This paper presents a single, integrated model to explain the persistence and volatility characteristics of the U.S. inflation time series. Policymakers learning about a Markov-switching natural rate of unemployment in a neoclassical Phillips curve model with time-varying preferences produces inflation persistence, volatility clustering, and mean/variance correlation. The interaction between the policymaker’s preferences and the Phillips curve generates the first and last results. Policymaker learning produces clusters of volatility as the monetary authority resets the learning algorithm whenever a shock to the Phillips curve occurs. Simulations using parameters estimated via Gibbs sampling confirms the theory.