A1252
Title: Responsibly emboldening predictions via boldness-recalibration
Authors: Adeline Guthrie - Virginia Tech (United States) [presenting]
Christopher Franck - Virginia Tech (United States)
Abstract: Probability predictions are essential for informing decision-making across many fields. The purpose is to enable better decision-making by improving probability forecasts in terms of their calibration and boldness - which are essential properties of useful forecasts. A Bayesian model selection approach is proposed to assess calibration and a strategy for boldness-recalibration that enables practitioners to responsibly embolden predictions subject to their required level of calibration. Specifically, the user is allowed to pre-specify their desired posterior probability of calibration and then maximally embolden predictions subject to this constraint. The key of these methods is demonstrated via real-world case studies pertaining to the prediction of housing foreclosures and ice hockey games.
