US mortgage lending, loan level
What the loan-level register can prove, and where it stops
In 2024, on a single-product screen of 3,070,808 decisioned applications (first-lien, owner-occupied, site-built, one-unit home purchase, not business-purpose, not reverse, not open-end), Black applicants were denied at 16.07% and White non-Hispanic applicants at 6.26%. Reweighting Black applications to the White non-Hispanic joint distribution of debt-to-income bucket, leverage bucket, loan type and income bucket moves the Black rate to 11.75%. So 4.32 pp of the 9.81 pp raw gap (44.0%) is composition, and 5.49 pp is not.
That residual is not evidence of discrimination, and this page will not claim it is. The public register carries no credit score, no assets or reserves, no employment history and no automated-underwriting recommendation. What HMDA can do is rule out debt load, leverage, product and income as a full explanation. What it cannot do is settle the rest, because the fields that would settle it are the fields Congress did not require. The full finding, group by group and year by year.
The funnel, 2024
The register held 12,229,298 rows in 2024. Only 10,754,906 of them are applications: 1,273,313 rows are loans bought by a second reporter after someone else originated them (already counted once), and 201,079 are preapproval outcomes rather than applications. Of the applications, 8,636,578 reached a credit decision, 1,536,699 were withdrawn by the applicant and 581,629 were closed for incompleteness. Every denial rate on this layer uses decisioned applications as its denominator, the same definition /mortgage uses.
| Year | Applications | Decisioned | Originated | Denied | Denial rate | Volume | Median rate |
|---|---|---|---|---|---|---|---|
| 2018 | 12,939,736 | 10,614,504 | 7,720,995 | 2,554,198 | 24.06% | $1.99T | 4.75% |
| 2019 | 15,129,137 | 12,266,692 | 9,325,241 | 2,538,965 | 20.70% | $2.66T | 4.13% |
| 2020 | 22,666,973 | 18,024,085 | 14,541,302 | 2,863,286 | 15.89% | $4.36T | 3.13% |
| 2021 | 23,297,562 | 18,550,947 | 14,997,308 | 2,919,007 | 15.74% | $4.76T | 2.99% |
| 2022 | 14,322,093 | 11,310,522 | 8,394,278 | 2,499,110 | 22.10% | $2.84T | 4.74% |
| 2023 | 9,987,150 | 7,995,321 | 5,691,726 | 1,994,758 | 24.95% | $1.78T | 6.88% |
| 2024 | 10,754,906 | 8,636,578 | 6,176,052 | 2,099,966 | 24.31% | $2.03T | 6.88% |
Source: HMDA public LAR (CFPB/FFIEC), snapshot files Applications = action taken 1-5; decisioned = 1, 2, 3; originated = 1. Median rate is the median contract rate on originated loans, which carry a rate on 96.4% of 2024 originations (the rest are exempt or not reported and are excluded from the median, not treated as zero). Methodology
The refinance market inverted
Rate-and-term refinancing, which replaces a mortgage without taking equity out, went from 6,630,947 originations in 2020 to 434,606 in 2023, a fall of 93.4%. Cash-out refinancing, the product that converts home equity into cash, fell from 2,983,172 in 2021 to 649,094 in 2023, a fall of 78.2%. The mix flipped: cash-out was 24.6% of refinance originations in 2020 and 59.9% in 2023; in 2024 it was 52.4%, at a median contract rate of 6.99% against 3.25% in 2020.
Source: HMDA public LAR (CFPB/FFIEC), snapshot files Originated loans only (action taken 1), by reported loan purpose. Loan purpose 31 is a rate-and-term refinancing, 32 a cash-out refinancing; the two are reported separately only from 2018, which is why /mortgage charts them as one line. Methodology
Source: HMDA public LAR (CFPB/FFIEC), snapshot files Composition of refinance originations. Each column sums to that year's rate-and-term plus cash-out refinancings. Methodology
US mortgage origination is unconcentrated
4,878 distinct institutions originated a mortgage in 2024. The ten largest made 22.3% of the loans, the fifty largest 46.2%, and the Herfindahl index is 105 on a 10,000 scale: the concentration equivalent of about 95 equal-sized lenders sharing the whole country. The highest reading in the 7 modern-schema years is 139, in 2021, and the lowest is 79, in 2018. The lender layer takes this apart: 930 reporters decided 1,000 or more applications in 2024, and the 300 largest of them have a page each.
Source: HMDA public LAR (CFPB/FFIEC), snapshot files Shares of originated loans (counts, not dollars), by Legal Entity Identifier. The public register identifies the filer by LEI only; the HMDA panel file that maps an LEI to an institution name is not part of this data estate, so lenders are not named here. Methodology
The volume shock
Origination volume peaked at $4.76T in 2021 and fell to $1.78T by 2023, a drop of 62.7%. The denial rate moved the other way, from 15.74% in 2021 to 24.31% in 2024. The highest denial rate in the 7 modern-schema years is 24.95%, in 2023.
Source: HMDA public LAR (CFPB/FFIEC), snapshot files Sum of loan amounts on originated loans. Modern-era loan amounts are reported in dollars, rounded to the midpoint of a $10,000 interval, so a volume total is exact to within that rounding and no further. Methodology
The county layer
1,764 counties decided 500 or more applications in 2024 and have a FIPS code that validates against the county reference; together they cover 8,254,145 of 8,636,578 decisioned applications (95.6%). Each has a page. The denial rate across them runs from 8.8% to 58.0%. Of the 414 counties where both Black and White non-Hispanic applicants filed at least 100 screened applications, the Black denial rate is the higher of the two in 414.
| County | Decisioned | Denial rate | Originated | Volume | Median loan |
|---|---|---|---|---|---|
| Los Angeles CA | 140,655 | 27.2% | 94,659 | $63.3B | $485,000 |
| Maricopa AZ | 139,073 | 22.1% | 101,888 | $39.0B | $325,000 |
| Cook IL | 101,121 | 26.3% | 70,838 | $25.6B | $255,000 |
| Harris TX | 86,670 | 28.2% | 58,621 | $21.0B | $265,000 |
| Riverside CA | 68,702 | 26.3% | 47,106 | $19.6B | $395,000 |
| Clark NV | 67,939 | 24.8% | 48,471 | $18.0B | $325,000 |
| San Diego CA | 66,453 | 24.8% | 46,550 | $30.1B | $505,000 |
| Miami-Dade FL | 57,358 | 36.4% | 33,574 | $17.8B | $375,000 |
| Orange CA | 54,074 | 23.0% | 39,024 | $27.3B | $505,000 |
| Broward FL | 53,272 | 35.8% | 31,647 | $14.3B | $315,000 |
| Bexar TX | 50,127 | 25.4% | 35,292 | $10.4B | $245,000 |
| San Bernardino CA | 50,101 | 27.4% | 33,497 | $11.8B | $345,000 |
Source: HMDA public LAR (CFPB/FFIEC), snapshot files The twelve counties with the most decisioned applications in 2024. County names and the FIPS reference come from the FDIC Summary of Deposits county file. Methodology
The rest of the layer
- Denial gaps raw and composition-standardized denial rates for 11 applicant groups, every year, and a plain statement of what the public register does not carry.
- Denial reasons the primary-reason field on every denial, by year, by loan type and by state.
- Lenders concentration, and a page for each of the 300 largest reporters by decisioned applications.
- Methodology the screen, the standardization, the sentinels, the schema drift between the 2018 and 2019 files, and what this data cannot tell you.