There is a number I check almost out of reflex on any big multi-candidate Polymarket market: add up the YES price on every candidate. In theory it should come to a dollar, since exactly one of them wins. In practice, in any market with real retail flow, it drifts. Sometimes 101 cents, sometimes 104, occasionally worse early in a race. Each candidate trades on its own order book, nothing in the matching engine forces the set to sum to one, and longshot bias does the rest. People love a 3 cent lottery ticket, and every one of those buys pushes the aggregate a little further past 100.
When the sum sits meaningfully above a dollar there is, on paper, free money on the table. Buy NO on every candidate. One of them wins, which means every other NO in your basket pays out a full dollar. If the whole basket cost less than that guaranteed payout, you keep the difference no matter who takes it. This is the classic dutch book, and it genuinely shows up. The question is whether it survives contact with the order book, and most of the time it does not.
What the NO basket actually pays
Run the numbers on a five candidate market where YES trades at 40, 30, 18, 10 and 6 cents. That sums to 104. The corresponding NO prices are 60, 70, 82, 90 and 94, so the full basket costs about 3.96. At resolution one candidate wins and four lose, so four of your five NO positions pay a dollar each. Guaranteed payout of 4.00 against a cost of 3.96, roughly 4 cents per basket, about 1 percent on capital, whoever wins.
The part that makes this practical rather than academic is Polymarket's negative risk structure. In a true one-winner market, holding NO on candidate A is economically identical to holding YES on everyone else, and the negative risk adapter lets you convert between the two on-chain. The corollary I care about is that a complete set of NO, one share per candidate, can be converted straight into cash. You do not wait for the race to resolve. The four dollars comes out now instead of at settlement, which transforms the return math, because a 1 percent edge you can realize the same day and recycle is a very different asset from 1 percent locked up for months. It also sidesteps most resolution risk, since you are out of the market before the oracle ever gets involved.
Check the outcome set before you check the prices
Everything above assumes the outcomes are mutually exclusive, meaning at most one can resolve YES. That assumption fails more often than people expect, and it fails quietly.
Some multi-outcome pages on Polymarket are a single negative risk market with one guaranteed winner. Others are a stack of independent binary markets grouped under a heading, where two of them can resolve YES, or none can. The \"which of these will happen\" style groupings are the usual offenders. A sum past 100 across independent markets tells you nothing, and a NO basket across them can lose in ways the one-winner math never allows. If two listed outcomes both resolve YES, the basket pays a dollar less than you modeled and the whole trade goes underwater. So before touching prices, read the resolution rules on each market and confirm the event guarantees at most one YES. The negative risk conversion being available at all is a decent tell, since the adapter only applies to true one-winner markets.
Then read what happens if no listed outcome wins. For the NO basket this scenario is actually fine, every NO pays and you collect more than planned. But it matters for the mirror trade, buying YES on everything when the sum sits under 100, which only works if the list exhausts the possibilities. If the rules are vague about cancellation, postponement, or a winner from outside the list, treat that vagueness as a cost rather than assuming it away.
Price it off the bids, not the screener
The sum that made you look is almost always built from last trades or midpoints, and you cannot trade either of those. Buying NO on a candidate means crossing to the NO ask, and because the YES and NO sides of the book are mirrored, the NO ask equals one minus the YES bid. So the condition you can actually execute is that the YES bids, summed across every candidate, exceed a dollar at sizes you can genuinely hit. Profit per basket is exactly that sum minus one dollar. A screener showing 103 off last trades can easily sum to 99 on the bids of five thin books, and then there is nothing to do.
The explicit costs are small here. Order placement on Polymarket is relayed, so it typically costs no gas, most markets have historically charged no trading fee, and Polygon gas for the conversion and any redemption is typically cents. The costs that actually kill baskets are the spreads you cross, the price impact once you eat past the top of the book, and the capital lockup if for some reason you cannot convert. Budget for all three before the first order goes out.
Legging in is where it dies
The moment your first leg fills you are directional. Hold NO on everyone except one candidate and you are, funnily enough, long that candidate. If they win, every share you hold pays out, and if anyone else wins you come up a dollar short of the full basket. An incomplete basket is a position, and usually a position you never wanted. My rule is to fill the thinnest book first. If the illiquid longshot leg will not fill at your price, you find out before committing much capital, and unwinding one or two liquid legs costs almost nothing. Do it in the other order and you can end up holding four liquid legs while the fifth market moves away from you, usually because other people spotted the same mispricing and are eating the same book.
The checklist I actually run, in order:
- Confirm the event is a single one-winner market, with rules guaranteeing at most one outcome resolves YES.
- Read the edge cases in the rules: cancellation, postponement, a winner not on the list.
- Sum the YES bids at your intended size. Edge per basket is that sum minus 100 cents. Under roughly 2 cents, stop.
- Check depth on every leg and size the basket to the thinnest one.
- Leg in from least liquid to most liquid, with limit orders wherever the book allows.
- Once the set completes, convert through negative risk and free the capital, unless you have a deliberate reason to hold to resolution.
- Log realized edge after slippage and compare it to the number from step 3. If the two keep diverging, your fills are worse than you think.
These windows exist because retail keeps buying longshot YES and no mechanism ties the books together, so they reappear constantly, mostly too small or too thin to bother with. I pass on most of the ones I check. The few that clear every step pay something like 1 to 2 percent with essentially no market risk, which is modest, but it compounds fine when the conversion lets you recycle capital, and the habit of verifying the outcome set will save you from worse trades than this one.