The parameters of application scorecards are usually estimated using a sample that excludes rejected applicants which may prove biased when applied to all applicants. This paper uses a rare sample that includes those who would normally be rejected to examine the extent to which (1) the exclusion of rejected applicants undermines the predictive performance of a scorecard based only on accepted applicants, and (2) reject inference techniques can remedy the influence of this exclusion.

This study investigates the impacts of unobservable firm heterogeneity on modelling corporate bond recovery rates at the instrument level. Based on the recovery information over a long horizon from 1986 to 2012, we find that an obligor-varying linear factor model presents significant improvements in explaining the variations of recovery rates with a remarkably high intra-class correlation being observed. It emphasizes that the inclusion of an obligor-varying random effect term has effectively explained the unobservable firm level information shared by instruments of the same issuer and thus results in an improvement of predictive accuracy of recovery rates. The empirical results show that the latent economic cyclical effects have been well represented by firm level heterogeneity, and strong evidence is presented for the normal distributional assumption of the recovery rates. Finally, we demonstrate the choice of recovery rate models may influence portfolio risk with the obligor-varying factor model generating a more right clustered loss distribution than other regression methods on the aggregated portfolio.