by Ricardo Schechtman
Central Bank of Brazil
This study investigates new proposals of statistical tests for validating the PDs (probabilities of default) of credit rating models (CRMs). The proposed tests recognize the existence of default correlation, deal jointly with the default behavior of all the ratings, and, in contrast to previous literature, control the error of validating incorrect CRMs. Power-sensitivity analysis and strategies for power improvement are discussed for the calibration tests, whereas a non-typical goal is proposed for the tests of discriminatory power, leading to results of power dominance. Finally, Monte Carlo simulations investigate the finite sample bias for varying scenarios of parameters.
JEL Codes: C12, G21, G28.
Full article (PDF, 48 pages, 1,749 kb)