Bayesian Logistic Regression Models for Credit Scoring by Gregg Webster

 
 

---

62 
Chapter 4: Results 
 
4.1 Initial Data Analysis
Analysis is now illustrated on a real life home equity data set. This data set was first 
analysed by Wielenga 
et al
. (1999). Here, the methodology described in Section 3.1 is 
performed.
The data set contains the loan performance of 5,960 home equity loans. The target 
(dependent) variable is a dummy variable indicating whether or not a default occurred 
during the duration of the loan. If a default occurred the value is one; if no default occurred 
the value is zero. The data set consists of 12 input (independent) variables. The variables 
are summarized in Table 4.1. 
Table 4.1 
Variable type and description for each variable in the data set. 
Variable
Model Role Variables Type 
Description 
BAD 
Target 
Categorical - Nominal 
1 = defaulted on loan 
0 = paid back loan 
REASON 
Input 
Categorical - Nominal 
HomeImp = home improvement
DebtCon = debt consolidation 
JOB 
Input 
Categorical - Nominal 
Six occupational categories 
LOAN 
Input 
Numerical - Continuous Amount of loan request 
MORTDUE Input 
Numerical - Continuous Amount due on existing mortgage 
VALUE 
Input 
Numerical - Continuous Value of current property 
DEBTINC 
Input 
Numerical - Continuous Debt-to-income ratio 
YOJ 
Input 
Numerical - Continuous Years at present job 
DEROG 
Input 
Numerical - Discrete 
Number of major derogatory reports 
CLNO 
Input 
Numerical - Discrete 
Number of trade lines 
DELINQ 
Input 
Numerical - Discrete 
Number of delinquent trade lines 
CLAGE 
Input 
Numerical - Continuous Age of oldest trade line in months 
NINQ 
Input 
Numerical - Discrete 
Number of recent credit inquiries 

---

63 
The target variable is a binary variable consisting of ones and zeros. There are 1,189 ones 
and 4,771 zeros in the target variable. This means that close to 20% of the applicants 
defaulted during the duration of the loan or became seriously delinquent.
For the input variables, there are two categorical variables. The other ten input variables 
are all numerical of which four are discrete and six are continuous. Summary statistics for 
the input numerical variables are given in Table 4.2.