Prediction Market Arbitrage: Why the Price Gap Rarely Pays Out

TL;DR

• Kalshi's taker fee runs up to 1.75% of a contract's price near 50 cents (0.07 × price × (1 − price), capped at 3.5 cents per contract). Polymarket's taker fee on a politics market tops out at 1%. On an 8-cent Kalshi-Polymarket gap, entry fees on both sides alone can eat a third of it.
• A price gap that survives fees is usually one of two things: a temporary mispricing that closes as liquidity flows in, or two markets with different resolution rules wearing the same headline. Only the first one is arbitrage.
• Kalshi does not refund your stake if the underlying event is cancelled. It extends the settlement window or falls back to the last traded price or a committee ruling. Polymarket's 50/50 payout is real but its own docs call it rare. Only Limitless refunds at cost, and only when the market's own rules don't already cover the scenario.
• Acting on a cross-platform gap means already holding two funded, verified accounts. Kalshi and Polymarket's separate US arm require full identity verification; Polymarket's international platform, Limitless, Predict.Fun, and Opinion only need a connected wallet. Most retail traders don't have both in place before the gap closes on its own.

You spot Polymarket pricing a market at 62 cents while Kalshi prices the same event at 55 cents. That seven-point gap looks like free money. It usually isn't. By the time you account for fees, KYC mismatches, withdrawal friction, and settlement timing, most of that gap closes before you can capture it.

This piece breaks down why the same event prices differently across platforms, what it actually costs to act on the gap, and what a real cross-platform trade looks like once fees and risk are priced in.

Why Do Kalshi and Polymarket Price the Same Event Differently?

Kalshi, Polymarket, Limitless, Predict.Fun, and Opinion each run their own central limit order book, with their own liquidity and their own users. None of the five prices through an automated market maker the way early crypto prediction markets did. Every price reflects whichever traders happened to place orders on that specific platform, not a shared pool of opinion.

Kalshi's user base skews toward US retail traders funding accounts through a bank-linked, CFTC-regulated exchange. Polymarket's international platform draws a global, crypto-native audience trading through a self-custodied wallet.

Limitless is the largest prediction market built on Base, with more than $500 million in cumulative volume. Most of its markets run on a standard order book, but it also runs separate AMM-priced markets for short-duration hourly and daily crypto and stock-price contracts, a structure none of the other four platforms uses.

Predict.Fun routes every open position's collateral into Venus Protocol, a BNB Chain lending market, so its traders are partly optimizing for yield on idle capital, not just the outcome. Opinion is built around institutional-style macro contracts (Fed decisions, CPI, GDP) resolved through what it calls an AI oracle, a different resolution model from the other four.

Different pools of people, funding through different rails, pricing the same outcome, will not converge instantly. Sometimes they don't converge until the event resolves. We cover the mechanics of that divergence in more depth in why odds disagree across platforms.

That's the setup. The harder question, the one most explainers skip, is why you can't just buy the cheap side and sell the expensive side and pocket the difference.

Can You Actually Capture a Prediction Market Price Gap?

Say Kalshi prices a Yes contract at $0.55 and Polymarket prices the same outcome's Yes contract at $0.62. In theory, buying the Kalshi contract and taking the opposite side on Polymarket locks in seven cents regardless of outcome. That's the same logic behind arbitrage on any two exchanges trading the same instrument.

Prediction markets add frictions that arbitrage between two stock exchanges doesn't have. Four of them matter most: withdrawal friction, KYC mismatches, settlement-timing differences, and fees. Each one chips away at the edge before you can actually bank it.

How Much Does Withdrawal Friction Really Cost You?

Moving capital off one platform to redeploy on another isn't instant, and the two rails work differently.

Polymarket, Limitless, and Predict.Fun settle withdrawals on-chain. A withdrawal clears once the transaction confirms, which is fast, but you're still paying gas, and the funds still need to land in a wallet you can immediately redeploy.

Kalshi, as a CFTC-regulated exchange, moves fiat through ACH bank transfers, which clear in one to three business days, not minutes. Kalshi also accepts crypto deposits and withdrawals through Zero Hash, its crypto processing partner, which can settle much faster than ACH.

There's a catch. Kalshi requires a prior crypto deposit on file before it will process a crypto withdrawal. A trader who funded their account by bank transfer can't shortcut out through crypto rails without first sending crypto in.

If a strategy depends on capital being free and mobile between two platforms the moment a gap appears, the withdrawal window itself eats into the edge before a dollar gets redeployed.

Do You Need KYC on Both Platforms Before You Can Act?

Kalshi requires a legal name, date of birth, Social Security number, and a government-issued ID before you can fund an account, standard for a CFTC-regulated exchange. Polymarket's international platform, Limitless, Predict.Fun, and Opinion only need a connected wallet, no ID upload for basic access.

Polymarket isn't a single product on this point. Its separate, CFTC-regulated Polymarket US arm requires the same full identity verification Kalshi does. If the gap you spotted involves Polymarket US rather than the international platform, the KYC math changes entirely. Our KYC requirements breakdown covers the full split by platform.

If you don't already have verified, funded accounts on both sides of a price gap the moment you spot it, the gap has usually moved by the time you clear onboarding.

What Happens When the Two Platforms Don't Settle on the Same Rules?

Two platforms tracking "the same" event don't always resolve it the same way, and the differences run deeper than a shared headline suggests.

Kalshi's rulebook doesn't guarantee a refund if the underlying event is cancelled or postponed. Under Rule 7.2(b), Kalshi can extend the contract's settlement window rather than resolve it outright. If the outcome still can't be determined, Rule 6.3(c) settles the contract using the last traded price or, if that isn't fair, a binding ruling from Kalshi's Outcome Review Committee. A genuine refund only applies under Kalshi's emergency powers, covering things like system failures or extraordinary market threats, not routine postponements.

Polymarket resolves most markets through an unopposed or once-disputed proposal under its UMA-based oracle. Someone proposes an outcome, a two-hour window passes, and if nobody disputes it, that's the result. Its 50/50 payout, where every share redeems for $0.50, only appears after a proposal is disputed, a second proposal is also disputed, and the question escalates to a full UMA token-holder vote. Polymarket's own documentation calls this outcome rare, a real fallback, not the default result of an ambiguous event.

Limitless is the one platform with a documented, if conditional, refund-at-cost policy. Most of its markets now resolve automatically through the Pyth Network oracle rather than a dispute process. If a market is genuinely misresolved, meaning its published criteria can't be applied to what actually happened, Limitless refunds the original bet amount, not the payout, not zero. But if that specific market's own written rules already cover the scenario (its docs cite a cancelled game as exactly this kind of case), those rules control instead, and the refund fallback doesn't apply.

Predict.Fun and Opinion don't publish a platform-wide cancellation policy. Predict.Fun's infrastructure suggests a Polymarket-style dispute process, but that's an inference, not something either platform documents directly.

A price gap that looks arbitrable can actually be two contracts with different resolution paths wearing the same headline. If one platform runs a two-hour dispute clock and another can extend its settlement window indefinitely while a committee sorts things out, you're not locking in a spread. You're taking on timing and resolution risk you didn't price. See what happens when an event gets cancelled for the full platform-by-platform rundown.

How Much Do Trading Fees Really Eat Into the Spread?

Every platform charges takers something, and the rate depends on both the platform and the category, not a flat 1-2% you can assume across the board.

Kalshi charges 0.07 × price × (1 − price) per contract, capped at 3.5 cents. That formula peaks at a 50-cent price, where the fee hits its 1.75% ceiling, and shrinks toward both extremes. Maker orders, resting limit orders that add liquidity instead of taking it, cost about a quarter of that rate.

Polymarket's taker fee varies by category on the same price × (1 − price) curve: 4% for politics, finance, tech, and mentions; 5% for sports, economics, culture, and weather; 7% for crypto. Geopolitical markets are fee-free. Makers pay nothing and collect a rebate on filled limit orders.

Predict.Fun's taker fee slides between roughly 0.018% and 2%, peaking near a 50-cent price. It also takes a 10% cut of whatever yield your collateral earns on Venus Protocol while the position stays open, a cost that has nothing to do with the trade itself. Opinion charges takers a fee that scales toward the low single digits as price approaches 50 cents; makers pay nothing and get half the taker fee back as a rebate.

A trader entering Kalshi's side of a 50-cent political market pays a higher percentage fee than a trader entering Polymarket's side of the same category. The fee drag isn't symmetric just because the two platforms are comparable in size. For the full platform-by-platform breakdown, see how prediction market fees actually work.

What Does an Arbitrage Trade Actually Look Like After Fees?

Take a hypothetical "Candidate X wins" contract trading at $0.60 on Kalshi and $0.68 on Polymarket, in the politics category. That's an 8-cent gap, about 13% of Kalshi's price.

Run the fees first. Kalshi's taker fee: 0.07 × 0.60 × 0.40 = $0.017 per contract, rounding up to about 2 cents. Polymarket's politics-category taker fee: 0.04 × 0.68 × 0.32 = $0.009 per contract, under a penny. Entering both sides costs roughly 2.5 to 3 cents of the 8-cent gap before either position moves.

That leaves about 5 cents. Now price in the rest. If Kalshi and Polymarket define "wins" on slightly different terms (a certification date, a recount clause, which reporting source triggers settlement), you're not holding a matched pair. You're holding two directional bets that happen to look similar today. If the event gets contested or delayed, Kalshi can extend its settlement window or fall back to a committee ruling while Polymarket's UMA process runs its own dispute clock on a different timeline entirely. And if you want to close one side and redeploy that capital, Kalshi's ACH withdrawal takes one to three business days unless you funded via crypto first.

What started as an 8-cent, 13%-looking edge is, after fees alone, closer to a 5-cent edge, held across two accounts with different resolution mechanics and different withdrawal speeds, for however long the event takes to resolve. That's not arbitrage in the textbook sense. It's a directional position with extra steps and extra risk, dressed up as a locked-in spread.

What Does a Persistent Price Gap Actually Tell You?

This is the actual value of comparing platforms side by side: not racing to fill a gap, but reading why it exists. A gap driven by different resolution rules or different regulatory access is structural, and it won't close until the event resolves. A gap driven by one platform's order book being thinner than the other's is temporary, and it tends to close as liquidity finds it.

Reading Kalshi's order book next to Polymarket's tells you more about who's pricing each outcome, a US bank-account holder clearing full identity verification versus a global wallet-holder with none, than it tells you about a trade to execute. The same holds comparing Limitless, Predict.Fun, and Opinion against the two larger venues.

A spread that survives fees, KYC friction, and settlement-timing risk is a signal about who's trading where and why, not a standing invitation. For a fuller platform-by-platform breakdown beyond arbitrage mechanics, see our Kalshi vs. Polymarket comparison.

Sources

• Kalshi. "Fees." Help Center. help.kalshi.com
• Kalshi. Fee Schedule. kalshi.com/fee-schedule
• Kalshi. "Crypto Withdrawals." Help Center. help.kalshi.com
• Kalshi. DCM Rulebook (Rules 6.3, 7.2). kalshi.com/regulatory/rulebook
• Polymarket. "Trading Fees." Help Center. help.polymarket.com
• Polymarket. "Resolution." Documentation. docs.polymarket.com
• Polymarket. "How Are Markets Disputed?" Help Center. docs.polymarket.com
• Predict.Fun. "Predict Fees and Limits." docs.predict.fun
• Opinion. "Fees." Documentation. docs.opinion.trade
• Limitless Exchange. "Market Resolution." Documentation. docs.limitless.exchange

PredictionHero aggregates publicly available prediction market data for informational purposes only. This is not financial advice. Prediction markets may not be available in all jurisdictions.