Big-Data Modeling of Nonlinear Banking Fragility: Quantum Features, Machine Learning Validation, and Explainable Risk Signals
Main article
Abstract
Banking fragility in emerging markets emerges from interactions among credit risk, market volatility, concentration, and macroeconomic shocks that conventional linear panel models struggle to capture. This study develops a big-data analytical pipeline that fuses functional features derived from Quantum Field Theory (QFT)—including a double-well stochastic potential indicator and a Faddeev-Popov constrained quantization correction—with supervised and unsupervised machine learning to detect nonlinear, regime-switching dynamics in the Mexican banking system. Drawing on an annual panel of eleven multiple-banking institutions covering 2014 to 2023, we engineer 14 micro-prudential, macro-financial, and quantum-inspired features. We then validate the quantum indicators against observed insolvency proxies via logistic regression, Random Forest classification with SHAP-based interpretation, principal-component clustering, and bootstrap resampling. The quantum fragility feature is positively and significantly associated with the lower-quartile Z-score state (coefficient = 2.66, p = 0.003) and yields measurable lifts in extreme-event sensitivity, particularly around the 2016 emerging-market shock and the 2020 pandemic. Random Forest importance and SHAP attribution rank non-performing loans, return on assets, the Lerner index, and the capitalization ratio as the dominant risk drivers, with the Faddeev-Popov correction contributing complementary signal in transition periods. The framework offers a reproducible, explainable big-data architecture for prudential supervision, early-warning systems, and research on financial fragility in emerging markets.
