An
Illustrative
Example:
•
A
typical
scorecard
might
look
like
this
for
the
variable
current
debt
to
annual
income
ratio.
Range Points
0.00 to 0.05 ---------------------------------- 20
0.05+ to 0.25 ---------------------------------- 30
0.25+ to 0.75 ---------------------------------- 40
0.75+ to 1.25 ---------------------------------- 50
1.25+ to 2.00 ---------------------------------- 35
2.00+ to 3.00 ---------------------------------- 25
3.00+ to 5.00 ---------------------------------- 15
5.00+ to 8.00 ---------------------------------- 5
8.00+ to max ---------------------------------- 0
If
the
current
debt
is
$50,000
and
the
annual
income
is
$40,000
for
a
particular
loan
applicant,
the
debt
to
income
ratio
is
1.25
and
the
number
of
points
attributed
to
this
variable
is
50.
With typical scorecards, a single dollar increase in the current debt will push the applicant to the next bracket and reduce the points by 15. Scorecards will cause abrupt changes at the data bracket boundaries.
Neural
Networks do not suffer from this shortcoming. They allow smooth
transitions throughout the range. Continuous variables are treated as
truly continuous.
This is a great operational
advantage.
