Study of credit rating models and design of model based on predictive analystics

Abstract

While much work has happened in creation of a Credit-score for individuals for trying to judge their suitability for clearance of loans etc not much effort has gone into using the same for post-disbursement collection of money. Given that you have more information about the individuals/decision makers now rather than when the loan was sanctioned, the use of such a model becomes more robust. This study was undertaken to look at the current existing models on credit scoring and make an attempt to create a credit scoring model which helps in better allocation of resources towards collection strategies. The question the study tries to address is how to classify the decision makers i.e. Can the individuals who have to repay credit be segmented in a manner which makes it easier to direct energies into more efficient collection strategies? We are in effect saying that individual might be willing to pay but as it may turn he/she is not always able to. But where can we draw this line. How to classify between the “willingness to pay” & the “ability to pay”? One way of doing it is to use past credit history of the individual to classify him into different segments. Each of these segments can then take a different way to address the collection strategies. Existing literature was surveyed to gain understanding of the need and scope of credit scoring and to understand the many existing credit scoring models. Then a data sample of individuals with automobile loan with a maximum of 24 EMI’s was cleaned and divided into a test sample and a validation sample. Using MATLAB a binary logit model was applied on the said data to find for ways to increase the optimization of effort going into collection strategies of retail credit industry. Also other uses of Logit models are discussed in Brand Choice Models, transit models etc