IFRS 9 and Modelling of Expected Credit Losses

The aim of the thesis is to compare three approaches to estimating the probability of default within the calculation of expected credit losses under the IFRS 9 standard. The theoretical part describes the IFRS 9 standard, the structure of the ECL calculation, and the methodology of the Cox proportional hazards model, Markov transition matrices with a PiT adjustment using the Vasicek model, and the XGBoost model. In the empirical part, these models are estimated on a synthetic portfolio and evaluated in terms of discrimination, calibration, and PD prediction across three macroeconomic scenarios. The models achieve similar calibration, but the Markov approach is weaker in discrimination. In the scenario-based prediction, the models diverge significantly. XGBoost predicts a lower PD, while Markov-PiT exhibits a wider scenario range.