NPR looks at pros and cons of digital lending and machine learning

Newswire: March 31, 2017.

Dateline: Washington.

Digital lending poses a significant challenge to government regulation, not just in the pace and variety of its growth but also in new methods it uses to evaluate creditworthiness. Today, NPR’s Morning Edition examined the matter of alternative data, and the questions of fairness it poses in lending practices.

For now, alternative data is mainly used in approving short-term, high-interest loans. Alternative data informs machine learning, or complex algorithms that factor in such disparate data as SAT scores, public records and purchase histories. In 10 years, says one current digital lender, hardly any credit decision will be made without it.

This presents questions of fairness and privacy, of course, particularly if someone’s browsing history, contacts list or diligence in punctuating text messages is believed to be a sign of a bad credit risk.

For its part, online lender Upstart believes the current basis of an applicant’s FICO score and reported income to be “quite biased against people.” Dave Girouard argued that machine learning could make other factors, such as where one lives, irrelevant in access to credit.

A fomer government regulator, Jo An Barefoot, countered that other habits could lead to all sorts of conclusions about someone’s lifestyle that, paired with existing data, could unjustifiably thumb them down for a loan.

In sum, Barefoot believes digital lending to be a public benefit, but it will be incumbent on regulators to ensure that it is fairly applied.

Read the full story here, or listen to it below: