Learning from Experience with Delayed Feedback

摘要

Many important settings in individual and organizational life involve allocating resourcesbetween different types of activities with different delays between allocation and results.Examples include factory managers choosing to spend time on production now or on processimprovement that may boost output later, and individuals choosing to get a job now or stay inschool to get a better job later. Empirical studies show that learning is difficult in dynamicsystems and people often fail to learn from experience in the presence of delays. Understandingthe processes that hinder learning from experience is central to improving learning and decisionmaking.We use a formal model to examine the effect of time delays on learning fromexperience. The model represents a decision-maker engaged in a continuous time allocation taskwho learns from his performance as he tries to improve the payoff determined by his ownactions. Our analysis shows that in an easy learning task where the payoff landscape is smoothand has only a single peak, we can still observe sub-optimal performance. Moreover, thedecision maker can learn to believe that the sub-optimal performance is really the best she cando.