Embed
Email

Automobile Loan Credit Model

Document Sample

Shared by: huanglianjiang1
Categories
Tags
Stats
views:
0
posted:
11/21/2011
language:
English
pages:
9
Automobile Loan

Credit Model

Matt Bean

Matt Caspermeyer

Venkat Sakruti

Methodology

 50% – 50% Data Split into training and validation

 Dummy variable creation

 Performed discriminant analysis via regression

in SAS

 Removed various variables based on following

analysis:

 Ensured coefficients of variables were logical

 Eliminated highly correlated variables

 Eliminated insignificant variables



 Final model consisted of 48 dummy values

derived from 13 variables

Scorecard

Variable Name Interval Points Variable Name Interval Points

Intercept (all applicants start with this score) Intercept 668 0 83

$0-1200 -31 1-3 32

$1201-$1600 0 4-6 0

Amount of down payment $1601 + 40 # of trades opened in last 24 months 7+ -62

0-1 0 0 35

2-3 -18 1-5 0

# of derogatory public records 4+ -40 6-15 -35

0-6 -48 # of trades ever rated as bad debt 16+ -88

7-48 0 .01% - 40% 45

Time at residence (in months) 49+ 10 40.01% - 75% 0

0-9 -51 Ratio of balance to credit limit for open auto 75.01% - 90% -69

10-59 0 trades 90.01% + -88

Time at job (in months) 60+ 14 Car 0

0-23 -32 Auto Type Truck 50

24-35 0 0% -39

36-49 39 .01% - 40% 0

50-54 47 40.01% - 50% 42

Customer age (in years) 55+ 99 Ratio of satisfactory trades to total trades 50.01% + 75

0-4 87 0-1 -31

5-6 48 2-5 0

7 21 6-7 43

8 0 # of revolving trades 8+ 93

9 -46 $0 - $1000 0

10-12 -69 Value of trade in vehicle $1001 + 38

Age of car (in years) 13+ -131

K-S Statistic

 Measures maximum separation between

the cumulative percentage distributions of

known good/bad applicants based on

model score.

K-S Test – Training Dataset

KS Test for Training Dataset



100





90





80





70





60

Cumulative %









Cumulative % good

50

Cumulative % bad



40





30





20





10





0

E



5



5



5



5



5



5



5



5



5



5



5



5



5



5



5



0

97



92



87



82



77



72



67



62



57



52



47



42



37



32



27



20

R

O



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO

M

R



1



1



1



1



1



1



1



1



1



1



1



1



1



0



1



0

O



95



90



85



80



75



70



65



60



55



50



45



40



35



30



25

01

10









Score Ranges







Maximum separation of .2497 at score range of 651 to 675

K-S Test – Validation Dataset KS Test for Validation Data Set



100





90





80





70





60

Cumulative %









Cumulative % good

50

Cumulative % bad



40





30





20





10





0

E





5



5



5



5



5



5



5



5



5



5



5



5



5



5



5

97



92



87



82



77



72



67



62



57



52



47



42



37



32



22

R

O



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO

M

R



1



1



1



1



1



1



1



1



1



1



1



1



1



0



0

O



95



90



85



80



75



70



65



60



55



50



45



40



35



30



20

01

10









Score Ranges





•Maximum separation of .2345 at score range of 651 to 675.

•Separation value is 6% lower than training dataset.

K-S Test – Combined Dataset

KS Test for Combined Dataset



100





90





80





70





60

Cumulative %









Cumulative % good

50

Cumulative % bad



40





30





20





10





0

E



5



5



5



5



5



5



5



5



5



5



5



5



5



5



5



0

97



92



87



82



77



72



67



62



57



52



47



42



37



32



27



20

R

O



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO



TO

M

R



1



1



1



1



1



1



1



1



1



1



1



1



1



0



1



0

O



95



90



85



80



75



70



65



60



55



50



45



40



35



30



25

01

10









Score Ranges





Maximum separation of .2362 at score range of 651 to 675

How did the model perform against

known behavior types?

651+ Actual Data

Good Bad Totals



n

io

ct

Good 6147 1908 8055

i

ed

Pr





Bad 3295 2692 5987

Totals 9442 4600 14042







Success Rate 651+ 62.9%



Accepted at 651+ 8055

% Bad at 651+ 23.7%



If Accepted all 14042

% Bad would be 32.8%

Shift acceptance to 626+

Results in 301 additional Good customers with

1.7% more Bad customers than 651+



626+ Actual Data

Good Bad Totals

n

io

ct









Good 6875 2335 9210

i

ed

Pr









Bad 2567 2265 4832

Totals 9442 4600 14042



Success Rate 626+ 65.1%



Accepted at 626+ 9210

% Bad at 626+ 25.4%



Related docs
Other docs by huanglianjiang...
Property Loss Damage Claim Form
Views: 0  |  Downloads: 0
Hurt Hawks
Views: 0  |  Downloads: 0
MEIJI UNIVERSITY_ TOKYO
Views: 4  |  Downloads: 0
Noam Gill GBAS System
Views: 0  |  Downloads: 0
GCD
Views: 0  |  Downloads: 0
FREE COMMUNITY SECTOR JOB ADVERTISING TRIAL
Views: 1  |  Downloads: 0
Constraints
Views: 0  |  Downloads: 0
Monroe Co Benton twp AR 2009
Views: 0  |  Downloads: 0
By registering with docstoc.com you agree to our
privacy policy

You are almost ready to download!

You are almost ready to download!