Limitless Contract Study Finds Close Calibration in Nine of 10 Bands

Prices in a study of 2,681 resolved Limitless contracts aligned closely with outcomes in most probability bands, though a sports-heavy sample and thin trading limit what the results establish.

Published Thursday, Oct. 22, the analysis used public Limitless API data from October 2026. It counted 556 open markets in its snapshot and separately tested 2,681 resolved contracts, comparing each contract’s price one day before scheduled resolution with its outcome. About 85% of the resolved sample consisted of sports contracts.

In nine of 10 probability bands, the difference between prices and the observed share of “yes” outcomes was within four percentage points. Of 165 contracts priced from 70% to 80%, 74.5% resolved “yes.” Among 113 contracts priced at 90% or higher, 97.3% did.

A binary event contract generally pays $1 if an event occurs and $0 if it does not. Its price can therefore serve as an implied probability, but calibration describes how prices compare with outcomes across groups of contracts. It does not show that an individual contract priced at 75% will resolve “yes” three times out of four.

The 50%–60% band had a 48.8% “yes” rate. About one-third of contracts in that range, mostly smaller sports markets, still showed their initial 50% listing price one day before scheduled resolution. That pattern suggests some quotes may not reflect active trading close to resolution; the larger error in the band was tied mainly to these thinly traded markets.

The results do not establish that sports-heavy performance will carry over to economic or political events. Limitless partially funded the analysis. The underlying list of 2,681 contracts and enough detail about the API queries to reproduce the calculations were not published with the findings, limiting review of the aggregate results.

The findings add a new sample to previous work on event-market accuracy. That review and the Limitless analysis examined different markets and samples.

Other studies measure different questions

Separate research on Kalshi examined U.S. interest-rate forecasts rather than sports contracts. In its sample since 2022, the median and modal federal funds rate forecasts had zero average absolute error by the day before Federal Open Market Committee (FOMC) meetings. The study also found a statistically significant improvement over federal funds futures, a benchmark used to gauge expected interest rates.

The September 2024 meeting drove much of the difference: Kalshi placed more weight on a 50-basis-point rate cut, which occurred. The working paper’s findings are preliminary and do not represent the Federal Reserve Board’s views. The results are specific to the study’s forecast measures and meeting sample.

Election-market research has produced mixed comparisons with polls. One study compared Iowa Electronic Markets forecasts with 964 polls across five U.S. presidential elections from 1988 through 2004. The market was closer to the final two-party vote split in 74% of the comparisons.

A second study of the same elections found poll-based forecasts performed better after adjusting polls to predict Election Day results. Together, the studies do not establish that either market prices or polls are consistently more accurate across elections; they compare particular methods and historical samples.

Market-implied probabilities also need interpretation. Risk premiums can make them differ from real-world probabilities, so a quoted price is not an objective measure of certainty. Across these examples, the evidence covers distinct tests: grouped contract outcomes on Limitless, rate forecasts on Kalshi and election-market prices compared with polls. The results describe those samples, rather than a single accuracy rate for prediction markets overall.