HMDA reporting.

Who must report, what goes on the loan/application register, the deadlines and edits, and how examiners test the data.

How AI coworkers check the LAR against the file

HMDA reporting is the yearly filing of mortgage data that the Home Mortgage Disclosure Act requires of most mortgage lenders in metropolitan areas. Each covered application, origination and purchase becomes a row on a loan/application register, or LAR, and the data are made public.

About 4,768 lenders filed HMDA data for 2025.⁠[1] Each row carries 110 data fields, and examiners test them against the loan file.⁠[2] This guide covers who must report in 2026, what goes on the LAR, the deadlines, edits and resubmission, and what errors have cost lenders.

What is HMDA reporting?

HMDA reporting is how a lender complies with the Home Mortgage Disclosure Act of 1975: it records each covered application, origination and purchase on the LAR every quarter and files the LAR once a year, under Regulation C, 12 CFR part 1003.⁠[3]⁠[4]

Its purposes are public: to show whether lenders serve their communities' housing needs, to guide public investment, and to help identify possible discriminatory lending patterns.⁠[3] Lenders file with the CFPB on the HMDA Platform, which counts as filing with their own federal regulator.⁠[1]⁠[5]

Is every HMDA error a violation?

No. An unintentional error isn't a violation if it happened despite procedures reasonably adapted to avoid it, and quarterly data recorded in good faith can be corrected before the annual filing (12 CFR 1003.6(b)–(c)).⁠[3] The CFPB found Washington Federal Bank hadn't kept such procedures, so its errors were “intentional and not bona fide errors.”⁠[6]

How is HMDA data used in fair lending reviews?

As a screen. The agencies use HMDA data in fair lending and Community Reinvestment Act exams,⁠[7] and a disparity is a reason to pull loan files, not proof of discrimination. Since July 21, 2026, Regulation B says the Equal Credit Opportunity Act doesn't authorize disparate-impact liability; the CFPB's rule leaves HMDA unchanged.⁠[8]

HMDA reporting statistics

In 2023, most closed-end HMDA reporters were small banks and credit unions, but most of the loans came from mortgage companies.

Small banks and credit unions were nearly two in three closed-end HMDA reporters in 2023

Institutions that reported closed-end data for 2023, by type. Small banks: under $1 billion in banking-organization assets.⁠[9]

Show the data for Small banks and credit unions were nearly two in three closed-end HMDA reporters in 2023
Closed-end reporters and originations by institution type, 2023 (CFPB Table 5A)
TypeInstitutionsOriginations (thousands)
Small banks (under $1B)1,677234
Credit unions1,574529
Independent mortgage companies9202,598
Large banks ($1B and over)8391,060
Affiliated mortgage companies91208
All reporters5,1034,627
  • 42.2% of closed-end reporters in 2023, 2,152 of 5,103, originated fewer than 100 closed-end loans. Together they made about 2.2% of reported originations.⁠[9]
  • 56.1% of reported closed-end originations in 2023 came from independent mortgage companies. Credit unions made 11.4%.⁠[9]
  • 14.6% more institutions reported for 2023 than for 2022, an increase the CFPB says is likely due to the lower closed-end threshold.⁠[9]

Who must report HMDA data?

A bank, savings association or credit union must report the HMDA data it collects in 2026 if it passes all five tests below; a for-profit nondepository lender needs only the location and loan-volume tests (12 CFR 1003.2(g)).⁠[3]

HMDA coverage tests for data collected in 2026

An institution must meet every test in its column.⁠[3]⁠[4]

AssetsMore than $59 million on December 31, 2025None
LocationA home or branch office in a metropolitan statistical area (MSA) on December 31, 2025Same test
Loan activityOriginated at least one home purchase loan, or refinancing of a home purchase loan, secured by a first lien on a one- to four-unit dwelling in 2025None
Federal linkFederally insured or regulated, or such a loan was federally insured, guaranteed or supplemented, or intended for sale to Fannie Mae or Freddie MacNone
Loan volumeAt least 25 closed-end mortgage loans in each of 2024 and 2025, or at least 200 open-end lines of credit in each yearSame test

Each product is tested on its own. A lender with 200 open-end lines but fewer than 25 closed-end loans reports only its lines of credit (12 CFR 1003.3(c)(11)–(12)).⁠[3]

What is the HMDA asset threshold for 2026?

$59 million. Banks, savings associations and credit unions at or below it on December 31, 2025, don't collect HMDA data in 2026. The CFPB adjusts the threshold every year by the CPI-W; it was $58 million for 2025. The exemption doesn't excuse reporting data the institution had to collect in 2025.⁠[4]

Will the HMDA threshold change?

Possibly. Executive Order 14393, signed March 13, 2026, directs the CFPB to consider proposing Regulation C amendments that would raise the asset threshold for smaller banks, exclude inquiries from HMDA, and reduce reporting burdens.⁠[10] The order doesn't amend the rule; until the CFPB finalizes one, the tests above apply.

What is the HMDA partial exemption?

Insured banks, savings associations and credit unions that originated fewer than 500 closed-end loans in each of the two preceding years can skip 26 of Regulation C's 48 data points for those loans. Open-end lines are tested separately.⁠[11] The skipped items include the universal loan identifier, property address, rate spread, credit score, loan costs and the underwriting ratios. A bank rated “needs to improve” on its two most recent CRA exams, or “substantial noncompliance” on the latest, doesn't qualify (12 CFR 1003.3(d)).⁠[3]

What data goes on the HMDA LAR?

The HMDA LAR has one row for each covered application, origination and purchase. Regulation C requires 48 data points (12 CFR 1003.4(a)), reported in 110 data fields.⁠[3]⁠[11]⁠[2]

Identifiers and dates

The universal loan identifier, the application date, the action taken and the date of that action.

The loan

Loan type, purpose, preapproval, amount, lien status, term, interest rate, prepayment penalty and payment features, and reverse-mortgage, open-end and business-purpose flags.

The property

Address and census tract, construction method, occupancy, the property value relied on, total units, and multifamily and manufactured home details.

The applicant

Ethnicity, race and sex, and whether they were noted by visual observation or surname; age; the income relied on; and the credit score relied on, with its scoring model.

Pricing and costs

Rate spread over the average prime offer rate (APOR), HOEPA status, total loan costs or points and fees, origination charges, discount points and lender credits.

Underwriting, channel and sale

Debt-to-income and combined loan-to-value ratios, denial reasons, the automated underwriting result, the application channel, the originator's NMLSR ID and, for a loan sold that year, the purchaser type.⁠[3]

Which HMDA fields do examiners test first?

The 37 key fields. The Federal Reserve, FDIC and OCC focus transaction testing on them; they include the application date, loan amount, action taken and its date, census tract, applicant demographics and income, credit score, interest rate, the underwriting ratios and property value. Examiners may review other fields too.⁠[2]

When is HMDA data due?

HMDA data are due by March 1 of the following year: data collected in 2026 are due Monday, March 1, 2027. A deadline that falls on a Saturday or Sunday moves to the Monday, which is why 2025 data were due March 2, 2026 (12 CFR 1003.5(a)(1)).⁠[3] Three other clocks run alongside it.

Record each quarter's data30 calendar days after quarter-end

Data go on the register within 30 calendar days after the end of the quarter in which you took final action. Good-faith errors corrected before the annual filing aren't violations.

12 CFR 1003.4(f), 1003.6(c)(1)⁠[3]

Quarterly filing for large reporters60 calendar days after Q1, Q2 and Q3

An institution that reported at least 60,000 covered loans and applications for the preceding year, excluding purchased loans, also files its data for each of the first three quarters.

12 CFR 1003.5(a)(1)(ii)⁠[3]

The annual filingMarch 1 of the following year

File on the HMDA Platform with your Legal Entity Identifier, correct or confirm every edit, and have an authorized representative with knowledge of the data certify that it is accurate and complete. Keep a copy of the register for at least three years.

12 CFR 1003.5(a)(1)(i), (iii), (a)(3)⁠[3]⁠[5]

Public notices3 business days after the FFIEC's notice

Once the FFIEC says your disclosure statement is available, make a written notice available at your home office and each branch in an MSA, saying it can be obtained on the CFPB's website. A similar notice covers your modified LAR, and a general notice about your HMDA data goes up in those lobbies.

12 CFR 1003.5(b)–(e)⁠[3]

What are HMDA edits?

HMDA edits are the checks the HMDA Platform runs on a register before it can be submitted.⁠[5] Clearing them doesn't make the data right: an income of $85,000 passes every edit whether the underwriter relied on $85,000 or $58,000. Only the loan file can tell.

The four types of HMDA edits

From the Filing Instructions Guide for data collected in 2026.⁠[5]

SyntacticalFile format and filing yearCorrect every error6
ValidityAn allowed value in each fieldCorrect every error118
QualityExpected values in a field or combination of fields, such as an applicant age between 18 and 100Confirm or correct52
Macro qualityExpected patterns across the register, such as no more than 30% of entries reported as withdrawnConfirm or correct7

Edit counts are ours, from the guide's edit tables. The data fields didn't change from 2025.

When must HMDA data be resubmitted?

Examiners require a resubmission when their sample finds too many errors in one field. Under the FFIEC's HMDA Examiner Transaction Testing Guidelines, they check a random sample of LAR rows against the loan files, and a field that reaches the threshold must be corrected across the whole register and resubmitted.⁠[7]

HMDA transaction testing: sample sizes and thresholds

Two errors in any field in the initial sample send examiners on to the rest of the sample.⁠[7]

25–5030*153 errors (10.0%)
51–10030203 errors (10.0%)
101–13047293 errors (6.4%)
131–19056293 errors (5.4%)
191–50059303 errors (5.1%)
501–100,00079354 errors (5.1%)
100,001+159614 errors (2.5%)

*With fewer than 30 rows, examiners review every row and the threshold stays at 3 errors. Each applicant's ethnicity fields count as one group, and so do race fields.

Dates within three calendar days, loan amounts within $1,000 and income rounding don't count toward the thresholds. Examiners can still order a resubmission below the threshold when errors would make the data unreliable, such as withdrawn applications coded as denials, or when reportable loans were left off.⁠[7]

What have HMDA errors cost lenders?

A HMDA violation can bring civil money penalties (12 CFR 1003.6(a)).⁠[3] Four CFPB actions show what drew them.

Freedom Mortgage, 2024$3.95 millionAllegedly widespread errors across multiple fields in its 2020 data, while it was under a 2019 order.⁠[12]
Bank of America, 2023$12 millionFor at least four years, hundreds of loan officers didn't ask applicants the required demographic questions and reported that they had chosen not to answer.⁠[13]
Washington Federal Bank, 2020$200,000Errors in 40 of 100 files in its own audit of its 2016 register and in 27 of 84 files the CFPB reviewed from 2017, traced to staffing, training, quality control and board oversight.⁠[6]
Freedom Mortgage, 2019$1.75 millionInaccurate race, ethnicity and sex data for 2014 to 2017, much of it intentional. Some loan officers were told to select non-Hispanic white when applicants didn't answer.⁠[14]

The 2024 Freedom Mortgage findings are allegations from a complaint resolved by a stipulated judgment.

How to keep HMDA data accurate

  • Write your procedures down. The bona fide error defense depends on procedures reasonably adapted to avoid errors.⁠[3]
  • Test fields against the file, not just the edits. Start with the 37 key fields.⁠[2]
  • Run the examiners' sample on yourself. The FFIEC's sample sizes and thresholds show which fields would fail while there's still time to fix them across the register.⁠[7]
  • Train for the demographic questions. Ask however the application is taken, read the questions aloud on the phone, and use visual observation or surname only for in-person applicants who don't answer.⁠[3]

Why HMDA data reviews take so long

Checking one row means opening the loan file in several places. When every row takes that long, review gets rationed to a sample, and the errors outside it are filed under your certification. Compliance operations covers where a compliance team's other hours go.

  1. 1Loan origination system
  2. 2Application and demographic form
  3. 3Credit report and AUS findings
  4. 4Underwriting worksheet
  5. 5Appraisal
  6. 6Closing Disclosure
  7. 7Rate lock and APOR table

How AI coworkers help

Socratix AI coworkers take on the gathering. For every row on the register, they pull the source documents from each system, compare them with the LAR, and hand your analysts each mismatch with the evidence and a recommendation. Your team still makes every call — it just stops spending filing season assembling the picture.

Examiners test a register field by field against the loan file whatever the lender's size.⁠[7] AI coworkers check every row instead of a sample, so the people you have spend their time on the judgment calls. Regulators work the same way: in GAO's 2025 review, most said AI outputs inform their staff's decisions but aren't the sole basis for them.⁠[15]

  1. GatherPulls the documents behind each row, from the application to the rate lock, into one view per loan.
  2. AssessCompares every reported field with those documents and Regulation C's definitions, and flags each mismatch with the document it came from.
  3. RecommendProposes the correction with the evidence and a plain-language reason. Your analyst decides.
  4. LearnLearns from each decision — your policies, your loan products, your team's judgment — and every step is logged for the examiner.

Related guides

HMDA reporting FAQ

Do credit unions have to report HMDA data?

Yes, if they pass Regulation C's depository tests: for 2026, assets over $59 million, an office in a metropolitan area, a first-lien home purchase loan (or refinancing of one) on a one- to four-unit dwelling in 2025, a federal connection, and 25 closed-end loans or 200 open-end lines in each of 2024 and 2025. Insured credit unions under 500 loans of a type in each of those years get the partial exemption for it. In 2023, 1,574 credit unions reported closed-end data.

12 CFR 1003.2(g)(1), 1003.3(d)⁠[3]⁠[4]⁠[9]

Which loans are excluded from HMDA reporting?

Among others: loans made in a fiduciary capacity, loans secured by unimproved land, temporary financing, purchases of servicing rights or partial interests, loans under $500, loans primarily for agricultural purposes, and business- or commercial-purpose loans unless they're a home purchase, home improvement or refinancing. Loans bought in a merger or branch acquisition are excluded too.

12 CFR 1003.3(c)⁠[3]

Do denied and withdrawn applications go on the LAR?

Yes. The register covers applications as well as loans, with the action taken and its date. Preapproval requests go on only if they're denied, approved but not accepted, or turned into a home purchase loan. Denials carry the principal reasons, unless the transaction is partially exempt.

12 CFR 1003.4(a)(8), (a)(16)⁠[3]

What is a universal loan identifier (ULI)?

The ULI identifies each loan or application on the LAR. It starts with your Legal Entity Identifier, adds up to 23 characters unique within your institution, and ends with a two-character check digit; it can't include anything that directly identifies the borrower. A purchaser keeps the ULI already assigned, and partially exempt transactions may use a non-universal identifier of up to 22 characters.

12 CFR 1003.4(a)(1)(i), 1003.3(d)(5)⁠[3]

How is HMDA rate spread calculated?

Rate spread is the loan's annual percentage rate minus the average prime offer rate (APOR) for a comparable transaction as of the date the interest rate was set. It's reported for covered loans and approved-but-not-accepted applications subject to Regulation Z, other than assumptions, purchased loans and reverse mortgages. The CFPB publishes APOR tables at least weekly.

12 CFR 1003.4(a)(12)⁠[3]

Can AI check HMDA data before filing?

It can help when it checks each field against the loan file, the way examiners do, rather than only running the Platform's edits. Socratix AI coworkers gather the source documents for every row, flag the fields that don't match with the evidence, and recommend a correction. Your team decides on each change, and an authorized representative still certifies the register.⁠[7]⁠[3]

Sources

  1. 1.CFPB, 2025 HMDA Data on Mortgage Lending Now Available · Mar 31, 2026
  2. 2.Federal Reserve, FDIC and OCC, Designated Key HMDA Data Fields (37 of 110 data fields, data collected from Jan 1, 2018) · Oct 2017
  3. 3.eCFR, 12 CFR Part 1003, Home Mortgage Disclosure (Regulation C), including Appendix B · current
  4. 4.CFPB, Home Mortgage Disclosure (Regulation C) Adjustment to Asset-Size Exemption Threshold, 91 FR 445 · Jan 7, 2026
  5. 5.FFIEC and CFPB, Filing Instructions Guide for HMDA Data Collected in 2026 (version 1.0) · Oct 2025
  6. 6.CFPB, Consent Order, In the Matter of Washington Federal Bank, N.A., 2020-BCFP-0019 · Oct 27, 2020
  7. 7.FFIEC, HMDA Examiner Transaction Testing Guidelines (attachment to Federal Reserve CA 17-2) · Aug 2017
  8. 8.CFPB, Equal Credit Opportunity Act (Regulation B), final rule, 91 FR 21620 (effective Jul 21, 2026) · Apr 22, 2026
  9. 9.CFPB Office of Research, 2023 Mortgage Market Activity and Trends (HMDA data for 2023; Table 5A, lending institutions) · Dec 2024
  10. 10.The White House, Executive Order 14393, Promoting Access to Mortgage Credit, Sec. 3 · Mar 13, 2026
  11. 11.CFPB, 2018 HMDA Rule: interpretive and procedural rule implementing section 104(a) of the Economic Growth Act, executive summary · Aug 31, 2018
  12. 12.CFPB, Enforcement action: Freedom Mortgage Corporation (HMDA, complaint Oct 10, 2023; stipulated judgment Jun 26, 2024) · Jun 26, 2024
  13. 13.CFPB, CFPB Orders Bank of America to Pay $12 Million for Reporting False Mortgage Data (press release, as originally published) · Nov 28, 2023
  14. 14.CFPB, Enforcement action: Freedom Mortgage Corporation, 2019-BCFP-0007 · Jun 5, 2019
  15. 15.GAO, Artificial Intelligence: Use and Oversight in Financial Services (GAO-25-107197) · May 19, 2025

This guide is general information, not legal advice.

Get started

Ready to see what Socratix can do?

Fraud is dynamic. You need systems that adapt, learn, and act in real time — our agents do more than automate workflows, they make intelligent decisions.