Bayesian Logistic Regression Models for Credit Scoring by Gregg Webster

Table 4.14
Geweke diagnostic statistics for each variable of the Bayesian logistic 
regression model with informative prior. 
Variable 
z 
(Intercept) 
1.3294 
LOAN 
0.8905 
MORTDUE 
-0.7683 
VALUE 
-0.5705 
REASONHomeImp 
-0.4904 
JOBOffice 
1.9190 
JOBOther 
0.8031 
JOBProfExe 
-0.5658 
JOBSales 
0.4618 
JOBSelf 
-2.1442 
YOJ 
-2.5343 
DEROG 
0.6239 
DELINQ 
1.8627 
CLAGE 
0.4984 
NINQ 
0.9205 
CLNO 
-1.6003 
DEBTINC 
-0.6575 
Table 4.14 shows that the variables YOJ and JOBSelf have 
| |
. Therefore, these two 
variables have not converged. All the other variables have converged according to the 
Geweke diagnostic.
Bayesian logistic regression model with non-informative prior 
 
An improper uniform prior is now used as a prior. This is a non-informative prior and 
provides no prior information for the parameters. Again a tuning parameter of 0.6 was 
used in order to get the acceptance rate in the lower end of the 20-40% range. The 
acceptance rate was 24.5%. The model is summarized in Table 4.15.

---

90 
Table 4.15
Bayesian logistic regression model with non-informative prior on the “new” 
data. 
The parameter estimates still have the same interpretation and interpretations for the 
parameters of LOAN, DEROG and DEBTINC are: 
-
The parameter of LOAN is 5.25E-06 and is insignificant at the 5% significance level. The 
reason for this is that the 95% credibility interval does contain zero. A unit increase in 
LOAN with all other variables held fixed, means that there will be a 5.25E-06 increase in 
the log-odds of default. 
-
The parameter of DEROG is 7.69E-01
and is significant at the 5% significance level since 
its credibility interval does not contain zero. A unit increase in DEROG with all other 
variables held fixed, means that there will be a 7.69E-01
increase in the log-odds of 
default.
-
The parameter of DEBTINC is 2.52E-01
and is significant at the 5% significance level 
because its credibility interval does not contain zero. A unit increase in DEBTINC with all