ChanceIndex

Track Record

How accurate are prediction markets? · Updated daily

A prediction market is only as useful as it is accurate. So we check. Every market we track is recorded when it resolves, alongside what it priced the outcome at one day, one week, and one month before. This page is the running scorecard.

The headline test is calibration: of all the outcomes a market priced at around 70%, roughly 70% should actually happen. A well-calibrated source sits on the diagonal below — not because every call is right, but because its confidence matches reality over many events. An outcome priced at 30% that happens is not a miss; it is exactly what "30%" means.

Calibration

0%0%25%25%50%50%75%75%100%100%Priced 2% → happened 7% (14)Priced 16% → happened 11% (9)Priced 25% → happened 0% (3)Priced 34% → happened 42% (12)Priced 45% → happened 54% (13)Priced 55% → happened 63% (8)Priced 64% → happened 83% (6)Priced 75% → happened 100% (7)Priced 84% → happened 100% (4)Priced 99% → happened 100% (3)Market-implied probability

0.153Brier score (lower is better; 0.25 is a coin flip)

79resolved markets at this horizon

270 resolved markets with a tracked price history. 100.0% of the 13 priced above 90% a day before came true.

By category

CategoryMarketsBrier (7d)
Sports570.176
Politics7—
Geopolitics4—
Economy4—
Elections2—
Crypto2—

Recent resolutions

Methodology

We include markets with at least $10,000 in volume that had a price at each horizon shown — one-day novelty markets that resolve trivially would flatter the numbers without measuring anything. Probabilities are the market-implied prices from our sources. Calibration groups outcomes into ten-point buckets; the Brier score is the mean squared error between the priced probability and the 0/1 result.

An event can have many outcomes (a race with forty candidates); to avoid stuffing the numbers with dozens of near-certain "no"s, each event contributes exactly one observation — the outcome with the highest trading volume. This rule is fixed in advance, so there is no room to pick which outcomes look good after the fact.

Each horizon is measured on a different set of markets: a market that only existed for three days appears in the 1-day figures but not the 7- or 30-day ones. So the horizons are not directly comparable — a lower Brier at one horizon reflects which markets had that much history, not that markets get more accurate further out.

Past calibration does not guarantee future accuracy. Nothing here is financial advice.