77
In Figure 4.16 a cut-off probability of 0.3 gives the lowest value of the error function. This
cut-off probability is lower than the cut-off probability when
only the total error is
minimized. Because we are risk averse, a cut-off probability of 0.3 will be used. This means
that anyone with a probability of being bad less than 0.3 will be classified as good, and any
with a probability of being bad greater than 0.3 will be classified as bad. Both cut-off
probabilities 0.48 and 0.3 were used and the results compared.
4.4 Logistic Regression Model on “new” Data
Six Fisher scoring iterations were needed for the parameters to converge. The estimated
parameters are given in Table 4.7.
Table 4.7
Logistic regression model fitted on the “new” data.
Variable
Estimate
Std. Error z value Pr(>|z|)
Significance
(Intercept)
-8.62E+00 1.36E+00 -6.317
2.67E-10
Significant
LOAN
5.85E-06
1.38E-05 0.423
0.67203
Insignificant
MORTDUE
-6.50E-06
6.97E-06
-0.933
0.350944 Insignificant
VALUE
1.62E-06
5.85E-06
0.277
0.781878 Insignificant
REASONHomeImp 1.09E-01
3.22E-01
0.337
0.736028 Insignificant
JOBOffice
-9.80E-01
5.82E-01
-1.684
0.09211
Insignificant
JOBOther
1.62E-01
4.55E-01
0.357
0.721458 Insignificant
JOBProfExe
1.06E-01
5.29E-01
0.2
0.841722 Insignificant
JOBSales
3.33E+00
9.42E-01
3.535
0.000408 Significant
JOBSelf
-1.44E-01
9.04E-01
-0.159
0.873381 Insignificant
YOJ
-2.68E-02
2.15E-02
-1.244
0.213402 Insignificant
DEROG
6.56E-01
2.10E-01
3.125
0.001779 Significant
DELINQ
1.16E+00
1.68E-01
6.904
5.05E-12 Significant
CLAGE
-6.65E-03
2.08E-03
-3.196
0.001394 Significant
NINQ
2.06E-01
6.59E-02
3.122
0.001798 Significant
CLNO
-4.08E-02
1.63E-02
-2.501
0.012374 Significant
DEBTINC
2.33E-01
3.33E-02
6.985
2.86E-12 Significant
78
What is interesting now is that the variable LOAN has gone from being significant on the
“old” data to insignificant on the “new” data, and the JOB variable has a different significant
dummy variable. Other than this, the models on the “new” and “old” data are similar. The
residual deviance of the model is 341.18 with 549 degrees of freedom.
Interpretation is now given
for the parameters of LOAN, DEROG and DEBTINC.
-
The parameter of LOAN is 5.85E-06
and is insignificant at the 5% significance level. A unit
increase in LOAN with all other variables held fixed, means that there will be a 5.85E-06
increase in the log-odds of default.
-
The parameter of DEROG is 6.56E-01 and is significant at the 5% significance level. A unit
increase in DEROG with all other variables held fixed, means that there will be a 6.56E-01
increase in the log-odds of default.
-
The parameter of DEBTINC is 2.33E-01 and is significant at the 5% significance level. A
unit increase in DEBTINC with all other variables held fixed, means that there will be a
2.33E-01 increase in the log-odds of default.
In order to check
the adequacy of the model, collinearity of the independent variables,
outliers and influential observations are now considered. The correlation matrix of the
numerical independent variables is given in Table 4.8.