Polymarket efficiency: two tests, updated monthly
Article details
AI-assisted drafting, research and editing. Published by OmniOS OÜ. How we use AI
The same slope, month by month
A dollar on every longshot
Key findings
- In the 61 baskets where every outcome had a bid, the best bids summed to a median of 87.1¢, and none came to more than 100¢. Counting missing bids as zero, 3 of 555 baskets did before fees and none after.
- The favourite-longshot slope is 0.91 across 3,663 resolved contracts in 1,298 events, January 2025 to September 2026. Its 95% bootstrap range is 0.84 to 0.97 clustered by event and 0.84 to 0.98 clustered by end date. Both ranges are below 1, the value for prices that match outcomes; above 1 would point to overpriced longshots.
- A dollar on every contract priced at 20¢ or less a week before its scheduled end came out at +23.5¢ per dollar after the stake and before fees, across 2,567 contracts (95% range +6.67¢ to +40.29¢). This is payoff minus a public price, not a trading result.
Bid sums: one order-book snapshot, 15 October 2026, 13:14 UTC. Slope and dollar figures: resolved political contracts priced a week before their scheduled end, January 2025 to September 2026.
| Test | Sample | Result | Range and detail |
|---|---|---|---|
| Bid sum, every outcome had a bid | 61 baskets | Median 87.1¢ | 0 above 100¢ before fees, 0 after; highest 99.7¢ |
| Bid sum, missing bids counted as zero | 555 baskets | Median 90.8¢ | Most active tenth (56 baskets) 96.25¢; 3 above 100¢ before fees, 0 after; highest 101.5¢ |
| Favourite-longshot slope, pooled | 3,663 contracts in 1,298 events | 0.91 | 95% bootstrap range 0.84 to 0.97 clustered by event; 0.84 to 0.98 clustered by end date (328 dates) |
| A dollar on every longshot, 20¢ or less | 2,567 contracts | +23.5¢ per dollar | 95% bootstrap range +6.67¢ to +40.29¢, by event; before fees |
| A dollar on every favourite, 80¢ or more | 268 contracts | −2.92¢ per dollar | 95% bootstrap range −7.47¢ to +1.6¢, by event; before fees |
The two bid-sum counts have different denominators and are never pooled. 1,141 of 1,696 baskets tied to the 2026 US midterms were left out first. After-fee counts subtract each market's published taker fee on one share; book depth, minimum order sizes, rebates and gas are not included, so a sum above 100¢ is not by itself an executable profit. Read the definitions.
The series tracks two questions: how bids add up across the outcomes of an event, and how prices compare with later results. The basket snapshot is dated 15 October 2026, 13:14 UTC; the resolved-contract analysis covers January 2025 to September 2026.
These are related tests with stated implementation gaps, rather than a matched comparison of venues. Read the definitions and limitations.
Related Vultax research: Polymarket price calibration and deadline-market accuracy.