MarketTruth: Prediction Market Mispricing Detector

Detects mispriced prediction markets by comparing live news sentiment vs Polymarket odds. Find underpriced YES markets before the crowd does.

Comment

Inspiration

Prediction markets are supposed to be the most efficient markets in the world. But are they really? We wanted to find out where news sentiment and market probability diverge — and build a system to detect it automatically.

What it does

MarketTruth fetches live data from Polymarket's CLOB API (1000+ active markets), extracts keywords from each market question, pulls relevant news headlines via Google News RSS, scores sentiment using VADER NLP, and calculates a mispricing score showing where news strongly contradicts market odds.

Markets with high positive scores = news is bullish but market says unlikely (underpriced YES) Markets with negative scores = news is bearish but market says likely (overpriced YES)

How we built it

  • Data: Polymarket CLOB API for live prediction market odds
  • NLP: VADER sentiment analysis on Google News RSS headlines
  • Analysis: Custom mispricing score formula normalised to [-1, 1]
  • Visualisation: Publication-quality matplotlib charts on dark theme
  • Deployment: FastAPI endpoint with 10-minute caching serving live alerts

Challenges

Getting consistent news data for niche prediction market questions was tricky. We solved it with keyword extraction and fallback query strategies.

What we learned

Markets misprice events surprisingly often — especially resolved events where odds lag behind news by hours.

What's next

Real-time alerts via Discord/Telegram when mispricing exceeds threshold. Integration with trading systems for automated arbitrage execution.

Built With

Share this project:

Updates