Optimal guessing: Choice in complex environments

In this paper, we extend the analysis of our earlier work on boundedly rational learning in an i.i.d. setting [Easley and Rustichini, Econometrica 67 (1999) 1157–1184] to complex decision problems. We show that the axioms from our earlier analysis can be applied in this more complex setting, and along with some new axioms, they asymptotically yield expected utility maximization. Perhaps most important is our demonstration of a simple procedure that insures expected payoff maximization no matter what Markov process the underlying process on states follows. We view this result as providing a positive learning result for all worlds in which learning is possible.