Classical time varying factor-augmented vector auto-regressive models-estimation, forecasting and structural analysis

Abstract

Summary: We propose a classical approach to estimate factor-augmented vector auto-regressive (FAVAR) models with time variation in the parameters. When the time varying FAVAR model is estimated by using a large quarterly data set of US variables from 1972 to 2012, the results indicate some changes in the factor dynamics, and more marked variation in the factors' shock volatility and their loading parameters. Forecasts from the time varying FAVAR model are more accurate, in particular over the global financial crisis period, than forecasts from other benchmark models. Finally, we use the time varying FAVAR model to assess how monetary transmission to the economy has changed.