Prediction market arbitrage: how it works, how it doesn't
Published 2026-05-22 · Last reviewed 2026-05-22
TL;DR
- Cross-venue arbitrage (Kalshi vs Polymarket) is real but constrained by KYC, ACH, and gas friction.
- Same-venue arbitrage on logically related contracts (sum > $1 or < $1) is the cleanest opportunity.
- Most apparent edges disappear once fees and execution slippage are accounted for.
Free money? Not quite. But when Kalshi and Polymarket disagree on the same event, the spread can be real — and yes, traders have absolutely paid rent off it.
The three flavors of arbitrage
- Cross-venue. Same event, different prices on different exchanges.
- Same-venue logical. Related contracts on one venue whose prices imply impossible joint probabilities.
- No-vig. YES + NO bid/ask spread crossing the $1 line.
Cross-venue: Kalshi vs Polymarket
The classic example. Both venues list the same event (e.g. a presidential election). YES trades at 58¢ on Kalshi and 55¢ on Polymarket — a 3¢ spread. Theoretical edge: buy on Polymarket, sell on Kalshi, lock in 3¢ minus fees on every share.
Practical reality: you need pre-funded accounts on both venues. Polymarket users outside the US have to bridge USDC. Kalshi users pay 1–7% trading fees. By the time you execute, the spread often collapses. Cross-venue arb is mostly a market-maker game.
Same-venue logical
Cleanest retail opportunity. Two contracts on Kalshi: "Will Harvard admit rate be below 3.5%?" and "Will Harvard admit rate be below 4.0%?". The second must be priced higher than the first (it's strictly more likely). When the prices invert or compress implausibly, you can trade the spread.
Same applies to categorical markets: the prices of all outcomes must sum to ~$1. If they sum to $1.05, you can short every outcome and pocket the difference. If they sum to $0.95, you can buy every outcome and pocket the difference. These do appear on thin markets.
No-vig opportunities
On a single market, the bid on YES + bid on NO occasionally exceeds $1, meaning you can simultaneously short both and pocket the difference. These are rare and live for seconds because market-maker bots sweep them. Retail almost never catches them.
Costs that kill apparent edge
- Trading fees. Kalshi 1–7% per side; Polymarket 0% trading but gas on funding.
- Slippage. Hitting the bid at size moves it.
- Capital cost. Capital tied up on both venues earns nothing.
- Resolution divergence. Same event, different resolution sources — they occasionally diverge. That's arb risk, not arb profit.
Who should bother
Traders with serious size, automation, and willingness to manage two venues. Retail arb at small size is rarely worth the effort versus directional trading. The biggest arb you can do as a retail trader is on same-venue logical inconsistencies in thin markets you understand better than other participants.
Related
Frequently asked questions
Can you actually arbitrage Kalshi and Polymarket?
In theory yes; in practice it's hard. The same event listed on both venues will often show a 2–5¢ spread, but you need an account on both, capital pre-funded on both, and willingness to take execution and resolution-difference risk. Real arb profits require size and infrastructure.
What is a no-vig arbitrage?
When the bid + ask on YES and NO on a single market sum to less than $1 (a free dollar) or more than $1 (a free dollar on the other side). These are rare and usually disappear in seconds.
Are there bots doing arb on prediction markets?
Yes. Both Kalshi and Polymarket have professional market-makers running cross-venue and intra-venue arb. Retail arb opportunities are usually leftovers the bots can't capture due to fee structure or capital limits.
Can you arb correlated political markets?
Sometimes. 'Republican wins Senate' and 'Republican wins majority of Senate seats up' should price almost identically; when they don't, the spread is tradable. Resolution criteria differences create the apparent edge — read them carefully before assuming arb.
Independent coverage. Some outbound links are affiliate links — see footer disclosure.