Learning benevolent leadership in a heterogenous agents economy

This paper studies the potential commitment value of cheap talk announcements in an agent-based dynamic extension of the Kydland-Prescott model. In every period, the policy maker makes a non-binding inflation announcement before setting the actual inflation rate. It updates its decisions using individual evolutionary learning. The private agents can choose between two different forecasting strategies: They can either set their forecast equal to the announcement or use an adaptive learning scheme to (potentially) forecast the true inflation. They switch between these two strategies as a function of information about the associated payoffs they obtain through word-of-mouth, choosing always the currently most favorable one. While all agents using the first strategy make the same forecast, those using the second strategy may generate different individual forecasts. In spite of the complexity of the environment, the boundedly rational policy maker learns to sustain a situation with a positive but fluctuating fraction of believers. This outcome is Pareto superior to the outcome predicted by standard theory. Interestingly enough, the actions taken by the policy maker undergo marked qualitative changes as a function of the prevailing heterogeneity among and learning characteristics of the private agents.