Abstract. The main substantial features of the PD curve (default probability) formed in practical modeling are substantiated in the articles. It is proved that the main characteristics of the PD curve are that it is based on data on the actually restored default rate in each of the risk classes over a period of time and has a shape that approximate for coincides with the exposure function. It is shown that the best aspect that affects the calibration is the number of rating classes and ways to build them. It is determined that the slope of the curve demonstrates the classification model of efficiency. It is determined that the slope of the curve demonstrates the classification efficiency of the model. Models with high discriminant properties are characterized by a curve shape that has a slow increase in the rating classes of the upper part of the scale and a significant acceleration of growth in the last risk classes. Two main approaches to determining the number of risk classes are analyzed: the percentile-based approach and the equal score range approach. It is shown that when forming classes, it is necessary to take into account the total amount of sample observations, the proportion of «good» and «bad», and choose the number of classes so that it is not too large and not too small. Calibration practice shave been shown to be influenced by data, purpose, and study limitations. The application of the least squares method and the extrapolation method is considered on practical examples. The least squares method and in particular the derived extrapolation method allow to build a calibration curve on the basis of data on the relative frequency of defaults. It is determined that the mathematical apparatus of the family of nonlinear curves allows to model the process of exponential growth with different levels of intensity. The exponential curve and related functions may be useful in modeling more conservative PD estimates or for models with highly discriminatory properties, while the Weibull function, S-curve, and power function may be better adapted to moderate growth processes. The application of practical methods of constructing the PD scale is important for many domestic banking professionals who deal with internal models of credit risk.
Keywords: Calibration, Default, Probability, Curves, Probability of default curve calibration, Least squares method, Extrapolation method.
JEL Classіfіcatіon С44
Formulas: 21; fig.: 1; tabl.: 7; bibl.: 10.