Su alcune classi complete di funzioni di decisione

Bayesian solutions are a complete class under quite strong assumptions. Under weaker assumptions, in particular, removing the assumption that the parameter space is compact, one needs to include generalized Bayesian solutions, or limits of Bayesian solutions. A conjecture is that Bayesian solutions w.r.t. a finitely additive prior are a complete class. We give a first answer to this conjecture. We show that extended Bayesian solutions, as defined by Wald, are Bayesian solutions w.r.t. appropriate previsions of loss; and we discussed the problem of representing an admissible solution as a coherent Bayesian decision rule.