In credit scoring algorithms, payment history is a pretty solid proxy for future payment behavior. Past behavior is an effective predictor of future behavior, and all that. Although, it seems another part of American Express’s credit model apparently involves the places where you shop and the payment history of other patrons there (non-annoying, non-paged version). That is, your credit risk apparently increases if you use the card in a store where other customers are having trouble making their payments.

Not the most intuitive or obvious credit factor. I’m sure the data mining gods working at the credit card companies have decreed it so, and maybe they have the data for it. Still, I’d love to see the error matrix on that one.

On the one hand, it makes some sense that there are areas of a city, state, or country more prone to customer defaults or economic effects. On the other hand, this sort of profiling might create quite a number of false positives. One fellow in the story believes his credit limit was affected after he used a card at an out-of-state shop. Whether he was on the level or not, the moral there is that location and type of business where someone shops are dependent on a myriad of other factors, which makes them rather poor features for modeling (in theory). How about online retailers or aggregated processors like Paypal? It seems strange to use rather murky assessments such as these in any significant way when the companies surely already apply more direct data about the card user (occupation? place of employment?) to the calculation.

But nowadays no one has any credit anyway, so perhaps that doesn’t matter in the long run. What other great/reasonable/dubious features go into modeling one’s credit profile…