5 Prediction Market Arbitrage Opportunities Most People Miss

In liquid financial markets, arbitrage opportunities close in milliseconds. Algorithmic traders with co-located servers ensure that the same stock trades at the same price on NYSE and Nasdaq within fractions of a penny. Prediction markets are nothing like that.

Liquidity is fragmented across platforms that don't talk to each other. Polymarket runs on Polygon. Kalshi is a CFTC-regulated exchange with its own infrastructure. Betfair operates a UK-regulated exchange. PredictIt (when active) was a CFTC no-action-letter market with strict position limits. Each platform has its own user base, its own market makers, and its own order flow. Prices diverge because there's no mechanism to force convergence — no cross-platform market makers with the speed and capital to close every gap.

Add regulatory barriers (US residents can't easily trade on Betfair; Polymarket restricts US users; Kalshi has limited market selection) and the result is persistent price differences of 3-10% on identical events. Those differences are your opportunity. But they're smaller, harder to capture, and riskier than they look on paper.

The textbook arb. Same event, different prices on two platforms. Buy the cheap side on Platform A, buy the opposite side on Platform B, and one of them pays $1.00 regardless of the outcome.

Example: "Will the next UK Prime Minister be from the Labour Party?" trades at $0.55 YES on Polymarket and $0.50 YES on Betfair (which means NO on Betfair is $0.50). You buy YES on Betfair at $0.50 and NO on Polymarket at $0.45 ($1.00 minus $0.55). Total cost: $0.95 per pair. Guaranteed payout: $1.00. Gross profit: $0.05 per pair, or 5.3%.

Now subtract fees. Polymarket charges 2% on winnings. Betfair charges 2-5% commission on net winnings. If YES wins, your Betfair contract pays $1.00 minus ~3% commission = $0.97. Your total return on the $0.95 outlay is $0.97, or 2.1%. If NO wins, your Polymarket contract pays $1.00 minus 2% fee = $0.98. Return: 3.2%. Either way, you profit — but it's 2-3%, not 5%.

Execution steps: (1) Fund accounts on both platforms simultaneously. (2) Monitor price differences using an aggregator or custom script polling both APIs. (3) When the combined cost of YES on one platform plus NO on the other drops below $0.97 (your fee-adjusted break-even), execute both sides within seconds. (4) Wait for resolution. Capital is locked on both platforms until the event resolves.

Risks: execution lag (one side fills, the other moves before you can fill), platform default (unlikely but non-zero on unregulated platforms), and settlement differences (the two platforms might define the event slightly differently). That last one has burned arbers before — a contract that resolves YES on Polymarket might resolve N/A on Kalshi if their criteria differ by one word.

This is subtler and more profitable. Two markets are logically related, but priced inconsistently. You don't need two platforms — this can happen on a single exchange.

Example: On Polymarket, "Party X wins the presidency" trades at $0.45. But "Party X wins Pennsylvania," "Party X wins Michigan," and "Party X wins Wisconsin" all trade above $0.65. If you believe (correctly, based on historical data) that winning all three states gives Party X a ~90% chance of winning the presidency, then $0.45 for the presidency is underpriced relative to the state-level contracts.

The trade: buy "Party X wins presidency" at $0.45. If the state-level prices are right, the "fair" price for the presidency contract is closer to $0.55-0.60. You're buying at a discount to the implied probability from related markets.

This isn't risk-free like cross-platform arb. The correlation might not hold — maybe the candidate wins all three states but loses the presidency through some other path. But the expected value is positive if your correlation model is accurate.

More examples: "Company X IPOs in 2026" at $0.30 while "Company X files S-1 in Q1 2026" trades at $0.40. If filing an S-1 leads to IPO within the year 80% of the time, the IPO contract is underpriced at $0.30 (fair value: $0.32 from the S-1 path alone, plus probability of IPO without Q1 filing).

Finding these requires domain expertise. You need to know the causal and statistical relationships between events well enough to spot inconsistencies. The market doesn't correct these quickly because each market has its own liquidity pool, and cross-market analysis requires effort most casual traders don't invest.

News breaks in one time zone while traders in another are asleep or inattentive. Prices on globally accessible platforms adjust, but smaller or region-specific platforms lag by hours.

Example: A European court ruling drops at 9 AM CET (3 AM Eastern). Betfair reprices the relevant contract within minutes because its UK/European user base is active. Polymarket's US-dominated user base doesn't react until 7-8 AM Eastern, four to five hours later. In that window, there's a price gap.

The trade: buy the correctly-priced side on the slow platform before its users wake up and reprice. If Betfair has already moved "EU regulation passes" from $0.40 to $0.70 based on the ruling, and Polymarket still shows $0.45, you buy on Polymarket and wait for the US morning catch-up.

This is directional, not risk-free. You're betting that the slow platform will converge to the fast platform's price. Usually it does, but sometimes the fast platform overreacted and the slow platform's price was actually more accurate.

Profit: 2-6% gross, but the window is extremely short. As more traders adopt cross-platform monitoring and as Polymarket's user base becomes more international, these windows have compressed from hours to minutes. You need to be pre-funded on both platforms and watching news feeds at off-hours. This is not a casual strategy.

Realistic expectation: you'll find 1-3 time-zone arb opportunities per month if you're actively monitoring. After fees and the occasional misfire, net returns are modest. This works as a supplement to other strategies, not as a standalone approach.

Sometimes what looks like an arbitrage opportunity disappears when you properly account for all fees. And sometimes it doesn't — which is the opportunity.

Example: "Event Y happens" trades at $0.60 on Platform A (2% fee on winnings) and $0.42 NO on Platform B (no fee but wider spread). Combined cost: $1.02. Looks like a loss. But Platform A's fee only applies if you win, and you're guaranteed one side wins. The fee-adjusted cost: you pay $0.60 on A and $0.42 on B. If YES: A pays $1.00 minus 2% fee = $0.98. Net: $0.98 - $1.02 = -$0.04. If NO: B pays $1.00, no fee. Net: $1.00 - $1.02 = -$0.02. This one actually is a loss on both sides. Bad arb.

But change the numbers: YES at $0.55 on A, NO at $0.40 on B. Cost: $0.95. If YES: A pays $0.98 (after 2%). Profit: $0.03. If NO: B pays $1.00. Profit: $0.05. Real arb — but only because you accounted for the fee asymmetry.

The point: most "arbs" that appear to exist at a 3-5% spread are actually losers after fees. The ones that survive fee adjustment typically need a 5%+ raw spread. Most casual screeners don't adjust for fees correctly, which is why these get flagged as opportunities but don't actually work.

Where to find real fee-adjusted arbs: look for platform combinations where one side charges no fee or a very low fee. Betfair with a low commission tier (high-volume traders get 2%) paired with a no-fee DeFi platform creates the best fee-adjusted math. But the no-fee platforms tend to have poor liquidity, creating a different problem.

This is the sneakiest arb type and the one most people miss entirely. Two platforms list contracts on what appears to be the same event, but their resolution criteria differ in ways that create exploitable outcomes.

Example: Platform A lists "US enters recession in 2026" defined as two consecutive quarters of negative GDP growth (the textbook definition). Platform B lists "US enters recession in 2026" defined as the NBER officially declaring a recession. These are different things. The NBER uses a broader set of indicators and typically declares recessions 6-12 months after they technically begin. GDP could go negative for two quarters, resolving Platform A's contract as YES, while the NBER hasn't declared anything yet, keeping Platform B's contract in limbo or resolving NO for 2026.

The trade: if you believe a technical recession is likely but an NBER declaration in the same calendar year is unlikely, you buy YES on Platform A and NO on Platform B. Both could pay out. That's not an arb in the traditional sense — it's a correlated bet that exploits definitional differences. But if the combined cost is less than $1.00 and both paying out is a plausible scenario, the expected value can be strongly positive.

More common: contracts on "Person X is elected" where one platform defines "elected" as winning the vote count and another defines it as being inaugurated. If there's a legal challenge between the vote and inauguration, these contracts can diverge.

Finding these requires reading the fine print. Not the contract title — the actual resolution criteria, which are often buried three clicks deep. The traders who profit from resolution arb are the ones who read settlement rules the way lawyers read contracts: every word matters.

Risks: you might be wrong about the definitional difference. "I thought Platform B defined it differently" is not a defense when both contracts resolve the same way. Do your research before deploying capital.

In traditional markets, arbitrage is competed away by well-capitalized firms with superior technology. In prediction markets, it persists for three reasons.

First, fragmented liquidity. There's no cross-platform order routing. Each platform is an island. A single entity would need accounts funded on 3-4 platforms simultaneously, with enough capital on each to execute meaningful positions. Most traders don't bother.

Second, regulatory barriers. A US-based trader can easily use Kalshi but faces legal questions on Polymarket and practical barriers on Betfair. A UK-based trader can use Betfair easily but can't access Kalshi. Regulatory fragmentation prevents the natural arbitrageurs from operating across all platforms.

Third, the economics are thin. After fees, a typical cross-platform arb yields 2-3% with capital locked for weeks or months. For a hedge fund, that's not worth the operational overhead. For an individual, it's meaningful but not life-changing. The market is in a sweet spot where the opportunity is too small for institutions but large enough for dedicated individuals.

Manual scanning is possible but tedious. Most serious arb traders use some combination of these tools.

API access: Polymarket, Kalshi, and Betfair all offer APIs. You can pull current prices programmatically and run cross-platform comparisons. Polymarket's API is open and well-documented. Kalshi's requires an account. Betfair's requires a paid subscription for streaming data.

Aggregator sites: OddsJam, Oddschecker, and a handful of crypto-specific aggregators (like markets.wiki) show prices across platforms side by side. They don't account for fees, so you'll need to do that math yourself. These sites are better for spotting candidates than for executing trades.

Custom scripts: a Python script polling two platform APIs every 60 seconds, comparing prices, and alerting you when a spread exceeds your fee-adjusted threshold is table stakes for regular arb traders. The code is straightforward — the hard part is handling edge cases (API rate limits, order book depth, partial fills).

Don't expect a turnkey "arb bot" that prints money. If such a thing existed and worked reliably, the arbs would close instantly. The tools help you find candidates faster. The judgment — whether the arb is real, whether the resolution criteria match, whether the liquidity is there — is still yours.