News

Improving Borrowing Behavior Through Social Media Analysis

Methods to help people anticipate and avoid poor borrowing decisions

Have you ever borrowed money to buy something and later regretted the decision? Maybe even wondered if you had temporarily lost your mind? If so, you’re not alone, at least judging by recent data on credit card debt, loan defaults, and home foreclosures. In research conducted for the TFI and U.S. National Science Foundation, data scientists Rich Colbaugh and Kristin Glass have developed methods to help people anticipate and avoid poor borrowing decisions by leveraging social media and machine learning.

Their report shows that using social media to understand borrowing behavior has advantages over other forms of data: social media posts are inexpensive to collect, reveal underlying decision-drivers even in settings that may be embarrassing or distressing, and contain user-specific information, opening up the possibility to give personalized decision-support. The research also demonstrates that machine learning, the technology which helped Watson win at Jeopardy, enables good models for borrowing to be built by examining many examples of loan outcomes, both good and bad, and learning what patterns of social media activity are predictive of defaults. Importantly, this approach does not depend on preconceived ideas of what drives borrowing behavior

Twitter-based prediction models have been learned for several poor borrowing outcomes, including credit card delinquency, auto loan default, student loan delinquency, borrowing at high interest rates, and having debt in collections. These models predict loan results with accuracies approaching or exceeding 90%. Additionally, the project has identified concrete ways to improve borrowing decisions.

"by supplying early warning that a contemplated loan is likely to produce an undesirable outcome, the system can help users anticipate and avoid these outcomes"

Rich Colbaugh and Kristin Glass

For instance, by supplying early warning that a contemplated loan is likely to produce an undesirable outcome, the system can help users anticipate and avoid these outcomes. Pinpointing drivers of poor borrowing facilitates design and implementation of behavior-change strategies aimed at decreasing the likelihood of poor borrowing in the future; a mobile “app” is being designed to deliver such strategies in a convenient format.

The finding that emotions in the weeks leading up to a borrowing event are more important than rational considerations gives a new, empirically-grounded perspective on personal borrowing. Finally, it is worth noting that, although the present project focuses on borrowing, many of results may be applicable to other financial activity, such as savings, investing, and budgeting.