Abstract

The Fed’s policy rule switches during the different phases of the business cycle. This finding is established using a dynamic mixture model to estimate regime-dependent Taylor-type rules on US quarterly data from 1960 to 2021. Instead of exogenously partitioning the data based on tenures of the Fed chairs, a Bayesian framework is introduced in order to endogenously select timing and number of regimes in a data-driven way. This agnostic approach favors a partitioning of the data based on two regimes related to business cycle phases. Estimated policy rule coefficients differ in two important ways over the two regimes: the degree of gradualism is substantially higher during normal times than in recessionary periods while the output gap coefficient is higher in the recessionary regime than in the normal one. The estimate of the inflation coefficient largely satisfies the Taylor principle in both regimes. The results are substantially reinforced when using real-time data.