Unrealistic Expectations and Misguided Learning” (joint with Paul Heidhues and Philipp Strack).

Abstract: We explore the learning process and behavior of an individual with unrealistically high expectations about ability when outcomes also depend on an external state of the world that affects the optimal action. We identify a broad class of situations in which ``learning'' leads the individual systematically away from a correct understanding of the world and to lower and lower performance. Even if the agent has correct initial beliefs about the state, due to her unrealistic expectations she is negatively surprised about her outcome. As a result, she becomes more pessimistic about the state of the world, and adjusts her behavior---suboptimally. This adjustment lowers outcomes, perpetuating the misdirected learning further. The greater is the loss from choosing a suboptimal action, the further the agent's action ends up from optimal. If the agent has an outside option, then she is initially too willing to enter a task in which performance depends on her ability and may persist in it for too long (as described in the existing literature), but she eventually tends to quit even superior tasks and to switch too much between tasks. We argue that our model provides one mechanism for self-defeating behavior described in the psychology literature, and that the decision situations in question are common in economic settings, including bureaucratic decisionmaking, delegation, public policy, and labor-leisure choices.