Structural models provide little information about the near term and,
as a result. policy-making institutions spend too much time producing
short-term forecasts. Here at the Fund, Machine Learning and Big Data
are changing this.
By combining time-series forecasting models, we are now able to
deliver forecasts for 38 countries on a weekly basis. Using several
activity indicator series, including some with high frequency, we are
applying various machine learning algorithms to produce promising
results. The system is even smart enough to dynamically shift its
machine learning algorithms to improve its performance. We explore
this technology and how it will impact the work of the Fund and add
value to our member countries.