Nowcasting World Trade With Machine Learning: A Three‐Step Approach

ABSTRACT We nowcast world trade using machine learning, distinguishing between tree ‐based methods (random forest and gradient boosting) and their linear ‐ regression ‐based counterparts (macroeconomic random forest and gradient boosting—linear). While much less used in the literature, the latter are found to outperform not only the tree ‐based techniques but also more “traditional” linear and nonlinear techniques (OLS, Markov‐switching, and quantile regression). They do so significantly and consistently across different horizons and real‐time datasets. When doing so, we find that using preselection and factor extraction significantly improves the accuracy of machine learning predictions. This approach also outperforms two workhorse nowcasting methods: PCA‐OLS and dynamic factor models. Replication code can be found at https://github.com/baptiste‐meunier/NowcastingML_3step .