Title Posterior-Based estimation and benchmark-based validation of probability of default in low default portfolios
Translation of Title Aposterioriniu skirstiniu pagrįstas įsipareigojimų nevykdymo tikimybės įvertinimas ir palyginamaisiais įverčiais pagrįstas validavimas žemo nemokumo portfeliuose.
Authors Komarovaitė, Enrika
Full Text Download
Pages 68
Keywords [eng] Credit Risk, Probability of Default, Low Default Portfolio, Bayesian Estimation, Probability of Default Validation, Kredito rizika, Įsipareigojimų nevykdymo tikimybė, Žemo nemokumo portfelis, Bajeso vertinimas, Įsipareigojimų nevykdymo tikimybės validavimas
Abstract [eng] Under the internal ratings-based approach introduced in Basel II, banks rely on internal probability of default (PD) estimates as key inputs to regulatory capital requirements to cover unexpected credit losses. In low-default portfolios, default events are rare and may even be zero over the years, complicating the estimation and validation of banks' internal PD estimates. In this thesis, PD estimators designed explicitly for low default portfolios are reviewed and used as benchmarks to evaluate the conservatism of internal bank PD estimates. The PD estimation methods reviewed in this thesis are the upper confidence bound approach (Pluto \& Tasche, 2005), considered at different confidence levels, and the Bayesian approach (Tasche, 2013), considered under conservative and uniform priors. To assess the conservatism of banks’ internal PD estimates, two measures are applied: a log-ratio of the upper confidence bound to the bank’s internal PD estimate and the posterior tail probability that the true PD exceeds the bank’s internal PD estimate. For better interpretability and comparability across portfolios and years, both raw measures are transformed to a standard [0,3] score scale and mapped to traffic-light colors. The results show that estimates produced by estimators specifically designed for low default portfolios remain well-defined in cases of rare default events, including when no defaults are observed. The proposed validation framework offers a transparent, adjustable, and interpretable way to flag potential underestimation of the bank’s internal PD estimate.
Dissertation Institution Vilniaus universitetas.
Type Master thesis
Language English
Publication date 2026