Basel III introduced the Internal Rating Based (IRB) and IRB-Advanced (IRBA) approaches, which allow banks to use their own internal estimates of risk parameters to calculate the necessary regulatory capital requirements for credit risk. While the IRB approach enable banks to create and utilize sophisticated risk models adapted to their unique experiences and data, the IRBA methodology grants banks even greater discretion, allowing them to estimate all risk components independently, provided they meet specific criteria and obtain regulatory approval. Backtesting is a crucial process in financial risk management, employed to assess the performance and reliability of models over time. This practice is essential for maintaining robust risk management systems and ensuring compliance with regulatory requirements. By comparing predicted risk estimates with actual outcomes, backtesting helps in identifying discrepancies, ensuring that models remain accurate and relevant under changing market conditions. The Probability of Default (PD) parameter is a risk input that measures the likelihood that a borrower will default on their debt obligations in a specific date. This report focuses on the development of a PD model and its subsequent validation through Backtesting, ensuring its alignment with regulatory standards. The PD model development followed a structured approach, utilizing logistic regression combined with K-means clustering to form distinct risk classes, each assigned a specific PD. A scoring system was designed to rank obligors by risk, incorporating the Margin of Conservatism (MoC) to provide a buffer against potential risk underestimations, thereby enhancing model reliability. The backtesting framework was evaluated on four dimensions: stability, discriminatory power, calibration accuracy, and conservatism. Three scenarios were simulated to test the model's robustness. Results indicated that the PD model generally maintained stability and discriminatory power, though calibration issues and heterogeneity in clusters were observed. The model was conservative, overestimating risk.

info:eu-repo/semantics/publishedVersion