A comparison of MIDAS and bridge equations

This paper compares two single-equation approaches from the recent nowcasting literature: mixed-data sampling (MIDAS) regressions and bridge equations. Both approaches are suitable for nowcasting low-frequency variables such as the quarterly GDP using higher-frequency business cycle indicators. Three differences between the approaches are identified: (1) MIDAS is a direct multi-step nowcasting tool, whereas bridge equations provide iterated forecasts; (2) the weighting of high-frequency predictor observations in MIDAS is based on functional lag polynomials, whereas the bridge equation weights are fixed partly by time aggregation; (3) for parameter estimation, the MIDAS equations consider current-quarter leads of high-frequency indicators, whereas bridge equations typically do not. To assist in discussing the differences between the approaches in isolation, intermediate specifications between MIDAS and bridge equations are provided. The alternative models are compared in an empirical application to nowcasting GDP growth in the Euro area, given a large set of business cycle indicators.