Introduction
Data aggregation is ubiquitous. To widen credit access, lenders now use nonconventional sources of personal technological information to measure borrower creditworthiness. Alternative data credit scoring is touted as a useful solution for borrowers with little or no credit history or “thin credit files.” The supposedly neutral algorithm provides a predictive analysis of the borrower’s ability to repay, thus allowing the borrower to obtain credit within the formal banking network.
Alternative data has the potential to expand access to financial services for underserved populations and make credit markets more competitive. Machine learning, or predictive behavioral analytics, collects and sorts the borrower’s social and business network identity data to assess their risk. The lender’s perception (and assessment) of tangible factors such as, educational level, internet browsing history, social media associations, health status, past and current employment, or even movies, all become relevant in assessing the borrower’s repayment of credit. Proponents argue that this digital footprint of an individual’s personal networks, choices, and habits is fairer, more transparent, or even “color‑blind,” reducing discrimination in the decision-making process.
The algorithmic formulae and machine learning models that produce digital footprint technologies are protected as trade secrets. Regulators routinely evaluate lenders for compliance with fair lending laws. Lenders, however, assert trade secret protections to shield their algorithmic scoring models. By directing, adopting, and using technologies created in-house or purchased from private companies, these lenders may intentionally or unintentionally obscure discriminatory conduct. How, then, do regulators determine if digital footprint algorithms fairly assess creditworthiness? What if the underlying data of the algorithm is incomplete or implicitly biased? What if a lender impermissibly designs the digital footprint algorithm to segment markets in legally prohibited ways, thereby perpetuating credit inequality? This Article questions lenders’ broad ability to keep secret the alternative data relied upon and offers policy recommendations for regulating this new world of digital footprint algorithmic scoring.
This Article makes three contributions to the existing literature. First, it shows how the lender’s choice of alterative data and its interpretation of that data may result in technological redlining in violation of existing fair lending laws. Second, the Article participates in the ongoing critical race theory debate about algorithmic bias and how law and technology must combine to create algorithmic justice. Specifically, it posits that failure to police algorithms for bias prior to their use can contribute to systemic discrimination in lending.
Third, the Article proposes protections for consumers’ algorithmic network identities and recommends policies that regulators are uniquely positioned to implement. It recommends a specific, transparent lending disclosure when lenders use algorithmic network identity data. Unlike in the European Union, American consumers may be unaware that a lender has used alterative data in its creditworthiness evaluation. Lenders should disclose when alternative data has been used and how it is used. Similar to other areas of law where prophylactic antidiscrimination measures are appropriate, the history of redlining and sub-prime lending in minority communities dictates a similar control in this context. Consequently, a lender will bear the burden of ensuring fairness before the lending process begins instead of providing the individual with an ineffectual post hoc remedy.