Abstract
Banking regulation stipulates that to calculate minimum capital requirements a long-term average of annual default probability (PD) should be used. Typically, logistic regression is applied with a 12-month sample period to obtain retail PD estimates. Thus the output will reflect the default rate in the sample, and not the long-term average. The ensuing calibration problem is addressed in the paper by a ‘variable scalar methodology’, based on an actual application in a commercial bank. Using quarterly intra-bank loss data over 15 years, a state-space model of the credit cycle is estimated by a Kalman filter, resulting in a structural decomposition of the credit cycle. This yields an adjustment factor for each point in the cycle for each of two client segments. The regulatory compliance aspects of such a framework, as well as some practical issues are presented and discussed.