Banks eye you carefully before stamping “Approved” on your credit card application. Using your credit score and details on your credit report, they calculate the odds you’ll end up sticking them with an unpaid balance.
But months or years later, when an unpaid debt does go into collection, the card issuer knows relatively little about who is likely to pay.
Now researchers from Texas and Switzerland have built a model that predicts which overdue credit card debts are more likely to be paid. It’s aimed at helping banks — but that could also spare consumers endless collection calls, or give them more access to payback plans.
“One key takeaway is, don’t count on history too much,” said Naveed Chehrazi, an assistant professor at the McCombs School of Business at the University of Texas. “What matters is the current pattern of behavior.”
Using data from an unnamed bank, Chehrazi and his co-researcher, Thomas Weber of the Ecole Polytechnique Federale in Lausanne, Switzerland, took a hard look at who pays and who doesn’t. They examined 6,000 accounts worth about $6 million that went into collections between 2004 and 2006.
They studied what happens after a card balance is “charged off,” meaning it no longer counts as an asset for the bank, and it no longer builds up interest. They also looked at the debtors’ credit scores, credit reports and payment histories for clues about the likelihood of payment.
It turns out that more recent information predicted repayment better than the standard credit data banks traditionally rely on. For example, a high FICO score at the time of charge-off isn’t very relevant to the odds of repayment, the research found. And the bigger the unpaid balance, the less relevant a borrower’s past financial history was.
“What matters is the kind of signaling taking place between the account holder and bank,” Chehrazi said. “When making a payment, I’m communicating that I’m serious about paying back my debt.”
The paper is titled “Dynamic Valuation of Delinquent Credit-Card Accounts,” and it has been accepted for publication in the journal Management Science. Before you click through, a warning: It’s pretty dense. It was sent back to the authors to add more text explaining their detailed mathematical constructs. However, much of the paper still looks like this:
The model gains its predictive power by factoring in developments that take place as the collection process moves along. For example, even a bounced check can be an important sign. “The payment is zero, but the signal you’re sending the bank … is you want to pay back your debt,” Chehrazi said. If you’re serious about wanting to pay but lack resources, the bank might collect more by offering a repayment plan, instead of dunning for the full amount.
Will banks use the prediction tool? Risk management consultants have contacted Chehrazi about using the model, he said, so it might wind up in use by card issuers. Any consulting fees would go to the university and fund more research. His and Weber’s work was preceded by a repayment model developed by a trio of international researchers and described in a 2010 paper in Production and Operations Management, “Optimizing the Collections Process in Consumer Credit.”
While the repayment model is aimed at banks, it could also help ease collection headaches for some consumers by tailoring collection efforts. Armed with the odds of collecting an account, banks should be able to choose better options — whether that means sending the bill out to a collection agency, offering a settlement or repayment plan, or simply dropping collection efforts in order to focus on better prospects.
Once a loan has soured, Chehrazi said, “Those are the actions under bank control.”