A0396
Title: Nowcasting GDP in Switzerland: What are the gains from machine learning algorithms?
Authors: Rolf Scheufele - Swiss National Bank (Switzerland)
Milen Arro-Cannarsa - University of Bern (Switzerland) [presenting]
Abstract: Several machine learning methods for nowcasting GDP in Switzerland are compared. Based on a large mixed-frequency data set, we investigate the predictive ability of regression-based methods (Ridge, LASSO, Elastic net), tree-based methods, bagging and SVR. As benchmarks, we use univariate models, forward selection algorithms and factor models. For the period between the Financial Crisis and the COVID-19 crisis, which is particularly challenging in terms of nowcasting, we find that all considered ML techniques beat the univariate benchmark and the forward selection algorithms. Ride regression and SVR turn out to be most successful and outperform the factor model based on principle components by more than 10\% in terms of RMSE.
