Credit risk is a relevant problem faced by banks and financial institutions.

The traditional statistical models which are generally used to quantify the credit

risk present several drawbacks. First, in their standard versions they are not robust

and do not take into account that the data may be corrupted by several outliers.

Second, when a parametric model is employed, the variable selection procedure

might be severely affected by the so called masking and swamping effects. This

work extends robust statistical methods to credit risk analysis, showing how the

traditional approach can be greatly improved through robust methods.