摘要：The Basel II Accord offers banks the opportunity to estimate Loss Given Default (LGD) to calculate their own value for the capital required to cover credit loss when extreme events occur. This paper evaluates alternative methods of modelling LGD to provide explanatory models of LGD at the level of the account for a large sample of credit card loans which are in default. The models are also tested for estimation of LGD within a cross-validation framework using several alternative predictive performance measures. Both square error and absolute error are used to measure accuracy and we find that the choice of measure is important for determining the best model. We show that simple Ordinary Least Squares regression is as good and usually better for estimating LGD when compared with Tobit and a decision tree approach. Absolute error may be a good alternative performance measure for LGD estimations and we show that a fractional logit model performs best in this context.
Loading...
