Modeling Bank Failure with System Dynamics and Genetic-Algorithm Optimization: Evidence from Bank Melli Iran

Abstract

Bank failure develops through interacting balance-sheet, operational, and macroeconomic mechanisms, yet conventional early-warning models often represent these mechanisms as static predictors. This study models the drivers of bank failure from the perspective of capital distress and accumulated losses and evaluates policy combinations using an integrated system-dynamics and genetic-algorithm approach. An applied mixed-method design was used. Candidate factors were identified through a literature review and semi-structured interviews with 11 senior banking managers, researchers, and academics selected purposively and by availability. Three rounds of fuzzy Delphi screening were followed by expert validation of the causal structure. Institutional data for Bank Melli Iran covered Iranian years 1390–1403 (approximately 2011–2024). A Vensim model represented resource mobilization, funding composition, credit efficiency, income, costs, capital, and accumulated losses. A genetic algorithm then assessed two multi-lever policy scenarios. The multi-lever scenario that jointly optimized resource mobilization, funding composition, and credit efficiency generated a feasible solution for all 13 simulated transitions within the defined policy domain; its minimum accumulated-loss-to-capital ratio was approximately −0.49996. When final lever effectiveness was capped at 1, no feasible solution existed for 1395–1398, even at the maximum permissible intervention. Inflation had the largest mean absolute sensitivity (0.178), and credit efficiency was the most influential policy lever. Historical reproduction yielded a correlation of approximately 0.92 and a mean error below 9%. The integrated model supports time-dependent diagnosis and comparison of intervention pathways. Its outputs should be interpreted as conditional policy evidence rather than a guarantee that bank failure will be prevented.