Prediction-market arbitrage

A binary prediction-market contract pays $1 if an outcome happens and $0 if it does not. If you can buy every mutually exclusive outcome of an event for less than $1 in total, on one venue or across two, you lock in the difference whichever outcome wins. The edge is real but usually small. Your capital is tied up until the event resolves, and the trade only works if both contracts resolve on the same facts.

How it works

Single venue: complete sets

On Polymarket, the YES and NO shares of a binary market are tokens that can be split from $1 of collateral, and merged back into it, at any time before resolution. The order book uses the same mechanism to match complementary orders. Kalshi's book also links the two sides: a bid for YES at one price is equivalent to an offer of NO at $1 minus that price. As a result, the YES ask plus the NO ask of a single binary market almost never sums below $1. The matching engine or a bot closes that gap within moments.

The version you actually see is the multi-outcome event ("which of N candidates wins"). Exactly one YES pays $1, so the YES asks across all outcomes should add up to at least $1. If they add up to less after fees, buying one YES of each locks in the gap, provided the listed outcomes cover every possibility, including any "Other" bucket. Polymarket's negative-risk markets link multi-outcome events so that a NO in one outcome can be converted into YES shares in all the others. Read the market type before assuming a set is complete.

Cross-venue: the same event on two venues

Buy YES on one venue and NO on the other when the combined price plus fees is below $1.

Venue facts that matter for this trade:

• Polymarket (international) runs a central limit order book with positions held as tokens on Polygon, collateralized by a dollar stablecoin. Markets resolve through UMA's optimistic oracle. A proposer posts a bond, a 2-hour challenge window follows, and a disputed proposal escalates to a vote of UMA token holders. Polymarket also operates a separate CFTC-regulated US exchange with its own market list and rules.
• Kalshi is a CFTC-regulated designated contract market that trades in US dollars. The exchange determines each outcome under the contract's written rules and the source they name.
• Fees. Both venues use a taker fee of the form contracts × rate × price × (1 − price). It is largest at 50¢ and shrinks toward the extremes. Kalshi's standard rate is 0.07, rounded up to the next cent per order. Polymarket's rate varies by market category, and makers pay no fee. Both schedules change, so recheck them before each trade.

Where the yield comes from

The yield comes from fragmentation. The two venues have different user bases, access rules, fee structures and funding rails. Dollars in a US bank account and stablecoins on Polygon do not move between venues instantly, and few traders can legally use both. Prices for the same event therefore drift apart, and the spread pays whoever bridges the gap. It pays for three things: the operational work, the capital you tie up until resolution, and the risk that the two contracts are not quite the same bet.

On a single venue, the edge comes from attention. Multi-outcome markets with many thin order books misprice briefly when news hits. Bots compete for these gaps within seconds, so manual traders rarely win them.

Worked example

Assume the following. These are illustrative prices and fee rates, not current market data.

• 1,000 contracts on each leg.
• Polymarket YES ask: $0.50. Taker fee rate: 0.04 (the rate depends on market category).
• Kalshi NO ask: $0.44. Fee: 0.07 × contracts × price × (1 − price), rounded up to the cent.
• Both markets resolve on identical criteria, 122 days from now.
• Deposit, withdrawal and currency-conversion costs are ignored, and all capital is assumed to be deployed.

Annualize by time to resolution. Holding-period return is 3.39%. APR = 3.39% × 365 / days held. APY = (1.0339)^(365 / days) − 1, which assumes you can reinvest at the same edge every time a trade resolves.

The same 3.39% is a strong trade over one month and a weak one over a year. A 3% edge over 6 months is roughly 6% APR, not 3%. The APY column is optimistic, because matching edges are rarely waiting the day your capital frees up. Use APR when you compare long-dated trades.

Fees can erase a thin gap. Take YES at $0.55 and NO at $0.42, a 3¢ gross gap. The Polymarket fee is 1,000 × 0.04 × 0.55 × 0.45 = $9.90. The Kalshi fee is 0.07 × 1,000 × 0.42 × 0.58 = $17.052, rounded up to $17.06. Total cost is $996.96, leaving $3.04 of profit (0.3%) before any transfer costs. Nearly the whole gap goes to fees.

Step-by-step execution

1. Confirm legal access to each venue from where you live and complete verification. Do not plan around a venue you cannot lawfully use.
2. Fund both venues in advance. Opportunities close faster than bank transfers or on-ramps settle. Count idle balances in your return on capital.
3. Match the contracts line by line. Compare the event definition, resolution source, deadline and its time zone, what counts as "announced" or "confirmed", and how each venue handles cancellation, postponement, ties and revisions.
4. Compute net edge after both venues' fees at your intended size, using the actual order book depth.
5. Execute the less liquid leg first with a limit order, then take the other leg immediately. Never leave one leg open deliberately.
6. Record the pair with both rule texts, and set alerts for any rule clarification, dispute or change in market status.
7. Check for an early exit. If the spread closes well before resolution, selling both legs can free capital at a better annualized return than holding to the end.
8. After resolution, redeem, withdraw if needed, and reconcile the actual return against the plan, including transfer costs.

Risks

Resolution-criteria mismatch is the main risk. Two markets with the same headline can resolve differently. Hypothetical but typical examples: one closes at midnight Eastern time and the other at midnight UTC. One resolves on an official announcement and the other on the event occurring. They cite different data sources for the same number. One voids or splits a cancelled event while the other resolves it NO. If the legs resolve inconsistently, you can lose both, which means up to 100% of the position against a 3% edge. Trade only pairs with near-identical rules, demand more edge for any ambiguity, and size each pair so a total loss is survivable.

Oracle and dispute risk. On Polymarket, a disputed outcome goes to a token-holder vote, which can take time. On ambiguous questions the vote can land on a reading you did not expect. Kalshi settles under its own rules and can also delay settlement while an outcome is under review. Either way, delay stretches your holding period and lowers the annualized return.

Capital lockup and opportunity cost. Every dollar in the trade earns nothing else until resolution. Compare the APR with what the same capital earns in a low-risk stablecoin or cash product. Some venues have offered interest on balances or rewards for holding long-dated positions. Where that applies, include it, but check the current terms.

Execution and legging risk. Prices move between the first and second leg, fills can be partial, and thin books cap your size. A half-filled pair is a directional bet.

Liquidity and exit. Order book depth limits how much you can trade at the quoted edge. Exiting before resolution can cost more than the edge you were capturing.

Transfer costs. Wire fees, card or on-ramp fees, currency conversion, stablecoin spreads and gas all come out of a return measured in single-digit percentages.

Platform and regulatory access. Polymarket's international venue blocks US users and a number of other jurisdictions. Polymarket's US exchange is a separate venue with different listings. Kalshi is regulated in the US and restricts access from several countries. Some governments have moved to block prediction markets entirely. Using a VPN to get around a geoblock typically breaches the venue's terms and can lead to frozen funds. Rule changes can also suspend a market while you hold a position in it.

When it stops working

Stand aside when any of these apply:

• The net edge, annualized over the realistic time to resolution, is below what low-risk cash or stablecoin yield pays.
• The two venues' rules differ on anything that could plausibly decide the outcome.
• Your access to either venue is uncertain, or a regulatory change is pending that could suspend the market.
• Bots are clearing the single-venue gaps to zero. Manual traders should focus on cross-venue and longer-dated pairs, where speed matters less than reading the rules carefully.

Our 5–20% range assumes a mix of trades that resolve within one to six months, at a 2–5% net edge each, with some idle capital between them. Returns are opportunistic, not continuous.

Key takeaways

• Buying every outcome of an event for under $1 locks in the gap only if the outcomes are exhaustive and both venues resolve on the same facts.
• Annualize by time to resolution. A 3% edge over 6 months is about 6% APR, and fees can erase a 3¢ gap entirely.
• Resolution mismatch is a tail risk that can take the whole position. The rules text is the contract.
• Single-venue gaps are mostly taken by bots. Cross-venue spreads persist because few traders can legally access both venues.
• Budget for idle balances, transfer costs and dispute delays. They are what turn a paper edge into a mediocre APR.