Mortgage denial disparities by race in 2024 HMDA data
In 2024 HMDA home-purchase data, Black and American Indian applicants were denied at roughly 2.2x the White rate — a stark but unadjusted gap that screens for review, not proof of discrimination.
Denial-rate disparity is the first thing a fair-lending team looks at in a new HMDA vintage, and the 2024 home-purchase file makes the reason obvious. Across first-lien, principal-residence purchase applications that lenders actually decisioned, the denial rate for Black applicants is more than double the rate for White applicants. The pattern is not subtle, it is not new, and — read correctly — it is not a conclusion. It is the opening question.
What the 2024 numbers show
Grouping applicants by HMDA race code (with Hispanic assigned by ethnicity first, so the buckets
don't overlap), and defining the denial rate as denied / (originated + approved-but-not-accepted + denied):
| Applicant group | Denial rate | Denied / decisioned |
|---|---|---|
| Asian | 9.5% | 25,989 / 273,364 |
| White | 11.7% | 233,452 / 1,999,459 |
| Native Hawaiian / Pacific Islander | 16.1% | 1,021 / 6,360 |
| Hispanic (any race) | 19.4% | 115,272 / 592,901 |
| American Indian / Alaska Native | 25.7% | 7,436 / 28,986 |
| Black | 25.7% | 83,973 / 327,272 |
Three features stand out. Black and American Indian / Alaska Native applicants sit at 25.7% — roughly 2.2x the 11.7% White rate. Hispanic applicants are denied at 19.4%, about 1.7x White. And the lowest denial rate belongs not to White applicants but to Asian applicants at 9.5% — a reminder that the disparity structure isn't a simple minority-versus-White dichotomy. Native Hawaiian / Pacific Islander sits at 16.1%, though on a thin base of 6,360 decisioned applications, so treat that cell as directional.
The volumes matter for how much weight each rate carries. White applicants dominate the file at roughly 2.0 million decisioned applications; Black (327k), Hispanic (593k), and Asian (273k) buckets are all large enough that the gaps are statistically stable, not artifacts of small samples.
The caveat that governs everything above
Every rate on this page is unadjusted. HMDA has no credit score, only a coarse (public-file-binned) DTI, and nothing on wealth, reserves, or assets — the variables that actually drive underwriting, and that correlate with race in the U.S. population. So these gaps are a screening signal: consistent with disparate treatment, and equally consistent with credit and collateral differences the data can't see. They flag where to look, not what you'll find.
How a fair-lending team actually uses this
The disparity table is step one of a funnel, not a verdict. In practice the workflow runs:
- Benchmark. Compute the unadjusted gaps at the level that matters — nationally, by MSA, and by institution — and flag outliers against peer lenders in the same markets. This is the HMDA-native step, and it's what the numbers above support.
- Control for what HMDA does hold. Stratify by loan amount, income, LTV proxy, loan type (conventional vs. FHA/VA), and the coarse DTI bucket. Gaps that survive coarse stratification are worth escalating; gaps that mostly close are lower priority.
- Drill into underwriting data. For the flagged lenders and markets, move off HMDA entirely and run a matched-pair analysis or a regression with the actual underwriting file — credit score, full DTI, reserves, AUS findings, and adverse-action reasons. Only there can a gap be attributed to treatment rather than to unobserved risk.
The 2024 numbers are a strong instance of step one: large, stable, and pointed. But their job is to tell a compliance analyst where to spend the expensive analysis, not to close the case. Read as a map to the questions worth asking, HMDA is doing precisely what it was built for.