Abstract for: Eliciting the Dynamics of Regional Banking from Banking Supervisors

Banking supervisors develop rich mental models through daily interaction with financial institutions, yet translating this expertise into analyzable frameworks remains challenging. While causal loop diagrams (CLDs) effectively capture expert knowledge, they traditionally require extensive data and offline processing for dynamic simulation. This research bridges this methodological gap by enabling direct simulation of expert mental models without such prerequisites. Through structured debriefing of Federal Reserve Bank supervisors, we construct a multimodal directed signed graph representing key financial variables and their relationships. Our novel approach employs graph-algebraic simulation techniques to analyze network structure via centrality metrics and cycle detection, combined with Gaussian Mixture Models for uncertainty quantification. We validate the methodology through progressive assessment of financial system archetypes, demonstrating simulation fidelity to known system dynamic archetypes. Network analysis reveals a hierarchical financial system structure with distinct tiers of influence. Core variables (earning assets, asset prices, capital) exhibit high cycle participation and strong interconnectivity, functioning as primary transmission channels. Statistical analysis of system dynamics indicates supervisors' mental models systematically overestimate long-term capital and earnings decay, suggesting systems prone to severe crises. Optimization identifies critical parameter values where system stability emerges from balancing dampening and responsiveness. Our findings reveal potential supervisory bias that systematically overestimates crisis likelihood and severity, potentially overlooking stabilizing factors like robust local economies and adaptive management practices. This methodology offers a framework for analyzing complex financial networks, quantifying uncertainty in expert assessments, and identifying regulatory biases. The results contribute to understanding how expert mental models shape stability assessment and suggest more nuanced approaches to financial supervision that balance prudential oversight with market dynamism. AI was used to brainstorm, draft, and edit text within word constraints. Intellectual contributions are wholly attributable to the authors.