Predictive power of Bayesian CAR models on scale free networks: an application for credit risk

The monitoring of loans’ life-cycle has received the increasing attention of the scientific community after the 2008 global financial crisis. A number of aspects of this broad topic have been addressed by means of several regulatory, statistical and economical tools. However, many issues still require further investigation. In this work, we are interested in the monitoring phase of granted loans to anticipate possible defaults and to investigate whether there is evidence of a liquidity contagion effect within a trade network of firms. To this end, we apply a Bayesian spatial model to a proprietary dataset, and assess its out-of-time predictive performance.