Applications of Credit Scoring Models

The application of credit scoring on consumer lending is an automated, objective and consistent tool which helps lenders to provide quick loan decisions. In order to apply for a loan, applicants must provide their attributes by filling out an application form. Certain attributes are then selected as inputs to a credit scoring model which generates a credit score. The magnitude of this credit score is proved to be related to the credit quality of the loan applicant. As such, it is used to determine whether the loan will be granted, and also the amount of interest being charged. Currently, little effort has been devoted to verifying the correctness of the reported attributes provided by prospective borrowers. Moreover, with a long history of use of the same credit scoring model, borrowers will learn about the characteristics being used by the lender to make loan decisions, and may be motivated to lie about their attributes in order to increase their chances of loan approval. This thesis examines the effect on consumer lending if some borrowers strategically falsify their attributes on the application form. Under normal circumstances, analysts believe that using a larger dataset to develop credit scoring models will increase model accuracy. We will show that if some borrowers respond dishonestly to some questions on the application form, using higher dimensional data to build models will increase the associated accumulated error, and may result in having a more complex model but with low predictive power compared against using a dataset with lower dimensions. Furthermore, we will show that it is profitable for lenders to spend extra cost to directly eliminate lies in the dataset. In particular, we will examine the optimal amount of effort that lenders should spend on identifying liars in order to equilibrate between risk and return. However, we will also show that it is still possible for fraudulent loan applicants to eventually adjust their lies to escape from credit checks and get loans. Indeed the business of consumer lending may usefully be modeled as a game performed between the lender and the borrower. We will explore the cost to make a clever lie on the attributes and the cost to verify the correctness of the reported data towards the interaction between the lender and bad liars. The impact of having liars in the business not only affects the profitability of lenders, but also lowers the utility of those borrowers who always repay their loans and the utility of the economy as a whole. The proposed issues will be studied using discriminant analysis on simulated data, and then further assume the characteristics of borrowers follow half triangular distribution to present theoretical results. This research shows the importance of enriching data before making loan decisions. It can help lending financial institutions to reduce risk and maximize profit, and it also shows that it is feasible for customers to lie intelligently so as to evade credit checks and get loans