Towards a credit network based early warning indicator for crises

This paper presents an agent based model which underlines the importance of credit network and leverage dynamics in determining the resilience of the system, defining an early warning indicator for crises. The model reproduces macroeconomic dynamics emerging from the interactions of heterogeneous banks and firms in an endogenous credit network. Banks and firms are linked through multiple credit relations, which derive from individual target leverage choices: agents choose the more convenient leverage level, according to a basic reinforcement learning algorithm. Simulations are calibrated on balance sheet data of banks and firms quoted in the Japanese stock-exchange markets from 1980 to 2012.