Regularized Aggregation of One-Off Probability Predictions
提出一种无需用户干预的贝叶斯聚合器，能高效处理大量一次性概率预测，在美国情报界四年预测锦标赛数据上比简单平均法降低约20%的平方误差，主要得益于校准改进。
How much can rational people really disagree? If we can understand the limits of such disagreement, can we remove noise by labeling excess disagreement as irrational and then construct a group belief based on everyone's rational beliefs? Based on this idea, “Regularized Aggregation of One-Off Probability Predictions” by Satopää proposes a Bayesian aggregator that requires no user intervention and can be computed efficiently even for a large number of one-off probability predictions. To illustrate, the aggregator is evaluated on predictions collected during a four-year forecasting tournament sponsored by the U.S. intelligence community. The aggregator improves the squared error (a.k.a., the Brier score) of simple averaging by around 20% and other commonly used aggregators by 10%−25%. This advantage stems almost exclusively from improved calibration. An R package called braggR implements the method and is available on CRAN.