As the aim of large banks has been changing to select customers of highest benefits, it is important for banks to know not only if but also when a customer will default. Survival analyses have been used to estimate over time risk of default or early payoff, two major risks for banks. The major benefit of this method is that it can easily handle censoring and competing risks. An ROC curve, as a statistical tool, was applied to evaluate credit scoring systems. Traditional ROC analyses allow banks to evaluate if a credit-scoring system can correctly classify customers based on their cross-sectional default status, but will fail when assessing a credit-scoring system at a series of future time points, especially when there are censorings or competing risks. The time-dependent ROC analysis was introduced by Hu and Zhou to evaluate credit-scoring systems in a time-varying fashion and it allows us to assess credit scoring systems for predicting default by any time within study periods.
