EconoGNN: A graph neural network framework for temporal economic resilience insights

Area of Science:

• Economics
• Computer Science
• Network Science

Background:

• Global economic shocks highlight the need for systemic analysis of regional economic resilience (RER).
• Existing RER studies are limited by localized data, inconsistent definitions, and static models, hindering generalizability.
• Complexity theory, economic modeling, and machine learning offer potential for advanced RER analysis.

Purpose of the Study:

• To introduce EconoGNN, a Graph Neural Network (GNN) framework for predicting and explaining RER.
• To enable reproducible, interpretable, global-scale RER analysis across 183 countries over 25 years.
• To integrate diverse datasets and theoretical frameworks for a comprehensive RER model.

Main Methods:

• Developed EconoGNN, a GNN framework combining complexity theory, economic modeling, and machine learning.
• Integrated 81 million trade records (UN COMTRADE) and 500,000 macroeconomic observations (Penn World Table).
• Utilized a World Bank resilience metric and GNNExplainer for validation and interpretability.

Main Results:

• EconoGNN achieved high performance: F1-scores of 0.750, AUC-ROC of 0.792, and PR-AUC of 0.757.
• Demonstrated robust performance across various recovery thresholds (F1: 0.730-0.771).
• GNNExplainer confirmed explanation reliability (Fidelity+: 0.827, Characterization: 0.913).

Conclusions:

• EconoGNN provides a powerful, interpretable tool for global RER analysis.
• The framework effectively models national and cross-border economic interactions, reducing bias.
• Enables country-customized understanding of resilience drivers, improving economic policy insights.