IMF Team Develops Predictive Economic Models for Sub-Saharan Africa
Project Context
As part of a Covid-19 response effort, IMF Africa Department and Harvard College Data Analytics Group partnered to develop country-level nowcast models of economic activity for the Sub-Saharan African region. The project goal was to build a set of high frequency economic indicator tools to contribute to the IMF’s nowcast and short-term forecast framework. The improved nowcast would help closely monitor economic activity in the Sub-Saharan Africa (SSA) region in a time of stress.
Detailed, real-time information about economic challenges in SSA countries is imperative for evidence-based policymaking, particularly in the context of the rapidly evolving Covid-19 pandemic. Data limitations pose substantial challenges to evaluating economic activity in real-time for some SSA countries. There is no central repository of high frequency data available for SSA, with data availability constrained by publishing delays.
Project team members included:
Reda Cherif — IMF, Africa Department, Senior Economist
Jiawen Tang (Project Lead) — Harvard Kennedy School (MPA-International Development), Stanford (MBA)
Brandon Buell — Harvard College 2020, Statistics and Economics
Carissa Chen — Harvard College 2021, Economics and History
Hyeon-Jae Seo — Harvard College 2020, Applied Mathematics with a focus in Economics & Computer Science
Nils Wendt — Harvard SEAS Visual Computing Group, Exchange Scholar and Technical University Munich (MS in Robotics, Cognition and Intelligence)
Jerry Huang (Harvard College 2022) served as the Engagement Coordinator for this team.
Approach & Contributions
Our Harvard project team’s work had four main contributions to the IMF Africa Department’s nowcasting strategy:
1) Expanded dataset of high-frequency variables for the top 10 economies in SSA, including alternative sources such as shipping, mobile payments, and Google Search Trends. The latter is a novel dataset created by the team based on relevant terms in local languages and which turned out to be a key predictor of GDP.
2) Implementation of two types of modelling approaches (Machine Learning and parametric Factor Model) that incorporate mixed-frequency data variables to generate Quarterly GDP nowcasts and forecasts, with flexibility to adapt to the data availability context. The factor model implementation is generalizable for other SSA countries, particularly for countries with limited data over shorter time horizons.
3) Developed granular, city-level geospatial Covid-19 vulnerability to contagion risk maps, leveraging geospatial data on population density, population over 60 years of age, and access to handwashing water and soap, generously shared by Fraym.
4) Constructed set of economic indicator trackers for countries with limited publicly available data using geospatial, mobility trends, and nitrogen oxide concentration data.
In addition to presenting results to the IMF Africa Department, the project team delivered a final report and handover documentation including code scripts, replication data files, and data dictionary. By providing the data blueprint and a suite of prediction tools for nowcasting, the project team’s work contributes to expanding country-level nowcasting to other countries in Sub-Saharan Africa.
This article reports the work of Harvard College Data Analytics Group’s COVID-19 Crisis Response Team. Edited by Kelsey Wu.