Statistical Techniques in Scorecard Modeling

Topic:

Statistical Techniques in Scorecard Modeling

Index:

A. Objective P.3

B. Credit Scoring P.3

C. Introduction P.4

D. Data Sampling and Processing

i) Sample random sampling P.6

ii) Data Structure P.6

E. Linear Regression P.8

F. Linear Programming P.11

G. Decision Trees P.16

H. Discussion P.20

A.

Objective:

This report is to evaluate the accurateness and the effectiveness of the three methods (decision tree, linear programming and linear regression) in developing credit scorecard based the data of potential clients provided.

B. Credit Scoring:

Credit scoring, which has become progressively more significant with the remarkable growth in consumer credit recently, can be defined as a quantitative method in assessing the credit risk of retail clients. The basic principle of that is to allot a score to each individual borrower based on the information obtained from their loan applications. The major use of credit scorecard is to make lending decisions based on the value of score. A good model should give a higher score to borrowers who will likely repay loan on time while a lower score should be given to those with a higher probability of default or delinquent.

C. Introduction:

The true error rate is the best indicator that evaluates accuracy of the model as that is the error rate which tested on the true distribution of cases in the underlying population. A set of sample data is to build the model and a different part of independent sample is used for testing. The observations for predicting and testing should be randomly taken from the population in order to ensure the test error rate can estimate the true error rate reliably....