Studying the lifetime of a credit card account is an interesting topic for credit card issuers. What is the natural cause of death for a credit card account? It is credit card fraud, which eventually results in the account to be killed and replaced with a new number.
Source for picture: credit card fraud
Credit card accounts can die from various causes:
Yet, given enough time, a credit card account will eventually meet its natural fate: being hijacked by criminals, and die - I call this dying from old age. From a business point of view, this brings an interesting question: how do you predict, based on user segment, when a credit card will die, and how to leverage your forecast to limit losses.
Here's my proposal, based on some anecdotal evidence:
It is an interesting problem from a predictive modeling viewpoint. It is sometimes referred to as time-to-crime statistical prediction, in which exponential distributions used to model lifetime of an account, are combined with censored data (censored because future is not known) for parameter fitting, to predict time-to-crime.
Here's a quick way for an issuer to further reduce losses, based on my above example:
You might as well automatically change the account number after x years, but unless all issuers agree on this scheme, as an issuer, you are likely to lose users if you implement this strategy.
Question: do you know any study providing details about the lifetime of credit card accounts?
Most card issuers model account attrition due to various causes. The attrition rate due to fraud is quite small compared to the other main effects: charge-off, bankrupt, deceased, closed by request. The problem with fraud is that it is primarily an exogenous effect (random point in time), rather than an age-related effect. Examples of exogenous effects that would have to be filtered out in any model are accounts involved in the Target and Home Depot breaches. Once filtered out, is there enough data left to build a plausible model?
Resources are better spent predicting major breaches. Larger issuers may be able to spot patterns of suspicious activity at their infancy and determine the set of their cards that are at risk.