Forecasting US real GDP using oil prices: A time-varying parameter MIDAS model

In this paper, we introduce the functional coefficient to existing mixed-frequency data sampling (MIDAS) regression to make the parameter change over time. The proposed time-varying parameter MIDAS (TVPMIDAS) is employed to forecast the U.S. real GDP growth using crude oil prices. We find the out-of-sample predictability of GDP growth across different forecasting horizons. The percent reduction of mean squared predictive error achieves 14% when the nonlinear oil price measure is employed. The TVP-MIDAS can outperform a series of competing models including the OLS regression with quarterly oil price, the constant coefficient and Markov regime switching MIDAS regressions. (C) 2018 Elsevier B.V. All rights reserved.