Lenders historically underserve disadvantaged groups like minorities and low-income individuals. I’m not claiming it’s intentional. I’m not arguing that people are actively trying to discriminate against minorities. Recent analysis1 shows it’s an accidental side effect from the way our credit markets operate. And the good news is, there’s a practical fix. Let’s take a close look, starting with the basics.

Economic growth is fueled by business growth. Business growth is fueled by lending. Lending is metered through credit scores. Credit scores reflect past economic activity of the borrower. This cycle is self-reinforcing. (Figure 1) When the sequence is efficient and effective, we enjoy a virtuous cycle of economic growth. The inverse is also true. Ineffective credit allocation yesterday leads to ineffective credit allocation today and limited economic growth tomorrow. Either way, credit scores are used as a signal of applicant quality in many high-stakes screening decisions, from hiring to insurance underwriting to rental housing to lending.

Figure 1- equity & economics

Lenders face more uncertainty when assessing the creditworthiness of minority and low-income individuals. Specifically, under-served groups have so-called thin credit files, characterized by sparser credit report data. This information disparity is a statistically significant driver of inequality in credit market outcomes. For example, the standard deviation of credit score noise of minority applicants is 2.2 times higher than that of non-minority applicants. This cannot be remedied by changing how we train credit scoring models or improving the predictive models. Mathematicians and computer scientists have long recognized GIGO ― garbage in, garbage out. The fix is to address the underlying data quality issue that drives the problem. Equalizing credit score signal noise can shrink disparities in approval rates by up to 50%.

The negative consequences of the current situation are still substantial. In 2019, the homeownership rate among white non-Hispanic Americans was 73.3%, compared to 42.1% among Black Americans. This 31.2 percentage point difference was the largest gap since the Census’ time series began in 19942. The mortgage market plays a prominent role in the persistence of wealth gaps across generations3. Historically disadvantaged groups are less likely to transition to homeownership and build home equity4. This is a vicious cycle. Disparities in credit access at one point in time then translate into disparities in credit report information in the future. Meanwhile, modern FinTech underwriting continues to play a role in perpetuating cross-group disparities in loan terms5 without any concern for the consequences of their operations. FinTech is now attracting $1 of every $5 of venture capital investment. We’ve got to get FinTech on a path to efficient & equitable lending now6.

This situation is unacceptable from a humanitarian perspective. It’s also unacceptable from a data science perspective. The limited attention paid to these data quality issues, outlined above, is nothing better than professional malpractice by the data professionals involved. Blattner and Nelson’s ground-breaking analysis is a wake-up call to the Financial Services sector to adopt the practical disciplines and methods of ethical digital.

Let’s look at a practical way forward. Remember that the root cause of the problem is that sparser credit report data prevents lenders from accurately calculating the probability of default for minority and low-income individuals. The solution is to begin to absorb new sources of data: social media.

As one example, Yelp provides a rich set of data on minority-owned businesses7. The data includes both a) trailing indicators of performance like ratings8, reviews, and search activity9; and more importantly to lenders b) leading indicators of future performance like rate of community engagement through review responses and resilience (as measured by the rate of reopening after COVID-19 closures). Do you think this is a far-fetched idea? Think again. For several years Federal government agencies have leveraged social media data when assessing individuals applying for security clearance, visas, and citizenship. It’s time for the banking sector to catch up with best practices in the Federal government.

ProTips

1. If you see something, say something
2. Act with honesty and integrity
3. Focus on solutions
4. Treat others how you want to be treated
5. Make it right

On a positive note, on 25 January 2021, President Biden issued Executive Order EO 13985: Advancing Racial Equity and Support for Underserved Communities Through the Federal Government. It calls for:

“a comprehensive approach to advancing equity for all, including people of color and others who have been historically underserved, marginalized, and adversely affected by persistent poverty and inequality…Because advancing equity requires a systematic approach to embedding fairness in decision-making processes, executive departments and agencies (agencies) must recognize and work to redress inequities in their policies and programs that serve as barriers to equal opportunity…By advancing equity across the Federal Government, we can create opportunities for the improvement of communities that have been historically underserved, which benefits everyone. For example, an analysis shows that closing racial gaps in wages, housing credit, lending opportunities, and access to higher education would amount to an additional $5 trillion in gross domestic product in the American economy over the next 5 years.”

That’s good news for all of us. We can’t undo the past. We most definitely can create a better future.

Parting shot: Let’s look at lending efficiency and equity through a National Security lens. We are in a vicious whole-of-country competition with China. Military strength is grounded in a foundation of economic strength. At DL, we want every American to have the opportunity to contribute to economic growth and benefit from it.

Michael Conlin

Michael Conlin​

Chief Technology Officer​
Phone: (703) 216-5856​
michael.conlin@definitivelogic.com​

1 “How costly is noise? Data and disparities in Consumer Credit”, Laura Blattner and Scott Nelson, May 5, 2021 2usafacts.org/articles/homeownership-rates-by-race/ 3 Charles and Hurst, 2003; Kuhn et al., 2020 4 Charles and Hurst, 2002 5Definitive Logic Credo Principles 6$1 of every $5 in venture last quarter went to fintech 7yelpeconomicaverage.com/diverse-business-report.html 8Yelp: “Consumers also rate many businesses owned by women, Black people, and Latinx people higher than other businesses, but the ratings edge existed prior to when the businesses indicated their identity attributes – demonstrating that the higher rating more likely reflects a great customer service experience than any correlation to the identity attribute(s). 9Based on Yelp’s identity attributes, not only have consumers increasingly sought out Black, Latinx, and women-owned businesses they’ve also engaged with many of these businesses more compared to other businesses.