Looking Backwards, Looking Inwards: Priors and Introspection

The three notions studied here are Bayesian priors, invariant priors and introspection. A prior for an agent is Bayesian, if it agrees with the agent's posterior beliefs when conditioned on them. A prior is invariant, if it is the average, with respect to itself, of the posterior beliefs. Finally, an agent is introspective, if he is certain of his own beliefs. We show that a prior is Bayesian, if and only if it is invariant, and the agent is almost surely introspective. We show how to edogenize priors, and how to express the events that an agent has a Bayesian or invariant prior. Finally, we study properties of the endogenized common prior.