Polymarket vs Kalshi: Why the Same Question Can Trade at Two Different Prices

The same question can trade at two different prices on Polymarket and Kalshi because the venues differ in fees, settlement rules, liquidity and locked capital.

When one prediction market venue prices "Will X happen by December 31?" at 55 cents and another lists what looks like the same question at 58 cents, it is tempting to read the gap as free money or as proof that one crowd is wrong. Often it is neither. This page explains what has to be true before two contracts are really the same question, the frictions that let a price gap sit open (fees, bid and ask, locked capital, who can trade where), why arbitrage does not close every gap, and what a gap can and cannot tell you. It uses Polymarket and Kalshi as the two venues most readers will meet, and every number in the worked example is hypothetical. It is descriptive throughout.

What "the same question" has to mean

A prediction market contract pays a fixed amount if an event happens and nothing if it does not, so its price can be read on a probability scale (Wolfers and Zitzewitz, 2004). Two contracts are only comparable if they pay out in exactly the same states of the world, and that is decided by the rules, not the headline.

Polymarket's documentation is explicit on the point: "The market title describes the question, but the rules define how it resolves," and the rules specify the resolution source, the end date and the edge cases. Kalshi's help centre says the same thing in its own terms: each market operates under its own rules, documented on the market's page, with a named source used to verify the outcome. Two listings with near-identical titles can still differ on details that move the price:

• Resolution source. One contract may settle on an official government release, another on a specific data provider or a media call. If the sources can disagree, the contracts are different bets.
• Deadline and time zone. "By December 31" can mean the end of the day in New York or in UTC, and a contract that settles on a figure published in January is not the same as one that settles on December 31.
• Touch versus close. A price contract can settle on whether a level is touched at any point or only on where it closes on a date. The first pays in more states of the world, so its YES side should cost at least as much.
• Edge cases. What happens if the event is postponed, if a candidate withdraws, or if the answer is unclear? Polymarket's rules provide for a market to resolve 50/50, each share redeeming for $0.50, when neither outcome applies. A venue without that clause, or with a different one, prices the edge case differently.

A January 2026 study of more than 100,000 events across ten venues, by Gebele and Matthes, calls this problem "semantic non-fungibility": economically identical events are listed independently, with no shared notion of event identity, so liquidity does not pool and prices are free to drift apart. The first step in reading any gap is to confirm it compares the same payoff.

Fees change what each price is worth

Both venues charge trading fees that depend on the price. Polymarket's fee documentation states that makers are never charged and only takers pay, with a fee of contracts times a category fee rate times p times (1 minus p), where p is the share price, so the fee is symmetric around 50 cents and largest there. The rate depends on the market category, from zero on geopolitical and world events markets (which Polymarket says are fee-free) to 0.07 on crypto markets. Kalshi publishes its own fee schedule, which also charges a fee that scales with price times one minus price, so it too is largest for contracts near 50 cents and shrinks toward the 1 cent and 99 cent extremes, and its help centre notes that resting maker orders can carry their own fee, charged only when they trade.

Two consequences follow. First, a 55 cent quote on one venue and a 55 cent quote on another do not cost the same to trade if their fee rates differ, so the break-even price differs too. Second, because both fee curves peak in the middle, a gap of a few cents on a contract near 50 cents has to clear more fee than the same gap on a contract near 5 or 95 cents. Fee schedules change, so any comparison has to use the schedule in force on the day.

A quoted gap is not a tradeable gap

The price a market page shows is usually a last trade or a midpoint, not a price anyone can act on in size. To buy, you pay the ask; to sell, you receive the bid; and the gap between them, the bid-ask spread, is a cost on every leg. A "gap" between two venues' mid prices can disappear entirely once you compare one venue's ask with the other venue's bid.

Depth matters as much as the top of the book. A thin book may show a good price for the first few hundred contracts and a much worse one after that, so the average fill on a larger order drifts away from the quote. How to read a prediction market order book walks through that on a real example. Stale quotes are a further trap: a resting order that nobody has traded against for days can sit at an old price while the other venue moves, which makes a gap look stable when one side is simply not trading. Kresmion's public market pages under /odds carry a liquidity block for the more actively traded Polymarket markets (best bid, best ask, the spread and the dollar depth resting near the mid, with the capture time), which is the information needed to tell a quote from an executable price on that venue.

Locked capital and the time value of money

A binary contract ties up its full collateral until it resolves. On Polymarket, one dollar of collateral creates one YES share and one NO share, and a winning share redeems for $1.00, so the money behind a position is committed until the market settles. Holding YES on one venue and NO on the other locks capital on both, possibly for months.

That money could earn a return elsewhere, and whoever buys the expensive side of a contract commits more of it, so on long-dated contracts the wait tends to pull prices toward 50 cents: the likelier side can trade a little below its probability and the unlikely side a little above it. Caleb Maresca's 2026 paper, which describes that asymmetry, calls this the "long-horizon problem": the opportunity cost of committed capital reduces liquidity and price accuracy on distant events. In his simulations with language-model traders the measured bias was smaller than earlier estimates, and paying interest on positions removed most of it. Both venues now pay something on some positions, which shrinks that cost without removing it, and the terms differ. Kalshi's help centre says it pays a variable interest rate on cash and open positions for accounts above a minimum balance, and Polymarket's help centre describes holding rewards at a variable rate on a listed set of long-dated markets only. A gap between two venues can partly reflect that one side earns something on its collateral and the other does not.

Who can trade where, and in what money

The two venues sit under different legal frameworks and admit different users. The Commodity Futures Trading Commission designated KalshiEX LLC as a contract market in November 2020, and Kalshi settles winning contracts to cash balances in US dollars. Polymarket's documentation describes its markets as collateralised onchain in pUSD, a token it says is backed by USDC, with the backing enforced onchain. Which venue a given person can open an account on depends on where they live and on each venue's own account checks.

That matters for prices because arbitrage needs the same trader, or the same pool of money, on both sides. If the people who could close a gap cannot easily hold accounts on both venues, or would have to convert between dollars in a bank account and dollars on a blockchain to do it, the two books can serve largely separate crowds and settle at different prices. Moving money between the two adds transfer time, conversion costs and operational risk.

A worked example (hypothetical)

All of the following numbers are invented to show the arithmetic; none is a market price. Suppose venue A offers YES on a question at 55 cents and venue B offers NO on what appears to be the same question at 42 cents. Buying 100 YES on A and 100 NO on B costs $97.00, and if both contracts resolve YES, or both resolve NO, exactly one leg pays $100.00. On quoted prices that is a $3.00 gap.

Now apply hypothetical costs. Assume both venues charge a taker fee of 0.07 times the number of contracts times p times (1 minus p), the shape both venues' fee documents describe, before any rounding. The YES leg costs 0.07 x 100 x 0.55 x 0.45, about $1.73, and the NO leg 0.07 x 100 x 0.42 x 0.58, about $1.71, so fees alone take about $3.44, more than the whole gap. If the market resolves in six months and the $97.00 could otherwise earn a hypothetical 4 percent a year, the wait costs another $1.94.

Two further risks sit outside the table. If either price is a stale quote or the books are thin, the fills can be worse than 55 and 42 cents, or the quote can be gone before it fills. And if the two contracts' rules differ, the payoff is no longer fixed at $100.00: should venue A resolve NO while venue B resolves YES, the YES on A and the NO on B both expire worthless and the whole $97.00 is gone. Changing the inputs changes the answer (a fee-free category, maker orders instead of taker orders, interest paid on the collateral, or a shorter wait all shrink the costs), which is why the same quoted gap leaves a different amount, or nothing, for traders facing different costs.

Why arbitrage does not close every gap

Textbook arbitrage needs the same payoff, no cost, no risk and simultaneous execution in size. Andrei Shleifer and Robert Vishny's 1997 paper "The Limits of Arbitrage" showed why real arbitrage falls short of that: it is done by a limited number of specialists with limited capital, and when a gap can widen before it closes, they may not be able to hold the position long enough to profit. Between two prediction market venues several conditions can fail at once: the payoffs may not match exactly, both legs pay fees and spreads, capital is locked until resolution, and the trader needs accounts and funds on both sides.

The academic evidence fits that picture. Gebele and Matthes report that roughly 6 percent of the events in their dataset were listed on more than one platform at the same time, and that semantically equivalent markets showed "persistent execution-aware price deviations" of 2 to 4 percent on average, "even in highly liquid and information-rich settings." Execution-aware means the authors subtracted standardised allowances for fees, spreads and other execution costs, so these are not raw quote gaps. The authors read what remains as arbitrage opportunities that persist because of "structural frictions rather than informational disagreement": the gaps survive their cost assumptions, but not because the two crowds disagree about the event.

How to read a gap between two venues

A short checklist turns a headline gap into something interpretable:

1. Same payoff? Read both rule sets: the resolution source, the deadline and time zone, touch versus close, and the edge cases. 2. Executable prices? Compare one venue's ask with the other venue's bid, not two mids, and check the depth at those prices and when each quote last changed. 3. Costs on both legs? Apply each venue's fee schedule at the prices and order type being compared. 4. Carry and access? Allow for how long the collateral is locked, whether either side earns anything on it, and whether the same person can trade both venues at all. 5. Which one is right? Neither, necessarily. A relative gap does not say which venue is closer to the truth; only resolved outcomes can, which is what calibration studies measure. Kresmion's study of how well prediction market prices matched outcomes a week before resolution shows what that kind of test looks like on one venue.

Limitations

• This page measures nothing. It explains the mechanisms; it does not report how large or how frequent cross-venue gaps are on any given day. The only measured magnitudes quoted are from the cited academic study, which uses its own matching method and sample (2018 to 2025).
• The example is hypothetical. The prices, fee rate, carry rate and horizon are invented to show the arithmetic. Real fee rates vary by venue, category, order type and date.
• Venue terms change. Fee schedules, interest and holding reward rates, collateral arrangements and which users each venue accepts have all changed over time and can change again; check the venue's own documents on the day.
• Mechanisms overlap. In any real gap, rule differences, fees, spreads, carry and access can all be at work at once, and a quote alone cannot separate them.

Key takeaways

Frequently asked questions

Why do Polymarket and Kalshi show different prices for the same event?

Often because the two contracts are not perfectly identical, or because frictions keep the prices apart: different fees, bid-ask spreads and depth, capital locked until resolution, and different users and settlement money on each venue. Genuine disagreement between the two crowds can sit on top of that, but a quoted gap alone cannot show how much of it is disagreement.

Is a price gap between two prediction markets free money?

No. A gap between quoted prices is an upper bound before costs. Fees on both legs, the spread, thin depth, the time value of locked collateral and the risk that the two contracts resolve differently can each consume it, and in the hypothetical example on this page fees alone exceed a 3 cent gap. Kresmion makes no trading recommendation.

Which venue's price is more accurate?

A gap does not answer that. Accuracy can only be judged against resolved outcomes across many markets, which is what calibration and scoring rules such as the Brier score measure. Prices are also shaped by fees and the time value of money, so neither venue's price is a pure probability.

Can the same question resolve differently on the two venues?

Yes, if the rules differ. Each venue resolves under its own written rules and named sources, and Polymarket, for example, can resolve a market 50/50 when neither outcome applies. Two contracts with similar titles but different sources, deadlines or edge-case clauses can end up paying out in different states of the world.

Why can a long-dated contract's price differ from its probability?

Because the money behind it is locked until resolution and could earn a return elsewhere. The side that pays more up front (YES on a likely event, NO on an unlikely one) bears more of that cost, which tends to pull long-dated prices toward 50 cents: the likelier side can trade a little below its probability and the unlikely side a little above it. Interest or holding rewards on positions reduce that cost where a venue offers them, but the terms differ by venue and market.

This page is information, not investment advice.

---

Source: Polymarket documentation, Fees, https://docs.polymarket.com/trading/fees ; Polymarket documentation, Resolution, https://docs.polymarket.com/concepts/resolution ; Polymarket documentation, Polymarket 101 (shares, redemption and pUSD collateral), https://docs.polymarket.com/polymarket-101 ; Polymarket help centre, Holding Rewards, https://help.polymarket.com/en/articles/13364459-holding-rewards ; Kalshi Fee Schedule, https://kalshi.com/docs/kalshi-fee-schedule.pdf ; Kalshi help centre, Fees, https://help.kalshi.com/trading/fees ; Kalshi help centre, Market Rules, https://help.kalshi.com/en/articles/13823822-market-rules ; Kalshi help centre, APY on Kalshi, https://help.kalshi.com/en/articles/13823847-apy-on-kalshi ; CFTC, "CFTC Designates KalshiEX LLC as a Contract Market", release 8302-20, https://www.cftc.gov/PressRoom/PressReleases/8302-20 ; Gebele and Matthes, "Semantic Non-Fungibility and Violations of the Law of One Price in Prediction Markets", arXiv 2601.01706, 2026, https://arxiv.org/abs/2601.01706 ; Shleifer and Vishny, "The Limits of Arbitrage", Journal of Finance 52(1), 1997, https://onlinelibrary.wiley.com/doi/full/10.1111/j.1540-6261.1997.tb03807.x ; Maresca, "Can Interest-Bearing Positions Solve the Long-Horizon Problem in Prediction Markets?", arXiv 2602.21091, 2026, https://arxiv.org/abs/2602.21091 ; Wolfers and Zitzewitz, "Prediction Markets", Journal of Economic Perspectives 18(2), 2004, https://www.nber.org/papers/w10504 . Venue documents read 2026-10-03. The worked example uses hypothetical prices, fees and rates.

Kresmion Research.