Our email addresses, the devices we use, the ads we click on, and even our spelling ability all say something about how likely we are to default on a loan, according to research from Duke University’s Fuqua School of Business.
Manju Puri, a finance professor at Fuqua, looked at data for more than 270,000 purchases from a German furniture website similar to Wayfair. Her research, published in The Review of Financial Studies, found the information we share when we visit a website can predict credit worthiness just as effectively as a traditional credit score.
In her work, Puri has focused on the financial system, bank runs, how the 14 million-plus Americans without bank accounts can be integrated into financial markets, and the role technology can play in that process. This research, "On the Rise of FinTechs: Credit Scoring Using Digital Footprints," began as a working paper for the National Bureau of Economic Research and was co-authored with Tobias Berg of the Frankfurt School of Finance and Management, and Valentin Burg and Ana Gombović of Humboldt University of Berlin.
Puri hosted a live discussion on LinkedIn about the findings and also described the research in a Q&A.
What kind of information commonly left on websites is predicting people would default on their loans?
When we visit websites, we hand over large amounts of information, often without thinking about it. That information provides an enormous amount of predictive power. In our study we used only very basic data that any commercial website has on its users.
We first looked at what devices people used to access the site. We then studied how and when people shopped – whether they browsed during the day or late at night, and whether they searched for the site independently or clicked on a paid ad. We also analyzed the email address shoppers used when registering, and whether they typed in all lower case or made spelling errors.
Because the site we studied also offered in-house credit to customers, we were able to correlate these variables with default rates.
What specific correlation did you find between these behaviors and default rates?
Each of our measures told us something useful:
- People who used their own name in their email address were much less likely to default than people who did not. Not only that, but for people who input their email address incorrectly, the default rate is five times higher than for those who don’t make a mistake. People who don’t capitalize the first letters in their shipping address also have higher default rates.
- People who shopped between midnight and 6 a.m. were more likely to default on a loan. The same was true of people who came to the site via a paid ad rather than a price comparison website. Late night, ad-driven purchases suggest impulsive behavior, which is indicative of personality traits correlated to defaults.
- People on Apple’s operating system – an iPhone or an iPad – had lower default rates than those of Android users. This suggests that ownership of an Apple device helps predict income, a finding also supported by other research.
Many websites ask for much more information than this, such as income level or what kind of car you drive, all of which would normally predict income. These sites can say even more about you.
Overall, how does the digital footprint compare to a traditional credit score?
Collectively, this information is as good as or better than a credit score at predicting the likelihood of loan default – even though it doesn’t include any of the data that a bank would normally ask for. This very basic information, collected by surfing a web site, does as well as a credit score that looks at your outstanding loans, credit card limits, repayments, and past dues. That’s pretty amazing.
We also find that when you combine the digital footprint and a traditional score, you do better than just using the credit score.
What can companies, regulators and individuals take from this research?
This is a reminder to individuals to be thoughtful about impulsive behavior, and what we are revealing of ourselves when we are online. Maybe we think we’re anonymized for not having our names in our email, but we are still telling websites something about ourselves. Of course, there is also the possibility that as we become aware of that, we can manipulate our digital footprint. People once dressed up and put on a formal suit when they went to banks to apply for a loan. Perhaps we could now don digital “suits” when we go online, to try and influence people’s impression of us.
For the financial system more broadly, this shows a way for traditional banks and financial service firms to reach customers without bank accounts. Usually, someone opens a bank account, gets a loan and is monitored as they pay it back. But if you don’t even have a bank account in the first place then you don’t have a traditional credit bureau score, you’re out of the system. This paper shows you can give unbanked people credit using the digital footprint as a predictor of their credit worthiness.
For firms, we’re showing that you can harvest your own website and get a lot of data. Whether you’re selling physical goods, financial advice or anything else, you have a digital footprint for everyone who visits you. At the very least, companies could use it for their own customer credit analysis when deciding whether to do business with people. If you don’t have a credit score but you’re using an iPhone or iPad, if you didn’t visit the site in dead of night, and you spelled and capitalized your registration properly, chances are I can sell to you and you’re not going to default.
Finally, for regulators there are privacy issues. Who owns this data? Is it the consumer or is it the business? Should there be limits on how this data can be used? These are important and interesting questions. Fintech is becoming so important, and it’s everywhere, but still we know so little about its reach.