Back to Articles

HMDA Reporting Automation: Fewer Errors Before You File

11/15/2026
16 min read
HMDA Reporting Automation: Fewer Errors Before You File

HMDA reporting automation means capturing loan application data once, at the point of origination, validating it continuously against Regulation C and the FFIEC edit rules, and routing every exception to a named owner with a documented resolution, so the annual filing becomes a certification of work already done rather than a scramble in February. For credit unions and community banks, the case is practical: the regulator tests your loan/application register (LAR) against your loan files, and unexplained differences are treated as errors. Automation moves error detection upstream, where correcting a field costs minutes instead of a resubmission. RiskInMind's platform includes regulatory compliance monitoring, regulatory reporting automation and audit-ready reporting for institutions in exactly this position.

TL;DR

  • HMDA automation works best when it starts at application intake, not at the filing file: every covered data point is captured, validated and logged as the loan moves.
  • Regulation C requires LAR entries within 30 calendar days after the end of the quarter of final action, and the annual submission is due by March 1 (moved to the next Monday when it falls on a weekend).
  • The CFPB's resubmission thresholds, 10% of a sampled LAR for smaller filers and 4% for filers with 100,000 or more entries, make pre-submission testing the core control.
  • Automation does not replace certification: an authorized representative with knowledge of the data still certifies the filing, so exception routing and audit trails matter as much as speed.
  • Evaluate any tool on integration, security, edit handling, change management for yearly rule updates, and evidence an examiner can read.
Model Governance

Automate Regulatory Model Risk Governance

Examine models against 32 qualitative criteria and resolve risk Tiers with pre-deployment checklists per OCC 2011-12 guidelines.

Table of Contents

Who has to report, and what the threshold means for your build

Before you automate anything, settle scope, because the automation has to mirror the coverage tests exactly. Under the FDIC's HMDA examination manual, a bank, savings association or credit union is subject to Regulation C only if it meets all five tests: assets above the annually published asset-size threshold on the preceding December 31, a home or branch office in a metropolitan statistical area, at least one first-lien home purchase loan or refinancing in the preceding year, federally related status, and loan volume of at least 25 closed-end mortgage loans or 200 open-end lines of credit in each of the two preceding calendar years.

That last test is the one most credit unions watch. The closed-end threshold is 25 loans in each of the two preceding years, and open-end lines have their own 200-line threshold. A rules engine that tracks both counts through the year tells you in advance whether you will be a filer next cycle, which affects staffing and vendor decisions long before January.

Coverage also depends on the transaction. Regulation C excludes categories listed in 12 CFR 1003.3(c), including fiduciary loans, loans secured by unimproved land, temporary financing, loans under $500 and agricultural loans. Business- or commercial-purpose loans are excluded unless they are for home purchase, home improvement or refinancing. Those exclusions, per the same FDIC manual, are where manual intake most often goes wrong: a loan officer decides on the fly, and the decision is never recorded. An automated intake step should ask the coverage questions in a fixed order and store the answer with the application.

What the register has to hold

The LAR is a row per covered application or loan. The CFPB's 2024 reference chart lists each data point with its format and the codes for "not applicable" and "exempt." Application date is YYYYMMDD, or "NA" for purchased covered loans. Loan amount is in dollars. Action taken uses codes 1 through 8, covering originated, approved but not accepted, denied, withdrawn, closed for incompleteness, purchased, preapproval denied and preapproval approved but not accepted, and each carries an action taken date.

Smaller institutions may qualify for a partial exemption. The FDIC manual explains that eligible insured depository institutions and credit unions must still collect, record and report 22 data points, among them ethnicity, race, sex, age, income, LEI, loan amount, action taken, census tract and type of purchaser, while 26 others, such as rate spread, credit score, property value, DTI and CLTV, may be reported as exempt. For automation, this means your data model needs an exemption flag by field. Otherwise the system either over-collects or marks a required field as optional.

Where manual HMDA reporting breaks down

The failures repeat across institutions, and each maps to a point in the workflow:

  • Re-keying. Data moves from the origination system to a spreadsheet to the filing file. Each hop is a chance for a transposed census tract or a date in the wrong format.
  • Late recording. Regulation C requires covered-loan data on the register within 30 calendar days after the end of the calendar quarter in which final action is taken. When recording waits for year-end, withdrawn and incomplete applications are the ones that go missing.
  • Undocumented judgment calls. Whether an application was denied or withdrawn, whether a loan is covered at all, whether a broker-originated file is yours to report. The answer lives in someone's head.
  • Rule drift. Thresholds adjust annually and the Filing Instructions Guide is reissued each year. A hard-coded spreadsheet silently goes stale.
  • Edit overload at the deadline. Teams meet the platform's edits for the first time in the final days before the filing window closes.

None of these is solved by a faster file generator. They are solved by moving validation and documentation to the moment of data entry.

A workflow you can automate end to end

The sequence below follows the loan, not the filing. Each step names what it takes and where it usually goes wrong.

  1. Capture at intake. Pull applicant and property data from the application, the loan origination system and supporting documents into one structured record. Where it goes wrong: free-text fields and scanned documents that nobody parses. Document extraction closes that gap, and RiskInMind's loan application and bank statement analysis products are built for structured extraction from application files.
  2. Determine coverage. Apply the Regulation C coverage rules and exclusions to each application, and store the reason. Where it goes wrong: undocumented exceptions for commercial-purpose and temporary-financing loans.
  3. Map to the register layout. Convert internal fields to HMDA data points, with the right code lists, the "NA" and exempt values, and the partial-exemption flags. Where it goes wrong: code tables that were correct for last year's rule.
  4. Validate continuously. Run the same checks the filing platform will run, on every record, as it is created or changed. The next section covers this in detail.
  5. Record within the quarterly window. Post final-action records to the register on a schedule that beats the 30-day limit by a comfortable margin.
  6. Test in the sandbox. The CFPB offers a Beta Filing Platform for testing data, reviewing and resolving edits and correcting formatting errors before the official filing period, and the Bureau does not retain the data. A file that has passed your internal edits should arrive there clean.
  7. Certify and submit. The authorized representative certifies, and the file is submitted through the HMDA Platform. This step belongs to the filer and to the regulator's system. Automation prepares the evidence; it does not sign for you.
  8. Retain and resubmit if needed. Keep the filed register for at least three years and keep the correction trail, since resubmission windows run three years from January 1 of the following year.

Validation and exception routing: the part examiners read

The 2026 Filing Instructions Guide requires filers to address all edits before submitting, and the HMDA Platform lets you view and download edit reports. It also states that there are no significant changes to the 2026 submission process or data fields versus 2025. The submission is a pipe-delimited .txt file through the web-based platform, and legacy methods such as email, diskette and paper are not permitted. The same filing FAQ points to a File Format Verification Tool that checks pipe-delimiting, field count and integer formatting, and to the Guide's sections on syntactical, validity, quality and macro quality edits.

A good automated check layer mirrors those four edit families:

  • Syntactical and validity checks run on every record as it is saved: format, length, allowed codes.
  • Quality checks flag values that are valid but suspicious, such as an income far out of line with the loan amount. These need a human decision, so they should go to a queue, not be silently overwritten.
  • Macro quality checks look at the whole register, for example the ratio of denials to originations. They belong in a monthly or quarterly review rather than at record level.

Exception routing is what turns a check into a control. Each flagged record should carry an owner, a status, the corrected value, a reason and a timestamp. When an examiner compares the LAR with the loan files, that history is the explanation for any difference. The FFIEC examiner guidelines direct examiners to compare LAR data against loan and application files and have the institution correct and resubmit when differences are unexplained.

The error thresholds that set your testing standard

The CFPB's resubmission guidelines are the number to design against. Institutions with fewer than 100,000 LAR entries are expected to correct and resubmit when 10% or more of a sampled LAR contains errors. For institutions with 100,000 or more entries the threshold is 4% of the sample, or below 2% in an individual field if errors prevent accurate analysis of lending. Examination teams draw a random sample sized to the institution's LAR, and can begin with a smaller streamlined sample, completing the full sample once errors exceed the streamlined threshold.

A worked example: a credit union reports 1,200 LAR entries. If an examiner's sample finds errors in 10% of it, the register is a resubmission candidate. If your own pre-submission audit samples the same way and finds 3%, you have a correction list and a written finding, instead of a surprise. Running that sample every quarter, not once a year, is what makes it a control. Automation helps because it can draw the sample, compare each sampled record with source documents and log the finding without anyone assembling files by hand.

The error thresholds that set your testing standard

Controls and governance around the automation

The CFPB's HMDA compliance bulletin describes what a sound program contains: policies, procedures and internal controls; regular pre-submission audits with transactional analysis and written findings; review of regulatory changes; reporting systems suited to lending volume; assigned staff responsibilities; periodic training; corrective action for prior deficiencies; and board and management oversight. It also expects the compliance management system to be calibrated to each institution.

When you add automation, each of those items gets a concrete counterpart:

Program elementWhat the automated version looks like
Policies and proceduresDocumented rules in the system, versioned when Regulation C or the Filing Instructions Guide changes
Pre-submission auditScheduled sampling against source documents with written findings stored automatically
Regulatory change reviewAnnual update of thresholds, code lists and edits, with a recorded approval
Staff responsibilityNamed owners on every exception and every sign-off
TrainingRole-based access and a log of who corrected what
Board oversightDashboard of error rates, open exceptions and filing readiness

The Federal Reserve's CA 21-17 letter issued revised HMDA examination procedures in December 2021 that include a compliance management system review alongside the transaction testing guidelines. An examiner will therefore look at how the process is governed as well as at the numbers. A system that produces a clean file but cannot show who approved a rule change or resolved a flag will not satisfy that review. For a broader view of how testing programs fit together, see our piece on the role of compliance testing programs.

Security belongs in the same conversation, since the register contains applicant demographic data. Ask where the data is processed, who can see it and whether the audit trail is complete. Our overview of banking data residency and GLBA and FFIEC rules covers how to turn those expectations into exam evidence.

Filing calendar: the recurring pattern

The annual LAR is submitted electronically between January 1 and March 1 of the following year, certified by an authorized representative with knowledge of the data, and a copy is retained for at least three years, per Regulation C as summarized in the FDIC manual. When March 1 falls on a weekend, the deadline moves to the next Monday, which is why calendar-year 2025 data was due March 2, 2026.

Quarterly reporting applies only to institutions that reported at least 60,000 applications and covered loans, excluding purchased loans, in the preceding year, according to the CFPB's data collection timelines. Those filers report within 60 calendar days after each quarter-end, except the fourth quarter, which is filed with the annual LAR, and the supplemental guide for quarterly filers allows discrete or cumulative submissions and corrections to prior-quarter records in a cumulative filing. Most credit unions are annual filers, but the 30-day recording rule applies to everyone. Check the HMDA Platform's timelines page each year for the exact filing periods.

HMDA annual and quarterly reporting timeline

How to evaluate an HMDA automation tool

Use criteria a risk technology committee can score:

  • Source integration. Does it read from your loan origination system and documents, or does it depend on exports?
  • Edit coverage. Does it run syntactical, validity, quality and macro checks equivalent to the current Filing Instructions Guide, and update them yearly?
  • Exception workflow. Are owners, reasons and timestamps captured by default?
  • Audit trail. Can you reproduce, for any record, what was entered, changed, by whom and why?
  • Security posture. Does it handle applicant data inside controls your examiner will accept?
  • Speed and scale. Can it validate in real time during origination, not in an overnight batch?
  • Adjacent coverage. HMDA data feeds fair lending and CRA work. A platform that serves those too avoids a second data pipeline. See our guides on fair lending monitoring and CRA compliance analytics.

What RiskInMind brings to HMDA workflows

RiskInMind is an AI risk management platform built for credit unions, community and regional banks and lenders. It covers regulatory compliance monitoring, regulatory reporting automation, audit-ready reporting, loan application processing and document analysis, with SOC 2® certification, bank-grade security and real-time processing with response times under half a second. Its regulatory agent, Erina, is the compliance automation component, and the Loan Application product handles intake. Plans are Starter, Professional and Enterprise, and you can see them on the pricing page. For a deeper look at the regulatory side of the filing, read our HMDA reporting guide and our note on the role of automation in regulatory reporting.

Riskinmind

If your team is mapping its HMDA process this year, the RiskInMind platform is the place to start a demo conversation.

Our perspective

HMDA is a data quality problem wearing a compliance costume. Institutions rarely fail because they misunderstand Regulation C. They fail because the data was keyed three times, the exception was fixed in a spreadsheet nobody can find, and the person who made the call has moved on.

That is why we build around the audit trail. Our platform is designed for credit unions and community banks that need enterprise-grade controls without enterprise-sized teams: specialized AI agents coordinated by a central director, real-time processing, and granular records of every action. When validation happens as the loan moves, the annual filing becomes a certification of work already evidenced.

Our view is simple: automate capture, validation and documentation, and keep people for judgment and sign-off. The examiner's question is never only whether the file passed the edits. It is whether you can show how the data got there.

Raj

FAQ

Does automation remove the need for an authorized representative to certify the HMDA filing?

No. Regulation C requires the annual LAR to be certified by an authorized representative with knowledge of the data. Automation supplies validated data and the evidence behind it, which makes that certification better supported, but the signature stays with a person.

How often should we test HMDA data before filing?

Quarterly is a practical rhythm, matching the 30-day recording rule. The CFPB's compliance bulletin points to regular pre-submission audits with transactional analysis and written findings. Sampling against the 10% and 4% resubmission thresholds gives you a measurable standard.

Which edits must be cleared before the HMDA Platform accepts a submission?

The Filing Instructions Guide requires filers to address all edits before submitting, covering syntactical, validity, quality and macro quality edits. The platform produces edit reports you can view and download, and the beta sandbox lets you rehearse the process before the official filing period.

Do credit unions have to file quarterly?

Only institutions that reported at least 60,000 applications and covered loans, excluding purchased loans, in the preceding year file quarterly. Other covered institutions file once a year, but all must record data on the register within 30 calendar days after the end of the quarter of final action.

Sources

Recommended

HMDA reporting automation