Re: DM: Credit Score

[Alexandr A. Savinov]

Musial Jerry wrote:
> 
> Hi,
>         I am looking for help/resources related to credit scoring of
> 'existing' customers.  In particular, I am interested in how 
>researchers
> deal with tenure as it relates to whether or not one of our 
>customers will
> continue to pay their bills in a timely manner.  We intuitively 
>feel that
> the longer a customer has been with us (and paying us) the better 
>credit
> risk.  However, my initial results show that the longer a person 
>has been
> with us, the odds of his not paying increase.  Anyone have any 
>ideas?

You try to predict some (bad) event (failure to pay, churning, 
death, etc.) analyzing time sequence, i.e., one dimensional 
distribution of previous payments. This is like if you were 
trying to predict some disease proceeding only from the patient 
age or exchange rate proceeding only from its history. Generally, 
for detailed analysis more dimensions should be taken into account. 
Thus we come to the problem of multidimensional analysis of some 
distribution. What concretely you will try to discover in this 
distribution depends on the situation and methods applied. For 
example, you could try to find groups (clusters) of customers with 
similar behavior, in particular, the group of potentially "bad", 
"dangerous" in some sense customers. More complicated analysis 
consists in finding dependencies in transitions of customers 
between groups. Or, an interesting problem is finding dependencies 
between dimensions (=attributes=variables=...). 

So the key point here is _multidimensional_ mode of thinking about 
the problem. Of course, you always can project your results and the 
whole distribution onto a single dimension, but obviously this 
result is not very informative (this is why your question arises).

Regards,

Sasha Savinov

--
Alexandr A. Savinov, PhD
Senior Scientific Collaborator, Laboratory of AI Systems
Inst. Math., Moldavian Acad. Sci. 
str. Academiei 5,  MD-2028 Kishinev, Moldavia
Tel: +3732-73-81-30, Fax: +3732-73-80-27
mailto:savinov@math.md
http://www.geocities.com/ResearchTriangle/7220/