Model comparisons using tournaments: likes, "dislikes," and challenges

Abstract

We examine the value added of tournaments that compare the predictive power of behavioral models. The advantages and disadvantages of this particular method have not been systematically discussed in the literature; here we aspire to do so. We conclude that tournaments are a useful addition to our arsenal of methods and a significant step forward in the study of the predictive power of behavioral models. Such tournaments are not without limitations, however, and we identify and illustrate current limitations, propose solutions to overcome some of them, and identify a research agenda to address the remaining.