.

Why it would persist The argument, as stated before testing

The belief is that the contracts are many, settle every week and draw small retail stakes, so a professional trader may not find it worth the effort to correct each one.

02 · How it was tested

Each film has a ladder of contracts: "score above 60", "above 70", and so on, settled on the Monday after wide release from the aggregator's published score. 24 hours before settlement we looked at the bid and ask of every rung. Where YES cost 3 to 40 cents (the long-shot side) we simulated buying NO at the price a taker would pay and holding to settlement. We counted the exchange fee. We also compared prices with outcomes in broad price bands, without a trading rule, using the mid price (the midpoint of bid and ask). Intervals are "film-clustered": each film counts as one unit, because the contracts of one film share one score.

Pass required

At least 40 distinct films, mean net at least +1.5 cents per contract, film-clustered bootstrap 95% interval lower bound above 0, mean above 0 without the 3 best films, and the same sign in both chronological halves.

Killed if

Any pass condition not met.

Costs

Net per contract = 1 if the result is NO, minus the NO price, minus the exchange taker fee 0.07 x series multiplier x p x (1 - p), with p the NO price.

Controls

No trading control. A descriptive calibration table compares the YES mid price with the outcome in five price bands, at 24 hours and at 72 hours before settlement (the release Friday).

03 · What happened

Why it was rejected

Only 14 films produced a trade, not the 40 required; the mean after costs was -11.3 cents, and the interval (-29.8 to +6.1 cents) does not exclude zero.

28 films had settled contracts; 15 contracts from 14 films met the rule (YES ask 3 to 40 cents, spread at most 10 cents).
Buying NO lost 11.3 cents per contract after spread and fee, with a film-clustered 95% interval from -29.8 to +6.1 cents.
NO won 80% of the time, while the average NO price was 91 cents. The NO side did not win often enough to cover its price, so the long-shot side showed no sign of being overpriced in this sample.
In the calibration table at 24 hours, all 10 contracts priced 70 to 90 cents and all 29 priced 90 to 100 cents (17 films) resolved YES. At the cells' own average prices, 10 of 10 would happen about 8% of the time and 29 of 29 about 44% (approximate, from 0.78^10 and 0.972^29), so both are consistent with ordinary chance; the comparison is gross of spread and fee and has no trading rule.
At 72 hours before settlement, prices were within sampling error of outcomes in every band.

No sign the long-shot side was overpriced

Mean net result of buying NO on contracts priced 3 to 40 cents for YES, after spread and fee, with a film-clustered 95% interval. The pass bar was +1.5 cents with the interval above zero; the interval spans zero because only 14 films qualified.

Chart data (cents per contract)

	Mean	95% low	95% high
Buy NO, 24 h before settlement	-11.3	-29.8	6.1

NO won less often than it cost

How often the NO side won against the average price paid for it, on the 15 contracts. The price (0.91) is above the win rate (0.80), the gap behind the -11.3 cents.

Chart data (share / price in dollars)

	Value
NO hit rate	0.8
Average NO price paid	0.91

Far fewer films than the rule required

Films with settled contracts, films that produced a qualifying trade, and the number the pass rule required.

Chart data (films)

	Films
Required by the pass rule	40
Films with settled contracts	28
Films with a qualifying trade	14

04 · What it means

Why it failed Interpretation · not established

The sample was too small to say much: 14 films and 15 contracts, with an interval wider than the effects usually looked for. What it does show is that the opposite side lost money, so the belief that fans overpay did not appear in this sample. Prices on these ladders were within sampling error of outcomes at 72 hours.

Lesson

A pre-set sample size is part of the rule: a market with a short history of settled contracts cannot produce 40 independent events, and a result from 14 can be killed by its mean without proving that no mispricing exists. A spotless record such as 29 of 29 is also ordinary chance at those prices, so it says little about a rare miss.

Scope of this result

Covers settled critic-score contracts of one US event-contract exchange available through October 2026 (28 films), the quotes 24 hours and 72 hours before settlement, and a taker strategy held to settlement. It does not cover audience scores, films whose contracts were archived, or any quote outside the hour candles used.

A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.

1. Test design

Pre-registered

Yes: rule and criteria fixed before the test data was examined

Data

Settled film review score contracts (score above N) and their hourly bid and ask quotes — a US event-contract exchange, public anonymous API; hourly quote candles, one settlement per contract; settled contracts available through 2026-10 (28 films); public API, no login

Universe

Every settled contract in the exchange's film review score series whose rules refer to the critics' score with a strike 'above N' (audience score excluded). The score is an integer and the test is strictly above N. Older one-film series are archived and returned no settled contracts.

Signal

At T = settlement time minus 24 hours, take the last hourly bid and ask close within the previous 6 hours; skip a contract with a missing or one-sided quote or a spread above 10 cents. Trade where the YES ask is between 3 and 40 cents.

Execution

Buy NO at 1 minus the YES bid (a taker crossing the spread) and hold to settlement.

Costs

Net per contract = 1 if the result is NO, minus the NO price, minus the exchange taker fee 0.07 x series multiplier x p x (1 - p), with p the NO price.

Controls

No trading control. A descriptive calibration table compares the YES mid price with the outcome in five price bands, at 24 hours and at 72 hours before settlement (the release Friday).

Sample split

Pass rule required the same sign in both chronological halves of the films; the half results are not reported.

Pass criterion

Kill criterion

Any pass condition not met.

Power

14 films against the 40 required. The film-clustered interval is 35.9 cents wide, so half its width is about 18 cents: an effect of that size would be detected only about half the time (rough approximation).

2. Result

Observations

Instruments

Net mean

-11.3

95% interval

-29.8 to +6.1

Median

—

Worst

—

Finding 1

28 films had settled contracts; 15 contracts from 14 films met the rule (YES ask 3 to 40 cents, spread at most 10 cents).

Finding 2

Buying NO lost 11.3 cents per contract after spread and fee, with a film-clustered 95% interval from -29.8 to +6.1 cents.

Finding 3

NO won 80% of the time, while the average NO price was 91 cents. The NO side did not win often enough to cover its price, so the long-shot side showed no sign of being overpriced in this sample.

Finding 4

In the calibration table at 24 hours, all 10 contracts priced 70 to 90 cents and all 29 priced 90 to 100 cents (17 films) resolved YES. At the cells' own average prices, 10 of 10 would happen about 8% of the time and 29 of 29 about 44% (approximate, from 0.78^10 and 0.972^29), so both are consistent with ordinary chance; the comparison is gross of spread and fee and has no trading rule.

Finding 5

At 72 hours before settlement, prices were within sampling error of outcomes in every band.

3. Interpretation

Why it failed

Lesson

Scope

4. Reproducibility

Steps

1) List the settled contracts of the exchange's film review score series through its public API; keep those whose rules mention the critics' score and a strike 'above N'. 2) For each film take the hourly bid and ask candles of each contract and read the last close within the 6 hours before settlement minus 24 hours; skip missing, one-sided or spreads above 10 cents. 3) Where the YES ask is 0.03 to 0.40 buy NO at 1 minus the YES bid, hold to settlement, net of the taker fee. 4) Bootstrap the mean over films. 5) Tabulate outcomes against YES mid price in five bands at 24 and 72 hours.

Grade

COMPLETE