Prediction Market Handbook

Handbook / Data

How accurate are prediction markets?

Computed November 11, 2026 from 1451 resolved Polymarket markets (Sep 12 to Nov 10, 2026)

Short answer

Close to their prices. One day before close, the favorite won 66.8% of 1187 markets we could score, and in most price bands the share that resolved YES landed within a few points of the price. Contracts priced 40 to 50% resolved YES 46.4% of the time.

Price versus outcome

0%0%25%25%50%50%75%75%100%100%
Filled dots: price one day before close. Rings: one week before. The dashed line is perfect calibration. Bands with fewer than 10 markets are not plotted.

The numbers

Price bandMarkets (1 day)Avg priceResolved YESMarkets (1 week)Resolved YES
0-10%1161.9%1.7%1031.9%
10-20%4215.8%23.8%3619.4%
20-30%9425.4%30.9%7443.2%
30-40%13235.4%37.9%9034.4%
40-50%26545.5%46.4%16346.6%
50-60%22554.1%53.8%15857.0%
60-70%14564.4%69.0%7968.4%
70-80%10074.6%76.0%6484.4%
80-90%4084.6%85.0%2781.5%
90-100%2896.2%96.4%11100.0%

Brier score (lower is better; always guessing 50% scores 0.25): 0.200 one day out, 0.196 one week out. Favorite win rate one week out: 68.6% of 805.

Method

  1. For each UTC day from 2026-09-12 to 2026-11-10, we took Polymarket markets that resolved that day with at least $100,000 in volume, up to the 60 largest, leaving out minute- and hour-long crypto contracts.
  2. For each market we read the hourly price history of the first outcome (YES, or the first team in a head-to-head) and took the last price at or before 24 hours, and 7 days, before the market closed.
  3. We grouped those prices into 10-point bands and compared the average price with the share of markets that resolved in favor of that outcome.

Limits

Markets that opened less than a day (or a week) before closing have no price at that horizon and are left out, which removes many single-game sports markets. High-volume markets are over-represented. Markets inside one event (several games of one series, several rungs of one price ladder) resolve together, so the observations are not independent. Bands with few markets, especially at the extremes, move a lot from one run to the next.

How to read prediction market odds covers what a price means and what fees do to it.