R. Peláez, R. CAO, J. VILAR, I. Van Keilegom
This paper discusses techniques for estimating the probability of default (PD) based on cure models. A nonparametric survival estimator proposed by Lopez-Cheda et al. (2017a) and Lopez-Cheda et al. (2017b) is considered and transformed to obtain an estimator of the probability of default (NPCM estimator). Its behaviour is compared by simulation with Beran's PD estimator and parametric methods based on cure models. The results obtained show that the NPCM estimator provides good estimations of PD and reduces the error committed by the parametric alternatives. Beran's PD estimator is competitive with the NPCM estimator in most scenarios. The asymptotic properties of the NPCM PD estimator are analysed: an almost sure representation of the estimator and asymptotic expressions of the bias and variance are obtained, as well as its asymptotic normality. Finally, to illustrate the use of Beran's and NPCM estimators, a statistical analysis of German bank loans is carried out.
Keywords: Survival analysis, cure models, kernel estimator, probability of default, credit risk
Scheduled
GT15 Risk Analysis
June 7, 2022 12:00 PM
A03