A carregar...
Publicação
Probability of Default Estimation in Low-Default Portfolios: Calibration, Uncertainty, and Capital Implications under an IRB-Inspired Framework
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 4.49 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Low-default portfolios (LDPs) pose a fundamental challenge for probability of default (PD) estimation because the very limited number of observed adverse events constrains both statistical reliability and prudential calibration. This study addresses a gap in the literature: existing research rarely examines, within a single corporate LDP framework, whether models that improve discrimination also deliver stable, calibrated, and capital-relevant PDs. Using Moody’s Orbis data for EU and US non-financial corporates, the final sample contains 766,775 firm-year observations and 317 oneyear-ahead adverse legal events. The empirical framework combines lagged logistic regression, Bayesian regularisation, machine-learning and oversampling benchmarks, calibration diagnostics, external anchoring, macro-financial stress testing, and IRBinspired loss and capital translation. Results show that Bayesian and machine-learning models improve ranking, but none resolves the calibration gap. Oversampling enhances class separation while severely distorting probability levels. The study contributes an integrated framework that treats risk ranking, probability calibration, stress sensitivity, and prudential translation as distinct but complementary stages in corporate low-default modelling.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Statistics and Information Management, specialization in Risk Analysis and Management
Palavras-chave
Low-Default Portfolios Probability of Default Calibration Bayesian Methods Benchmarks IRB Capital Requirements