Machine Learning Approach for PD Term Structure Modeling Under IFRS 9 Regulatory Framework

As a prerequisite for lending and the preservation of healthy portfolios, banks must estimate the Probability of Default (PD) correctly. There are multiple methods available to assess credit risk, however reliance on single predictive models, which is predominantly used, ignores the multifaceted aspects of credit risk. At the same time, the compliance of IFRS 9 has become a focal concern. It goes without saying that one of these regulations is the estimation of PD for the whole life of a credit contract, which presupposes the calculation of incremental PDs during the contract's life-the PD term structure. Accurate PD term structure forecasts are important for business planning within the boundaries of the firm's risk appetite and the ever-changing regulations. This paper focuses on the application of machine learning algorithms XGBoost and Random Boosting Forest (RBF) to enhance the accuracy of PD term structure forecasting. A profound assessment of results is carried out based on specified performance metrics, and the models are compared to the traditional paradigm.