Superforecasting reality check: Evidence from a small pool of experts and expedited identification

Abstract

Superforecasting has drawn the attention of academics - despite earlier contradictory findings in the literature, arguing that humans can consistently and successfully forecast over long periods. It has also enthused practitioners, due to the major implications for improving forecast-driven decision-making. The evidence in support of the superforecasting hypothesis was provided via a 4-year project led by Tetlock and Mellers, which was based on an exhaustive experiment with more than 5000 experts across the globe, resulting in identifying 260 superforecasters. The result, however, jeopardizes the applicability of the proposition, as exciting as it may be for the academic world; if every company in the world needs to rely on the aforementioned 260 experts, then this will end up an impractical and expensive endeavor. Thus, it would make sense to test the superforecasting hypothesis in real-life conditions: when only a small pool of experts is available, and there is limited time to identify the superforecasters. If under these constrained conditions the hypothesis still holds, then many small and medium-sized organizations could identify fast and consequently utilize their own superforecasters. In this study, we provide supportive empirical evidence from an experiment with an initial (small) pool of 314 experts and an identification phase of (just) 9 months. Furthermore - and corroborating to the superforecasting literature, we also find preliminary evidence that even an additional training of just 20 minutes, can influence positively the number of superforecasters identified.