Prediction market arbitrage: how it works and why most of it isn’t

Every binary contract pays $1 to whoever holds the winning side. If you can buy YES somewhere and NO somewhere else for a combined cost below $1 — fees included — you collect $1 whatever happens, and the difference is profit. That is the whole idea; everything else is about whether the gap is real.

• Cross-venue: YES on one exchange and NO on another, on what has to be the same event. The most common kind, and the one this lesson is mostly about.
• Within one market: YES and NO on the same contract offered for less than $1 combined. On a working order book this almost never lasts, because market makers close it in seconds.
• Multi-outcome: buying YES on every candidate in an election, say, when the prices add to less than $1. Polymarket links such markets through what it calls negative risk — a NO on one outcome converts to a YES on every other — which keeps the outcomes priced consistently and leaves little to capture.

Illustrative prices, not live quotes. Kalshi offers YES on an event at 46¢; Polymarket US offers NO on the same event at 51¢. That is 97¢ for a guaranteed $1 — it looks like 3¢ of free money per contract.

• Kalshi’s taker fee is 0.07 × price × (1 − price) per contract, rounded up: about $1.74 on 100 contracts at 46¢.
• Polymarket US’s taker fee is 0.0695 × price × (1 − price): about $1.74 on 100 contracts at 51¢.
• Total cost for 100 pairs: $97 + $3.48 = $100.48, for a $100 payout. The “arbitrage” loses 48 cents.

Widen the gap to 44¢ and 50¢ and it works: $94 plus about $3.47 in fees is $97.47, returning $2.53 — 2.6% — when the event settles. That is the realistic size of a genuine cross-venue arbitrage: a few percent, before you count the time your money is locked up. Fees are largest at 50¢ and shrink towards 1¢ and 99¢, so the same gap is worth more near the extremes. The arbitrage calculator runs these numbers for any two prices and stake.

A price gap between two venues is usually a sign that the two contracts are not quite the same, not that the market has missed something. The checks we run before calling anything arbitrage:

• The rules differ. Two contracts on “the same” election can settle on different sources or dates — one on the called result, one on certification — so one leg can win while the other also loses. How contract rules differ shows how far apart “identical” markets can be.
• The rules match except in an edge case — a postponement, a tie, a data revision — and in that case both legs lose.
• The fees eat the gap, as in the example above.
• The quote is stale or the depth is not there: the price you saw was for five contracts, and the next level up erases the edge. Liquidity and spreads explains why the top of the book is not the price of a large order.
• One market has closed, or the two settle months apart, which changes what the return is worth.
• You cannot use both venues. polymarket.com does not accept US residents and Kalshi limits international users, so a gap between them is not one a single person can always trade. Polymarket vs Polymarket US covers who can use which.

An arbitrage pays when the event settles, and until then both legs tie up cash on two platforms. A 2.6% return that settles in 60 days is roughly 17% a year; the same 2.6% over ten months is about 3% a year, less than a savings account. Compare the annualised figure, not the headline one.

There are practical costs too: deposits and withdrawals on two venues, the time money takes to move between them, and in some cases deposit fees — Novig, for example, charges 3% on card deposits. Fees and the true cost of a trade covers the whole bill.

The gaps that survive fees are small and short-lived, so the traders who capture them mostly use software that watches several order books at once. Doing it by hand is realistic on slower markets — long-dated politics or economics contracts — where a gap can persist for hours. Whatever you use, check the two contracts’ rules side by side before you trade; no scanner can decide that two markets settle on the same fact.

We are building an arbitrage scanner that applies the checks above, and it stays off until each venue’s data terms allow us to show their prices side by side. Until then, the arbitrage calculator, the expected value calculator and Kalshi vs Polymarket US cover the manual route.