CFPB seeks to prevent algorithmic bias
On February 23, the CFPB released an outline of possible options for upcoming rulemaking to prevent algorithmic bias in automated home valuation models (AVMs). Dodd-Frank mandates that the Bureau, Federal Reserve Board, OCC, FDIC, NCUA, and FHFA engage in joint agency rulemaking to strengthen the oversight of AVMs, which requires (i) ensuring a high level of confidence in the estimates; (ii) protecting against data manipulation; (iii) avoiding conflicts of interest; (iv) requiring random sample testing and reviews; and (v) accounting for other factors deemed “appropriate” by the agencies. The Small Business Advisory Review Panel’s Outline of Proposals and Alternatives Under Consideration details options for ensuring computer models used to determine home valuations are accurate and fair. While recognizing that AVMs “have the potential to contribute to lower costs and shorter turnaround times in the performance of property valuations” and are increasingly being used—in part due to advances in database and modeling technology and the availability of larger property datasets—the Bureau cautioned that using AVMs may introduce several risks that can impact data integrity and accuracy. The outline also expressed concerns that AVMs may “reflect bias in design and function or through the use of biased data and may introduce fair lending risk.” To mitigate potential fair lending risks in AVMs, the Bureau stated it is considering proposing “a requirement that covered institutions establish policies, practices, procedures, and control systems to ensure that their AVMs comply with applicable nondiscrimination laws.” The Bureau added that it “preliminarily believe[s] standards designed to ensure compliance with applicable nondiscrimination laws may help ensure the accuracy, reliability, and independence of AVMs for all consumers and users.” Without proper safeguards, the Bureau warned in its announcement that “flawed” AVMs “could digitally redline certain neighborhoods and further embed and perpetuate historical lending, wealth, and home value disparities.”
Among other things, the outline also previewed definitions under consideration for terms such as “mortgage originator,” “mortgage,” and “consumer’s principal dwelling,” and noted that the Bureau is considering a “principles-based option” to allow regulated institutions more flexibility to set their own AVM quality control standards, as well as a “prescriptive option” with a more detailed set of requirements for institutions to reduce potential compliance uncertainty. “It is tempting to think that machines crunching numbers can take bias out of the equation, but they can’t,” CFPB Director Rohit Chopra said. “This initiative is one of many steps we are taking to ensure that in-person and algorithmic appraisals are fairer and more accurate.”
The Bureau stated that the next step will be to review the options to determine their potential impact on small business stakeholders as required by the Small Business Regulatory Enforcement Fairness Act of 1996. Feedback will be used to inform the Bureau’s efforts on developing a formal proposal with the other agencies.