Learning Rational Expectations: The Finite State Case