Nowcasting the world trade index

Forecasting the evolution of world trade can be extremely useful as it is a good proxy for global growth. To this end, we have developed a machine-learning nowcasting method for forecasting global trade. Our work is based on a working paper by Menzie Chinn, Baptiste Meunier and Sebastian Stumpner, 2023, “Nowcasting World Trade with Machine Learning: a Three-Step Approach”, published by Banque de France. This article describes a three-step approach that consists in the pre-selection of variables, extraction of factors, and finally regression. Compared with the work of the Banque de France, we have revised the overall structure, opting for an ensemble learning model. In addition, we have developed our own method for optimising the model parameters, the choice of these parameters being decisive for forecast accuracy.