Recently, the credit assessment models used in practice have been based on the simple scoring functions estimated by discriminant analysis or logistic regression. These models are designed to distinguish whether or not a loan applicant belongs to the population of ‘would be‘ defaulters.
Poirier[1980] proved that the traditional credit scoring models maintain the sample selection bias problem which would cause a bias in the parameter estimation.
In this paper, it is recommended that different loan policies be applied to the firms according to the existence of a transaction data. In the case of no transaction data, the default probability model that considers the characteristics of industries and market event is considered. And, in the case of transaction data, the model is revised to correct the sample selection bias.
