Forecast Aggregation

The simplest aggregation approach is an arithmetic mean of all individual probability forecasts. This exploits the Wisdom of the Crowd effect and typically beats the median individual forecaster. More sophisticated methods weight contributors by their historical Brier Score or Log Score performance, giving influence proportional to demonstrated accuracy. Extremizing is an important refinement: when averaging shrinks probabilities toward 50%, the composite understates certainty that exists in the aggregate. Multiplying log-odds by a factor greater than one pushes the result back toward the tails and has been shown empirically to improve accuracy on competitive forecasting platforms like Metaculus and Good Judgment Open. The choice of aggregation method has outsized effects on final Calibration.

A Prediction Market performs continuous implicit aggregation through its Order Book. Every trade is a vote weighted by both capital and conviction, so better-informed traders with larger stakes have greater influence on the final Market Price. This differs from simple averaging: it is a form of incentive-weighted aggregation where participants bear real financial consequences for errors. Research comparing explicit algorithmic aggregation against Polymarket and Kalshi prices generally finds that well-run markets are competitive with the best algorithmic aggregates on short-horizon binary events. Combining market prices with prior probabilities from historical base rates and explicitly elicited forecasts can outperform either source alone, particularly for low-volume markets with thin Liquidity.