Title: A New Approach To Time Varying Parameters in Vector Autoregressive Models

Abstract: We propose a flexible means of estimating vector autoregressions with time-varying parameters (TVP-VARs) by introducing a latent threshold process that is driven by the absolute size of parameter changes. This enables us to dynamically detect whether a given regression coefficient is constant or time-varying. When applied to a medium-scale macroeconomic US dataset our model yields precise density and turning point predictions, especially during economic downturns, and provides new insights on the changing effects of increases in short-term interest rates over time.