Christophe Abraham
ENSAM-INRA, Montpellier, France

We propose a class of loss functions for modeling the incomplete preferences of the decision maker.  The class is built upon two extreme losses.  We show how to compute the set of Bayes actions associated with that class.  This result is applied to study the legitimacy of the approximation of the subjective loss by a classical loss (quadratic, LINEX and entropy losses).  Then, the asymptotic behavior of the Bayes actions set is investigated:  it is shown that the Bayes actions set can be reduced up to a limit set by adding observations.  Finally, we conclude by connecting these results to the loss-prior robustness.

Seminar to be held in Room 107, 24 Hillhouse Avenue at 4:15 pm