Bayesian Networks and Financial Stress Testing - Assessing the Probability of Default for a Credit Institution

Abstract

The thesis is focused on providing a view relative to Bayesian Networks as specific data analysis tools that may find application in the conduction of Financial Stress Testing exercises. These practices are mainly implemented by credit institutions to assess their current economic healthiness and possibly predict future trends relative to specific key performance indicators. In particular, with concern to the financial context, such exercises may be useful as for regulatory compliance purposes, other than being of guidance for the implementation of crisis-prevention actions. In order to adequately respond to the needs of credit institutions, Bayesian Networks are instruments deemed to be capable of providing accurate indications on causal connections persisting between and among business-specific factors. The assessment of such relations, via simulation procedures, may allow the identification of criticalities relative to the single credit institution, which could consequently be able to decide where to focus efforts and, in case necessary, evaluate the execution of corrective actions. Therefore, in this sense, Bayesian Networks are considered to be useful and adequate instruments in supporting Financial Stress Testing exercises. Furthermore, to this end, the present document also provides a case study analysis, based on real-world data, on the application of the previously-mentioned practices.