Modelling Social Learning in an Agent-Based New Keynesian Macroeconomic Model

We propose an agent-based macroeconomic model (ABM) inspired by the New Keynesian general equilibrium model (NKM, Woodford 2003). We analyse the aggregate economic dynamics resulting from social learning of agents (households and firms). Households’ labour supply and consumption demand, as well as firms\' labour demand and wage offers evolve through imitation and random experimenting by the agents. We study, in this setting, the aggregate properties of the economy and the ability of those learning agents to coordinate on the intra-temporal equilibrium of the original model. Our approach is quite different from the existing learning literature in the NKM (à la Evans & Honkapohja, that mainly focuses on learning for testing local stability of equilibria), since learning is directly embedded in the behaviour of the individual agents. This original approach opens new perspectives about the NKM, and allows us to ask new questions about the coordination problems that can result from social learning. First, our computational analysis (Monte Carlo simulations) shows that social learning does not allow the agents to correctly learn about the interdependence between markets, because of the emergence of coordination problems that result in insufficient labour supply and depressive dynamics. Second, we shed light on the general properties of social learning that are behind these results in a general (dis)equilibrium setting, and prove that their neutralisation, at least on the one side of the markets, can significantly improve the performance of the economy. Our results point to the importance of carefully modelling learning mechanisms within macroeconomic ABMs.