Forecasting Dutch inflation using machine learning methods

Despite the benefits of forecasting inflation accurately, improving simple models has proved challenging. This research explores advances in machine learning (ML) methods to forecast Dutch inflation. We investigate whether ML models with a large number of covariates are systematically more accurate than simple benchmark models, such as AR and RW. Amongst other we will investigate (polynomial) shrinkage methods, (targeted/boosted) factor models, ensemble methods, random forests and neural networks.

Joint with Robert-Paul Berben (De Nederlandsche Bank).