← all strategiesNegRisk Multi-Outcome Arbitrage Scanner
Arbitragescore 7found Jun 27, 2026github.com/matthewnyc2/arbitrage
Scan Polymarket's NegRisk multi-outcome markets for mispricings where the sum of all outcome probabilities deviates from 1.0, allowing risk-free or positive-EV hedged positions.
How It Works
Polymarket's NegRisk mechanism is used for multi-outcome markets (e.g., 'Which team wins the tournament?') where exactly one outcome resolves YES. In a fairly priced market, all outcome prices sum to approximately 1.0 plus fees. When this sum deviates materially — either above (overpriced, sell all) or below (underpriced, buy all) — an arbitrage exists. This scanner uses Python asyncio for concurrent scanning of multiple markets simultaneously to detect these violations in near-real-time.
Entry/Exit Conditions
Entry is triggered when the sum of outcome prices in a NegRisk market deviates from the no-arbitrage boundary by enough to cover transaction costs and still yield profit. The bot would buy all outcomes (if underpriced sum) or sell all outcomes (if overpriced sum). Exit occurs at market resolution — since one outcome pays 1.0 and the rest pay 0, the hedged book locks in the spread at entry.
Verified Results
The repository notes 60 passing tests and a live browser demo, indicating a production-grade implementation with meaningful engineering validation. Paper-trading mode is available for live forward testing.
Implementation Notes
Async Python architecture is well-suited to scanning many markets concurrently without blocking I/O. The browser demo suggests a real-time monitoring interface. Production-grade framing implies error handling, reconnection logic, and position tracking are implemented.
Risks
NegRisk arbitrage windows are typically very short-lived and are actively hunted by professional market makers. Execution risk is high — if one leg of the multi-outcome hedge fails to fill at the modeled price, the position becomes directional. Gas/transaction fees on-chain and Polymarket's own fees must be carefully modeled. Liquidity in individual outcome legs may be insufficient for meaningful size.