Penalized Utility Bayes Estimators
Carlos Carvalho, University of Texas at Austin
We describe how utility functions with an explicit parsimony term can be used to extract accessible summaries from complex posterior distributions. We first develop the idea in the context of variable selection in linear and non-linear regression models. We then demonstrate the use of the ideas in an empirical application where we seek to select a small set of exchange traded funds for a passive, long term investment strategy. This is joint work with Richard Hahn, Robert McCulloch and David Puelz.