Rule-based monetary policy under central bank learning

This paper evaluates the performance of three popular monetary policy rules where the central bank is learning about the parameter values of a simple New Keynesian model. The three policies are: (1) the optimal non-inertial rule; (2) the optimal history-dependent rule; (3) the optimal price level targeting rule. Under rational expectations rules (2) and (3) both implement the fully optimal equilibrium by improving the output/inflation trade-off. When imperfect information about the model parameters is introduced, the central bank makes monetary policy mistakes, which affect welfare to a different degree under the three rules. The optimal history-dependent rule is worst affected and delivers the lowest welfare. Price level targeting performs best under learning and maintains the advantages of conducting policy under commitment. These findings are related to the literature on feedback control and robustness. The paper argues that adopting integral representations of rules designed under full information is desirable, because these rules deliver the beneficial output/inflation trade-off of commitment policy, while being robust to implementation errors.