| Abstract [eng] |
The purpose of this thesis is to investigate the impact of non-normally distributed components on the estimation of the through-the-cycle probability of default (TTC PD) parameters in the Asymptotic Single Risk Factor (ASRF) model, which is used by financial institutions to calculate capital requirements under the Basel III regime. The study uses a controlled Monte Carlo simulation experiment involving 10,000 simulations for two countries, Lithuania and Germany, associated credit cycles. Each simulation involves generating a credit cycle, creating portfolios with specific parameters, and evaluating the parameter estimates for each portfolio using the General Method of Moments (GMM), a two-step regression (TSR), and a combination of them. The results show that deviations from normality in the credit cycle significantly affect the accuracy of parameter estimates, with larger deviations resulting in greater errors. In the more realistic scenario, the quality of the portfolio also plays a role in the accuracy of the estimates. The study also finds that the GMM most of the time performs better than TSR in estimating the TTC PD parameters and in certain cases, the most accurate results are provided by a combination of these two methods. Additionally, the results show that neither the GMM nor the TSR method is suitable for low default portfolios. |
