Prediction Market Arbitrage: How to Find Pricing Gaps and Profit Without Picking Winners
A bot made $150,000 running prediction market trades — without predicting anything.
Academic research documented over $40 million in arbitrage profits extracted from Polymarket alone between April 2024 and April 2025. That's money left on the table by the market — captured by traders who knew where to look.
Here's the thing nobody talks about in prediction markets: you don't always need to be right about the outcome to make money. Sometimes the market just misprices two related things, and the trade is just arithmetic.
Let's break down how this actually works — from the simplest version to the more sophisticated strategies that most participants never try.
The Core Idea: The Market Isn't Always Right
Prediction markets aggregate crowd wisdom. But crowds are human — and humans make inconsistent pricing decisions, especially across different platforms or across logically related markets.
When the same event is priced on Polymarket and Kalshi at different probabilities, or when two logically related markets on the same platform imply contradictory outcomes, a pricing gap exists. That gap is profit — if you can capture it before it closes.
Delta-neutral is the concept of structuring a position so you don't care which direction the outcome resolves. You win either way because you're capturing the spread between prices, not betting on direction.
Type 1: Cross-Platform Arbitrage (Polymarket vs. Kalshi)
The simplest version: same event, two different prices.
How it works:
Polymarket prices "Bitcoin above $100k by March" YES at $0.60.
Kalshi prices the same event YES at $0.55 (meaning NO is $0.45).
You buy YES on Kalshi at $0.55 and NO on Polymarket at $0.40. Total spent: $0.95. Payout regardless of outcome: $1.00. That's a 5.3% locked-in return.
Real examples from the market: when Polymarket prices an event at 60% and Kalshi at 55%, you buy YES on Kalshi and NO on Polymarket — locking in profit whatever happens.
The documented scale: IMDEA Networks Institute documented $40 million+ in arbitrage profits extracted from Polymarket alone over 12 months. A single automated bot executing 8,894 trades on Polymarket's 5-minute crypto markets reportedly generated nearly $150,000 without any directional prediction required.
The fee reality check:
This is where most retail traders get burned. Fees destroy margins at small spreads.
| Platform | Fee Structure |
|---|
| Polymarket (US) | 0.01% per trade |
| Polymarket (International) | 2% on net winnings |
| Kalshi | 0.7% |
If you're using Polymarket International + Kalshi, combined fees of 2.7% mean any spread under 2.7% is actually a loss. The practical rule: target spreads of 6%+ for reliable returns after fees, execution slippage, and capital costs.
There's also a capital efficiency problem: you need funds deployed on both platforms simultaneously. That doubles your capital requirement and cuts effective return rates roughly in half.
Type 2: Intra-Platform Arbitrage (Single Platform, Same Market)
One of the cleanest opportunities — and the most systematically exploited.
The rule: In any given market, YES + NO should equal $1.00. Always.
If YES trades at $0.52 and NO at $0.51, the total is $1.03. That's overpriced relative to the fundamental constraint. If YES is $0.44 and NO at $0.53, total is $0.97 — buy both and collect $0.03 per dollar when the event resolves.
In Polymarket's 5-minute crypto markets specifically, AI trading systems have exploited exactly this — YES + NO briefly dipping below $1 during volatile moments, allowing bots to buy both sides and lock in instant profit.
Why does this happen? Market makers update YES and NO prices asynchronously. During high-volatility moments — a sudden news event, a price spike — the bid/ask on the two sides can temporarily create these gaps before algorithms close them.
Speed is everything here. These opportunities typically last seconds. This is primarily a bot game.
Type 3: Correlated Market Arbitrage
This is where it gets more interesting — and harder to automate. Markets on logically related outcomes can imply contradictory probabilities.
Political combinatorial arbitrage:
Consider three Polymarket markets during an election cycle:
- "Democrats win Senate" — priced at 45%
- "Democrats win House" — priced at 40%
- "Democrats win both chambers" — priced at 30%
Wait. That's wrong. The probability of winning both must be lower than the probability of winning either individually. If "win both" is priced at 30% while "win Senate" is 45% and "win House" is 40%, there's an internal inconsistency.
QuantPedia documented systematic edges in exactly these kinds of correlated mispricings in prediction markets. A 2020 study on political markets found PredictIt regularly showed these internal contradictions, with the sum of all candidate win contracts sometimes exceeding $1.00.
Sports correlations:
"Team A wins championship" + "Team A's star player scores in the final" + "Final goes to overtime" — these are correlated but independently priced. Referencing them against each other creates combinatorial opportunities, especially when breaking news changes one market faster than related markets update.
Cross-asset correlation:
If Bitcoin options markets imply a 75% probability of BTC above $80k by July, but Polymarket prices the same outcome at 65%, you can buy the prediction market contract (underpriced by 10 percentage points) and hedge with an offsetting options position. The profit comes from the pricing discrepancy, not from Bitcoin's direction.
Type 4: Delta-Neutral Hedging with Crypto Derivatives
The most sophisticated version of the strategy: pair a prediction market position with an offsetting derivatives trade to isolate the pricing discrepancy while neutralizing directional exposure.
The setup:
You believe Polymarket is underpricing "Bitcoin above $80k by Q2 2026" at 65% — your analysis or options market data suggests 75% is more accurate.
- Buy YES on Polymarket at $0.65 (deploy, say, $1,000)
- Simultaneously short BTC futures in a notional amount that approximates your prediction market exposure
What happens:
- If Bitcoin rises, your prediction market YES position gains value AND moves deeper in the money. Your BTC short loses some — but you're exposed to the mispricing, not the direction.
- If Bitcoin falls, your futures short profits, partially offsetting prediction market losses.
The goal isn't to eliminate all risk — it's to isolate the pricing discrepancy as your source of return while hedging the underlying directional noise.
This is how institutional traders operate in traditional markets. Prediction markets are relatively new to this kind of sophisticated hedging, which is precisely why the opportunity exists.
Tools for Finding and Executing Arbitrage
Detection and monitoring:
For manual traders:
Focus on slower markets where bots have less advantage:
- Major political events (months out) — markets are liquid enough to arb but update slowly enough to give humans time
- Long-dated economic outcome markets — macro events where information spreads slowly
- Markets during breaking news — prices update asynchronously across platforms, creating temporary gaps
The Risks Nobody Mentions
Settlement risk: The two platforms might resolve the same event differently. Kalshi might use official government data; Polymarket uses a crowd-voted oracle (UMA). If resolution criteria diverge on an ambiguous outcome, your "risk-free" arb suddenly isn't.
Liquidity risk: You might see a 7% spread but only be able to buy $50 worth before the price moves. Large visible arbs are often tiny in absolute size, and scaling them requires either deep liquidity or many simultaneous small positions.
Platform risk: Both platforms need to function for your position to pay out. Smart contract risk, USDC depegging, platform insolvency — these are real in crypto in a way they aren't on traditional exchanges.
Timing/execution risk: You need both legs of the arb filled at the expected price. If leg one fills and leg two's price moves before you execute, you're now directional — with the opposite position to what you wanted.
Regulatory risk: Polymarket's international platform restricts US users. Cross-platform strategies may require using platforms with different geographic rules. Operating outside your jurisdiction's legal framework adds risk that doesn't exist in the spread calculation.
Is Retail Still Viable?
Honest answer: pure mechanical cross-platform arb is dominated by bots. Most prediction market arbitrage is done by automated programs that monitor prices and execute trades in milliseconds. By the time a human notices a 5% spread on 5-minute crypto markets, it's probably gone.
But retail traders still have real advantages in specific areas:
Slower markets: Political events months away, long-dated macro outcomes, and obscure sports markets have less bot competition. A persistent 3% spread on a 2027 election might hold for hours.
Complex correlations: Bots optimize for simple paired arb between two prices. Combinatorial opportunities across 3–4 logically related markets require more sophisticated analysis — this is where human reasoning can still find genuine edge.
Information-based mispricings: When breaking news should materially change a probability but markets haven't updated yet, you're not doing pure arbitrage — you're doing directional trading with a strong edge. That's a different category, but it's related.
The $40 million in documented arbitrage profits isn't a secret. It's just math, applied consistently, before the spread closes. The question is whether you can find and execute it in the relevant timeframe.
Start Simple, Then Build
If you're new to prediction market arbitrage, the progression:
Learn the platforms. Understand how Polymarket and Kalshi price markets, how resolution works, and where fees apply. Our Prediction Markets Overview has a side-by-side comparison.
Watch for YES+NO mispricing. Check a few liquid markets: does YES + NO ≈ $1.00? If you consistently see it below in fast-moving markets, that's your first live opportunity.
Study correlated markets. Start with political outcome markets where multiple related events are priced. Map the logical dependencies and look for internal contradictions.
Automate alerts. Even a simple notification when Polymarket vs. Kalshi spreads exceed your threshold is more efficient than manual price-checking.
Prediction markets are still early. The bots have captured the obvious opportunities. But the volume, number of markets, and complexity of correlations is growing faster than the sophistication of the automated traders — which means there's still room for thoughtful players.
Have you tried prediction market arbitrage? What worked, what failed, and what opportunities are you watching right now? The most useful edge discoveries in this community come from practitioners, not textbooks.
Related: The $POLY Token Analysis · How to Farm the Polymarket Airdrop · Prediction Markets Overview
Sources:
