The increasing intensity and frequency of natural disasters, amplified by climate change, are profoundly reshaping the risk profile of financial institutions. A growing body of empirical evidence shows that events such as hurricanes, floods, droughts, and earthquakes deteriorate loan portfolios,
increase default probabilities, and weaken banks’ ability to maintain adequate capital buffers (IPCC, 2023; NGFS, 2024; Bitar et al., 2022). Studies from diverse contexts — including rural China (Li et al., 2021), India (Singh et al., 2020), European microfinance (Di Bella, 2025), U.S. banks exposed to
hurricanes (Brei et al., 2019; Cortés & Strahan, 2017), and seismic events in Italy (Banca d’Italia, 2023) — consistently highlight the emergence of nonlinear, geographically asymmetric mechanisms that traditional econometric models struggle to capture. Such models typically rely on equilibrium
assumptions and representative agents, making them inadequate for modelling behavioural adaptation, credit contagion, and endogenous feedback loops (Colander et al., 2009; Farmer & Foley, 2009; Lux, 2009). To address these limitations, this study introduces an agent-based model (ABM) developed to
simulate bank solvency under physical climate shocks, integrating NGFS scenarios and an empirical calibration grounded in both micro- and macro-financial evidence. The model architecture consists of four interconnected modules: (i) households and firms, characterized by climate vulnerability,
income dynamics, and adaptive default behaviour; (ii) banks, featuring regulatory capital, provisioning rules, and endogenous credit rationing; (iii) the climate environment, generating physical shocks based on distributions aligned with NGFS (2024); and (iv) the regulator, equipped
with dynamic macroprudential tools and mitigation instruments (moratoria, countercyclical buffers). Calibration relies on elasticity estimates and empirical relationships drawn from key contributions: post-hurricane capital deterioration in Florida (Brei et al., 2019), rural credit sensitivity to climate
shocks in China (Li et al., 2021), liquidity contraction in India (Singh et al., 2020), climate-induced microfinance vulnerability (Di Bella, 2025), seismic impacts on credit and deposits in Italy (Banca d’Italia, 2023), and the most comprehensive cross-country analysis available, covering over 9,000
banks (Bitar et al., 2022).
Simulation results highlight three core dynamics: (1) a nonlinear relationship between shock severity and default probability; (2) substantial deterioration of capital ratios, with amplified effects under multiple shocks; and (3) emergent forms of endogenous credit rationing. Furthermore, the model
shows that adaptive prudential instruments — such as climate-sensitive capital requirements, territorially differentiated risk weights, and temporary moratoria — can mitigate systemic fragility. Overall, the model provides a computational platform for intelligent climate stress testing, cognitive
simulation applied to financial stability, and regulatory assessment based on complex adaptive system modelling. This aligns closely with the conference’s thematic areas on AI-driven modelling, agentbased cognition, and complex adaptive systems.
