The novel coronavirus upended American lives, changing how we work, how we socialize, how we shop, and - for millions of us - whether we are still employed. For many creditors, it also changed patterns of consumer default. In the four decades I have worked in consumer finance law, I have watched a dramatic change in the sophistication and reliability of credit underwriting systems. In the 1970s, when my career began, credit scoring systems were becoming popular, supplementing and even replacing rules like the "Four Cs - Character, Capacity, Capital, and Collateral" or some variation of these "C" terms.
The biggest limitation of credit scoring systems in those days was the lack of an easy way to quantify a consumer's credit history. The credit industry then developed credit scores - a single number that predicted risk based on a multitude of credit bureau factors - and the credit scoring industry took a giant leap forward and is continually improving.
An element that all credit scoring systems have in common is that they are based on the repayment histories of customers who have come before the current applicant. Certain traits have been good predictors of repayment or default, based on the performance of past credit customers. These models are regularly revalidated and updated to keep them at peak performance. And this approach works well - except when it doesn't.
Many creditors have seen surprising spikes in default rates. Customers who looked like good risks are suddenly missing payments. But these default spikes are not occurring evenly across the country; in fact, they seem quite localized. Merely raising required scores for approval is too blunt a tool to address this. The newly delinquent customers are not necessarily the ones with marginally acceptable approval scores.
One client took a close look at where spikes in defaults were occurring and determined that the defaults seemed concentrated in areas where the industries most affected by COVID-19 played a big role in local economies - tourism, retail sales, transportation, plus the oil and gas business, which recently experienced big layoffs due to a sudden drop in energy prices. The client realized that its historically focused models weren't catching the new losses due to COVID-19. The company needed an enhanced model that could predict the effect of these radically new times on default rates. Where were the layoffs going to grow and not recover right away?
The answer is not just industry specific. Economists tell us that every 10 job layoffs in a single industry will result in another 15 or more layoffs in other industries. Five of those additional layoffs will be workers who were part of the industry's supply chain, such as raw materials, parts, and transportation. Ten more will be affected by reduced consumer spending, from the local supermarket and car dealership to the little car wash and sandwich shop.
This client correctly understood that knowing a credit applicant's employment industry sector was not enough. It needed to consider all the jobs that would also be lost by a big downturn in an area where the affected industries provided lots of jobs. This meant considering at-risk geographic areas.
My clients who have talked with me over the years about the Equal Credit Opportunity Act and disparate impact have heard me say that considering geography is the "third rail" of credit discrimination - touch it and you die. Our country has a long history of segregated housing patterns in many areas, and racial "redlining" remains a real concern with regulators.
Some creditors have used geography as a shorthand to assess credit risk. This can work, but usually it does not work very well. Creditworthiness can be associated with socioeconomic factors, and these factors can be associated with geography and race. This means that creditors (and their credit underwriting model developers) can find an association between creditworthiness and geography, and this association usually has a racial effect. But researchers have found that this association is not especially strong and that looking at individual attributes, such as income, debts, and credit history, are better predictors of creditworthiness.
When I have examined a creditor's plan to consider geography in some aspect of underwriting, I have found that it had an adverse impact on racial and ethnic minorities. This was especially true if the geographic unit being considered was small, such as a ZIP code. Considering large areas, such as a metropolitan statistical area ("MSA") was much safer because an MSA includes both the urban core of an area, the outer suburbs, and all the neighborhoods and towns in between. MSAs tend to have a good racial mix that resembles the larger region.
That advice worked for this client because it was concerned about the economic health of the metropolitan region, not just a few ZIP codes. But the lawyers asked me, "Won't the regulators still blow a fuse if they see the creditor considering geography?" It was a good question.
Time for a quick reminder on how disparate impact liability works in discrimination law. First, the plaintiff or government identifies a "facially neutral" factor, like geography, that has a disproportionate adverse impact on a protected group. (Plain English translation: the factor hurts minorities more than non-Hispanic whites.) The creditor may still consider the factor if it can prove a "business justification." But a war over whether a particular practice is adequately "business justified" can be long and bloody. Finally, even when the practice is declared justified, the creditor still loses if it could have used other factors that would have met its business needs just as well and had a lesser adverse impact.
These are battles creditors are eager to avoid and understandably so. But this client had a real dilemma. Continuing to rack up huge losses in specific areas could quickly put it out of business. That's when I had an idea.
What if the old conventional wisdom about considering geography hurting minorities wasn't true in this case? Normally, a creditor lacks information on the racial impact of a factor, and using proxies can be a tedious and unsatisfying task. But this seemed simpler. Using Census Bureau data on the racial and ethnic composition of every county in the nation, we could compare the impact of excluding certain geographic areas from the creditor's footprint until the economy in each area improved.
I won't give away the results, but I will say the conventional wisdom did not hold up, and the client was delighted with our findings. It took only a few days of hard work to get the results, and the cost was modest. Now the client has a solid report to show its regulator, with findings the regulator can easily replicate if it chooses. No bloody fight over business justifications or less discriminatory alternatives should occur.
What's the takeaway here? I still caution creditors about the risks of using geography in credit evaluation systems. But sometimes conventional wisdom is wrong, and finding out the real impact of a policy might be quicker and less expensive than you thought. And in economic times when the usual creditworthiness metrics are not working, investigating whether a disparate impact even exists might just save the day.