Monetary Policy Rules in a Non-Rational World: A Macroeconomic Experiment

Abstract This paper introduces a learning-to-forecast laboratory experiment based on a New-Keynesian macroeconomy that is particularly close to the model's microfoundations. In this setup, subjects forecast their individual optimal consumption and prices instead of aggregate outcomes. Due to different personal experiences, coordination of forecasting behavior does not occur naturally, and there is considerable randomness in subjects' responses. Thompson Sampling, a learning heuristic from operations research that links randomness to the Bayesian posterior uncertainty, describes subjects' individual forecasting data well, and explains the observed patterns in the experiments. The experimental results show that a particularly aggressive anti-inflation response by the central bank is needed to achieve coordination on rational expectations and macroeconomic stability.