Bayesian learning and expectations formation : anything goes

The paper proves that the Bayesian approach to learning and expectations formation implies no propositions that could conceivably be refuted by observation. For a (non-expanding) universe infinite in time but finite at any point of time, it is shown that by a suitable choice of priors, any betting strategy can be rationalized. Consequently, no sequence of future choices can be excluded on the basis of a decision maker's v. Neumann-Morgenstern utility function, her knowledge of past observations, and the assumption of Bayesian rationality. Since no betting strategy is irrational, Bayesianism is useless as a normative theory as well. -- Bayesianism ; Bayesian learning ; decision theory ; expectations ; learning ; rational expectations ; rationality ; rationalizability ; subjectively expected utility theory ; subjective probability