A new method for comparing Loss Given Default (LGD) models has been developed. This method focuses on loss functions related to regulatory capital charges, penalizing LGD forecast errors more for high-exposure and long-term credits. By using asymmetric loss functions, only errors underestimating regulatory capital are penalized. The approach ranks models differently from traditional methods, which only consider LGD forecast errors. When applied to six LGD models using real data, the new method showed that models' rankings based on capital charge losses differ from those based on current LGD loss functions.