A data gold mine

Yesterday I posted about American Express data-mining their cardholder's shopping habits in order to determine who might be at risk of defaulting, and cut their credit lines.  The comment thread has gotten quite lively, with commenters variously arguing that it's unfair to use a black-box correlation matrix that the consumer can't manipulate, that the risk of Type II error is quite high, that credit card companies have as much income as they need from your income and payment history, that this violates privacy concerns.

I think there's some misunderstanding about what the card companies are trying to do when they do this kind of data-mining:  why they trim your credit card limits when your shopping patterns change, or raise your rates when you're late on another card but current on theirs.  They do these things because they can (or so the card companies believe) be reliable indicators of someone in financial distress.

A common pattern before bankruptcy is to run all your cards up to the limit, using cash advances and balance transfers to kite the payments until you can't keep the thing going, and then file.  American Express is, quite sensibly, worried that more people will be doing this over the next few years.  (And in a few of the comments, I detect a certain panic at the thought of having this option shut off). 

By the time you are a cardholder for a while, American Express has only one piece of information about you:  your payment history.  But as the fine print on the financial statements always says, "Past performance no guarantee of future results"--especially in this environment.  They don't know whether you've lost your job, or gotten divorced, or gotten into a nasty car accident.  That's why they're mining the data; right now, the information asymmetry runs heavily in favor of the cardholder.

It is true that there will be false positives, and that some people will suffer a minor ding to their FICO score because their outstanding balance will now be an unsatisfactorily high percentage of their current credit limit.  But this damage is easily rectified by paying down some of the balance.  If you can't pay down your balance to below 50% of your credit limit, then you have no business borrowing any more money, which is the only reason this temporary decrease would hurt you.

There's also a hidden assumption that if American Express can't do this sort of data mining, they won't be able to reduce the credit limits of people who frequent divorce lawyers and Monster.com.  But the credit card companies are well aware that they face massive exposure to losss on people using credit to ride out the downturn.  If you deny them the ability to fine-tune their credit management, they will simply use blunter instruments, like cutting the credit limits of everyone below a certain income, or whose card is new, or whose job application listed a risky occupation like, er, journalist.

This article available online at:

http://www.theatlantic.com/business/archive/2009/02/a-data-gold-mine/4654/