Hyper-parameterised dynamic regressions for nowcasting Spanish GDP growth in real time

Abstract

This paper analyses the nowcasting performance of hyper-parameterised dynamic regression models with a large number of variables in log levels, and compares it with state-of-the-art methods for nowcasting. We deal with the 'curse of dimensionality' by exploiting prior information originating in the Bayesian VAR literature. The real-time forecast simulation conducted over the most severe phase of the Great Recession shows that our method yields reliable GDP predictions almost one and a half months before the official figures are published. The usefulness of our approach is confirmed in a genuine out-of-sample evaluation over the European sovereign debt crisis and subsequent recovery.