Abstract
We propose an elicitation method, Reciprocal Scoring (RS), that challenges forecasters to predict the forecasts of other forecasters. Two studies show how RS can generate accurate forecasts of otherwise unanswerable questions. Study 1 establishes the epistemic credibility of RS: forecasters randomly assigned to use RS were as accurate as forecasters predicting objectively resolvable outcomes using a proper scoring rule—and both groups were more accurate than a control group that felt accountable to neither intersubjective RS metrics nor objective metrics. Study 2 establishes the practical value of RS. We ask highly accurate forecasters to predict each other’s forecasts of the effect of government policies on COVID-19 mortality, yielding a real-time ranking of the expected effectiveness of pandemic-containment policies. As in Study 1, RS forecasters converged but in this case on policy recommendations that stand up to scrutiny, even with the benefit of hindsight. The core contribution of RS is its power to create accountability for accuracy in policy debates that have long been stalemated by the absence of accountability.
