Working Paper: CEPR ID: DP15867
Authors: Massimiliano Marcellino; Dalibor Stevanovic; Philippe Goulet Coulombe
Abstract: Based on evidence gathered from a newly built large macroeconomic dataset for the UK, labeled UK-MD and comparable to similar datasets for theUS and Canada, it seems the most promising avenue for forecasting duringthe pandemic is to allow for general forms of nonlinearity by using machinelearning (ML) methods. But not all nonlinear ML methods are alike. Forinstance, some do not allow to extrapolate (like regular trees and forests)and some do (when complemented with linear dynamic components). Thisand other crucial aspects of ML-based forecasting in unprecedented timesare studied in an extensive pseudo-out-of-sample exercise.
Keywords: Machine Learning; Big Data; Forecasting; COVID-19
JEL Codes: C53; C55; E37
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| machine learning methods (C45) | improved forecasting of macroeconomic indicators (E37) |
| macroeconomic random forests (E19) | outperform traditional autoregressive models (C22) |
| macroeconomic random forests (E19) | successfully predicting unprecedented values for variables like hours worked (C29) |
| macroeconomic random forests (E19) | providing timely forecasts that align better with actual outcomes (C53) |
| traditional linear models (C20) | struggle to adapt to rapid changes during the pandemic (O00) |