Brief
How far mortgage refinancing fell from its peak
By Kian O'Connor
Published
Data retrieved
Lenders originated 8,866,976 refinance loans in 2020, the most of any year this hub covers, and 1,468,797 in 2024: 83.4% fewer. Refinancing was 60.6% of the 14,637,334 loans made in 2020 and 23.7% of the 6,197,085 made in 2024.
Refinance loans fell 83.4% from 2020 to 2024
Loans originated each year for a refinance or cash-out refinance, and for a home purchase, as lenders reported them under the Home Mortgage Disclosure Act. The newest year is the CFPB's Snapshot, its first release of the year.
Source: The CFPB
About this chart
The figures
| Year | Refinancing | Home purchase |
|---|---|---|
| 2018 | 2,300,222 | 4,351,463 |
| 2019 | 3,784,768 | 4,522,861 |
| 2020 | 8,866,976 | 4,924,799 |
| 2021 | 8,712,028 | 5,406,314 |
| 2022 | 2,632,650 | 4,381,012 |
| 2023 | 1,084,479 | 3,452,630 |
| 2024 | 1,468,797 | 3,534,113 |
| 2025 | 1,995,427 | 3,592,033 |
Embed this chart
Paste this where you want the chart. The picture carries its title, source and credit, under the site’s license.
The low came in 2023, at 1,084,479, and 2024 was above it.
Home purchase loans, the same years
Loans to buy a home fell less over the same years: 4,924,799 in 2020 and 3,534,113 in 2024, 28.2% fewer. The two are set side by side, not as cause and effect: the HMDA register records what each loan was for, and nothing about why a borrower took it.
A refinance and a cash-out refinance
The register reports the two apart. A refinance replaces a loan; a cash-out refinance also pays the borrower part of the home's value. Refinances fell from 6,695,811 to 699,166, 89.6% fewer. Cash-out refinances fell from 2,171,165 to 769,631, 64.6% fewer, so in 2024 lenders made more cash-out refinances than refinances. Every refinancing figure above is the two together.
The newest year
The CFPB's Snapshot of 2025 records 1,995,427 refinance loans, 29.2% of the 6,827,891 it records as made. The Snapshot is the year's first release, and a later release can add loans to it, so 2025 is given here and left out of the comparison above.
How this is counted
A loan is counted in the year a lender reported it originated, and only by the lender that made it: a loan bought from another lender is reported by both, so purchased loans are left out. The years before 2025 are each read from the CFPB's fullest release of that year.
What this data cannot show
The register has no interest rates, so it cannot say why refinancing fell or rose; it shows when, and by how many loans. It counts loans, not households, and a borrower can refinance more than once. The loan amounts it publishes are the midpoints of bands, not the amounts lent, so this brief counts loans rather than adding up their amounts.
What this cannot show
- A loan is counted where a lender reported it as originated. Loans a lender bought from another are left out, because the lender that made the loan reports it too.
- The register says what each loan was for, not why it was made. It has no interest rates, so it cannot say what moved refinancing.
- Each year before the newest is read from the CFPB's fuller release of it (the Three Year file for the earlier years, the One Year file after). The newest year is the Snapshot, the first release, which a later release can add loans to; it is shown beside the comparison, not in it.
- A count is of loans, not of households or homes: one household can refinance more than once in a year, and a loan can be for a home its borrower does not live in.
Sources
- HMDA public loan/application register, each year's fullest national release, Consumer Financial Protection Bureau. Public domain, under the CFPB's website policy ("Legal notices", "Copyright and trademark"). Retrieved October 16, 2026.
Every figure above is the recorded result of a query checked into the repository alongside this text. Continuous integration re-runs those queries and refuses the change if a number here and a number there disagree. Read the methodology, or the datasets behind it.
Published by the American Factbook. An independent publication.