Expectation Coordination and Agent-level Learning

University Of Oregon Eugene, Eugene OR

Investigators

Abstract

The PIs propose work that will use agent- level modeling methods to address research questions in macroeconomics. The PIs want to examine the behavior of a computational model of an entire economy. In this model, firms and consumers are boundedly rational and use adaptive learning rules. The PIs plan to embed this framework into a dynamic stochastic general equilibrium model and to allow for heterogeneous agents; so-called DSGE models are widely used in macroeconomics, and the PIs will use both Real Business Cycle and New Keynesian models. These DSGE models often have multiple possible outcomes that can be supported by rational behavior on the part of individuals via self-fulfilling prophecies. Economists call these "sunspot" equilibria; if everyone in the economy thinks that a signal, perhaps in the form of some external event (even an astronomical event that has no effect on the planet), predicts an upcoming recession, then after seeing the signal, people may start spending less money in anticipation of possible layoffs, and employers may begin to lay off workers in anticipation of a drop in sales. The result is a recession, even though the signal itself has no direct effect on any individual in the economy. The PIs will examine when sunspot equilibria are stable over time when people learn from experience, thus formalizing the process by which fear of recession leads to recession Modeling people as computational agents with a simple adaptive learning rule is one way to examine the stability of sunspot equilibria. They will also examine the effect of government policies such as monetary and fiscal policy in models with this kind of adaptive learning. The goal here is to examine agent-level learning in general equilibrium environments. Recent work by the research team has demonstrated that boundedly rational agents can learn to solve dynamic optimization problems over time by in effect replacing them with repeated two-period problems, while updating in each period their forecasts of shadow prices of key state variables. The project will embed this and related implementations of learning into DSGE models. The agent-level approach facilitates solving computational models with heterogeneous firms and households; this allows both for the study of macroeconomic policy and for the examination of wealth inequality. The PIs will re-examine the learning stability of sunspot equilibria in Real Business Cycle and New Keynesian models. They will also examine models in which the zero-lower-bound on interest rates leads to multiple steady states. And they will examine the impact of fiscal policy and government spending multipliers in New Keynesian models with learning. The proposed research will have specific implications for both monetary and fiscal policy, in addition to demonstrating how the PI?s bounded optimality approach can be used in a wide range of applied macroeconomic models.