Credit scores and mortgage approvals serve as fundamental gatekeepers to financial opportunity, heavily influencing an individual's borrowing costs and access to essential resources. Yet, despite decades of regulatory scrutiny and the recent rise of seemingly sophisticated algorithmic tools, new research indicates that systemic biases and fundamental inefficiencies persist throughout the financial ecosystem.

A series of research projects led by Gies College of Business professors Zilong Liu and Hongyan Liang (right) show that these flaws manifest in multiple ways, impacting individual borrowers who face miscalibrated credit scores, and systematically disadvantaging minority groups who encounter unequal treatment in the massive U.S. mortgage market. Furthermore, even where technology (Fintech) promises superior efficiency, structural regulations in the conforming mortgage market appear to neutralize these analytical advantages, leading to misaligned pricing that fails to accurately differentiate risk. Collectively, these findings underscore a critical and urgent need for continuous auditing, recalibration, and enhanced transparency to achieve genuinely equitable and efficient financial inclusion.

Algorithmic bias embedded in credit scores disadvantages women

In their paper, “Are credit scores gender-neutral? Evidence of mis-calibration from alternative and traditional borrowing data,” Liu and Liang study whether credit scoring systems inherently disadvantage women, particularly within the context of subprime borrowing. The findings, which were published in Journal of Behavioral and Experimental Finance, demonstrate clear evidence of a systemic gender bias embedded within current scoring models.

The study revealed that female borrowers consistently receive lower credit scores—by approximately 6 to 8 points—than men, even after controlling rigorously for an extensive set of credit risk variables, including payment history, amounts owed, and credit history length. This disparity is particularly striking because, upon closer examination, Liu and Liang found that women generally exhibit equal or lower default rates compared to men when matched for the same credit score level. This suggests a systematic miscalibration in the credit scoring models that unjustly penalizes creditworthy female borrowers.

Beyond the score itself, the predictive efficacy of these credit scores also significantly differs by gender, consistently favoring males. The analysis of the Area Under the Curve (AUC), a measure of predictive accuracy, showed that credit scores were slightly better at predicting default among male borrowers across most loan types (e.g., bill pay, installment loans, lines of credit). This differential predictive power further highlights an embedded algorithmic bias that disproportionately impacts women.

“Although a 6-to-8 point reduction might seem modest, this difference carries considerable economic significance,” said Liu, a clinical assistant professor of business administration. “It can lead to substantial repercussions, including higher interest rates, restricted credit limits, and decreased opportunities for loan approval, ultimately diminishing economic welfare over multiple borrowing cycles.”

“Our findings should encourage regulators and lenders to recalibrate their models, and it is critical that they continually audit scorecards to mitigate these biases and promote equitable financial inclusion,” said Liang, a senior lecturer of business administration.

Racial disparities persist across mortgage lender types

Another comprehensive study utilized Home Mortgage Disclosure Act (HMDA) data from 2018 to 2023 to analyze racial and ethnic disparities in approval rates, rate spreads, and origination charges across five distinct lender types: large banks, fintech lenders, non-bank lenders, small banks, and credit unions. The analysis found that persistent disparities exist, and that no single lender type uniformly eliminates disadvantages for minority borrowers. Their findings were released in “Racial Disparities in Conforming Mortgage Lending: A Comparative Study of Fintech and Traditional Lenders Under Regulatory Oversight,” which was published in the journal FinTech.

Large banks, which are subject to stringent regulatory oversight, emerged as a relatively equitable benchmark in terms of pricing, often offering equitable or even slightly more favorable rate spreads to minority borrowers once they are approved. However, these institutions impose the most stringent credit access criteria, resulting in significantly higher loan denial rates for Black and Hispanic applicants compared to White borrowers.

Fintech lenders showed greater inclusivity in terms of access, demonstrating narrower denial disparities for minority borrowers relative to large banks. However, this broader access often came at a cost, as fintech lenders typically charge higher overall rate spreads and fees.

“While Fintech lenders may reduce the minority denial gap, they still impose higher rate spreads on approved minority borrowers, suggesting the cost of borrowing remains a challenge,” said Liu (right).

Perhaps the most surprising finding concerned credit unions. While credit unions offered the lowest overall rate spreads to all borrowers, they paradoxically imposed the steepest pricing penalties on minority borrowers—particularly Black and Hispanic applicants—when considering the interaction effects after controlling for risk factors. This outcome challenges their reputation for consumer-friendly practices.

“In our research, it has become clear that neither technological innovation nor alternative lending models alone sufficiently eliminate systemic inequities,” said Liang. “Disparities unfortunately persist, underscoring the necessity for enhanced regulatory oversight and greater transparency in pricing and approval algorithms.”

Fintech’s risk assessment limited in highly regulated mortgage market

A third piece of research by Liu and Liang investigated the presumed advantage of fintech firms in risk assessment within the highly regulated environment of U.S. 30-year conforming mortgages (loans sold to Fannie Mae and Freddie Mac, the GSEs). While fintech lenders are celebrated for superior risk screening in unsecured consumer lending, this study questioned whether that advantage translates to the conforming mortgage market, which imposes stringent underwriting standards and rapid risk transfer.

In “Do Fintech Firms Excel in Risk Assessment for U.S. 30-Year Conforming Residential Mortgages? Liu and Liang examined how well lenders align their initial interest rates with actual loan performance (default or prepayment). The study, which was published in FinTech, concluded that fintech lenders do not demonstrate superior predictive accuracy compared to traditional banks and non-fintech non-banks. Fintech lenders' interest rates sorted borrowers less effectively, resulting in lower predictive accuracy for delinquency risks.

Furthermore, the research identified pricing misalignment within fintech portfolios. Fintechs charged higher rates to borrowers who remained current (the safest borrowers) and lower rates to those who eventually defaulted or prepaid early.

“Essentially, their pricing models have created a ‘flatter’ rate curve, meaning that the safest borrowers effectively paid a convenience premium, while higher-risk borrowers received an implicit discount relative to traditional bank pricing,” said Liu.

Liu and Liang attribute this underperformance and pricing inefficiency to structural frictions specific to the conforming market. The GSEs impose rigid underwriting standards that rely heavily on conventional metrics like FICO scores, Loan-to-Value (LTV), and Debt-to-Income (DTI), limiting fintechs’ ability to leverage their advanced analytics and alternative data. Additionally, because credit risk is quickly transferred through securitization, lenders prioritize loan origination volume and speed over granular risk differentiation. These constraints jointly weaken the incentive for fintechs to fine-tune risk-based pricing, demonstrating that technology's advantage is context-dependent and constrained by the regulatory environment.