A nonorthodox Bayesian approach to statistical inference

In this paper we discuss by examples an atypical approach to statistical inference based on the optimization of the predictive performances of a parametric model. We show how to presented approach permits to deal with several statistical problems in an unified way