How Accurate Are Prediction Markets? Real Data from 486 Predictions
A prediction market price is meant to be a probability: a "Yes" trading at 70 cents says the crowd thinks there is about a 70% chance it happens. We checked how often that held up, using every prediction FakeBet players made with fake money on real markets between 25 August and 2 November 2026 — 486 predictions on 322 markets that have since resolved. Prices come from Polymarket's live order books. FakeBet is not affiliated with Polymarket.
The short answer
- Overall, the sides players backed came true 48.4% of the time. Their average price was 47.0%.
- Favourites — sides priced at 50% or more — came true 71.4% of the time (162 of 227) at an average price of 70.2%.
- Underdogs — sides priced under 50% — came true 28.2% of the time (73 of 259) at an average price of 26.6%.
- Of the 76 predictions priced under 10%, 1 came true. Of the 33 priced at 90% or more, all 33 did.
How often a price came true
| Price of the side | Predictions | Came true | Hit rate | Average price |
|---|---|---|---|---|
| 0–10% | 76 | 1 | 1.3% | 5.5% |
| 10–20% | 30 | 5 | 16.7% | 14.2% |
| 20–30% | 29 | 8 | 27.6% | 25.8% |
| 30–40% | 39 | 14 | 35.9% | 36.7% |
| 40–50% | 85 | 45 | 52.9% | 45.5% |
| 50–60% | 70 | 41 | 58.6% | 54.5% |
| 60–70% | 58 | 37 | 63.8% | 64.7% |
| 70–80% | 35 | 27 | 77.1% | 74.6% |
| 80–90% | 31 | 24 | 77.4% | 85.2% |
| 90–100% | 33 | 33 | 100.0% | 94.5% |
| All | 486 | 235 | 48.4% | 47.0% |
"Price of the side" is the price of whatever was backed at the moment of the bet, so a "No" bought at 70 cents sits in the 70–80% row. "Hit rate" is how many of those predictions came true. A perfectly calibrated market would put the hit rate right on top of the average price in every row.
In five of the ten rows the two numbers are within three points of each other. The rows that drift further — 40–50% came true 52.9% of the time, 80–90% only 77.4% — hold 31 to 85 predictions each. At that size, a handful of results either way moves a row by several points, so read single rows as noise and the whole table as the signal: prices and outcomes climb together, almost one for one.
Favourites and longshots
Split down the middle, the market's favourites came true 71.4% of the time against an average price of 70.2%, and its underdogs 28.2% against 26.6%. Both landed slightly above their prices — about a point for favourites and under two for underdogs. With 227 and 259 predictions, that is well inside what chance alone produces.
The extremes tilt the other way. Sides under 10 cents came true once in 76 tries, where their prices added up to about 4 wins; sides at 90 cents or more came true 33 times out of 33, against about 31. That is the favourite–longshot pattern our parlay data found for long-odds sports bets — cheap outcomes paying less often than their price says. Both rows are small, so treat it as a hint, not a finding.
One reason the prices behave: they are midpoints of live order books, the halfway point between the best offer to buy and the best offer to sell. There is no bookmaker margin built into them, so a fair price and the quoted price are the same number.
What it means if you play
- A price is a probability, not a tip. A side at 80% loses one time in five. In the 80–90% row it lost 7 times out of 31.
- Cheap sides are cheap for a reason. 76 predictions priced under 10 cents produced one winner.
- The edge, if there is one, is small. Nothing in the table says "the crowd is reliably wrong here". Beating these prices means knowing something the order book doesn't.
- Practise on the real thing. On FakeBet's prediction markets you trade the same markets at the same prices with fake money, and every position settles on the real outcome. Browse all markets or start with sport or crypto.
How we counted
Every prediction-market bet placed on FakeBet and settled won or lost before 3 November 2026, 00:00 UTC — 788 bets from 237 players, placed from 25 August 2026, the last one settled on 2 November. Left out: test accounts; 82 bets recorded as lost because a player topped an empty wallet back up (a forfeit, not a result); 10 void bets on markets that resolved 50/50; and every bet still open. The same player backing the same side of the same market more than once made one prediction, so it counts once, at the price of the first bet — that leaves 486 predictions on 322 markets, of which 151 resolved "Yes" and 171 "No". A prediction "came true" when the side backed is the side the market resolved to. These are players betting fake money on markets they chose, which is not the same as a random sample of every market listed. The figures come from one read-only query that returns totals only — no names, no player ids.
FakeBet is a free demo platform. No real money is wagered or won. Market prices and outcomes come from Polymarket; FakeBet is not affiliated with, endorsed by or connected to Polymarket. If real-world gambling affects your wellbeing, help is available at Gamblers Anonymous.
