Introducing PriceArb: Historical Edge Detection for Prediction Markets
Today we're launching PriceArb, a tool that identifies when prediction market prices diverge from historical probabilities.
The Problem
Prediction markets like Kalshi and Polymarket price event contracts based on crowd sentiment. But crowds are emotional. When a team scores, the market reacts. Sometimes it overreacts.
The question we're answering is simple: what does history say should happen from this exact game state?
How It Works
We compare two numbers:
- Market Probability: What the market currently prices the outcome at
- Historical Probability: What historically happens from this game state
The edge is the difference between these two numbers.
We calculate historical probabilities using two complementary methods:
- GameStateEdge™: Based on in-game situations (score, time, quarter)
- RecordEdge™: Based on team records entering the game
An Example
Say the Bills are down 7 with 8 minutes left. The market prices their win probability at 28%. But historically, the Bills in that exact position win 41% of the time.
That's a +13% edge.
| Metric | Value |
|---|---|
| Market Price | 28% |
| Historical Rate | 41% |
| Edge | +13% |
Further more, let's say the Bills went into that game with a 12-5 record and are playing against a team with a 3-14 record. Historically, teams with a 12-5 record trailing by 7 with 8 minutes left win 44% of the time. This is a +16% edge via RecordEdge™.
The Details
For more details on how we calculate each edge type, see our edges documentation.
Why This Matters
Other tools compare market to market (Kalshi vs Polymarket). We compare market to history.
This is a fundamentally different signal:
| Existing Tools | PriceArb |
|---|---|
| "Kalshi says 35%, Polymarket says 38%" | "Market says 35%, history says 48%" |
| Arbing actual market mispricing | Arbing real-time crowd psychology |
| Works when platforms disagree (will disappear rapidly as these markets get more efficient and more popular) | Works when the market is mispricing historical means, based on human psychology, and will work on any platform |
What's Next
We're starting with NFL games but the methodology works for any event type with historical precedent:
- Sports (all major leagues)
- Elections (with polling data)
- Economic events (Fed decisions, earnings)
Check out the dashboard to see live edges, or learn more about our edge calculations.
All trading involves risk. This is a research tool, not financial advice.