Re: DM: Credit Score

[Lynd D Bacon]

Actually, the effect of right-censoring is that you may not have observed all failures that will occur, as your data are collected over a finite time period.  So, some customers who have not yet been observed to pay late, may be late at some point.  The period varies by customer, since some customers are newer than others.  For some other variable to interact with this effect I suppose it would have to be related to how customers were selected to be customers at different times.  Alternatively, other variables could be related to the underlying hazard rate distribution(s).


Jerry, I'm sure you've thought about this, but it seems to me that time-to-first-late-payment by itself may not be very useful as far as financial policy decisions, and customer management, go.  Something related to the (risk-adjusted) present value of future payments, which would include the timing and size of payments as well as any associated fees, would probably be more useful.  But if what you're after is adjusting late-payment charges, that's a different story...


Andrew's point about the effects of things like differential mobility is a good one.  Another possibility is that customers who remain customers the longest do so because they can't get services from competitors due to their poor credit history. 8-)@


Seriously, this is an important point, since differential attrition can contribute to the association between tenure and payment failures. The idea is that the risk of late payment increases over time, and the rate of increase is perhaps related to factors that are associated with customer churn, as well as with other variables.


-lynd bacon

--------------------------------- 

LBA:Lynd Bacon & Associates, Ltd. 

       http://www.lba.com
 
       +1 708 957-0883 
       +1 708 957-1920 fax 
---------------------------------