Modeling Trust and Liquidity Under Payment System Stress: A Multi-Agent Approach
AI simulation reveals customer withdrawals can peak AFTER systems are fixed, driven by persistent trust scars.
Researcher Masoud Amouzgar's new paper presents a multi-agent AI model that simulates payment system stress. The model uses a Watts-Strogatz small-world network where AI agents (customers and merchants) interact with bounded memory variables tracking negative experiences. It proves withdrawals can peak during recovery phases, not outages, and shows payment substitution via instant transfers has non-monotonic effects. This demonstrates why technical fixes alone don't resolve systemic risk perception.
Why It Matters
Financial institutions need AI-powered behavioral models to design better crisis responses that address trust, not just technical recovery.