August 26, 2021
I believe we’ve reached a crossroads in consumer finance, where things will probably get either much better, or much worse, due to the technological transformation of products, providers, cost structures, business and market models, and infrastructure. Getting the good out of technology, and preventing its potential harms, will be hard, and much of the burden for getting it right will fall on policymakers. It will be incredibly challenging to look at so many novel activities and to get the regulatory response really right -- not to under-regulate, not to over-regulate, not to misregulate, not to overlap potentially conflicting regulatory strategies, and not to leave gaps where activities that should be regulated, aren’t. It will be especially hard to get all this right fast enough to keep up with exponential rates of change in financial markets.
The single most critical factor in meeting this challenge is to equip regulators with really sound, empirical, trustable information about how these technologies will affect the public. And that’s why I’m excited to welcome today’s guest. Melissa Koide is CEO of FinRegLab, and she joins me to talk about how she and her colleagues are meeting this need.
Full disclosure at the top: I serve on FinRegLab’s board of directors and chaired it for the first few years. In our talk, Melissa explains what inspired her to found the lab after leaving the Treasury Department in 2017, where she had seen the need for unbiased, empirical research on new financial technology and for insights on how public policy may need to change to address both the opportunities and risks arising.
The first technology change that FinRegLab has tackled is arguably the single most promising innovation anywhere for opening a chance to reach full financial inclusion -- namely, using new data and new techniques for assessing credit risk. Both of FRL’s first two projects explore this topic. The first one looked at “cash flow underwriting,” a technique being used by many fintechs and now by some banks, in which the lender seeks the customer’s permission to evaluate cash flow information in the person’s bank account, to supplement traditional credit history. The second project, underway now, builds on the first to evaluate what may prove to be an even more revolutionary change, namely bringing artificial intelligence techniques into credit underwriting. This one includes exploring the dicey issue of what to do to make AI models “explainable” to our human brains, so that we can be sure that they are both sound and, crucially, fair.
We don’t yet know the results of the second study, but the first one’s findings were very promising. As Melissa explains, current credit underwriting models rely heavily on credit history information, simply because this has traditionally been the only information that is easily available and easy to put to use. Credit history information is of course valuable, but it almost automatically excludes people who have little or no credit history, and also can penalize people who may have a complex credit background but who are, today, actually creditworthy. The system works well in many ways, but today, there’s no need to confine ourselves to using such a narrow data stream, because we now live in an ocean of data. The combination of digitization and new analytic techniques is making it possible to ingest vast volumes of information and to use it in finely tuned ways that were never possible before.
We’ve talked about this issue many times on Barefoot Innovation. See below for links to some of those episodes.
The challenge, of course, is to be sure that these new techniques are used responsibly -- that they assess risk accurately, and that they don’t import or develop biases and worsen unintentional discrimination in the credit system. And to do that, we need ways to evaluate them -- which means that, in some form or fashion, we need to make them “explainable,” or at least need to “police” them through methods we know for sure that we can trust. Melissa talks about the main options for solving this core explainability challenge. She also talks about the related issue of whether these models can be designed to produce clear reasons for denial in the adverse action notices that current law requires lenders to disclose when they decline an application.
Regulators are already encouraging lenders to explore using new kinds of data. Many banks are still hesitant to do so, however, partly due to regulatory uncertainty -- these models expose them to charges of illegal “disparate impact” discrimination, and policymakers have not yet developed clear guidance on how to avoid that pitfall. To unlock the upside potential of these new tools, regulators will have to figure out what the rules of the road should be. To do that, they will need the kind of objective, empirical information coming from FinRegLab.
If the evidence bears out the hope that these techniques can be more inclusive, with no loss of predictiveness (or even improved accuracy), I predict that the day will come when the regulatory paradigm will shift 180 degrees: rather than critiquing the new methods, regulators will be critiquing the old ones, as discriminatory. I also predict that, in the next few years, regulators will be incorporating their own version of this kind of analysis to find credit discrimination in their examinations.
Between that day and today, there is a huge amount of work to do, starting with deeply understanding what works, and what doesn’t. Watch the work of FinRegLab in the months and years ahead.
More on Melissa
Melissa Koide is the CEO of FinRegLab, a nonprofit financial innovation and research center that examines how technology and data can help achieve public policy aspirations, address regulatory requirements, and lead to a more efficient and inclusive financial marketplace. FinRegLab provides an independent platform for financial stakeholders and policymakers to dialogue and gain an evidence-based understanding of new financial technologies. FinRegLab’s research and experiments focus on matters such as how data and technology can be used to expand prudent access to credit for underserved consumers and small businesses and how new technologies can improve customer onboarding and meet KYC obligations.
Prior to establishing FinRegLab, Melissa served as the U.S. Treasury Department’s Deputy Assistant Secretary for Consumer Policy. In that role, Melissa helped to build the first government offered pre-retirement savings product, the myRA, and she established the $5 million Innovation Fund to support research and strategies to improve consumers’ financial health and their access to safe and affordable financial products and services. Before joining Treasury, she was the Vice President of Policy at the Center for Financial Services Innovation. Melissa is currently Vice Chair for the Milken Institute’s Fintech Advisory Council.
More for our Listeners
Next up on Barefoot Innovation we will have Stephanie Cohen, Global Co-Head of Consumer and Wealth Management at Goldman Sachs. And, happily, we will be making my 2021 SXSW session with Cleve Mesidor, on Diversifying Tech, into a podcast soon. Also watch for a show with Lisa Rice, CEO of the National Fair Housing Alliance.
Coming up, David will be speaking at AITE Group’s Financial Crime Forum. And don’t forget to register for my FDIC presentation, Banking on Data: Great Possibilities, Great Responsibilities!
In the fall, I’ll be speaking at SCxSC 21, hosted by the securities regulatory commission in Malaysia. And I’ll be at Money 2020, speaking with former acting Comptroller of the Currency Brian Brooks and Hummingbird Co-CEO Matt Van Buskirk on the future of cryptocurrency.
And be sure to sign up for AIR’s upcoming webinar with Google on Developing the Digital Regulator, on September 15. Among other things, I’ll be doing a fireside chat with the legendary Vint Cert, Presidential Medal of Freedom winner and one of the “fathers of the internet” (as well as the inspiration for the Architect character in the Matrix films. We will hope to see you there.
Stay informed by joining our mailing list