Aggregation mechanisms for crowd predictions
通过实验比较算术/几何平均、中位数、连续双向拍卖和密封报价拍卖等聚合机制，发现连续双向拍卖市场在聚合分散信息方面显著优于其他方法，且只有信息最充分的参与者能获得超额收益。
When the information of many individuals is pooled, the resulting aggregate often is a good predictor of unknown quantities or facts. This aggregate predictor frequently outperforms the forecasts of experts or even the best individual forecast included in the aggregation process ("wisdom of crowds"). However, an appropriate aggregation mechanism is considered crucial to reaping the benefits of a "wise crowd". Of the many possible ways to aggregate individual forecasts, we compare (uncensored and censored) arithmetic and geometric mean and median, continuous double auction market prices and sealed bid-offer call market prices in a controlled experiment. We use an asymmetric information structure, where participants know different sub-sets of the total information needed to exactly calculate the asset value to be estimated. We find that prices from continuous double auction markets clearly outperform all alternative approaches for aggregating dispersed information and that information lets only the best-informed participants generate excess returns.