Right Decisions from Wrong Predictions: A Mechanism Design Alternative to Individual Calibration

Decision makers often need to rely on imperfect Given these limitations, we study alternative mechanisms probabilistic forecasts. While average to convey confidence about individual predictions performance metrics are typically available, to decision-makers. it is difficult to assess the quality of individual forecasts and the corresponding utilities. To We consider settings where a single forecaster provides convey confidence about individual predictions predictions to many decision makers, each facing a potentially to decision-makers, we propose a compensation different decision making problem. For example, mechanism ensuring that the forecasted a personalized medicine service could predict utility matches the actually accrued whether a product is effective for thousands of individual utility. While a naive scheme to compensate patients [19, 20, 2]. If the prediction is accurate decision-makers for prediction errors can be for 70% of patients, it could be accurate for Alice exploited and might not be sustainable in the but not Bob, or vice-versa. Therefore, Alice might be long run, we propose a mechanism based on hesitant to make decisions based on the 70% average fair bets and online learning that provably accuracy. In this setting, we propose an insurance-like cannot be exploited. We demonstrate an application mechanism that 1) enables each decision maker to confidently showing how passengers could confidently make decisions as if the advertised probabilities optimize individual travel plans based were individually correct, and 2) is implementable on flight delay probabilities estimated by an by the forecaster with provably vanishing costs in the airline.