Short answer: HMDA data is the public record of mortgage applications that lenders report every year under the Home Mortgage Disclosure Act: who applied, for what, where, and what the lender decided. The 2025 file holds 6,827,891 originations totaling $2.41 trillion. It is the only public source that shows who makes the mortgages in a county, lender by lender. Read it with three rules: count originations and purchased loans separately, measure share against every lender and not only against banks, and treat a denial rate as a question, not a finding.
What HMDA is
Congress passed the Home Mortgage Disclosure Act in 1975 so the public could see whether lenders were serving the housing needs of the places they operate in. The rule that implements it is Regulation C, written by the Consumer Financial Protection Bureau. Each covered lender keeps a register of every mortgage application it handled during the year and files it by March 1 of the next. The FFIEC publishes the results, loan by loan, without names or addresses.
The file is large. For 2025 it records 6,827,891 originations for $2.41 trillion, against 6,176,052 for $2.03 trillion the year before: 10.6% more loans and 18.3% more dollars, with the average loan rising from about $329,000 to about $353,000.
Who has to report
Banks, savings associations, credit unions and mortgage companies all report, if they are active enough. The main test is volume: a lender reports closed-end mortgages if it originated at least 25 of them in each of the two preceding calendar years, and open-end lines of credit if it originated at least 200 in each of those years. A bank or credit union is also exempt if its assets are at or below a threshold that moves every year, $59 million for 2026 data, or if it has no office in a metropolitan area.
Two consequences follow. The smallest and the most rural lenders are not in the file, so a rural county's total understates what was really lent there. And banks and credit unions that originated fewer than 500 closed-end loans in each of the prior two years may leave out many of the newer fields, so pricing and underwriting detail is thinner for small lenders than for large ones.
What each record says
- What happened. Originated, approved but not accepted, denied, withdrawn by the applicant, closed as incomplete, or purchased from another lender.
- What kind of loan. Purpose: home purchase, refinance, cash-out refinance, home improvement or other. Type: conventional, FHA, VA or USDA. First or subordinate lien. Principal residence, second home or investment property.
- Where. State, county and census tract of the property.
- Who. The applicant's income, race, ethnicity, sex and age band, and the same for a co-applicant.
- On what terms. Loan amount, interest rate, the spread over a benchmark rate, fees, and bands for debt-to-income and loan-to-value.
- Why not. For most lenders, up to four reasons for each denial.
Some things are deliberately missing or blurred in the public file. There is no credit score. Loan amounts are rounded to the midpoint of a $10,000 band, ages and debt-to-income ratios are shown in bands, and there are no dates and no addresses.
Rule one: an origination is not a purchase
A lender that buys a closed loan from another lender reports it as a purchase. The lender that made the credit decision reports the same loan as an origination. Add the two together and a bank that buys loans looks like a far bigger mortgage lender than it is, and the same loan is counted twice in the county.
Keep them in separate columns. Market share, rank and denial rates all belong to originations and the applications behind them. Purchases tell you something else: who is buying mortgage assets, and from where.
Rule two: measure share against every lender
Mortgage companies that are not banks originate most home purchase loans in the United States. In many counties the largest mortgage lender has no branch and no deposits. A bank that measures its mortgage share only against other banks is using a denominator that leaves out most of the market, and can roughly double its apparent share by doing so.
The honest version sums every filer in the county: banks, credit unions and mortgage companies. Rank has to be measured the same way. A bank that ranks fourth among the banks in a county can rank far lower among all the lenders that made a loan there, and only the second rank describes the market a borrower sees.
County totals are nearly complete: 99.5% of 2025 originations carry a county. The rest are loans whose location was not reported, and they belong in the national total but in no county's.
Rule three: a denial rate is a question
The usual denial rate is denials divided by the applications the lender decided: originated, approved but not accepted, and denied. Applications the borrower withdrew, and files closed as incomplete, are left out because the lender never made a decision on them.
A gap in denial rates between groups of applicants is the most quoted number in the data and the easiest to misread. The public file has no credit scores, and debt-to-income and loan-to-value appear only in bands, so two applicants who look alike in the file can differ in ways an underwriter is required to weigh. Examiners treat a gap as a reason to pull loan files, not as a conclusion. A bank should read its own gaps the same way, and before an examiner does.
Higher-priced lending
Each record carries a rate spread: the loan's annual percentage rate less the average prime offer rate for a comparable loan. By convention a first-lien loan is higher-priced when the spread is at least 1.5 percentage points, and a subordinate lien at 3.5. A high share of higher-priced lending is not by itself a problem. It usually reflects what a lender makes, such as manufactured home loans or small second liens. It becomes a question when it differs sharply between groups of borrowers on the same product.
Reading a county
Put the pieces together and one county's file answers most of what a bank wants to know about its mortgage market.
- The league table. Every lender, ranked by originations, with dollars beside the count.
- Your rank among all lenders, and how it moved from one year to the next.
- The mix. Purchase against refinance, conventional against government-backed, and how your mix compares with the county's.
- Who is being served. The share of loans to low- and moderate-income borrowers and in low- and moderate-income tracts. These are the same comparisons a CRA examiner makes (CRA ratings explained).
- Where applications fail. Denial rates and denial reasons by product, against the lenders around you.
What HMDA cannot tell you
It is an annual file. The data for a year arrives in the spring and summer of the next, so it describes last year's market. It records applications, not inquiries, so it cannot see the borrower who was discouraged before applying. And it files each loan under the legal entity that made the decision, identified by a code called an LEI. A bank's mortgage subsidiary can file under its own. Matching those codes to banks takes a crosswalk: BankingLENS matches 2,535 banks, and 2,329 of them have mortgage data in the 2024 or 2025 files.
Frequently asked questions
What is HMDA data?
It is the public record of mortgage applications that lenders report each year under the Home Mortgage Disclosure Act of 1975. Each record shows the type and purpose of the loan, the amount, the county and census tract, the applicant's income, race, ethnicity, sex and age band, and what the lender decided. The 2025 file holds 6,827,891 originations totaling $2.41 trillion.
Which lenders have to report HMDA data?
Banks, savings associations, credit unions and mortgage companies that originated at least 25 closed-end mortgage loans, or at least 200 open-end lines of credit, in each of the two preceding calendar years. A bank or credit union is also exempt if its assets are at or below a threshold that is adjusted every year, $59 million for 2026 data, or if it has no office in a metropolitan area.
When is HMDA data released?
Lenders file the previous calendar year's data by March 1. Each lender's own file is made public in the spring, and the national loan-level file for the year follows, usually in June or July.
How do you calculate mortgage market share from HMDA?
Count a lender's originations in a county and divide by the originations of every lender that reported in that county, including mortgage companies and credit unions. Leave purchased loans out of both numbers. Measuring a bank only against other banks can roughly double its apparent share, because mortgage companies that are not banks originate most home purchase loans.
What is a high mortgage denial rate?
There is no standard, because denial rates depend on the kinds of loans a lender offers and on who applies. The usual formula is denials divided by the applications the lender decided: originated, approved but not accepted, and denied. HMDA has no credit scores, so a gap in denial rates between groups is a reason to look closer, not proof of discrimination.
What is a higher-priced mortgage loan in HMDA?
By convention, a first-lien loan whose annual percentage rate is at least 1.5 percentage points above the average prime offer rate for a comparable loan, or a subordinate lien at least 3.5 points above it. HMDA reports the rate spread for each loan, which is how the share of higher-priced lending is calculated.
Where these numbers come from
Every figure on this page comes from the national HMDA loan-level files for 2024 and 2025, counting originations and keeping purchased loans apart. County totals and ranks are measured across every lender that reported, not only banks. Reporting thresholds are the federal ones in effect in October 2026. In BankingLENS the Mortgage tab shows what happened to every application a bank took, its denial rates and reasons, higher-priced lending, borrower income, the neighborhoods it lent in and its rank in each county. You can see a real bank ranked against its FFIEC peer group on the sample scorecard, or open your own bank in the dashboard. Plans that cover your state or every bank start at $29 a month (pricing).