Risk of credit and lending in an artificial adaptive banking system

Risk of credit and lending in an artificial adaptive banking system

Abstract

We present a simulation tool we have developed to build artificial banking systems and to study the interaction among banks and firms under various conditions. In this application, we consider a banking system composed of artificial adaptive banks, which have to make decisions about the opportunity to lend money to prospective borrowers. Such borrowers are risky firms whose value evolve stochastically over time according to an heterogeneous (across firms), time-varying probability. Banks decide whether to give out loans or not on the basis of an information set which is partly firm-specific and partly of a macroeconomic nature. The evaluation of such information set takes place by means of neural networks which learn over time to distinguish among good and bad borrowers. We consider the model as a useful simulation instrument to analyze the dynamic evolution of an economy where some of the variables are not common knowledge. The results show that these learning techniques are effective and that banks learn to discriminate among borrowers. Moreover we can see that this simulation tool allows to study not only the effects of general macroeconomic conditions on such learning, but also the interactions among artificial agents and their behavior under different initial assumptions.