Polymarket arbitrage: how to find mispriced markets

Polymarket is the deepest prediction market, which makes it the anchor of most real arbitrage: its prices are the benchmark other venues drift away from. The edges live in those drifts.

What kinds of Polymarket arbitrage are there?

1 · Internal (multi-outcome) arbitrage. In an N-way market (e.g. "Who wins the nomination?"), the YES prices of all outcomes should sum to ≈$1.00. When they sum above it, shorting the overpriced set locks a spread. These windows are small, competitive and increasingly bot-arbitraged away.

2 · Cross-venue arbitrage. The same question trades on Kalshi, PredictIt or Limitless at a different price. Buy YES on the cheaper venue and NO on the other for a combined cost under $1.00. This is where the durable, human-findable edges are — fragmentation regenerates them every news cycle.

What does a Polymarket arbitrage trade look like?

Suppose "Will candidate X be the next prime minister?" asks YES at $0.07 on Polymarket, while PredictIt's market on the same succession prices X's NO at $0.85. The pair costs $0.92 and pays $1.00 — an 8% gross gap. After PredictIt's 10% fee on winning profit and depth limits, the net modeled edge is nearer 7% — and it only holds if both markets define "next prime minister" identically (caretaker PMs have burned people before).

How much can you make on Polymarket arbitrage?

Across a live board, most Polymarket cross-venue spreads run 0.5–3% gross; a handful reach 5–8% on thin books. Displayed prices are top-of-book — always check how many contracts the price actually holds before sizing a position.

Why do Polymarket arbitrage trades go wrong?

• Same words, different market — "win the nomination" ≠ "win the election"; "announce a run" ≠ "run"; "wins the first round" ≠ "finishes second". Read both resolution rules end-to-end.
• Different resolution sources — a BTC price market settling on Binance at noon is not the same market as one settling on a CF benchmark at 5pm.
• Stale quotes — Polymarket moves fast; a gap on a 10-minute-old snapshot may be gone. Verify against the live CLOB.
• US-person restriction — pairing Polymarket with Kalshi or PredictIt requires access to both, which one US-based person generally doesn't have.

The first item is not a theoretical risk. In our own settled history, every pair that lost both legs was a pair whose titles matched while the underlying questions did not — a candidate who finished second when one venue asked about winning, a tournament market paired with a per-match market. Nothing about the price warned us; only the rules did.

How do you check Polymarket order-book depth before trading?

The price you see on a market page is the best offer, not the available quantity. Polymarket runs a public central limit order book, so the depth is genuinely inspectable — and it frequently tells a different story from the headline. A market can quote an attractive ask with only a handful of contracts resting behind it; buy more than that and you walk into worse levels until the edge is gone.

The practical test is to price your intended size against the ladder rather than the top of book, on both legs, and take the worse result. We have repeatedly seen signals quoting a comfortable spread that turn negative once a hundred contracts are walked through the books — in one case a market quoting a 4% edge cost more than its $1.00 payout to fill, because a single contract sat at the advertised price.

What is negative-risk (multi-outcome) arbitrage on Polymarket?

Polymarket groups many questions into mutually exclusive sets — "who wins the nomination?", "how many rate cuts this year?" — where exactly one outcome can resolve YES. These are marked as negative-risk events, and they support a hedge that never leaves the venue.

Buying NO on every outcome of such an event costs the sum of the NO asks and pays out on all but one leg, so with N outcomes the guaranteed payout is N−1. Whenever the NO prices sum to less than N−1, the difference is locked in. Crucially this only requires that the outcomes are mutually exclusive: if some unlisted possibility occurs and nothing on the list resolves YES, every leg pays and you earn more.

The inverse trade — buying YES across every outcome — looks equivalent but is not. It only pays if the listed outcomes are exhaustive, and if an unlisted result wins, every leg loses. That asymmetry is why the NO direction is the sound one.

These same-venue baskets are the cleanest trades available: one account, one jurisdiction, and no question-matching risk at all, because every leg is an outcome of the same market. Their absolute returns are small — often a fraction of a percent — but many resolve within days, so the annualised figure can be far better than a long-dated cross-venue hedge. The catch is that all N legs must fill; a partial fill turns a hedge into a directional bet.

How to arbitrage Polymarket negative-risk markets, step by step

The whole strategy compresses into one comparison: the cost of a full NO set against its guaranteed payout of N−1.

1. Pick a negative-risk event. Polymarket labels them — the multi-outcome “who wins / which / how many” events where exactly one outcome can resolve YES.
2. Sum the best NO asks across every outcome and compare the total to N−1. Six outcomes with NO asks totalling $4.96 against a $5.00 guaranteed payout is a $0.04 basket — a 0.8% gross edge.
3. Reprice it at your size before trusting it. Walk each leg’s order book for the number of baskets you intend, not the top of book. The thinnest leg caps how many complete baskets exist: a displayed $0.04 edge backed by three contracts on one leg is a three-basket trade, whatever the other five books hold.
4. Fill the thin legs first, and complete every leg. The basket is only riskless when whole — holding N−1 legs out of N is a position on the missing outcome, not a hedge.
5. Hold to resolution, or convert. Where the event supports conversion, a complete NO set redeems for its guaranteed payout without waiting; otherwise the capital stays locked until the event resolves, so judge the edge on its annualised return, not the raw cents.

Worked through end to end: buy one NO on each of six outcomes for $4.96 total. At resolution at least five legs pay $1.00 — $5.00 back, $0.04 locked in regardless of which outcome wins; in the rare case an unlisted result wins, all six legs pay and the basket returns $6.00. The failure modes are execution, not direction — partial fills, and edges thin enough that walking one book erases them.

Can US residents trade Polymarket arbitrage?

Polymarket restricts US persons, so a US-based trader generally cannot hold a Polymarket leg at all. That rules out the Polymarket-versus-Kalshi and Polymarket-versus-PredictIt pairs that produce the largest headline spreads. For a US trader the workable cross-venue combination is Kalshi with PredictIt; for a non-US trader it is Polymarket with Limitless, or a same-venue Polymarket basket as described above. Using a VPN to bypass the restriction breaches Polymarket's terms and tends to surface at withdrawal, after your capital is committed.

Keep reading: Crypto arbitrage: real cross-exchange spreads, measured · Prediction market arbitrage: the complete guide · Polymarket vs Kalshi: fees, access & price gaps · Oracle risk: when both legs of a hedge lose · Liquidity: why the quoted edge is not the real one