Prediction Market Arbitrage: Hidden Costs, Fees, and the Real Math
The idea sounds beautiful: buy "Yes" on one platform for $0.40, buy "No" on another for $0.55, and collect $1.00 for a total outlay of $0.95. Five percent risk-free.
In practice, three layers of costs, several non-obvious risks, and different resolution rules across platforms stand between you and those 5%.
Three Layers of Costs
1. Platform Trading Fees
The key trap: many DEX platforms use dynamic fees that depend on event probability.
| Platform | Taker Fee | Maker Fee | Payout Fee | Notes |
|---|---|---|---|---|
| Polymarket | 0–1.8% (peaks at 50/50) | 0% + rebate | 0% | Geopolitics markets: 0% |
| Limitless | 0.03–3% | 0–1.5% | — | Lower fee for likelier outcomes |
| Predict.fun | 1–2% (standard) | — | — | Often zero-fee promos + cashback |
| Opinion | 1–2% | — | — | Fee goes to LP pool |
| Kalshi | Fixed | — | — | Regulated exchange, USD, not Web3. US residents only |
2. Network Gas Fees
| Network | Platforms | Gas per TX | Native Token |
|---|---|---|---|
| Polygon | Polymarket | Fractions of a cent | MATIC / POL |
| Base | Limitless, Opinion | $0.01–0.05 | ETH |
| BNB Chain | Predict.fun | $0.03–0.10 | BNB |
3. Cross-Chain Bridges
Bridge fees: $1–5 per transfer + percentage. Time: 1–15 minutes (spread may close while you wait).
Three Critical Risks
1. Slippage
AMM worsens your price on large orders due to thin liquidity.
2. Resolution Risk — Where 90% of Arbitrageurs Burn
Same event, different platforms, different resolution rules and data sources. Your positions can lose on both sides simultaneously.
3. Timing Lag
No atomic cross-chain execution. Second leg may execute after the spread closes.
Minimum Spread for Break-Even
| Cost Item | Approximate |
|---|---|
| Platform A fee (taker) | 1.0–1.8% |
| Platform B fee (taker) | 1.0–2.0% |
| Gas (4–6 transactions) | $0.10–0.50 |
| Slippage (both legs) | 0.5–2.0% |
| Total | 3–6% |
Profitable arbitrage realistically starts at 7–10% spreads.
Premarket.me — Aggregator for Arbitrage
Manually monitoring spreads across five platforms is impractical. Premarket.me solves this: it's a prediction market aggregator with a unified API that combines data from Polymarket, Limitless, Predict.fun and other platforms.
The key endpoint — /api/arbitrage — returns pre-calculated pairs: where to buy "Yes", where to buy "No", total cost, profit %, APR adjusted for time to expiry, and liquidity depth in USD.
Open-Source Arbitrage Bot
A ready-made Go arbitrage bot built on the Premarket API: suenot/premarket-arbitrage-bot.
Scanner (Premarket API) → Filter (profit/APR/depth) → Display → Executor (Polymarket CLOB)
The bot polls /api/arbitrage, filters pairs by configurable thresholds (MIN_PROFIT_PCT, MIN_APR, MIN_DEPTH_USD), and optionally executes trades via Polymarket CLOB with EIP-712 signing. Default mode: DRY_RUN=true (monitoring only).
Links
- 🌐 Polymarket · Limitless · Predict.fun · Opinion · Kalshi
- 🌐 Premarket.me: premarket.me · API Docs
- 💻 Arbitrage Bot: suenot/premarket-arbitrage-bot
Conclusion
Prediction market arbitrage is not "free money." It's systematic work with three layers of costs, cross-chain logistics, and resolution risk that can wipe out both legs of a trade. Minimum spread for break-even is 5%; for profit — 7–10%.
But the key insight: mechanical arbitrage alone is not enough. The real edge comes from building AI agent analytics around each specific bet: real-time news monitoring, tracking social signals, analyzing on-chain activity of large players, and correlating with macro events. The spread between platforms is a symptom; the cause is information asymmetry. The winner is the one who understands that event probability has shifted — not the one who clicks the button fastest.
Pre-funding wallets, maker orders, and API automation are table stakes. The real edge is in the analytical layer on top of market data.
MarketMaker.cc Team
Recherche quantitative et stratégie
Read More
Futures-Spot Arbitrage: From Cash-and-Carry to DeFi-CeFi
Statistical Arbitrage and Pairs Trading in Crypto Markets: From Cointegration to the Kalman Filter
