Fintech: Racial Discrimination in Virtual Mortgage Applications?
Sept 12, 2023
Sept 12, 2023
Article: The Racial Landscape of Fintech Mortgage Lending
Author: Tyler Haupert
DOI: https://doi.org/10.1080/10511482.2020.1825010
Published Online: 09 Nov 2020
In the last decade, traditional mortgage lending has entered into competition with a new digitized alternative. The growing industry of financial technology (fintech) lending distinguishes itself through completely virtual mortgage applications and the use of autonomous AI algorithms that sift through massive amounts of applicant data to predict default risk (Bartlett, Morse, Stanton, & Wallace, 2019; Buchak, Matvos, Piskorski, & Seru, 2018). Whereas traditional mortgage lending institutions require an applicant to meet in person with a loan officer as well as submit documents verifying their income, FICO credit score, and employment history to be considered for loan approval, fintech lenders often use alternative (and often purposefully undisclosed) sources of data such as, but not limited to, utility payments, social media, and credit card payments to underwrite loans.
The impacts of this new technology are not well understood, as the empirical work studying fintech mortgage lending is still in its early days. Some researchers and industry experts argue that the alternative sources of data collected by fintech lenders will better allow those who are at low risk of default but have poor credit scores to access credit (Kreiswirth, Schoenrock, & Singh, 2017); others argue that these data contain racial proxies that allow for covert discrimination (in contrast to traditional mortgage lending, where research suggests that an important locus of discrimination is at the in-person meeting with the loan officer) (Odinet, 2019; Student Borrower Protection Center, 2020). Demand for transparency around data collection and algorithm-determined consumer lending outcomes increase as the fintech industry grows (Hamilton, 2019; Hussain, 2019; O’Neil, 2017). Tyler Haupert’s article, “The Racial Landscape of Fintech Mortgage Lending,” brings new depth and clarity to this complicated debate by providing empirical evidence about how fintech affects the lending outcomes of Black, Latino, and Asian prospective homeowners relative to white applicants. Specifically, the article addresses two questions:
Haupert explores these questions through a series of binomial logistic regressions comparing the relationship of fintech vs. traditional lending on borrower race and neighborhood racial composition as well as loan approval and subprime terms conditional on approval. Haupert, following in the footsteps of researcher Buchak and colleagues (2018), categorizes the ambiguous category of “fintech” as businesses that advertise all-online applications, near-instant approvals, and the use of nontraditional data in underwriting. These analyses were conducted with data from a wide and inclusive pool of sources, including HMDA loan application register data, original data categorizing lenders as fintech or traditional, and data from the decennial census, ACS, the Federal Financial Institutions Examination Council (FFIEC) and the FHFA and spans from 2015-2017—a period in which the housing market had recovered from the 2008 crisis and during which fintech was rapidly growing.
Haupert hypothesizes that fintech lenders would approve more loans to racial minorities than traditional lenders, but would also offer more subprime loans to these same groups. His results support this hypothesis. Descriptive statistics show that 7.3% of loans approved by traditional lenders and 7.4% approved by fintech lenders were subprime despite both lender types serving neighborhoods of roughly similar aggregate racial composition and features. Additionally, fintech lenders have a much higher rate of applicants who don’t report their race: 20%, compared to 9.2% of applicants that don’t report their race to traditional lenders. This is a cause for concern as such information is critical to enforcing fair housing policies.
Haupert finds that while Black and Asian applicants were still marginally less likely to receive a loan from a fintech lender than a similarly qualified White applicant, both groups were more likely to have their applications approved by a fintech lender than a traditional one. In fact, Latino applicants were more likely than similarly qualified White applicants to be approved for a loan by a fintech lender, whereas they fared slightly worse when applying to traditional lenders. Neighborhood racial composition also has a different relationship with loan approval among fintech lenders than among traditional lenders. Peak fintech loan approval rates occur for applicants in neighborhoods with 40-60% residents of color, whereas traditional lenders distributed credit across neighborhoods with very little variation by racial composition.
Turning to subprime loans, Haupert shows that while traditional lenders meted out more such loans in total, the disparity in how these loans were distributed between racial minorities versus similarly qualified White applicants was greater among fintech lenders. Black applicants were more likely, Latino applicants roughly equally likely, and Asian applicants slightly less likely to receive a subprime loan from a fintech rather than a traditional lender. Disparities also appeared with respect to neighborhood racial composition: fintech lenders offered subprime loans at even higher rates than traditional lenders in neighborhoods with higher percentages of non-White residents.
How are these algorithmic decisions made? Haupert’s results indicate that applicants with poor credit are more likely to receive a loan from a fintech lender, and those with higher income are less likely to receive one, indicating that both credit and income are less important factors than in traditional lending. Alternative sources of data used in the underwriting process are likely given priority—just as fintech companies advertise. But what exactly those alternative sources of data are remains unclear due to the lack of industry regulation enforcing data transparency.
Thus, fintech appears to contribute to racial disparities in access to homeownership at a time when Black homeownership rates are already troublingly low. In the lingering shadow of the 2008 subprime housing crisis, Haupert urges for increased transparency around fintech’s data collection processes, and for the industry—which has capitalized on the same lack of scrutiny that many Big Tech companies have—to be held to stricter standards.
About the Author:
Tyler Haupert is an assistant professor of urban studies at NYU Shanghai. His research focuses on the social, economic, technological, and regulatory mechanisms contributing to racial segregation and exclusion in advanced economies, with particular interests in mortgage lending, housing policy, neighborhood change, and homelessness. He strives to design studies that inform policy and produce actionable results for legislators, regulators, planners, and advocacy organizations.
Citations:
Bartlett, R. P., Morse, A., Stanton, R. H., & Wallace, N. E. (2019). Consumer lending discrimination in the fintech era NBER Working Paper (No. 25943). Retrieved from https://www.nber.org/papers/w25943
Buchak, G., Matvos, G., Piskorski, T., & Seru, A. (2018). Fintech, regulatory arbitrage, and the rise of shadow banks. Journal
of Financial Economics, 130(3), 453–483.
Hamilton, E. (2019). AI perpetuating human bias in the lending space. Retrieved from https://www.techtimes.com/
articles/240769/20190402/ai-perpetuating-human-bias-in-the-lending-space.htm
Hussain, S. (2019). Tell HUD: Algorithms shouldn’t be an excuse to discriminate. Retrieved from https://www.eff.org/
deeplinks/2019/10/tell-hud-algorithms-are-no-excuse-discrimination
Kreiswirth, B., Schoenrock, P., & Singh, P. (2017). Using alternative data to evaluate creditworthiness. Retrieved from
https://www.consumerfinance.gov/about-us/blog/using-alternative-data-evaluate-creditworthiness/
O’Neil, C. (2017). Weapons of math destruction - how big data increases inequality and threatens democracy. New York:
Broadway Books.
Odinet, C. K. (2019). The new data of student debt. Southern California Law Review, 92, 1617–1692. Retrieved from
https://southerncalifornialawreview.com/wp-content/uploads/2019/12/92_6_Odinet.pdf
Student Borrower Protection Center. (2020). Educational redlining. Retrieved from https://assets.documentcloud.org/
documents/6768401/Education-Redlining-February-2020.pdf
Author: Tyler Haupert
DOI: https://doi.org/10.1080/10511482.2020.1825010
Published Online: 09 Nov 2020
In the last decade, traditional mortgage lending has entered into competition with a new digitized alternative. The growing industry of financial technology (fintech) lending distinguishes itself through completely virtual mortgage applications and the use of autonomous AI algorithms that sift through massive amounts of applicant data to predict default risk (Bartlett, Morse, Stanton, & Wallace, 2019; Buchak, Matvos, Piskorski, & Seru, 2018). Whereas traditional mortgage lending institutions require an applicant to meet in person with a loan officer as well as submit documents verifying their income, FICO credit score, and employment history to be considered for loan approval, fintech lenders often use alternative (and often purposefully undisclosed) sources of data such as, but not limited to, utility payments, social media, and credit card payments to underwrite loans.
The impacts of this new technology are not well understood, as the empirical work studying fintech mortgage lending is still in its early days. Some researchers and industry experts argue that the alternative sources of data collected by fintech lenders will better allow those who are at low risk of default but have poor credit scores to access credit (Kreiswirth, Schoenrock, & Singh, 2017); others argue that these data contain racial proxies that allow for covert discrimination (in contrast to traditional mortgage lending, where research suggests that an important locus of discrimination is at the in-person meeting with the loan officer) (Odinet, 2019; Student Borrower Protection Center, 2020). Demand for transparency around data collection and algorithm-determined consumer lending outcomes increase as the fintech industry grows (Hamilton, 2019; Hussain, 2019; O’Neil, 2017). Tyler Haupert’s article, “The Racial Landscape of Fintech Mortgage Lending,” brings new depth and clarity to this complicated debate by providing empirical evidence about how fintech affects the lending outcomes of Black, Latino, and Asian prospective homeowners relative to white applicants. Specifically, the article addresses two questions:
- Do different levels of racial disparities in loan approvals and terms (i.e. interest rates) exist among loans made by fintech lenders compared with traditional lenders?
- Do these disparities vary across neighborhoods of varying racial composition?
Haupert explores these questions through a series of binomial logistic regressions comparing the relationship of fintech vs. traditional lending on borrower race and neighborhood racial composition as well as loan approval and subprime terms conditional on approval. Haupert, following in the footsteps of researcher Buchak and colleagues (2018), categorizes the ambiguous category of “fintech” as businesses that advertise all-online applications, near-instant approvals, and the use of nontraditional data in underwriting. These analyses were conducted with data from a wide and inclusive pool of sources, including HMDA loan application register data, original data categorizing lenders as fintech or traditional, and data from the decennial census, ACS, the Federal Financial Institutions Examination Council (FFIEC) and the FHFA and spans from 2015-2017—a period in which the housing market had recovered from the 2008 crisis and during which fintech was rapidly growing.
Haupert hypothesizes that fintech lenders would approve more loans to racial minorities than traditional lenders, but would also offer more subprime loans to these same groups. His results support this hypothesis. Descriptive statistics show that 7.3% of loans approved by traditional lenders and 7.4% approved by fintech lenders were subprime despite both lender types serving neighborhoods of roughly similar aggregate racial composition and features. Additionally, fintech lenders have a much higher rate of applicants who don’t report their race: 20%, compared to 9.2% of applicants that don’t report their race to traditional lenders. This is a cause for concern as such information is critical to enforcing fair housing policies.
Haupert finds that while Black and Asian applicants were still marginally less likely to receive a loan from a fintech lender than a similarly qualified White applicant, both groups were more likely to have their applications approved by a fintech lender than a traditional one. In fact, Latino applicants were more likely than similarly qualified White applicants to be approved for a loan by a fintech lender, whereas they fared slightly worse when applying to traditional lenders. Neighborhood racial composition also has a different relationship with loan approval among fintech lenders than among traditional lenders. Peak fintech loan approval rates occur for applicants in neighborhoods with 40-60% residents of color, whereas traditional lenders distributed credit across neighborhoods with very little variation by racial composition.
Turning to subprime loans, Haupert shows that while traditional lenders meted out more such loans in total, the disparity in how these loans were distributed between racial minorities versus similarly qualified White applicants was greater among fintech lenders. Black applicants were more likely, Latino applicants roughly equally likely, and Asian applicants slightly less likely to receive a subprime loan from a fintech rather than a traditional lender. Disparities also appeared with respect to neighborhood racial composition: fintech lenders offered subprime loans at even higher rates than traditional lenders in neighborhoods with higher percentages of non-White residents.
How are these algorithmic decisions made? Haupert’s results indicate that applicants with poor credit are more likely to receive a loan from a fintech lender, and those with higher income are less likely to receive one, indicating that both credit and income are less important factors than in traditional lending. Alternative sources of data used in the underwriting process are likely given priority—just as fintech companies advertise. But what exactly those alternative sources of data are remains unclear due to the lack of industry regulation enforcing data transparency.
Thus, fintech appears to contribute to racial disparities in access to homeownership at a time when Black homeownership rates are already troublingly low. In the lingering shadow of the 2008 subprime housing crisis, Haupert urges for increased transparency around fintech’s data collection processes, and for the industry—which has capitalized on the same lack of scrutiny that many Big Tech companies have—to be held to stricter standards.
About the Author:
Tyler Haupert is an assistant professor of urban studies at NYU Shanghai. His research focuses on the social, economic, technological, and regulatory mechanisms contributing to racial segregation and exclusion in advanced economies, with particular interests in mortgage lending, housing policy, neighborhood change, and homelessness. He strives to design studies that inform policy and produce actionable results for legislators, regulators, planners, and advocacy organizations.
Citations:
Bartlett, R. P., Morse, A., Stanton, R. H., & Wallace, N. E. (2019). Consumer lending discrimination in the fintech era NBER Working Paper (No. 25943). Retrieved from https://www.nber.org/papers/w25943
Buchak, G., Matvos, G., Piskorski, T., & Seru, A. (2018). Fintech, regulatory arbitrage, and the rise of shadow banks. Journal
of Financial Economics, 130(3), 453–483.
Hamilton, E. (2019). AI perpetuating human bias in the lending space. Retrieved from https://www.techtimes.com/
articles/240769/20190402/ai-perpetuating-human-bias-in-the-lending-space.htm
Hussain, S. (2019). Tell HUD: Algorithms shouldn’t be an excuse to discriminate. Retrieved from https://www.eff.org/
deeplinks/2019/10/tell-hud-algorithms-are-no-excuse-discrimination
Kreiswirth, B., Schoenrock, P., & Singh, P. (2017). Using alternative data to evaluate creditworthiness. Retrieved from
https://www.consumerfinance.gov/about-us/blog/using-alternative-data-evaluate-creditworthiness/
O’Neil, C. (2017). Weapons of math destruction - how big data increases inequality and threatens democracy. New York:
Broadway Books.
Odinet, C. K. (2019). The new data of student debt. Southern California Law Review, 92, 1617–1692. Retrieved from
https://southerncalifornialawreview.com/wp-content/uploads/2019/12/92_6_Odinet.pdf
Student Borrower Protection Center. (2020). Educational redlining. Retrieved from https://assets.documentcloud.org/
documents/6768401/Education-Redlining-February-2020.pdf