ACTIVE LEARNING - MONTE-CARLO RESULTS

The revival of interest in learning in stochastic control models invites the examination of some unanswered issues from earlier work. Following in the tradition of Prescott and MacRae we consider Active Learning to examine the question of when these methods are superior to Passive Learning. In the earlier literature it was not possible to address this question, because computer speeds were not great enough. However, modem supercomputers make it possible to reopen this issue. In this paper we will ''amine two well-known stochastic control models, the MacRae model and the Abel model. Although no general conclusions can be drawn, the results from the MacRae model and the Abel model indicate that in certain cases Active Learning may be superior to Passive Learning and Certainty Equivalence. In the almost all cases we have analyzed so far, Active Learning is superior to Passive Learning which in turn is superior to Certainty Equivalence.