How accurate are prediction markets?
By Faizan Ahmad. Updated .
Close to their stated chances: Tradytics' study of its archive of 5.95 million Kalshi and Polymarket markets found that a 2% chance came true 2.6% of the time, 48% came true 48.2% and 97% came true 96.6%, that things priced as near-certain still failed about 1 in 15 times, and that chances on S&P 500 and Nasdaq closes match what same-day options price.
Stated chance against what happened
Polymarket finance questions, each read a day before it actually closed, on fresh prices only, leaving out questions already decided at 99% or 1% and questions that can settle early:
| The chance said | How often it happened |
|---|---|
| 2% | 2.6% |
| 8% | 8.2% |
| 48% | 48.2% |
| 78% | 78.2% |
| 88% | 84.2% |
| 97% | 96.6% |
Kalshi's S&P 500 4pm "or above" questions, read an hour before the close, land the same way: 4% happened 3.1%, 15% happened 13.0%, 45% happened 44.6%, 76% happened 80.4% and 96% happened 95.8%.
Where they miss
- Sure things. Questions priced 90 to 98% (94.4% on average) happened 93.3% of the time: about 1 in 15 failed. Unlikely ones priced at 4.9% happened 5.4%.
- Daily direction. Polymarket's daily "up or down" questions picked the right side 56% of the time and scored worse than calling every day 50/50.
Index closes: the same as options
On S&P 500 and Nasdaq 4pm closes since March 31, 2025, people's chance on each strike sat an average 4 to 6 points from the chance same-day options put on it, and 56 to 72% of strikes were within 5 points. The range people expect for the close is 1.01 to 1.03 times the range of same-day S&P options. A model built from the index's own price history did not beat people's chances on S&P closes at any time of day. Whoever quotes these questions can hedge in SPX options, so the two stay tied.
Earnings beats
On 1,307 Polymarket "Will X beat quarterly earnings?" questions with a price a day before, companies beat 73.9% of the time and the average chance was 72.4%. The chances also sorted which companies would beat:
| Chance a day before | Questions | Average chance | Beat |
|---|---|---|---|
| Under 50% | 225 | 29.5% | 34.2% |
| 50 to 70% | 208 | 61.0% | 64.9% |
| 70 to 85% | 357 | 78.4% | 79.8% |
| 85 to 95% | 414 | 90.3% | 89.9% |
| 95% and up | 103 | 96.6% | 94.2% |
How Tradytics shows this
Tradytics Probabilities reads every live finance question on Kalshi and Polymarket and shows the chance of each outcome, where it is likely to land, the index close questions scored against the last 30 sessions, and each earnings question with the stock's past report-day moves. The S&P 500 and Nasdaq are free at tradytics.com/probabilities; Plus and Pro show every stock and report.
Common questions
Are Kalshi's chances accurate?
On S&P 500 closes, yes: an hour before the close, 15% happened 13.0% of the time and 96% happened 95.8%, and the chances match same-day options.
What does a 95% chance on Polymarket really mean?
About what it says, a little less. Questions priced 90 to 98% failed about 1 in 15 times.
Are prediction markets more accurate than options on the S&P 500?
No, and not less. On index closes the two price the same chances.