A new working paper analyzing 1,200 geopolitical forecasting events across a five-year period finds that prediction market prices outperform structured expert panels by 18 percentage points on Brier score — a metric that penalizes both overconfidence and underconfidence.

Why this matters beyond the headline

The 18% figure is the average. The gap widens dramatically for long-horizon events (6–18 months out) where expert forecasters showed consistent overconfidence, especially in domains with high media coverage. Prediction markets, by contrast, corrected faster as new information arrived — sometimes within hours of a news event.

The paper's most striking finding: expert forecasters were most wrong on the events they were most confident about. Prediction markets showed no such correlation between confidence and accuracy.

The mechanism

This isn't magic — it's incentives. Experts face reputational risk for bold claims and career risk for being wrong publicly. Markets face financial loss for being wrong and financial gain for updating faster than others. The result is that markets aggregate information more efficiently, particularly at moments when conventional wisdom is most likely to be wrong.