Kurz gefasst
Fachartikel in der Zeitschrift IMF Working Papers, 2026-11. Verfasst von Hany Abdel-Latif und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Nowcasting GDP in low-income and fragile economies poses a set of problems that bind at once: short and noisy samples, ragged-edge releases, non-stationary regressors, model-selection uncertainty, the need to communicate uncertainty and attribution to non-technical audiences, and frequent within-quarter updates. Existing nowcasting toolkits, designed largely for data-rich economies, do not address this combination. This paper presents IMFNow, the open-source R package implementing the International Monetary Fund's nowcasting framework and how it addresses these constraints. In one reproducible pipeline, it integrates sixteen machine-learning and time-series learners, circular block-bootstrap cross-validation for short dependent samples, automatic stationarity handling with analyst override, expanding-window conformal prediction intervals calibrated to propagate model-selection uncertainty, the Model Confidence Set based on out-of-sample losses, SHAP decompositions, and a fast-update mode that reuses tuned hyperparameters within a quarter. The framework is delivered through two functions, macroNowcast and outputNowcast. We motivate each component by the forecasting problem it solves, document the architecture, and demonstrate the framework on quarterly GDP growth for Uganda and Tunisia using rolling-origin pseudo-real-time evaluation of point accuracy and interval coverage.
Zitieren
APA
Abdel-Latif, H., Centorrino, S., & Dovi, M. (2026). IMFNow. IMF Working Papers, 2026(240), 1. https://doi.org/10.5089/9798229065092.001
Vancouver
Abdel-Latif H, Centorrino S, Dovi M. IMFNow. IMF Working Papers. 2026;2026(240):1. doi:10.5089/9798229065092.001
BibTeX
@article{AbdelLatif2026,
author = {Abdel-Latif, Hany and Centorrino, Samuele and Dovi, Max-Sebastian},
title = {IMFNow},
journal = {IMF Working Papers},
year = {2026},
volume = {2026},
number = {240},
pages = {1},
doi = {10.5089/9798229065092.001}
}
EUVIMED