This paper tests the usefulness of time-varying parameters when forecasting with mixed-frequency data. For this we compare the forecast performance of bridge equations and unrestriced MIDAS models with constant and time-varying parameters. An out-of-sample forecasting exercise with US real-time data shows that the use of time-varying parameters does not improve forecasts significantly over all vintages. However, since the Great Recession, forecast errors are smaller when forecasting with bridge equations due to the ability of time-varying parameters to incorporate gradual structural changes faster.
forecasting Bayesian mixed frequency data time-varying parameters