| Banks and credit card lenders employ a system known as credit scoring to quantify the risk factors associated with each potential borrower. An excellent credit score usually assures the borrower's ability and willingness to pay her/his loan. Due to the massive number of applications received daily as well as an increasing number of governmental regulatory requirements, credit scoring has become a standard in the banking industry. In this thesis, the concept of credit scoring and the theory and statistics behind it are explained thoroughly. In the application sections, different statistical methods, such as logistic regression, discriminant function analysis, binary decision tree analysis, and artificial neural networks are used to analyze real data collected from a credit bureau. The results and models developed from these different analyses are then compared to determine the best method for developing a credit score model. Due to the inherently large number of attributes associated with each loan borrower provided by the credit bureau, a principal component analysis is first used to reduce significantly the number of variables that will be considered for inclusion in the credit score model. Three selection methods such as forward selection, backward elimination, and stepwise regression are also utilized to determine which subset of variables is to be included in the final model. The conclusion of the thesis discusses the best method among the four mentioned statistical methods used to analyze the data, and reveals the best final credit score model for this study. |
