Full-set arbitrage on multi-outcome markets is mostly gone
Of 1,100 neg-risk events scanned, 30 showed YES asks summing below $1 with real depth, but nearly all were incomplete events. The only real remainder was $47 net on $5,837 tied up for three weeks, 0.8%.
What we asked
In a multi-outcome event where exactly one outcome wins, the YES prices should sum to $1. If the lowest asks (the cheapest prices anyone is selling at) of all YES outcomes sum to less than $1, buying one share of each locks in a payout of $1 at resolution. We call this a full set. A neg-risk event is Polymarket's structure for mutually exclusive multi-outcome markets. The same logic applies to NO shares: for n outcomes, buying one NO of each pays n-1, so asks summing below n-1 would be an opportunity. We asked whether such opportunities exist, whether they survive fees, and whether they are large enough to matter.
Data
A live scan on 14 October 2026 at about 20:40 UTC, repeated 10 minutes later: 2,100 active events, of which 1,100 were neg-risk with at least two markets, and 24,122 order books. The taker fee comes from Polymarket's market configuration: 0.04 x p x (1 - p) per share, where p is the price. The Gamma API shows 1000 basis points, which is not the actual rate.
Result
- YES asks summing below 1: 46 to 47 events, of which 30 had an edge above 1 cent and depth of at least 10 sets.
- NO asks summing below n-1: 16 to 17 events, with an edge above 1 cent in 3 to 4. Each set needs n x $1 of capital, and the edge is hundredths of a percent.
- Most large edges (YES sums of 0.44 to 0.93, in events such as a sports leader, a UFC next fight, the Grammys, Billboard and governors) were in events flagged negRiskAugmented, with unfilled outcomes: only 3 to 9 of 14 to 50 markets were active. The remaining probability sits in outcomes not yet listed, so this is not arbitrage.
- Complete events (not augmented, all markets active) such as OpenAI IPO cap (sum 0.974), BTC vs gold vs S&P (0.980) and Fed January (0.994) left 0.4 to 0.6 cents per set after fees, with resolution in 2027 to 2028. Net result: zero.
- The only real residue was two-party US House races (Republican and Democrat outcomes plus an inactive placeholder "another party"), resolving 4 November:
| Event | Sets | Sum of prices | Edge | Fee | Net, $ | Capital |
|---|---|---|---|---|---|---|
| TX-25 | 2,000 | 0.980 | 2.0c | 0.32c | 33.1 | $1,960 |
| WI-05 | 2,000 | 0.993 | 0.7c | 0.24c | 9.7 | $1,985 |
| TX-26, TX-04, GA-10, NY-24, MN-06, IA-01, NY-23 | 10 to 500 | 0.990 to 0.997 | 0.3 to 1.0c | 0.1 to 0.8c | 3.6 | $1,400 |
In total, $47 net on $5,837 frozen for three weeks, or 0.8%.
Why the edge disappears
In complete events, the asks sum to nearly $1 and the fee takes what is left of any difference: after fees the margin was zero. Where the sum looks low, the event is usually incomplete, and the missing probability belongs to outcomes not yet listed. A sum below 1 there reflects the unlisted outcomes, not a free payout. We did not study who closes these gaps or how quickly.
Limits of this measurement
- This is a single scan repeated once 10 minutes later, on one evening, not a continuous record. Opportunities that appear and close within minutes, or at other times of day, are not captured.
- The House-race residue carries real risk. The placeholder "another party" outcome could be activated, and if an independent candidate won, both YES positions would lose. For augmented events, the oracle (UMA) might also interpret "another party" differently at resolution.
- Depth was read from displayed books and changes as orders are taken. The cost of tying up capital for three weeks is stated, not valued.
- The fee formula comes from the market configuration at scan time, and fees can change.
- The result is consistent with an earlier, separate measurement that found only small amounts available relative to thousands of dollars of capital; this scan repeats the finding rather than extending it.
What it means for a trader
In the complete events we scanned, buying every outcome left nothing after fees. The apparent large gaps were events with outcomes not yet listed. The remaining two-party House races offered about 1% over three weeks, with a placeholder risk that could turn the full set into a loss. This describes what the order books showed on the scan date, not a forecast.
Sources
- sourcePolymarket CLOB API
- sourcePolymarket Gamma API
Polytrading is an analytics tool. It does not trade for you, hold your keys, or promise returns.