Further investigation of parametric loss given default modeling

ABSTRACT We conduct a comprehensive study of some parametric models that are designed to;fit the unusual bounded and bimodal distribution of loss given default (LGD). We;first examine a smearing estimator, a Monte Carlo estimator and a global adjustment;approach to refine transformation regression models that address issues with LGD;boundary values. Although these refinements only marginally improve model performance,;the smearing and Monte Carlo estimators help to reduce the sensitivity;of transformation regressions to the adjustment factor. We then conduct a horse race;among the refined transformation methods, five parametric models that are specifically;suitable for LGD modeling (two-step, inflated beta, Tobit, censored gamma and;two-tiered gamma regressions), fractional response regression and standard linear;regression. We find that the sophisticated parametric models do not clearly outperform;the simpler ones in either predictive accuracy or rank-ordering ability, in-sample,;out-of-sample or out of time. Therefore, it is important for modelers and researchers;to choose the model that is appropriate for their particular data set, considering differences;in model complexity, computational burden, ease of implementation and model;performance.