Summary
Most HMDA errors don't come from typos. They come from reporting the wrong version of a number: the income from a later refresh, the second appraisal, the DTI from a revised debt schedule. Regulation C asks for the value the bank "relied on" in the credit decision for income, credit score, debt-to-income ratio (DTI), combined loan-to-value ratio (CLTV) and property value (12 CFR 1003.4), and the newest value in the system often isn't that one. This guide covers how to reconcile those fields against the decision record before filing, and how to log what you can't resolve.
For HMDA and fair lending compliance teams at U.S. banks reviewing their Loan/Application Register (LAR) before submission.
What Regulation C requires and what's bank practice
| Topic | What applies | Source |
|---|---|---|
| Income, DTI, CLTV, property value | The value relied on in the credit decision | 1003.4(a)(10), (23), (24), (28) |
| Credit score | The score relied on, even if it was only one factor, with the scoring model | 1003.4(a)(15) |
| Withdrawn or incomplete files | Report DTI, CLTV, property value and credit score as not applicable if no credit decision was made, even if the bank had calculated them | Official interpretations to 1003.4 |
| Recording | Within 30 calendar days after the end of the quarter in which final action was taken | 1003.4(f) |
| Filing | By March 1 after the collection year; quarterly too for banks that reported at least 60,000 covered loans and applications the year before | 1003.5(a) |
| Sampling method, exception register, correction approvals | Not set by the rule; this is how the bank runs its review | Bank procedures |
Before marking an optional field as missing, check whether the bank qualifies for a partial exemption. The exemption removes certain optional fields, not transactions, and never covers income or action taken (1003.3(d)).
Reconcile the LAR in five steps
1. Settle action taken first
Action taken decides which other fields apply, so start there. The distinctions that most often go wrong:
| Situation | Report as |
|---|---|
| Applicant expressly withdraws before a credit decision | Withdrawn |
| Conditional approval, applicant withdraws before meeting underwriting conditions | Withdrawn |
| Conditional approval, underwriting conditions not met, bank denies | Denied |
| Approved, applicant doesn't meet only customary closing conditions | Approved but not accepted |
| Bank sent a written incompleteness notice and got no response | File closed for incompleteness |
These come from the official interpretations of 1003.4(a)(8). Then set the action date from the record that shows the action, such as the denial notice or the written withdrawal, rather than whenever the system status changed.
2. Find the decision record for each field
For each sampled file, find the record that shows what the underwriter actually used: the income worksheet, the credit report and score selection, the DTI and CLTV calculations, and the valuation the decision was based on. When the bank got two valuations, it reports the one it relied on. Regulation C doesn't require a particular DTI or CLTV method, so keep the calculation with the file.
3. Compare the LAR with that record
Put the current LAR value next to the decision record value for each field. Report income in thousands, rounded to the nearest thousand with $500 rounding up, so $35,500 is reported as 36. A credit score still counts as relied on when the loan was denied for another reason.
If the decision record doesn't show which value was used, log the field as an open exception and request the record. Filling in the newest available value just hides the gap.
4. Correct a copy, not the source
Make approved corrections in a separate, versioned copy of the LAR, and never change source records to match it. Keep the original file, the exception register, the approvals and the corrected file together. Passing the FFIEC format checks confirms the file's structure, not its accuracy, so check each field's format against the Filing Instructions Guide for the collection year after the reconciliation is done.
5. Fix the cause before next year
Group exceptions by field and cause. If the same field fails for the same reason in the next cycle, the control didn't work. Examiners can direct changes to policies and procedures to prevent repeat errors, so record what you changed.
Worked example
Illustrative example. All figures are fictional.
A reviewer checks one originated loan and finds three fields where the file holds a newer value than the LAR.
| Field | Current LAR | Newer value in the file | Decision memo | Treatment |
|---|---|---|---|---|
| Income | 96 | $102,000 from a refresh after approval | $96,400 | Keep 96; the refresh wasn't relied on |
| DTI | 41.25 | 39.80 from a revised debt schedule | 41.25 | Keep 41.25; log the schedule as post-decision |
| Property value | 480000 | $495,000 from a second appraisal review | $480,000 | Keep 480000; the second review wasn't relied on |
All three LAR values are correct, but only because the decision memo shows what was used. If the memo were missing, each field would stay open until the reviewer found it. And if the file showed the applicant had withdrawn before any credit decision, action taken would change, and DTI, CLTV and property value would become not applicable.
Where HMDA reviews go wrong
The margin for error is small. Under the FFIEC's examiner testing guidelines, a bank with 501 to 100,000 LAR lines faces resubmission if examiners find 4 errors in any one field across a 79-file sample (FFIEC HMDA Examiner Transaction Testing Guidelines). Small differences, like income rounding or dates within three days, don't count toward that threshold, but they're tolerances for examiners, not a standard for your own corrections.
The guidelines single out withdrawn applications coded as denials. When that error is widespread enough to undermine fair lending analysis, examiners can require the action taken field to be corrected even if the bank is under the threshold. The other common failures are reporting the latest system value instead of the one relied on, and treating a clean format check as proof the data is right.
How Bretton fits in
HMDA is one of Bretton's ready-made workflows. Bretton checks mortgage application data for accuracy and internal consistency, gathers the source evidence behind each discrepancy, and prepares corrections with a documented review trail for your team to approve (Bretton HMDA). Your team still decides which value was relied on, approves each correction and certifies the filing.
Common questions
What does "relied on" mean for HMDA reporting?
It's the value the bank actually used in making the credit decision. If the file has a newer income figure or a second appraisal that wasn't used, the bank reports the earlier value it relied on.
How do you report income on the HMDA LAR?
In thousands of dollars, rounded to the nearest thousand, with $500 rounding up. $35,500 is reported as 36.
How many HMDA errors trigger resubmission?
It depends on LAR size. For a bank with 501 to 100,000 lines, examiners sample 79 files, and 4 errors in a field triggers resubmission.
Related guides
Sources
- CFPB, Regulation C, 12 CFR 1003.3, 1003.4 and 1003.5, with official interpretations
- FFIEC, HMDA Examiner Transaction Testing Guidelines
- FFIEC, 2026 HMDA Filing Instructions Guide
- Bretton, HMDA
Book a demo to see how Bretton reviews HMDA data before you file.
Discuss the evidence handoffs in your back office.
Talk to Bretton