Substack post uses Brazil election to test prediction market calibration against polls
A Substack post by Boz uses Brazil's upcoming presidential election as a case study for prediction market calibration. The post opens with a hypothetical where forecasters assign an 80% probability to Lula winning. In actual pricing, prediction markets favor Flávio Bolsonaro for the presidency even though polls show a runoff tie and Lula holds a first-round lead. The divergence echoes the 2024 US split between prediction market pricing and polling data. Boz examines how prediction markets perform in near-coin-flip electoral environments in Latin America.
Brazil's election is becoming a live experiment for whether prediction markets outperform polls in volatile democracies. Boz's framing matters because it treats Latin American elections as a calibration stress test, not a sideshow to US politics. Bolsonaro wins while polls called a tie, the case for market superiority gains a major emerging-market data point.
That would embolden platforms like Polymarket and Kalshi to pitch event contracts as more reliable than survey research in politically polarized countries. For traders, the divergence signals pricing opportunity; for academics, it creates a publishable natural experiment. A Bolsonaro victory would likely accelerate allocation to prediction-market data in election-risk models for Brazil and neighboring economies.