A Mixed-Frequency Factor Model for Nowcasting French GDP

This article presents a new nowcasting model for quarterly real GDP growth in France, developed at the Banque de France. The model is designed to forecast the first release of GDP growth at the end of each month within the quarter in question. The model belongs to the class of targeted factor models and it is estimated using the mixed-frequency three-pass regression filter. We estimate the model on a large set of monthly indicators. The Banque de France survey variables on manufacturing and services are particularly useful for estimating the factors. We extend the formulae for the contributions of the predictors in the mixed-frequency case, and show that, beyond a positive constant level of growth (the intercept), all groups of normalised supply-side and demand-side variables have contributed negatively to GDP growth since the onset of COVID-19 pandemic. A pseudo-real-time evaluation of the method shows the good performance of the model compared to several simple benchmarks and the existing MIBA tool used at the Banque de France, especially during the critical first two months of each quarter. The forecasting combination of the MIBA tool and the new model also performs well at the shortest horizon. In the robustness analysis, we show that this model outperforms a large set of alternative specifications.