Fairness audits of credit-scoring models (research idea)

Fairness audits of credit-scoring models

Fairness and algorithmic bias are growing in the AI journals, and credit scoring is one of the areas where AI rules are strictest. Yet only 19 of 473 recent fairness and bias papers carry a credit or lending keyword. What would an audit that lenders and regulators can both use look like, and what does fairness cost in default risk?

What prompted this

In the AI journals, "fairness" had 16 papers in July–September 2026 against about 10 expected and "algorithmic bias" 16 against about 11 (see the AI dashboard). Since October 2025, 473 papers carry a fairness, algorithmic-bias or discrimination keyword (186 in AI journals, 155 in economics), and only 19 of them carry a credit-scoring, lending, loan or credit-risk keyword.

Questions

On public credit data, how do approval rates and default rates change across groups when common fairness constraints are imposed?
Do the fairness fixes popular in AI papers hold up under the legal tests used in lending, such as disparate impact?
Can an audit work without protected attributes, which lenders are often not allowed to collect?

Data that could help

Mortgage applications with applicant demographics (HMDA), public credit datasets such as German Credit and LendingClub (the credit category of the dataset ranking), and the models of recent papers.

Next step

Train a standard credit model on LendingClub loans and on HMDA applications, apply three common fairness constraints, and report how approval gaps and predicted default rates change.

Related research activity

The keywords this note follows, in the tracked finance, accounting, economics, management, operations research and AI journals as of 2026-10-26
Keyword	Papers, last 30 days	Rank	Papers, last 36 months
algorithmic fairness	2	#1303	159
credit scoring	2	#1303	82
algorithmic bias	0	–	182

@misc{scholar4ai-ideas-fairness-audits-of-credit-scoring, author = {{Scholar4AI}}, title = {Fairness audits of credit-scoring models}, year = {2026}, howpublished = {\url{https://scholar4ai.com/ideas/fairness-audits-of-credit-scoring}}, urldate = {2026-10-23}, note = {Journal Research Rankings; data from OpenAlex (CC0); licensed CC BY-NC 4.0} }