Modelling credit products acceptance rate probabilities using dynamic programming and cross top application characteristics remainder offer characteristics tree

Credit scoring is a method of credit evaluation in the aspect of predicting credit acceptance especially in finance and banking industries. There are two types of credit scoring methods which are deductive (or judgemental) credit scoring, and empirical (or statistical) credit scoring. This thesis studies empirical credit scoring techniques, in particular, for estimating take-up probability distribution. As the market becomes saturated with many products and services, competitors are concerned about deepening the relationship with their customers. Cross-selling is the method used by plenty of competitors to achieve their purpose. It is an activity of selling multiple additional products or services to the present or existing customers. An enduring relationship between a customer and the organisation might be created by promoting to the customers the most appropriate products and explaining how the products may help them to achieve long-term financial needs. However, matching the customers with the right credit products is challenging because it involves risk on the financial provider's side while it is important to ensure the credit products offered are within the customer's purchasing power and satisfy the customer's credit eligibility. This requires financial providers to have effective credit scoring instrument which enables them to be able to make decisions effectively (easily, quickly, yet safety). The credit scoring instrument provides them with the best estimate of each customer's financial creditworthiness according to the products to be recommended. This study proposes a modified credit scoring model for credit products in cross selling. The focus of the credit scoring techniques in this thesis is on modifying a Classification and Regression Trees (CART) method, namely the Top Application characteristics Remainder Offer characteristics Trees (TAROT) and improving a dynamic programming model for predicting the best offer to be extended to the next customer. TAROT is used to classify which questions from the dataset to be asked for the purpose of cross-selling activities. The proposed dynamic programming model, with Bayesian updating to include the probabilities of acceptance of the previous customers, is used to match a suitable product to the suitable customers for the three and four variants of the product. Both techniques, the modified TAROT and the improved dynamic programming, were attempted to estimate the acceptance rate of credit cards products. Based on the study, it has been found that there is only one switch of offer occurs regardless of number of the credit products. The number of questions to be asked can be kept as minimal as possible in the decision trees.

Preview Text FS 2014 76 - IR.pdf Download (1MB) | Preview

Actions (login required)

View Item