Development of default models under limited data access conditions

Abstract

Since 2008 in banking sector capital requirements are set by "Basel II" approach. The internal rating models (IRB) capital calculation approach is based on probability of default (PD) models.

Historical data availability to develop PD statistical models often is limited. The model development must take into account the statistical regularities, including possible model overloading.

To develop probability of default statistical model for small and medium enterprises the modelling data set from the 2800 financial statements was used. The data set include 54 events of default.

The 34 debt coverage, liquidity, profitability, activity and other financial indicators were examined. Model overloading effects on the statistical reliability was investigated. By various statistical tests and methods the optimal number of risk factors was established and several models with different number of risk factors were compared.

It was shown that the model with optimal number of risk factors demonstrated the best statistical reliability. The results were analyzed in relation to minimum sample size criteria available in literature.