Developing an Early Warning System for Financial Networks: An Explainable Machine Learning Approach

Area of Science:

• Financial Network Analysis
• Econometrics
• Machine Learning

Background:

• Financial networks exhibit complex, nonlinear, and time-varying relationships, making early detection of instability challenging.
• Identifying influential variables for predicting financial network instability requires robust data-driven approaches.

Purpose of the Study:

• To develop a novel methodology for selecting variables that indicate financial network instability.
• To create an explainable linear model for predicting relationship value weights between network participants.

Main Methods:

• Leveraged Shapley values and modified Borda counts combined with statistical and machine learning techniques.
• Developed a data-driven variable selection process for financial network analysis.
• Utilized data from the March 2023 Silicon Valley Bank Failure for validation.

Main Results:

• The novel method successfully identified instability trends using only 14 input variables out of 3160.
• The approach demonstrated the ability to pinpoint key indicators of financial network instability.
• Generated parsimonious linear models for increased transparency.

Conclusions:

• The proposed methodology offers a powerful tool for early warning of financial network instability.
• This approach enhances the transparency and interpretability of complex financial systems.
• The variable selection technique has significant implications for financial stability monitoring.