Although non-traditional data could broaden the borrower pool, it could also serve to perpetuate existing inequality.
Lenders would like to draw deeper into the pool of borrower prospects in a bid to expand their business.
That’s causing the use of alternative data to become more of an input in credit scoring and lending. Even the major credit bureaus have begun looking to alternative data inputs.
For instance, Experian’s Clarity Services uses alternative financial data, and Experian Boost looks to utility and phone bill payments, to help some consumers hike up their credit scores. Also, TransUnion acquired FactorTrust, a provider of alternative data, to provide credit input on a deeper consumer pool.
Not to be left behind, Equifax bought alternative data provider DataX, and entered into a partnership with Urjanet to access consumers’ utility, cable and telecom payments.
The Consumer Financial Protection Bureau estimates that as of 2010, 26 million U.S consumers were “credit invisible,” or lacking any credit records, while another 19 million had scanty credit records that the bureaus could not use to generate a credit score.
According to the House Committee on Financial Services, the two largest groups in this credit-challenged population include Hispanic and African-American millennials who live in low-income neighborhoods.
While the use of alternative data could help extend credit to such people, this is still an emerging practice, and the use of alternative data comes with certain risks.
Alternative data present a broader picture
Alternative data refers to input that credit bureaus have typically not looked to in developing a consumer’s credit score. This could include utility, rent and cellphone bill payments, banking account activity, social media activity, education and online behavior.
Dave Shellenberger, vice president of scores and predictive analytics, FICO, observed, in emailed comments, “Alternative data can help provide additional insight into positive financial management behavior that may be missing from traditional credit information. We see this with the UltraFICO score in our development.”
In fact, input from additional checking, savings or money market account information, which UltraFICO uses, helps boost the FICO scores of more than 40 percent of those with thin credit files, who have responsibly managed their bank accounts, by 40 points or more, according to Shellenberger.
This sort of data has also boosted those who have experienced financial distress in the past, but are on the road to recovering their financial health, by getting a new job, for instance.
Shellenberger reports that the closer the alternative data is related to how consumers pay their bills or manage their finances, the better a predictor it is of which consumers will meet their financial obligations.
Emre Sahingur, VantageScore’s senior vice president of predictive analytics research and product management, also sees value in the use of additional data.
The VantageScore model uses data input from the national credit bureaus’ consumer files, including rent, utility and cellphone payment information. Sahingur doesn’t view this as alternative data, since it comes from the credit bureaus.
“Any data that can provide a more complete view of the consumer’s financial situation that is visible to the traditional files, that’s going to be useful,” Sahingur said. “And that applies to people who are not scorable, as well as people who already have scores.”
VantageScore doesn’t use other forms of alternative data, such as a person’s social media activity, that doesn’t hold up to scrutiny and could violate laws. FICO too has looked into the predictive ability of social media activity outside the U.S., and concluded that it “sees limited value,” Shellenberger said.
Drawing more people in
Some online lenders use alternative data in addition to traditional data in their lending decisions. Some use learning algorithms to analyze data in real time, and typically can approve consumers for a loan on their digital devices, without the need to meet loan officers in person.
In Congressional testimony at a July 2019 hearing that looked into the use of alternative data in credit scoring and lending decisions, Tulane University law professor Kristin Johnson noted that the absence of human loan officers in this process cuts down on targeted discrimination.
And the use of artificial intelligence in credit decisions could help cut down on defaults by better classifying consumers and pricing credit.
According to Johnson, “These process-oriented improvements enhance efficiency and accuracy, improve pricing, reduce operating and loan origination costs and enable fintech firms to offer credit to a greater diversity of consumers, in particular those who have struggled to obtain credit.”
And Lawrance Evans, Jr., managing director of financial markets and community investment for the U.S. Government Accountability Office, noted at the hearing that fintech lenders’ use of alternative data to confirm the identity of borrowers could also help them stay clear of fraud.
Also testifying at this hearing, Dave Girouard, CEO and co-founder of Upstart Network, said that the fintech lender’s computerized model (which uses machine learning algorithms, with inputs such as an applicant’s employment history and educational background, in addition to credit scores) approves 27 percent more consumers, while reducing interest rates by 3.57 percentage points, compared to traditional underwriting models.
For those with a “near prime” FICO score in the 620 to 660 range, Upstart even greenlights 95 percent more applicants, while providing interest rates that are 5.42 percentage points lower. Indeed, the model “provides higher approval rates and lower interest rates for every traditionally underserved demographic,” according to Upstart.
See related: Busted : 5 myths about alternative credit data
There are drawbacks too
The use of alternative data also poses certain risks. One concern is that the use of utility and electric payments could undermine certain government protections.
For instance, some states protect vulnerable consumers from shutdown of utilities for not paying their bills during peak usage months, and also when they face hardships. And use of rent payment data could penalize renters who withhold payments while their landlords address certain poor housing conditions.
Some of the alternative data also don’t seem to further the objective of expanding the creditworthy population.
For one, Johnson’s testimony for the Congressional hearing states, “It is also unclear how credit invisibles and unscorables who do not have conventional checking and savings accounts or credit cards will generate financial transaction data. Similarly, ranking consumers based on higher educational or professional accomplishments seems likely to replicate the current credit scoring patterns.”
And these people are also unlikely to reap the benefits of networking through social media, she added, that would be considered in giving them access to credit on better terms.
The GAO’s Evans also pointed out in his testimony that it may not be clear which alternative data lenders are using, and how they use them to make credit decisions. And the borrower may not have the ability to contest this input. Besides, it may be difficult to vouch for the accuracy of the alternative data input. There are also issues related to the privacy and security of data that is generated from online sources.
And, in his Congressional testimony, Aaron Rieke, managing director at Upturn, a nonprofit organization that advocates for a fair approach in the use of digital technology, noted, “Expansive data sets about people’s social connections, where they live, how they behave, where they shop and how they communicate are fraught with fair lending concerns.”
See related: Tech lobbying efforts likely to shape federal data privacy legislation outcomes
Inputs may run afoul of legal protections
Although there are legal protections that govern availability of credit, the sources of alternative data may be skirting some of these protections.
For instance, the Equal Credit Opportunity Act makes it illegal to discriminate against borrowers based on certain aspects such as their race, gender or religion.
Johnson noted, “Even if developers expressly program algorithms not to discriminate on the basis of a protected trait, the developers’ biases may creep in and influence the algorithm’s operation.”
Thus, if the input is garbage data, the output will be garbage too.
In her testimony at the Congressional hearing, Chi Chi Wu, staff attorney at the National Consumer Law Center, also said behavioral data input (which includes web browsing information) “has also shown indications of racial bias, despite relying on seemingly racially neutral algorithms.”
Thus, she would like all data input used in making credit decisions to be governed by the ECOA, as well as the Fair Credit Reporting Act.
The latter law addresses consumer reports, requiring them to use accurate inputs and giving consumers the right to speak out about any inaccurate input in the report.
Wu would like the laws to expressly state that the FCRA applies to “any third-party data used for credit evaluation purposes.” She observed, “Having ‘black boxes’ that evaluate creditworthiness should be a thing of the past, as a matter of both fairness and ensuring that consumers are fully educated about financial issues.”
See related: Sally Taylor-Shoff: Trended data a no-go for FICO
Making credit access more inclusive
There is legislation pending that would address the concerns that alternative credit data use fosters.
For one, the Credit Access and Inclusion Act would amend the Fair Credit Reporting Act and prohibit a utility company from reporting a customers’ delinquent balances if they have entered into a payment plan with the company and are meeting the requirements of the plan.
And the Clarity in Credit Score Formation Act seeks to amend the Fair Credit Reporting Act by requiring those who provide or use credit models that are used in making credit decisions to create standards that would prove that the models are providing accurate and valid outcomes. They would have to do this both before releasing these models, and at regular intervals after that.
It also requires the Consumer Financial Protection Bureau to review, at least once every two years, whether these models are using any inappropriate inputs. The CFPB could then prohibit the use of certain inputs if necessary.
As well, this law would mandate the consumer protection agency to study the impact of the use of “non-traditional” data on various groups, such as those without a credit history, or a thin credit file.
And the CFPB would be required to look into the impact on various other groups, such as immigrants, women, minorities and rural consumers, taking into account privacy security, discrimination and disparate impact aspects, among other effects.
The CFPB would then have to provide a report to Congress with any recommendations for change.
See related: Artificial intelligence shines new light on ‘credit invisibles’
Alternative data use is catching on
Irrespective of the controversies, it seems alternative data use is catching on, and putting in a legal framework would enhance its use.
Looking ahead, while alternative data use could serve to narrow the inequality in America, it could also serve to widen it.
As NCLC’s Wu concluded in her Congressional testimony, “We know that credit reports and scores can reinforce existing inequality. The question is whether we treat new sources of data, alternative data, in the same way or whether we develop algorithms and policies that allow the American dream to flourish once again.”