CIRJE-F-13. Matsushima, Hitoshi, "Efficient Entrepreneurship", July 1998.

This paper considers a situation in which a decision maker chooses between the safe action and the uncertain action infinitely many times. The decision maker knows the payoff for the safe action, but does not know the payoff for the uncertain action which is determined by an unknown probability function. The decision may be influenced by the payoff-irrelevant context in which the current decision problem is to be considered. The context fluctuates according to another unknown probability function. The decision maker is modeled by a Markov learning rule with reflecting barriers which determines a state of mind in every period on the basis of past experiences. We argue that the context-dependence of decision making plays an important role in finding out the efficient action in the long run, because it causes the decision maker to gather unbiased information at any time. We show that there exists a learning rule according to which the decision maker succeeds to choose the efficient action in the long run irrespective of how payoff-uncertainty and context-uncertainty are specified. We also characterize the class of such efficient learning rules, and argue that it is necessary that the upper reflecting barrier, regarded as the maximal strength of confidence that the uncertain action is more profitable than the safe action, greatly surpasses the negative of the lower reflecting barrier regarded as the minimal strength of confidence.