Macroeconomic Analysis without the Rational Expectations Hypothesis∗

This paper reviews a variety of alternative approaches to the specification of the expectations of economic decisionmakers in dynamic models, and reconsiders familiar results in the theory of monetary and fiscal policy when one allows for departures from the hypothesis of rational expectations. The various approaches are all illustrated in the context of a common model, a log-linearized New Keynesian model in which both households and firms solve infinite-horizon decision problems; under the hypothesis of rational expectations, the model reduces to the standard “3-equation model” used in studies such as Clarida et al. (1999). The alternative approaches considered include rationalizable equilibrium dynamics (Guesnerie, 2008); restricted perceptions equilibria (Branch, 2004); decreasing-gain and constant-gain variants of least-squares learning dynamics (Evans and Honkapohja, 2001); rational belief equilibria (Kurz, 2012); and near-rational expectations equilibria (Woodford, 2010). Issues treated include Ricardian equivalence; the determinacy of equilibrium under alternative interest-rate rules; non-fundamental sources of aggregate instability; the tradeoff between inflation stabilization and output-gap stabilization; and the possibility of a “deflation trap.” ∗Prepared for Annual Review of Economics, volume 5. I would like to thank Klaus Adam, Ben Hebert, Mordecai Kurz, David Laibson, and Bruce Preston for helpful discussions, Savitar Sundaresan for research assistance, and the Institute for New Economic Thinking and the Taussig Visiting Professorship, Harvard University, for supporting this research. A crucial methodological question in macroeconomic analysis is the way in which the expectations of decisionmakers about future conditions should be modeled. To the extent that behavior is modeled as goal-directed, it will depend (except in the most trivial cases) on expectations; and analyses of the effects of alternative governmental policies need to consider how expectations are endogenously influenced by one policy or another. Finding tractable ways to address this issue has been a key challenge for the extension of optimization-based economic analysis to the kinds of dynamic settings required for most questions of interest in macroeconomics. The dominant approach for the past several decades, of course, has made use of the hypothesis of model-consistent or “rational expectations” (RE): the assumption that people have probability beliefs that coincide with the probabilities predicted by one’s model. The RE benchmark is a natural one to consider, and its use has allowed a tremendous increase in the sophistication of the analysis of dynamics in the theoretical literature in macroeconomics. Nonetheless, the assumption is a strong one, and one may wonder if it should be relaxed, especially when considering relatively short-run responses to disturbances, or the consequences of newly adopted policies that have not been followed in the past — both of which are precisely the types of situations which macroeconomic analysis frequently seeks to address. While the assumption that an economy’s dynamics must necessarily correspond to an RE equilibrium may seem unjustifiably strong — and under some circumstances, is a heroic assumption indeed — it does not follow that we should then be equally willing to entertain all possible assumptions about the expectations of economic agents. It makes sense to assume that expectations should not be completely arbitrary, and have no relation to the kind of world in which the agents live; indeed, it is appealing to assume that people’s beliefs should be rational, in the ordinary-language sense, though there is a large step from this to the RE hypothesis. We should like, therefore, to replace the RE hypothesis by some weaker restriction, that nonetheless implies a substantial degree of conformity between people’s beliefs and reality — that implies, at the least, that people do not make obvious mistakes. The literature has explored two broad approaches to the formulation of a criterion for reasonableness of beliefs that is weaker than the RE hypothesis. One is to assume that people should correctly understand the economic model, and be able to form correct inferences from it about possible future outcomes. The other is to assume that that the probabilities that people assign to possible future outcomes should not be too This is stressed, for example, by Kurz (2012, p. 1).