Identifying Instability in U.S. Capital Markets: An Explainable Machine-Learning Based Approach for Assessing Risks in Complex Financial Systems

We introduce a novel methodology to capture and represent the diverse relationships between financial market institutions to prevent and mitigate negative economic shocks. An ensemble of machine learning models identifies influential features representing these relationships, with SHapley Additive exPlanations (SHAP) and a modified Borda count method selecting the most significant predictors. These features train an explainable linear model, providing a transparent and interpretable representation of inter-network relationships. The methodology is validated through case studies utilizing data from the FY20 U.S. recession and the failure of Silicon Valley Bank in March of 2023, demonstrating the model's accuracy in predicting financial stress and identifying institutional outliers less affected by economic shocks. This approach is significant for researchers in regulated fields, where model explainability is crucial, as it provides a framework to analyze high-dimensional, infrequently reported data and reveal valuable insights. As financial markets evolve and generate complex datasets, this methodology offers a robust tool for understanding and predicting financial system behavior, enhancing risk management, and financial stability.

Author

• Purnell Jr., Daren Lee

Language

• English

Keyword

• feature selection
• complex systems
• machine learning
• explainability
• financial stability
• economics

Date created

• 2025

Type of Work

• Dissertation

License

• All rights reserved

Rights statement

• In Copyright

GW Unit

• Systems Engineering

Degree

• Ph.D.

Advisor

• Kamp, John
• Etemadi, Amir

Committee Member(s)

• Mazzuchi, Thomas A
• Blackford, Joseph
• Sarkani, Shahram

Persistent URL

•  https://scholarspace.library.gwu.edu/etd/9880vr732