Chapter 11: Further Reading

Bias in Financial Services and Credit

A Note on Sources

The field of algorithmic fairness in financial services moves quickly; readers should consult the most current versions of regulatory guidance and the most recent HMDA data releases. Sources below are annotated to explain their significance and to orient readers to where each fits in the broader literature. Sources marked [Data Source] provide publicly accessible data useful for independent analysis.

Investigative Journalism and Primary Reporting

1. Martinez, Emmanuel and Kirchner, Lauren. "Secret Algorithm: The Racist Data Driving U.S. Mortgage Algorithms." The Markup, August 25, 2021.

The investigation that reignited the algorithmic redlining conversation. Martinez and Kirchner analyzed nearly 9 million mortgage applications from 2019 HMDA data, finding that Black applicants were 80% more likely to be denied than comparable white applicants, and that Latino and Asian applicants also faced significant disparities. The methodology note is particularly valuable for readers interested in HMDA data analysis. The piece named specific lenders with above-average disparities, producing significant industry and regulatory response. Available at: https://themarkup.org/denied/2021/08/25/the-secret-algorithm-that-is-driving-up-mortgage-rejections

2. Hansson, David Heinemeier. "Apple Card is such a fucking sexist program." Twitter/X, November 7, 2019.

The original tweet thread that sparked the Apple Card controversy. While this is a social media post rather than journalism, it documents the primary source event discussed throughout Chapter 11 and is notable as a case study in how viral social media can force regulatory attention to algorithmic discrimination that might otherwise have gone unaddressed. Archived at: https://web.archive.org/web/2019*/https://twitter.com/dhh/status/1192540900393705474

3. Faber, Jacob W. "We Built This: Consequences of New Deal Era Intervention in America's Racial Geography." American Sociological Review, 2020.

Rigorous empirical documentation of the long-term consequences of HOLC redlining maps on contemporary neighborhood racial composition, property values, and wealth. Faber traces the causal path from 1930s redlining decisions to current neighborhood inequality, providing the historical foundation for understanding why neighborhood-level variables in credit models can serve as racial proxies. Essential background reading for Section 11.4.

Academic Research on Racial Disparities in Lending

4. Bhutta, Neil and Hizmo, Aurel. "Do Minorities Pay More for Mortgages?" The Review of Financial Studies, Vol. 34, No. 2, 2021.

One of the most carefully designed studies of racial disparities in mortgage pricing. Using HMDA data merged with credit score and other financial data from a major credit bureau, Bhutta and Hizmo find that Black and Hispanic borrowers pay higher mortgage rates even after controlling for credit score, LTV, and other financial characteristics. The study quantifies the unexplained racial pricing gap at approximately 5-7 basis points — smaller than raw disparities but economically meaningful. Essential reading for understanding what HMDA data does and does not reveal about discrimination. Available at: https://academic.oup.com/rfs/article/34/2/763/5834060

5. Fuster, Andreas, Goldsmith-Pinkham, Paul, Ramadorai, Tarun, and Walther, Ansgar. "Predictably Unequal? The Effects of Machine Learning on Credit Markets." Journal of Finance, Vol. 77, No. 1, 2022.

Analyzes the effect of switching from traditional linear credit models to machine learning models on racial disparities in mortgage denial rates. The key finding: ML models improve predictive accuracy overall but do not systematically reduce racial disparities — and in some specifications, increase them. The paper is methodologically sophisticated and includes a careful analysis of how ML models change the distribution of approval rates across demographic groups. Required reading for anyone making claims about whether AI will reduce or increase lending discrimination. Available at: https://doi.org/10.1111/jofi.13090

6. Quillian, Lincoln, Lee, John J., and Honoré, Brandon. "Black-White Lending Disparities: Using Public Records to Understand Racial Disparities in Mortgage Lending." Proceedings of the National Academy of Sciences, 2020.

A national-scale analysis of racial disparities in mortgage lending using HMDA data combined with credit-related variables. The study documents that Black applicants face significantly higher denial rates than white applicants at similar income levels, with disparities that have persisted over time and across geographic areas. Provides historical context comparing current disparities to those documented in earlier decades, showing that while explicit discrimination has declined, algorithmic systems continue to produce racially disparate outcomes.

7. Koulayev, Sergei, Marc Rysman, Scott Schuh, and Joanna Stavins. "Explaining Adoption and Use of Payment Instruments by U.S. Consumers." RAND Journal of Economics, 2016.

Foundational research on differential patterns of payment instrument use across demographic groups. Relevant background for understanding why transaction data — a common alternative credit data source — encodes demographic differences that can produce proxy discrimination in alternative credit scoring models.

Regulatory Guidance and Enforcement Documents

8. Consumer Financial Protection Bureau. "CFPB Circular 2022-03: Adverse Action Notification Requirements and the Equal Credit Opportunity Act." August 10, 2022.

The CFPB's official guidance on how ECOA's adverse action notice requirement applies to AI and complex algorithmic credit models. Directly addresses the question of whether lenders can use standardized reason code lists when the codes may not accurately reflect the model's actual decision. The CFPB's answer is no — reason codes must accurately reflect the model's decision. Essential regulatory reading for compliance officers. Available at: https://www.consumerfinance.gov/compliance/circulars/circular-2022-03-adverse-action-notification-requirements-and-the-equal-credit-opportunity-act/

9. Federal Reserve. "SR 11-7: Guidance on Model Risk Management." April 4, 2011.

The baseline federal regulatory framework for model risk management at banks. While predating the era of modern ML, SR 11-7 establishes the principles — model inventory, independent validation, conceptual soundness, outcome analysis, governance — that form the foundation of fair lending model validation. The OCC's corresponding guidance is OCC Bulletin 2011-12. Together these documents are the primary regulatory reference for model risk management in U.S. banking. Available at: https://www.federalreserve.gov/supervisionreg/srletters/sr1107.htm

10. Consumer Financial Protection Bureau. "CFPB No-Action Letter Policy and Application." Updated guidance, 2019.

The CFPB's policy framework for issuing No-Action Letters, which provide regulatory comfort to financial institutions experimenting with innovative products that raise novel legal questions. The Upstart No-Action Letter (discussed in Section 11.3) is the most prominent example of the program applied to AI-driven credit. The application process and the conditions of the Upstart NAL are informative about how the CFPB approaches novel algorithmic credit models. Available at: https://www.consumerfinance.gov/compliance/no-action-letters/

11. New York Department of Financial Services. "Report on Investigation: Apple Inc. and Goldman Sachs Bank USA Credit Card Application Practices." March 2021.

The official regulatory report on the Apple Card gender discrimination investigation. Provides the DFS's methodology, findings, and recommendations, including the finding of no illegal discrimination and the identification of governance deficiencies. Reading the primary regulatory document alongside the secondary coverage is valuable for understanding how regulatory investigations of algorithmic discrimination actually work. Available at: https://www.dfs.ny.gov/reports_and_publications/press_releases/pr202103251

12. Consumer Financial Protection Bureau. "Data Point: Credit Invisibles." Patrice Alexander Ficklin and Paul Calem, May 2015.

The definitive documentation of the credit invisible population in the United States — 26 million Americans with no credit file and 19 million more with unscorable files. The report provides demographic breakdowns showing that credit invisibility disproportionately affects Black and Hispanic Americans and low-income communities. The foundational document for understanding the thin-file problem discussed in Section 11.3. Available at: https://www.consumerfinance.gov/data-research/research-reports/data-point-credit-invisibles/

Data Sources for Independent Analysis

13. [Data Source] Consumer Financial Protection Bureau. "Home Mortgage Disclosure Act (HMDA) Data." Annual releases.

The primary public data source for analyzing racial disparities in mortgage lending. Includes application-level data on race, ethnicity, sex, income, loan amount, LTV, DTI, interest rate, AUS recommendation, and loan outcome for nearly all mortgage applications in the United States. Updated annually. Available via the FFIEC Data Browser and CFPB's API. Available at: https://ffiec.cfpb.gov/ and https://www.consumerfinance.gov/data-research/hmda/

14. [Data Source] Federal Financial Institutions Examination Council. "HMDA Data Browser." Updated annually.

The interactive tool for accessing and filtering HMDA data without downloading raw files. Allows users to filter by lender, geography, loan type, and demographic group, and to generate summary statistics. The most accessible entry point for students and practitioners who want to explore HMDA data without advanced data processing skills. Available at: https://ffiec.cfpb.gov/data-browser/

15. [Data Source] National Community Reinvestment Coalition. "CRA Commitments Database."

Tracks Community Reinvestment Act commitments made by financial institutions during the merger approval process. Useful for understanding how CRA is applied in practice and for identifying the geographic distribution of CRA lending activity. Relevant to the discussion of digital lending and CRA obligations in Section 11.9. Available at: https://www.ncrc.org/

Books and Long-Form Resources

16. Baradaran, Mehrsa. "The Color of Money: Black Banks and the Racial Wealth Gap." Harvard University Press, 2017.

The definitive historical account of the relationship between Black financial institutions, the racial wealth gap, and the history of discriminatory financial policy. Essential background for understanding why the thin-file problem exists and why credit score disparities by race persist. Baradaran traces the accumulation of financial exclusion from Reconstruction through the present, providing the historical depth that algorithmic discussions often lack.

17. Eubanks, Virginia. "Automating Inequality: How High-Tech Tools Profile, Police, and Punish the Poor." St. Martin's Press, 2018.

Examines how algorithmic decision-making systems — including credit systems, public benefits algorithms, and predictive policing — disproportionately affect low-income communities and communities of color. While not focused exclusively on financial services, Chapter 3 on debt collection and financial systems is directly relevant. Eubanks brings a human-centered perspective that complements the regulatory and technical analysis in this chapter.

18. O'Neil, Cathy. "Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy." Crown, 2016.

The accessible introduction to algorithmic bias that brought the issue to a popular audience. Chapter 8, "Collateral Damage," addresses credit scoring specifically. While some of the technical details have been superseded by subsequent research, O'Neil's framework for identifying when algorithms are "weapons of math destruction" — opaque, unaccountable, and harmful at scale — remains a useful analytical lens.

19. Rugh, Jacob S. and Massey, Douglas S. "Racial Segregation and the American Foreclosure Crisis." American Sociological Review, Vol. 75, No. 5, 2010.

Documents the causal relationship between residential racial segregation — itself a product of historical redlining and discriminatory real estate practices — and the concentrated foreclosure crisis in minority neighborhoods during the 2008 housing crisis. Essential context for understanding why predatory lending and its algorithmic equivalents have such devastating consequences for communities that have already borne the costs of prior discrimination.

20. Gillis, Talia B. and Spiess, Jann L. "Big Data and Discrimination." University of Chicago Law Review, Vol. 86, 2019.

A careful legal and economic analysis of how big data and algorithmic models produce discrimination, focusing on the distinction between statistical discrimination and disparate impact discrimination. The paper's treatment of the interaction between predictive accuracy and discriminatory outcomes is particularly valuable — the authors show that maximizing predictive accuracy on historically biased data will tend to perpetuate discrimination even when protected characteristics are excluded. Accessible to readers without a legal background. Available at: https://lawreview.uchicago.edu/publication/big-data-and-discrimination

This reading list reflects sources available as of early 2026. Readers should check for updated versions of regulatory guidance, which is revised periodically, and for subsequent HMDA data releases, which are published annually by the CFPB.