Lending / denial gaps
Four controls explain 44% of the Black/White denial gap. 5.49 pp is left over.
The screen is 3,070,808 decisioned applications in 2024: first-lien, owner-occupied, site-built, one-unit home purchases that are not business-purpose, not reverse mortgages and not open-end lines of credit. Inside it, Black applicants were denied on 42,072 of 261,755 applications (16.07%) and White non-Hispanic applicants on 97,018 of 1,549,187 (6.26%). The raw gap is 9.81 pp.
Reweighting Black applications cell by cell to the White non-Hispanic joint distribution of debt-to-income bucket, leverage bucket, loan type and income bucket (7 x 6 x 4 x 7 = 1,176 cells, 1,069 of them occupied in 2024, covering 99.96% of the White non-Hispanic weight) moves the Black denial rate from 16.07% to 11.75%. Composition accounts for 4.32 pp of the 9.81 pp gap (44.0%). 5.49 pp is left over.
Every group, raw and standardized, 2024
The bar is the raw denial rate; the amber marker is the rate after reweighting that group to the White non-Hispanic composition; the dashed line is the White non-Hispanic raw rate of 6.26%. The standardization returns a figure for 11 of the 11 groups. The groups partition the screen: their application counts sum to 3,070,808, against 3,070,808 screened applications.
Source: HMDA public LAR (CFPB/FFIEC), snapshot files Screened decisioned applications, 2024. Groups are derived_race, except the three White rows, which are split on derived_ethnicity because the standardization's reference group is White non-Hispanic. The reference group's own bar is shown without a marker: reweighting it to itself returns itself. Methodology
| Applicant group | Screened apps | Denied | Raw rate | Standardized | Residual vs reference | White weight matched | Apps with no match |
|---|---|---|---|---|---|---|---|
| Free-form text only | 465 | 120 | 25.81% | 15.14% | 8.87 pp | 75.78% | 0 |
| Black or African American | 261,755 | 42,072 | 16.07% | 11.75% | 5.49 pp | 99.96% | 8 |
| Two or more minority races | 8,133 | 1,224 | 15.05% | 10.89% | 4.63 pp | 97.57% | 1 |
| Native Hawaiian or Pacific Islander | 5,907 | 859 | 14.54% | 11.54% | 5.27 pp | 97.13% | 0 |
| American Indian or Alaska Native | 20,848 | 2,667 | 12.79% | 9.69% | 3.43 pp | 99.40% | 0 |
| White, Hispanic or Latino | 333,665 | 39,785 | 11.92% | 9.68% | 3.42 pp | 99.97% | 11 |
| Race not available | 432,477 | 44,785 | 10.36% | 8.78% | 2.51 pp | 99.98% | 21 |
| Asian | 247,536 | 22,218 | 8.98% | 8.65% | 2.39 pp | 99.75% | 1 |
| White, ethnicity not reported | 121,241 | 8,777 | 7.24% | 7.02% | 0.76 pp | 99.92% | 2 |
| Joint (race) | 89,594 | 6,240 | 6.96% | 7.27% | 1.01 pp | 99.79% | 3 |
| White, non-Hispanicreference | 1,549,187 | 97,018 | 6.26% | n/a | n/a | n/a | n/a |
Source: HMDA public LAR (CFPB/FFIEC), snapshot files White weight matched is the share of the reference group's cell weight that lies in cells where the group has at least one application: the coverage of the reweighting. Apps with no match are the group's applications in cells the reference group never occupies; they drop out of the standardized figure and are shown so the reader can size what was dropped. Methodology
The residual, 2018 to 2024
Standardizing separately inside each year, against that year’s own White non-Hispanic composition, the Black residual runs from 5.38 pp to 6.49 pp across the 7 years, a range of 1.11 pp. The raw gap runs from 8.81 pp to 11.25 pp, a range of 2.44 pp. The residual is positive in 7 of the 7 years.
Source: HMDA public LAR (CFPB/FFIEC), snapshot files Each year is standardized to its own reference-group composition, so the series is not contaminated by changes in the reference group's mix over time. Methodology
| Year | Black apps | Black raw | White NH raw | Raw gap | Black standardized | Residual | White weight matched |
|---|---|---|---|---|---|---|---|
| 2018 | 291,319 | 17.06% | 7.74% | 9.33 pp | 13.59% | 5.85 pp | 99.97% |
| 2019 | 307,636 | 15.63% | 6.81% | 8.81 pp | 12.44% | 5.63 pp | 99.97% |
| 2020 | 364,942 | 17.95% | 6.71% | 11.25 pp | 13.20% | 6.49 pp | 99.98% |
| 2021 | 404,543 | 15.10% | 6.17% | 8.93 pp | 11.55% | 5.38 pp | 99.98% |
| 2022 | 342,366 | 16.53% | 6.54% | 9.99 pp | 12.43% | 5.89 pp | 99.97% |
| 2023 | 273,223 | 16.81% | 6.61% | 10.21 pp | 12.20% | 5.60 pp | 99.97% |
| 2024 | 261,755 | 16.07% | 6.26% | 9.81 pp | 11.75% | 5.49 pp | 99.96% |
Source: HMDA public LAR (CFPB/FFIEC), snapshot files Percentages are on a 0-100 scale. Residual = standardized Black rate minus the same year's raw White non-Hispanic rate. Methodology
Inside each bucket, one control at a time
The standardization above reweights all four controls jointly. These tables do the crudest possible version instead: compare the two groups inside a single bucket of a single control, where the applicants have, on that one dimension, the same reported profile. Counted bucket by bucket in 2024, the Black denial rate is the higher of the two in 7 of the 7 debt-to-income buckets, 6 of the 6 leverage buckets, 7 of the 7 income buckets, 4 of the 4 loan types.
Debt-to-income bucket
| Bucket | Black apps | Black denial rate | White NH apps | White NH denial rate | Difference |
|---|---|---|---|---|---|
| <30 | 21,616 | 14.65% | 301,203 | 4.61% | 10.04 pp |
| 30-36 | 26,448 | 9.54% | 252,448 | 3.49% | 6.05 pp |
| 36-42 | 57,983 | 9.31% | 393,883 | 3.82% | 5.49 pp |
| 43-49 | 89,249 | 9.86% | 443,251 | 4.30% | 5.56 pp |
| 50-60 | 51,657 | 19.50% | 119,652 | 13.91% | 5.59 pp |
| >60 | 12,660 | 86.85% | 27,877 | 73.23% | 13.62 pp |
| not reported | 2,142 | 52.01% | 10,873 | 29.04% | 22.96 pp |
Leverage (combined loan-to-value) bucket
| Bucket | Black apps | Black denial rate | White NH apps | White NH denial rate | Difference |
|---|---|---|---|---|---|
| <80 | 18,767 | 14.30% | 405,172 | 5.28% | 9.02 pp |
| 80-90 | 26,502 | 13.81% | 359,144 | 4.99% | 8.82 pp |
| 90-95 | 21,168 | 16.35% | 148,863 | 5.83% | 10.52 pp |
| 95-100 | 159,425 | 15.91% | 527,786 | 6.69% | 9.22 pp |
| >100 | 24,907 | 17.80% | 61,545 | 11.94% | 5.86 pp |
| not reported | 10,986 | 22.53% | 46,677 | 13.69% | 8.83 pp |
Applicant income bucket
| Bucket | Black apps | Black denial rate | White NH apps | White NH denial rate | Difference |
|---|---|---|---|---|---|
| <50k | 29,190 | 32.88% | 120,551 | 17.12% | 15.76 pp |
| 50-75k | 60,193 | 18.24% | 264,922 | 7.53% | 10.71 pp |
| 75-100k | 58,875 | 14.28% | 274,538 | 5.77% | 8.50 pp |
| 100-150k | 65,968 | 12.05% | 389,624 | 4.68% | 7.38 pp |
| 150-250k | 34,967 | 10.57% | 314,031 | 4.12% | 6.45 pp |
| 250k+ | 11,498 | 10.85% | 174,470 | 4.87% | 5.98 pp |
| not reported | 1,064 | 18.05% | 11,051 | 8.25% | 9.79 pp |
Loan type
| Bucket | Black apps | Black denial rate | White NH apps | White NH denial rate | Difference |
|---|---|---|---|---|---|
| Conventional | 106,871 | 15.30% | 1,114,861 | 5.60% | 9.70 pp |
| FHA-insured | 109,784 | 17.92% | 243,585 | 8.74% | 9.18 pp |
| VA-guaranteed | 42,377 | 13.04% | 165,240 | 6.43% | 6.60 pp |
| USDA Rural Housing / FSA | 2,723 | 19.28% | 25,501 | 10.30% | 8.98 pp |
Source: HMDA public LAR (CFPB/FFIEC), snapshot files Denial rates within one bucket of one control, 2024 screened applications. A "not reported" bucket is kept as its own row rather than dropped. Methodology
What this cannot tell you
- Race is not reported on 14.1% of the screen. 432,477 of the 3,070,808 screened 2024 applications carry no usable race, and that bucket is denied at 10.36%, between the White non-Hispanic rate of 6.26% and the Black rate of 16.07%. Every demographic figure on this page is conditional on race being reported.
- This is a cell reweighting, not a regression. It is direct standardization over a 7 x 6 x 4 x 7 grid of buckets, so it controls for the bucket an application falls in, not for the exact value inside the bucket. Two applications in the same debt-to-income bucket are treated as having the same debt load.
- Four controls, not all of them. The grid holds debt-to-income, leverage, loan type and income. The register also carries loan term, property value, census-tract characteristics, applicant age and sex, and the identity of the lender, and none of those are in the grid. At four dimensions the Black reweighting matches 99.96% of the reference weight.
- 8 Black applications have no White non-Hispanic comparison. They sit in cells the reference group never occupies in 2024 and drop out of the standardized figure entirely.
- It cannot be run before 2018. Debt-to-income, combined loan-to-value and the derived race and ethnicity fields enter the register with the 2018 schema, and the 2007-2017 file has none of them. The series on this page is therefore 7 years long and cannot be made longer. /mortgage carries the longer denial-rate series, but only by state and without any applicant controls at all.
- A county-level gap is a raw gap. Of the 414 counties where both groups filed at least 100 screened applications in 2024, the Black raw denial rate is the higher of the two in 414. Those are raw rates. The standardization is run nationally, over a 1,176-cell grid, and is not attempted county by county.
Source: HMDA public LAR (CFPB/FFIEC), snapshot files The standardization is reproducible from the published screened cell table without touching the loan-level files. Methodology