From Many Models, One: Macroeconomic Forecasting with Reservoir Ensembles

Giovanni Ballarin¹, Lyudmila Grigoryeva^1,2 and Yui Ching Li³

¹University of St. Gallen.
²Honorary Associate Professor, University of Warwick.
³Bank for International Settlements - The views expressed in this work are solely those of the authors and do not necessarily reflect those of the Bank for International Settlements.

Project structure:

• data: data loaded to construct forecasts.

• figures_multistep: all exported figures used in the paper.

• results: pickled (.pkl) files for all numerical experiments in the paper.

• python: model code; multi-frequency ESN model libraries; script with estimation, forecasting and simulations for the Medium-MD dataset.

  • Please check requirements.txt before running.

  • Run ensemble_multistep_medium_md.py to run forecast combinations and build tables/plots.

  • Flag DO_FORECASTS = False is required, as the full dataset cannot currently be shared due to licensing restrictions. Forecasts will be loaded from disk.

  • Set DO_PLOTS = True to also generate main plots.

  • Set OPTIONAL_PLOTS = True to generate optional plots.

  • Modify PATH_PREFIX as needed on your machine.

For details regarding data copyrights and sources, please refer to data/README.md.

Acknowledgements

YCL and LG acknowledge the financial support of the FoKo of the University of St. Gallen (Project Nr.1022324, Machine Learning Techniques for Macroeconomic Forecasting and Pricing of Financial Derivatives). GB thanks the Great Minds Postdoctoral Fellowship program of the University of St. Gallen, which made this research possible.