What Have Researchers Found About Race-based Discrimination in Algorithmic Credit Decisions?
Bartlett et al. (2022) found that algorithmic lenders do not discriminate in their decision to approve or reject minority loan applications, which is in strong contrast to face-to-face lenders who reject African-American and Latinx applications 6% more often than they reject non-minority applicants who are directly comparable in all material respects. Regarding loan pricing, Bartlett et al. (2022) find that although algorithms do not eliminate discrimination, they discriminate one-third less than face-to-face lenders. Bartlett et al. (2019)’s findings support the idea that algorithmic credit scoring reduces the impact of human discretion and mitigates disparate treatment in credit decisions.
Fuster et al. (2022) found that machine learning algorithms reduce disparity in mortgage acceptance rates across race and ethnic groups. Additionally, algorithmic credit scoring slightly increases the proportion of borrowers accepted for a mortgage for all race and ethnic groups, with the most significant increase observed for African-American borrowers. However, regarding credit pricing, Fuster et al. (2022) discovered that machine learning algorithms lead to greater interest rate disparity across race and ethnic groups. Furthermore, these algorithms result in higher and more dispersed interest rates for African-American and Hispanic borrowers, when benchmarked against traditional statistical techniques.
It seems that researchers’ findings on credit acceptance rates are quite consistent with each other. However, the findings on credit pricing are mixed and far from being conclusive. That said, algorithmic credit scoring has yet to be adopted by more lenders, and algorithms will continue to improve. It can be expected that as the prevalence of algorithmic credit scoring increases and data regarding its use becomes more abundant, there will be more empirical studies on this topic. Collectively, they will shed new light on how algorithms affect credit discrimination.
Future Directions for Investigating the Role of Algorithmic Credit Scoring in Affecting Lending Discrimination
While concerns about the potential for unintended discrimination in algorithmic credit scoring have been raised, there is still much to learn about the actual effects of these systems on lending practices. As such, there are a number of important areas for future research in this field.
- Developing new methods for measuring and identifying discrimination in algorithmic credit scoring models.
- Investigating the effect of different machine learning algorithms on discrimination in credit scoring.
- Examining the effectiveness of existing anti-discrimination laws and regulations in the context of algorithmic credit scoring.
- Studying the fairness and transparency of machine learning algorithms used in credit scoring.
- Investigating the role of human oversight in algorithmic credit scoring to ensure that the algorithms are not causing unintended discrimination.
- Examining the effect of different feature selection techniques on discrimination in credit scoring.
- Analyzing the impact of feedback loops and self-reinforcing mechanisms in algorithmic credit scoring on discrimination.
- Exploring the potential of explainable AI (XAI) techniques to increase transparency and reduce discrimination in credit scoring.
References:
Bartlett, R., Morse, A., Stanton, R., & Wallace, N. (2022). Consumer-lending discrimination in the FinTech Era. Journal of Financial Economics, 143(1), 30–56. https://doi.org/10.1016/j.jfineco.2021.05.047
Fuster, A., Goldsmith, P. P., Ramadorai, T., & Walther, A. (2022). Predictably Unequal? The Effects of Machine Learning on Credit Markets. Journal of Finance (John Wiley & Sons, Inc). 77(1), 5-47. doi: 10.1111/jofi.13090
Like (0)