Bayesian Logistic Regression Models for Credit Scoring by Gregg Webster

 
1.2 Objectives of the Study
The objectives of this study are as follows: 
-
Investigate credit scoring and the associated problems - such as reject inference.
-
Introduce the concepts and methods of the Bayesian logistic regression models for credit 
scoring. This includes an in-depth explanation of the Markov Chain Monte Carlo (MCMC) 
methods. 
-
Develop a standard logistic regression scorecard. 
-
Develop a Bayesian approach to the scorecard for when the bank enters a new market or 
there is a change in procedure. 
-
Compare the Bayesian approach to the standard logistic regression approach. This would 
involve comparing the models’ predictive powers on a test set.
-
Make recommendations on the Bayesian approach to credit scoring.
 
1.3 Organization of the Study 
Chapter 2 gives the history of credit scoring, problems with credit scoring and examines 
previous research on models used for credit scoring. The chapter provides a literature 
review on the models used for credit scoring focusing on the Bayesian logistic regression 
models. Chapter 3 examines the methods used in detail; it provides derivations and proofs 
of key results in order to gain an understanding of the models used. In Chapter 4 the results 
of the data analyses are presented and discussed. Chapter 5 summarizes the study, gives 
limitations and discusses areas for further research.

---

14 
Chapter 2: Literature Review
 
2.1 History of Credit Scoring 
Credit scoring is essentially a classification problem where applicants are classified into 
different groups. According to Thomas (2009) statistical classification techniques started 
when Fisher (1936) developed one of the first successful classification models to classify 
three different types of the iris flower. He used different physical measurements of the 
flower to discriminate between the three types of Iris flowers. Durand (1941) was then the 
first to recognise that these statistical classification techniques could be used to classify 
good and bad loans. Before this, Thomas (2009) states that financial institutions based 
decisions on whether to grant credit subjectively. When credit cards were introduced in the 
1960s, the usefulness of credit scoring started to be realized. Because of the large number 
of people applying for credit cards, automation of the credit application procedure seemed 
to be the only solution. When the financial institution introduced the credit scoring model 
they found that the model performed a lot better than the previous (subjective) judgment 
scheme. The result was that, as Thomas (2009) states, default rates dropped by 50% or 
more. In the 1980s the success of credit scoring in credit cards meant that financial 
institutions started using scoring methods for other products too such as personal loans, 
home loans and business loans.
The subprime mortgage crisis caused a global recession in 2007. This crisis proved that 
financial institutions did not fully understand the risks they were taking on. According to 
Rona-Tas and Hiß (2008) a credit score generally used by financial institutions in the 
U.S.A. is the Fair Isaac Co. (FICO®) score. They state that these FICO scores grew 
steadily from 2000 to 2005. This made subprime borrowers appear less risky. Possible 
reasons for these inflated FICO scores include the data used to construct the FICO scores 
are historical data, not necessarily only from subprime lenders, and banks putting pressure 
on credit rating agencies to inflate their credit rating scores. The reason why banks would 
put pressure on credit rating agencies is that they were able to sell their loans to investors. 
Thus, the banks would want to grant as many loans as possible and then sell them to 
investors.

---

15