Heikki Kauppi seminaariesitelmä 27.5

Professori Heikki Kauppi pitää seminaariesitelmän aiheesta Optimal Binary Prediction 27.5. klo 10-12 seminaarihuoneessa 469, Publicumin 4. krs.

Abstract: It is predicted whether one of two states realizes. Individuals receive different net utilities from correct predictions. The problem is to find the set of binary prediction rules given observable predictors such that everybody’s expected utility is maximized. I introduce a concept, the efficient frontier (EF), that characterizes the optimal prediction rules in a similar fashion as the ROC curve (commonly used in biometrics), but is more general than the latter. I study the properties of the EF and examine conditions under which the associated (optimal) prediction rules can be identified non-parametrically and semi-parametrically. I provide guidelines to handling the estimation problem in certain settings. The analysis makes useful insights to the binary prediction problem and provides tools for advancing methods for predicting binary outcomes in various fields.