IFRS 9 and Expected Credit Loss Modeling

This diploma thesis addresses the issue of expected credit losses under the IFRS 9 accounting standard. The aim of the thesis is to design and practically implement an approach for estimating both 12month and lifetime probabilities of default for a mortgage loan portfolio. The thesis utilizes a publicly available loan-level dataset provided by the American company Freddie Mac. Two approaches to probability of default modelling are proposed and applied in the thesis. The first approach is based on a Markov transition matrix model. This approach enables modelling the migration of the loan portfolio between individual risk states (rating pools) and subsequently capturing the dynamics of year-to-year transitions between states, allowing the derivation of probabilities of default in accordance with IFRS 9 requirements. The second approach employs the gradient boosting algorithm to model the monthly hazard rate of default. Based on this hazard rate, both 12month and lifetime probabilities of default are subsequently derived. The outcome of the thesis is a practical implementation and comparison of two distinct methodological approaches to estimating probability of default (PD) parameters in the context of IFRS 9. Based on the outputs of the developed models, both approaches are mutually compared. The contribution of the thesis lies in linking the accounting requirements of IFRS 9 with practical credit risk modelling methods applicable to credit risk management and the calculation of loan loss provisions.