Market structure
Cross-venue prediction market arbitrage: what is real and what is not
The same question at 61¢ on one venue and 66¢ on another looks like free money. It rarely is. A framework for telling the difference.
The naive arbitrage
Suppose "Will the central bank cut rates at its next meeting?" trades at 61¢ Yes on Venue A and 66¢ Yes on Venue B. Buy Yes on A at 61, buy No on B at 34 (since No ≈ 100 − 66), and you have paid 95¢ for a pair of contracts that should pay exactly 100¢ whatever happens. That is a 5.3% return with no risk.
Except that every one of those assumptions ("exactly 100¢", "whatever happens", "no risk") deserves scrutiny. In our experience, the majority of apparent cross-venue arbitrages either disappear once costs are counted or are compensation for a real risk.
Cost 1: fees on both legs
The arbitrage requires two trades, usually both as a taker, on venues with different fee schedules. A venue whose fees scale with P(1−P) charges most near 50¢, exactly where many contested questions trade. Rounding to whole cents per order punishes small sizes. Add any withdrawal or conversion costs and a 5¢ gross gap can shrink to 2–3¢, or to nothing.
Cost 2: carry and capital lock-up
Both legs tie up collateral until resolution. If the event is months away, the 5¢ profit has to beat the return on 95¢ of capital over that period. For a ten-month market at a 4–5% risk-free rate, roughly 3–4¢ of a 5¢ gap is simply the cost of money. Where one venue pays interest on balances and the other does not, the carry differs by leg, and the pair's breakeven shifts again.
Capital also has to be in two places at once. Moving money between fiat and stablecoin rails takes time and costs fees, and inventory stranded on the wrong venue cannot be used to close the next gap.
Risk 1: resolution basis
The largest risk in cross-venue trading is that the two contracts are not, in fact, the same contract. Examples of differences we see regularly:
- Source and timing. One contract resolves on the first official release; the other on a revised figure, or on a different publication.
- Edge cases. Postponed meetings, cancelled matches, ties, recounts, candidates who withdraw, "at least" versus "more than".
- Time windows. "By 31 December" in different time zones, or inclusive versus exclusive dates.
- Dispute processes. An exchange determination versus an oracle with a dispute window and a token-holder vote.
If both contracts can resolve No (or both Yes) under a plausible scenario, the "arbitrage" is actually a bet on that scenario not occurring. The price gap is often an accurate estimate of how likely it is.
Risk 2: execution and legging
The two legs cannot be executed atomically across venues. If the first leg fills and the second does not (because the price moved, depth was thinner than displayed, or one venue's API was slow), the trader is left with a naked directional position. In fast markets around news, legging risk can dominate everything else. Professional desks address it with conservative sizing, simultaneous order placement, pre-positioned inventory on both venues and strict rules for unwinding a broken pair.
A worked example
Return to the opening case: Yes at 61¢ on Venue A, No effectively at 34¢ on Venue B (because Yes trades at 66¢ there). Gross cost of the pair is 95¢; gross payoff is 100¢ if the contracts settle identically. Now apply the costs, using illustrative but realistic figures.
| Item | ¢ per pair |
|---|---|
| Gross gap | +5.0 |
| Taker fee, Venue A (fee scaling with P(1−P), near 61¢) | −1.7 |
| Taker fee, Venue B | −0.5 |
| Carry: 95¢ locked for 5 months at 4.5% | −1.8 |
| Conversion and transfer costs, amortised | −0.3 |
| Expected basis loss: 2% chance the contracts settle differently, losing ~95¢ | −1.9 |
| Net expected edge | −1.2 |
The "free" 5¢ is, on these assumptions, a slightly losing trade. Change any input (a cheaper fee tier, a shorter horizon, a basis memo that finds no plausible divergence) and it may become a good one. The point is not that cross-venue trades never work; it is that their value is entirely in the details.
Synthetic and correlated arbitrage
Some of the most interesting relative-value relationships are not between identical contracts but between related ones. A ladder of strikes must be monotone. A contract on "at least two rate cuts this year" must be priced below "at least one". The probability of a candidate winning the election should be consistent with the probabilities of their winning the states that determine it. A contract on a meeting-level decision should be consistent with what interest-rate futures imply.
These relationships can be exploited without identical contracts, but they introduce model risk: the trader's view of the joint distribution can be wrong. They are best treated as statistical trades with explicit risk limits, not as arbitrage in the textbook sense.
When the opportunity is real
Genuine relative-value opportunities do exist, usually for structural reasons that are not obvious from the price alone:
- Flow imbalances, when one venue's participant base leans heavily one way (home-team bias, partisan enthusiasm) and capital mobility limits correction.
- Speed differences, when one venue's prices lag another's on news because its participants are slower or its market makers are less active.
- Ladder inconsistencies, when strikes on one venue imply a distribution inconsistent with another's, or with related derivatives markets.
- Calendar mismatches, when contracts expire at different times and the gap between them can be modelled.
These are rarely risk-free. They are trades with a quantifiable edge and a quantifiable risk, sized accordingly. That is a respectable business. It is not free money.
Managing a pair until resolution
Entering a cross-venue position is the easy part. The pair then has to be managed for weeks or months, and the risks change as resolution approaches.
Rule changes are the first thing to monitor. Venues occasionally clarify or amend contract terms after listing, usually to address an edge case nobody anticipated. A clarification on one venue that is not mirrored on the other can turn a hedged pair into an open bet overnight. We track rule text and venue announcements for every contract we hold and re-run the basis memo whenever anything changes.
The second is the exit. A pair does not have to be held to resolution: if the gap closes early, unwinding both legs releases capital and removes basis risk. Whether that is worth doing depends on fees for the exit trades versus the carry saved. As a rule, the longer the remaining horizon and the larger the basis risk, the more attractive an early exit becomes.
Finally, the resolution itself. Disputed outcomes, delayed determinations and settlement timing differences can leave one leg paid and the other pending for days. Capital planning should assume this will happen.
A five-question framework
- Are the contracts identical under every plausible scenario? If not, price the basis.
- What is the all-in cost? Fees on both legs, rounding, conversion and withdrawal.
- What is the carry? Capital locked × time to resolution × cost of capital, per leg.
- Can both legs be executed at displayed size? If not, what happens if only one fills?
- Why does the gap exist? If you cannot answer this, assume someone else can, and that the gap is fair.
Applied honestly, this framework eliminates most apparent arbitrages and makes the remaining ones considerably more valuable, because you know exactly what you are being paid for.
Frequently asked questions
Is prediction market arbitrage risk-free?
Rarely. Fees, the cost of locking up capital until resolution, differences in resolution rules between venues, and the risk of only one leg filling usually explain most of the apparent gap.
Why do prediction markets on the same event have different prices?
Different fee schedules, collateral and interest, resolution wording and dispute processes, participant mixes, and the cost of moving capital between venues.
What is resolution basis risk?
The risk that two contracts that appear to describe the same event settle differently because of differences in source, timing or edge-case rules.
Related: Kalshi vs Polymarket · Pricing event contracts · Our relative-value approach