Learning with Model Uncertainty and Misspecification

点击翻译按钮获取中文摘要

Conventional economic theory and equilibrium concepts do not permit agents to experience doubts and ambiguity that would lead them to test the way their models of behavior are specified. Nor do they encompass alternative (sample) paths to escape from the status quo and switch among different models. These aepcets are crucial for explaining the dynamics of many types economic phenomena, particularly hwo we get to equilibrium outcomes. We propose an alternative approach to investigate the uncertainty and doubt in a dynamic context, without changing the preference of the decision maker. We aim to make theoretical and practical contributions, by developing models in which agents and economists are on an equal footing, in the sense that agents within the model confront the same doubts and ambiguity about their environment that confront an outside observer or econometrician anlyzing the data.This proposal contains three distinct projects designed to serve as laboratories for exploring the implications of endowing agents with model uncertainty and misspecification in a dynamic environment. First, the forecast combination is a way for a policy maker to aggregate dispersed information, and to hedge against model uncertainty, under the assumption that the data generating process is exogenous. We demonstrate the potential shortcomings of averaging forecasts, if the data generating process is endogenous, and develop a sensible way of combining different forecasts. Second, we investigate how how doubts can sow the seeds of financial crises. We first endogenize the heterogeneous beliefs by showing that doubts about the process generating fundamentals can produce heterogeneous beliefs and asset market bubbles. We then impose discipline on the degree of belief heterogeneity, to understand the source of recurrent bubbles. Third, a formal investigation of events generated by the presence of model uncertainty calls for a new solution concept. We plan to extend and then refine the notion of self-confirming equilibrium by focusing on the set of stochastic processes that satisfy a given bound on detection error probability rates.