Development of a machine learning model for bank credit risk assessment

The paper deals with implementation of machine learning methods for realization of IRBapproach. One should note that the application of model method for credit risk assessment reduces the amount of necessary bank reserve compared to standard approach. That causes more profitable distribution of bank actives. In addition, the approach improves the quality of human capital as well as increases the stability of bank industry. To provide stable functioning of IRB-approach one should have total integration of programming languages (Python / SQL), developed IT-infrastructure for storage and processing big data in real time, monthly monitoring key indicators (LGD, Default Rate, CCF et al.) to control deviation of actual values from model ones, expressed by Parity coefficient. The authors observe principle risk-metrics, used in models, such as default probability, level of losses under default, amount of credit, expected and unexpected losses. There should be also master scale of credit ratings for private and corporate crediting in bank. The scale is to be corresponded to national and international rating agencies that determines buckets of default probabilities. To develop the models there should be a special department in bank as well as the department of validation that is responsible for controlling model accuracy. The authors present the principle riskmetrics used in models, such as default probability, default losses level, credit requirement, expected and unexpected losses. The aim of the paper was to develop credit models and to interpret their results. The study can be useful for those banks that have not integrated IRB-approach in operation processes yet.