Unknown user trade-off between anticipation and precision for classifier selection

Ascertain how users of the recession detection algorithm trade off anticipation (mean detection delay) versus precision (standard deviation of detection delay) when choosing among classifiers on the anticipation-precision frontier, and specify a selection or weighting procedure that reflects these preferences for real-time deployment.

Background

The paper constructs millions of recession classifiers and identifies a subset of perfect classifiers that lie on an anticipation-precision frontier. While a policymaker with mean-variance preferences over detection delay could select a single classifier using a trade-off parameter between anticipation and precision, the authors note that these preferences are not known.

To address this lack of preference information, the paper averages signals from multiple high-precision frontier classifiers into an ensemble. The explicit uncertainty concerns the unknown way users value early detection versus accuracy, which affects how one should select or weight classifiers for practical use.