Bellwether is an automated researcher that studies political prediction markets.

The Research

Prediction markets are increasingly cited by journalists, pollsters, and campaigns as authoritative forecasts of political outcomes. But how accurate are they? Do markets systematically favor one party? Are they better at predicting presidential races than down-ballot contests?

Bellwether tracks thousands of political markets across platforms to answer these questions with data.

Unlike traditional academic research—which freezes findings in papers that go stale—Bellwether produces living estimates. As new elections resolve and new data arrives, analysis updates automatically. The goal isn’t a static publication but continuously maintained knowledge infrastructure.

What We Study

• Forecasting accuracy for presidential, Senate, House, and gubernatorial elections
• Calibration across probability ranges
• Platform comparison between Polymarket and Kalshi
• Partisan bias in market predictions

Methodology

Bellwether runs autonomously every day—discovering new markets, pulling price histories, classifying political categories, and recalculating accuracy metrics without human intervention. Markets are classified using a multi-stage verification pipeline to ensure high precision. We measure accuracy using Brier scores and calibration curves.

This approach treats research as infrastructure rather than a one-time effort: findings that maintain themselves and improve as more data becomes available.

Why Bellwether

Academic research typically works like this: spend months on a paper, publish it, and the findings freeze in time. By the time anyone reads it, the world has moved on.

Bellwether is an experiment in a different approach—research that operates at the speed of the phenomena it studies. When markets open, Bell tracks them. When elections resolve, Bell updates its accuracy estimates. When patterns emerge, they surface in the data automatically.

This is a prototype of what we believe research institutions could become: small teams of expert researchers directing automated systems to produce rigorous, replicable, and continuously updated knowledge on questions that matter.

Team

Acknowledgments

This research is supported by the Hoover Institution and the Stanford Graduate School of Business. Market data provided by Dome.