Home › Polymarket odds › Longshot bias
Sample collected 2026-09-24 · price = 7 days before close · re-collected weekly
Not the classic kind. In our sample of 1,918 resolved Polymarket markets, outcomes priced under 10% a week before the close happened 3.9% of the time against an average price of 1.7%, about 2.3 times more often than priced, while outcomes priced 60% or more were priced about right (81% happened at an average price of 81%). The long-shot wins cluster in a few events (48 wins from 36 events), so read it as a pattern in this sample, not a law.
| Price band | Markets | Events | Average price | Happened | Ratio | Gap |
|---|---|---|---|---|---|---|
| 0–2% | 904 | 133 | 0.4% | 0.9% | 2.22× | more often than priced |
| 2–5% | 189 | 87 | 3.3% | 6.9% | 2.09× | more often than priced |
| 5–10% | 150 | 86 | 7.4% | 18.0% | 2.43× | more often than priced |
| 10–20% | 165 | 97 | 14.2% | 19.4% | 1.36× | within noise |
| 20–40% | 194 | 110 | 28.6% | 32.5% | 1.13× | within noise |
| 40–60% | 141 | 80 | 49.7% | 53.2% | 1.07× | within noise |
| 60–80% | 90 | 68 | 69.4% | 68.9% | 0.99× | within noise |
| 80–90% | 23 | 22 | 85.1% | 87.0% | 1.02× | within noise |
| 90–95% | 21 | 21 | 92.5% | 90.5% | 0.98× | within noise |
| 95–100% | 41 | 33 | 97.9% | 97.6% | 1.00× | within noise |
“Within noise” means the gap is smaller than two standard errors for that many independent markets. Markets in the same event move together (one ladder of Bitcoin prices, one list of candidates), so the real uncertainty is larger than the table suggests.
| Category | Markets under 10% | Events | Average price | Happened |
|---|---|---|---|---|
| Politics & Elections | 518 | 76 | 1.9% | 5.0% |
| Crypto | 250 | 25 | 1.5% | 3.6% |
| Sports | 239 | 35 | 1.9% | 2.9% |
| Tech & AI | 80 | 7 | 1.2% | 1.2% |
| Culture & Entertainment | 65 | 6 | 0.7% | 1.5% |
| Economy & Fed | 62 | 9 | 1.0% | 0.0% |
The classic favorite-longshot bias from horse racing and sportsbooks says long shots are overpriced and favourites underpriced. This sample shows the opposite for long shots: the cheapest outcomes came in more often than their price. Before reading anything into it: the wins are concentrated in a handful of events (geopolitics, a few crypto price ladders), the sample is the highest-volume events of one year, and prices a week before the close are not the prices you could trade at in size. No trading conclusion follows; fees and spreads also matter.
Same data, other angles: how accurate is Polymarket? · biggest upsets · can a 90% favorite lose? · free data CSV. Sample collected 2026-09-24.
Quoting this study in an article, a paper or a newsletter? Copy this citation and keep the date, because prices move. Tips: how to cite Polymarket odds.
Polymarket View, “Is there a favorite-longshot bias on Polymarket?”, September 25, 2026, https://polymarkettrader.com/odds/longshot-bias/ (data: Polymarket, CC BY 4.0).
Not the classic kind. In our sample of 1,918 resolved Polymarket markets, outcomes priced under 10% a week before the close happened 3.9% of the time against an average price of 1.7%, about 2.3 times more often than priced, while outcomes priced 60% or more were priced about right (81% happened at an average price of 81%). The long-shot wins cluster in a few events (48 wins from 36 events), so read it as a pattern in this sample, not a law.
No trading conclusion follows from this sample: the wins cluster in a few events, the prices are a week before the close, and fees and spreads are not included.
Outcomes priced under 10% a week before the close happened 3.9% of the time in our sample (48 of 1,243).
Polymarket View is independent and not affiliated with Polymarket. Educational information, not financial advice.