We consider the problem of learning a probabilistic model from the viewpoint of an expected utility maximizing decision maker/investor who would use the model to make decisions (bets), which result in well defined payoffs. In our new approach, we seek good out-of-sample model performance by...