Prediction market arbitrage: where the margin goes

Prediction market arbitrage is the same idea as sportsbook arbitrage with a different set of ways to lose. You buy opposite outcomes of one event on two venues, the two prices add up to less than a dollar, and one side pays $1 whichever way the game goes. The hard part is everything between the price you see and the money you actually keep.

How prediction market arbitrage works

A prediction-market contract on a game is a yes/no share. Kalshi explains that a 70 percent chance means a Yes price of 70 cents and a No price of 30 cents, and that a matched Yes and No pair together put in $1. Polymarket pays $1 per share to holders of the winning outcome and nothing to the losing side.

So the arithmetic is simple. If you can buy Team A to win at 0.46 on one venue and Team B to win at 0.50 on another, you pay 0.96 for a set that pays 1.00 either way. That is 0.04 of gross margin, or 4.17 percent of the 0.96 you put up. These numbers are a hypothetical example, not a result anyone got.

The same check works against a sportsbook. Convert the book's American odds to a probability, convert the contract price to a probability, and see whether the two opposite sides sum to under 100 percent. The sports arbitrage calculator walks through that math for books, and the contract version only changes the units: cents instead of American odds.

Why the gap exists, and how long it lasts

Different venues are different crowds. Finance Magnates reports that monthly volume went from under $100 million in early 2024 to more than $8 billion in December 2025, and quotes a trader describing platforms as "still siloed" with fragmented liquidity. Prices on separate order books drift apart when each venue's flow is different.

They do not stay apart long. A study of political contracts on Polymarket, Kalshi and PredictIt, summarized by Quantpedia, found that such gaps "typically exist only for a few seconds, at best a few minutes," and that Polymarket generally led Kalshi because it was more liquid. That was political markets in late 2024. Sports contracts are a different market, so treat it as a reason for caution, not a forecast.

What eats the margin

Four things matter more than the headline gap.

1. Fees. Polymarket's sports taker fee is fee = shares x 0.05 x price x (1 - price), and makers pay nothing. At a price of 0.50 that is 1.25 cents per share, or 2.5 percent of the cost. At 0.20 it is 0.8 cents, or 4 percent of the cost. Kalshi charges a fee on the expected earnings of the contract, varies it for some events and charges maker fees on resting orders once they fill. Its full schedule is a separate document, so read it for the market you are trading.

Run the example above: Polymarket's leg at 0.50 costs 0.0125 in taker fee. Assume a Kalshi fee of 0.01 (hypothetical, check the real schedule). The 0.04 gross becomes 0.0175, about 1.8 percent of the 0.96 cost. A margin that looked fine at 4 percent is thin at 1.8 percent.

2. Resolution mismatch. Two contracts that look identical can settle on different rules. Moontower makes the point with wording: a "touch" contract resolves differently from a "settle" contract and prices differently for it. For a game, ask what each venue does if the game is postponed, shortened, tied or cancelled. If one venue refunds and the other waits, your hedged pair is not hedged. Polymarket resolves through the UMA optimistic oracle with a 2-hour challenge period, so the two legs can also settle at different times.

3. Liquidity and fills. A quoted price is only good for the size sitting at it. If you can buy 50 shares at 0.46 and the next 200 cost 0.48, the arbitrage is gone for any real stake. You need the order book depth, not only the top price, and your second leg has to fill after the first.

4. Capital and timing. Both legs are prepaid. A 1.8 percent return on a game tomorrow is a different thing from 1.8 percent on a season-long contract that ties up cash for months.

There is also a legal question. Whether sports event contracts can be offered under state law is unsettled in the courts: the Ninth Circuit ruled 3-0 against Kalshi on August 28, 2026, creating a split with the Third Circuit, and other circuits have cases pending. Availability for you depends on where you live, so check before you open an account.

A worked example with the fees counted

Say you want a 1,000-share position on the pair above, using the Polymarket fee formula.

• Buy 1,000 Team A shares at 0.46 on one venue: 460.
• Buy 1,000 Team B shares at 0.50 on the other: 500, plus a Polymarket taker fee of 12.50 if that is the venue (per the fee page).
• Assumed fee on the other leg: 10.
• Total outlay: 982.50 including fees. Payout: 1,000 whichever team wins.
• Net: 17.50, or about 1.8 percent.

That holds only if both orders fill at those prices, both contracts resolve the same way and both venues pay out. Miss any of the three and the result can be a loss. Everything here is hypothetical.

What a scanner can and cannot do for you

A scanner can put sportsbook and prediction-market prices on one board and flag pairs whose probabilities sum under 100 percent. The quant+ arbitrage scanner does that across sportsbooks and the prediction venues in its book list, including Kalshi, Polymarket (USA) and Novig, and shows the stake split with rounding for a total you choose. Its odds comparison board shows each venue's price next to a fair price.

It does not count each venue's fee for you, read each venue's settlement rules, or check how deep the order book is. Those three checks stay with you before you place anything, and quant+ never places bets or connects to your accounts. If you want the venue-by-venue differences, see Kalshi vs Polymarket and sportsbooks vs prediction markets.

Mistakes to avoid

• Trading the top-of-book price without checking size.
• Forgetting that a taker fee is charged on each leg, per venue.
• Assuming two "Team A wins" contracts settle the same way.
• Leaving one leg unhedged while waiting for the second to fill.
• Sizing so large that a single void or delay hurts.

Quant+ is analytics for informational purposes. 21+. If you or someone you know has a gambling problem, call 1-800-GAMBLER. Check that sports betting and prediction markets are legal where you live, and read the risk disclosure.