What Your HMDA LAR Looks Like to an Examiner: A Field-by-Field Risk Map

Published September 8, 2026 · Sources: the FFIEC Filing Instructions Guide for HMDA data collected in 2026 for every field number, name and code value, and the Interagency Fair Lending Examination Procedures (FFIEC, August 2009) for every examiner test. Both retrieved and parsed on September 8, 2026; see methodology and sources.

What this is. Your LAR has 110 fields and they do not carry equal exam risk. A handful are the axis of every disparity test an examiner runs. A larger group decides which loans get compared with which. Most of the rest can be wrong without changing a single analytical conclusion — they will fail an edit check and cost you a resubmission, which is a different kind of problem.

This page maps each field to the test it feeds. The field definitions are public and the examination procedures are public; what does not exist as a single reference is the join between them. For each field: which test it feeds, what a bad value looks like, and what happens downstream when it is wrong.

Tests are named by their risk factor numbers from the Interagency Procedures — P for pricing, U for underwriting, R for redlining, S for steering, O for overt indicators, C for compliance programme. Those numbers are the examiner’s own vocabulary, and quoting one settles an internal argument faster than describing it.

How examiners actually use the LAR

The LAR is not the examination. It is the screening layer that decides where the examination goes. The Interagency Procedures list it among the documents requested at scoping — “Home Mortgage Disclosure Act–Loan Application Register (HMDA–LAR) or loan registers and lists of declined applications” — and then use it in Part I, Step Four to identify the risk factors that set the scope of everything after.

That is why a defective field is expensive in a way that is easy to underestimate. It does not produce a wrong answer at the end of an examination. It produces the wrong examination: a risk factor that fires when it should not, or fails to fire when it should, and a scope built on that reading.

Two structural points before the field list. First, the prohibited basis fields are not one test among many, they are the axis of all of them. Every risk factor marked with an asterisk in the procedures is a comparison by prohibited basis characteristic, so ethnicity, race, sex and age are inputs to pricing, underwriting, steering and redlining alike. Second, several risk factors are computed within a category rather than across the register: U1 is measured “especially within income categories” and U2 “especially within denial reason groups.” A field you might file as merely descriptive is doing the stratifying.

What this page is not. A clean LAR is not a defence, and this map is not a compliance opinion. The Interagency Procedures are explicit that the HMDA data alone cannot be used to determine whether a lender is complying with fair lending laws, because the data omit “many potential determinants of loan application and pricing decisions, such as the applicant’s credit history, debt-to-income ratio, the loan-to-value ratio, and other considerations.” The same limit runs in the other direction: correct data does not make a disparity go away, it means the disparity you are looking at is real. For what the data cannot settle, see what regression can and cannot prove in a fair lending review.

The prohibited basis fields: the axis of every test

The Interagency Procedures open by listing what a lender may not discriminate on. Under ECOA: race or color, religion, national origin, sex, marital status, age, receipt of income from a public assistance program, and the good-faith exercise of a Consumer Credit Protection Act right. The Fair Housing Act adds familial status and handicap. Of all of those, the LAR carries four: ethnicity, race, sex and age. They are the dimension along which every asterisked risk factor is computed, which makes them the highest-consequence fields on the register.

Ethnicity of applicant and co-applicant — fields 19–32

Feeds: Every asterisked risk factor: U1–U3, P4, P6, P7, S4–S6, R1–R4.
Bad value: Five ordered slots per applicant plus a free-form field for other Hispanic or Latino origin, so the usual defects are a single aggregated value where multiple selections were made, and a refusal recorded as absence rather than as the code that means refusal.
Consequence: The population being compared is misconstituted. A disparity computed on a mis-assigned denominator is not a smaller finding, it is a different one, and it will not reconcile with the examiner’s recomputation.

Race of applicant and co-applicant — fields 33–50

Feeds: The same set as ethnicity, and it is the characteristic R1–R4 use when comparing areas by their concentration of minority group residents.
Bad value: Five slots per applicant with three conditional free-form fields (enrolled or principal tribe, other Asian, other Pacific Islander). Collapsing a multi-race selection to one code is the frequent defect.
Consequence: As above, and it additionally degrades any proxy work built on the reported values — see where BISG proxy methodology fails.

Sex of applicant and co-applicant — fields 51–54

Feeds: The asterisked risk factors, on the sex dimension.
Bad value: A joint application filed with only the primary applicant populated, which silently removes the co-applicant from every comparison.
Consequence: An under-counted prohibited basis population, which biases a disparity measure in a direction you cannot predict from the outside.

Age of applicant and co-applicant — fields 55–56

Feeds: Age is a prohibited basis under ECOA where the applicant has the capacity to contract. It also bears on O3, which flags “including variables in a credit scoring system that constitute a basis or factor prohibited by Regulation B” and directs the examiner to the automated underwriting and credit scoring appendix when a scoring system scores age.
Bad value: Derived from an application date and a date of birth that disagree, or defaulted where the applicant is not a natural person.
Consequence: An age distribution that invites an O3 look at your scorecard is an expensive thing to invite by accident.

Collected on the basis of visual observation or surname — fields 31, 32, 49, 50, 53, 54

Feeds: Data reliability rather than a disparity test. Bears on C2, “prohibited basis monitoring information required by applicable laws and regulations is nonexistent or incomplete,” and C3, where “data and/or recordkeeping problems compromised reliability of previous examination reviews.”
Bad value: Set inconsistently with the demographic fields they describe — observed data flagged as self-reported, or the reverse.
Consequence: These six flags are how an examiner judges whether your demographic data can be relied on at all. A C3 finding is worse than any single disparity, because it puts the whole submission in question.

Fields that drive pricing analysis

The pricing risk factors are P1 to P7. Two of them name the data directly. P4 is “substantial disparities among prices being quoted or charged to applicants who differ as to their monitored prohibited basis characteristics.” P6 is narrower, and it is the one computed straight off your filing:

“In mortgage pricing, disparities in the incidence or rate spreads of higher-priced lending by prohibited basis characteristics as reported in the HMDA data.” Interagency Fair Lending Examination Procedures, Part I, Step Four, risk factor P6

P1 and P2 both define price as “interest rate, fees and points,” which is what makes the whole pricing block of the LAR exam-relevant rather than rate spread alone. P2 is about discretion: it fires where loan officers or brokers “are permitted to deviate from those rates and fees without clear and objective criteria.”

Rate Spread — field 59

Feeds: P6 directly, and R4, which counts originations of higher-priced loans by area minority concentration.
Bad value: Reported where it should be NA or the reverse; computed against the wrong average prime offer rate date; or carrying a rounding or sign convention that differs from the calculation the regulator performs.
Consequence: This is the single field most likely to produce a pricing finding you cannot reproduce. P6 is measured on incidence and on spread, so one systematic error moves both halves at once.

Interest Rate — field 78

Feeds: P1 and P2, which name the interest rate as a price component and are concerned with discretion over it.
Bad value: The note rate where the fully-indexed rate was required, or a rate captured before a final re-lock.
Consequence: Feeds a terms-and-conditions comparison directly. Where P2 asks whether pricing discretion is governed, your own field is the evidence either way.

Total Loan Costs — field 73

Feeds: P1 to P4, as the fee component of price.
Bad value: Populated for a transaction where total points and fees applies instead, or taken from a superseded disclosure rather than the final one.
Consequence: Fee disparity is where discretion usually surfaces, because rate sheets are typically governed and fee waivers frequently are not.

Total Points and Fees — field 74

Feeds: P1 to P4, and it interacts with HOEPA status.
Bad value: Reported alongside total loan costs when only one of the two applies to the transaction type.
Consequence: Two mutually exclusive fields both populated is a validity failure, and it also makes your fee distribution incomparable across the register.

Origination Charges — field 75

Feeds: P1 and P2. This is the component most directly subject to loan officer discretion, which is what P2 exists to detect.
Bad value: Reported net of a credit rather than gross, so two files with identical economics report different charges.
Consequence: Creates or conceals an overage pattern. P1 specifically pairs financial incentives with broad discretion, and this field is where that pairing shows.

Discount Points — field 76

Feeds: P1 to P4 as a points component, and it is the field that explains an otherwise anomalous rate.
Bad value: Omitted, which makes a bought-down rate look like preferential pricing.
Consequence: Removes the legitimate explanation for a rate difference, turning a priced choice into an apparent disparity.

Lender Credits — field 77

Feeds: P1 to P4. The mirror image of discount points.
Bad value: Omitted where a credit offset a higher rate.
Consequence: The same failure as discount points in reverse: a rate that looks unexplained until the credit is visible.

HOEPA Status — field 60

Feeds: The higher-priced lending population P6 measures. Code 1 is high-cost mortgage, 2 is not a high-cost mortgage, 3 is not applicable.
Bad value: Code 3 used as a default rather than as a determination.
Consequence: Mis-sizes the high-cost population, which is the numerator of an incidence test.

Prepayment Penalty Term — field 79

Feeds: S4, which names prepayment penalties among the “products and features that have potentially negative consequences for applicants,” and R4, which counts such loans by area.
Bad value: Left blank rather than carrying the code for no penalty, which makes the feature unmeasurable rather than absent.
Consequence: S4 and R4 both compare the incidence of the feature, by prohibited basis and by geography. An unmeasurable feature does not score as zero, it scores as unknown.

Introductory Rate Period — field 83

Feeds: S2, which names payment option adjustable rate mortgages among the non-traditional products officers may have an incentive to place applicants into, and S4.
Bad value: Populated on a fixed-rate loan, or omitted on an adjustable one.
Consequence: Misclassifies the loan as traditional or non-traditional, which is the exact axis S2 and S4 test.

Balloon Payment, Interest-Only Payments, Negative Amortization, Other Non-Amortizing Features — fields 84–87

Feeds: S2 and S4. S2 names “negative amortization, ‘interest only’, ‘payment option’ adjustable rate mortgages” explicitly; R4 counts loans with potentially negative consequences by area.
Bad value: Derived from one internal flag rather than four independent determinations.
Consequence: Steering analysis works on feature incidence by group. Four fields collapsed into one cannot support it, and the risk factor is then evaluated on your worst-documented field.

Fields that drive geography and redlining analysis

The redlining risk factors run R1 to R12, and the first four are computed from the LAR against the minority composition of the areas in your market. R1 is the broadest:

“Significant differences, as revealed in HMDA data, in the number of applications received, withdrawn, approved not accepted, and closed for incompleteness or loans originated in those areas in the institution’s market that have relatively high concentrations of minority group residents compared with areas with relatively low concentrations of minority residents.” Interagency Fair Lending Examination Procedures, Part I, Step Four, risk factor R1

Read that list of actions against the Action Taken codes and the mapping is exact: R1 is computed from codes 1, 3, 4 and 5 tabulated by tract. R2 compares approval and denial rates by area, R3 compares denial rates “based on insufficient collateral” by area, and R4 compares higher-priced and negative-feature originations by area. R6 is different in kind: it looks for “explicit demarcation of credit product markets that excludes MSAs, political subdivisions, census tracts, or other geographic areas.”

Census Tract — field 18

Feeds: R1, R2, R3, R4 and R6. This is the join key for every geographic test: without it there is no area to compare against.
Bad value: A tract that is valid in isolation but does not belong to the reported county — the first five digits of the tract must match the reported five-digit county FIPS code. Also a tract from the wrong census vintage.
Consequence: The highest-leverage single defect on the register. A misassigned tract moves an application from one side of a redlining comparison to the other, and it does so silently, because both the wrong tract and the right one are real places. Assignment method matters here, which is why we publish how geocoding accuracy is measured and distinguish it from match rate.

County — field 17

Feeds: R6, and it is the consistency check on census tract.
Bad value: Reported NA where a tract is reported, or a county that contradicts the tract prefix.
Consequence: A county and tract that disagree is a validity failure, and until it is resolved the record cannot be placed on a map at all.

Street Address, City, State, ZIP Code — fields 13–16

Feeds: Not tested directly, but they are the inputs from which tract and county are derived, so they feed R1 to R4 and R6 through those fields.
Bad value: A mailing address rather than the property address; a PO box; a unit number carried into the street line in a way the geocoder cannot parse.
Consequence: These four are where a tract error is usually born. An address that geocodes to a centroid rather than a rooftop produces a plausible tract that is wrong, and nothing downstream flags it.

Action Taken, read geographically — field 11

Feeds: R1 and R2. R1 names four of the eight codes: originated (1), denied (3), withdrawn by applicant (4) and file closed for incompleteness (5).
Bad value: Withdrawn used where the file was actually closed for incompleteness, or either used where the application was denied.
Consequence: R1 counts these separately and by area, so substituting one for another changes the shape of the geographic pattern rather than just a total. The procedures also note that a withdrawal recorded after a prohibited basis applicant has received an offer of credit including pricing terms is itself a Regulation C reporting error.

Reason for Denial, read geographically — fields 68–71

Feeds: R3, which is specifically about “denial rates based on insufficient collateral” by area. Code 4 is Collateral.
Bad value: Code 9, Other, used where collateral was the operative reason; or only the first of up to four reasons populated.
Consequence: R3 is the one redlining factor that turns on a single code value. Under-reporting collateral denials suppresses the factor; over-reporting invents it.

Fields that drive underwriting disparity analysis

The underwriting risk factors are U1 to U9. Three are computed from the LAR, and each names the stratification it is computed within:

“U1. Substantial disparities among the approval/denial rates for applicants by monitored prohibited basis characteristic (especially within income categories)”
“U2. Substantial disparities among the application processing times for applicants by monitored prohibited basis characteristic (especially within denial reason groups)”
“U3. Substantially higher proportion of withdrawn/incomplete applications from prohibited basis group applicants than from other applicants” Interagency Fair Lending Examination Procedures, Part I, Step Four, risk factors U1–U3

Action Taken — field 11

Feeds: U1 as the numerator and denominator of the approval and denial rates, and U3 through codes 4 and 5.
Bad value: Code 2, approved but not accepted, used for a denial the applicant did not contest; or preapproval codes 7 and 8 mixed into the ordinary application population.
Consequence: U1 is the most frequently cited underwriting factor and it is arithmetic on this one field. An error here does not shade a finding, it creates or erases one.

Application Date and Action Taken Date — fields 4 and 12

Feeds: U2. Processing time is the interval between these two fields, and there is no other source for it.
Bad value: An action taken date earlier than the application date; a system date substituted for the decision date; or the date of the final adverse action notice rather than the decision.
Consequence: U2 is a disparity in duration by prohibited basis, so a systematic date convention that varies by channel or by officer will read as a processing-time disparity even when decisions were even-handed.

Reason for Denial — fields 68–71

Feeds: U2, which is measured within denial reason groups. Codes are debt-to-income ratio, employment history, credit history, collateral, insufficient cash, unverifiable information, incomplete application, mortgage insurance denied, other, and not applicable.
Bad value: Defaulting to code 9, Other. It is the honest answer to almost nothing and it removes the record from every within-reason comparison.
Consequence: An Other-heavy denial profile does not look neutral, it looks unexplained, and it also collapses the strata U2 depends on. Up to four reasons may be reported; reporting one where four applied loses the rest permanently.

Income — field 57

Feeds: U1, which is measured especially within income categories. It is the primary stratifier of the approval and denial comparison.
Bad value: Reported in dollars rather than thousands. A file carrying the full figure passes a format check and then sits three orders of magnitude away from its true band.
Consequence: Puts the application in the wrong income category, so it is compared against the wrong control group. This is the defect most likely to produce a disparity that dissolves the moment units are corrected.

Credit Score of applicant and co-applicant — fields 62–63

Feeds: U7, “relatively high percentages of either exceptions to underwriting criteria or overrides of credit score cutoffs.” It is also the control variable the Interagency Procedures name first when explaining what the HMDA data omit.
Bad value: Reserved codes used loosely: 7777 for a score not reported, 8888 for not applicable, 1111 for exempt. Choosing between them by habit rather than by fact.
Consequence: Without a usable score there is no creditworthiness control, so a comparison of two applications becomes a comparison of two rows. U7 is then evaluated against an override population you cannot describe.

Name and Version of Credit Scoring Model — fields 64–67

Feeds: O3, which flags a scoring system containing a factor prohibited by Regulation B, and it qualifies every score in fields 62 and 63.
Bad value: One model named across a register where two were in use, with the conditional free-form field left empty where code 8 was reported.
Consequence: Two scores from different models are not comparable numbers. Controlling on them as if they were is a methodological error an examiner will find, and it also obscures which scorecard an O3 review should look at.

Debt-to-Income Ratio — field 80

Feeds: Named by the Interagency Procedures as one of the determinants the HMDA data omit, which makes it a control in any comparison that uses it. Also the first denial reason code.
Bad value: A ratio computed on a different income definition from the one reported in field 57, so the two fields describe different applicants.
Consequence: A DTI that cannot be reconciled with reported income undermines both fields, and it is exactly the inconsistency a comparative file review surfaces.

Combined Loan-to-Value Ratio — field 81

Feeds: The loan-to-value control the procedures name, and it interacts with R3, the collateral-denial factor.
Bad value: Computed on the first lien only where subordinate financing existed, which is what makes it combined.
Consequence: Understates risk on exactly the files where a collateral denial would otherwise be explainable.

Property Value — field 88

Feeds: The denominator of CLTV, and it bears on R3.
Bad value: Purchase price substituted for appraised value, or the reverse, inconsistently across the register.
Consequence: Propagates into CLTV and therefore into any underwriting comparison that controls on it.

Loan Amount — field 10

Feeds: Named among the omitted determinants, and it is a control in pricing and underwriting comparisons alike.
Bad value: The approved amount where the applied-for amount was required, which quietly conflates request with outcome.
Consequence: A loan amount that reflects the decision cannot be used as a control for the decision. It also feeds the loan-amount-to-property-value relationship a quality edit checks.

Automated Underwriting System and Result — fields 96–107

Feeds: O3 and U5 to U7, the exception and override factors. Five AUS slots and five result slots, each with a conditional free-form field.
Bad value: The final system and result recorded where several runs occurred, so a recommendation that changed between runs is invisible.
Consequence: An override is the difference between what the system recommended and what was decided. If only the last run is filed, the override population is undercounted, which is precisely what U7 measures.

Fields that decide which loans get compared with which

This group is the one most often filed as merely descriptive, and it is the one that decides whether any of the tests above are computed on a sensible population. Part I, Step Three of the procedures tells the examiner how to divide the book before analysing it:

“Divide home mortgage loans into the following groupings: home purchase, home improvement, and refinancings. Subdivide those three groups further if an institution does a significant number of any of the following types or forms of residential lending, and consider them separately: Government-insured loans; Mobile home or manufactured housing loans; Wholesale, indirect and brokered loans; Portfolio lending…” Interagency Fair Lending Examination Procedures, Part I, Step Three

Every one of those groupings is read off a LAR field. Get one wrong and the disparity tests run on a population the examiner would not have assembled, which means your numbers and theirs will not agree even when both are computed correctly.

Loan Purpose — field 6

Feeds: The primary Step Three division. Codes are home purchase (1), home improvement (2), refinancing (31), cash-out refinancing (32), other purpose (4) and not applicable (5).
Bad value: Refinancing and cash-out refinancing used interchangeably. They are separate codes because they are different products with different pricing.
Consequence: Puts the application in the wrong comparison group at the very first cut, so every test downstream compares it against the wrong control population.

Loan Type — field 5

Feeds: The government-insured subdivision. Codes are conventional (1), FHA (2), VA (3), and RHS or FSA (4).
Bad value: Conventional as a default where the guarantee was not captured in the origination system.
Consequence: Government-insured loans price and underwrite differently. Mixing them into the conventional population manufactures disparities in both directions.

Construction Method — field 8

Feeds: The manufactured housing subdivision. Codes are site-built (1) and manufactured home (2).
Bad value: Site-built recorded for a manufactured home on owned land.
Consequence: Manufactured housing carries its own pricing and collateral profile, and Step Three says to consider it separately. Misclassified, it distorts whichever population absorbs it.

Submission of Application and Initially Payable to Your Institution — fields 93 and 94

Feeds: The wholesale, indirect and brokered subdivision, and they are how S6 identifies lending channels: “significant differences in the percentage of prohibited basis applicants in one of the lending channels compared to the percentage… of the other lending channel.”
Bad value: Both defaulted to code 1 for a broker-originated loan.
Consequence: Retail and broker channels typically have different discretion and different pricing outcomes, which is why S6 compares them. Collapse them and the comparison cannot be run; worse, broker discretion is attributed to your retail operation.

Occupancy Type — field 9

Feeds: Population definition. Codes are principal residence (1), second residence (2), investment property (3).
Bad value: Principal residence as a default for an investor purchase.
Consequence: Investment lending is priced differently and is not covered by the Fair Housing Act in the same way as owner-occupied lending, so an occupancy error changes both the comparison group and the statute in play.

Lien Status — field 61

Feeds: Population definition for pricing, since first and subordinate liens are not comparably priced. Codes are secured by a first lien (1) and secured by a subordinate lien (2).
Bad value: First lien reported for a piggyback second.
Consequence: Subordinate liens carry higher rates for reasons unrelated to the applicant. Filed as firsts, they inflate the high-priced incidence P6 measures.

Total Units — field 91

Feeds: Population definition, separating one-to-four family from multifamily. It also participates in a quality edit relating units, income and loan amount.
Bad value: A default of 1 where the property is a duplex or larger.
Consequence: Multifamily lending in a one-to-four family population distorts loan amount and income distributions, and it trips the quality edit that reads those three fields together.

Reverse Mortgage, Open-End Line of Credit, Business or Commercial Purpose — fields 108–110

Feeds: Population definition. Each identifies a product whose pricing and underwriting are not comparable to a standard closed-end forward mortgage.
Bad value: All three defaulted to the negative code, which is usually right and is therefore rarely checked.
Consequence: A reverse mortgage or a business-purpose loan sitting in the general population is a comparability failure that is invisible until someone asks why a small cluster of files behaves oddly.

Fields that carry filing risk but no analytical weight

These fields will not change a disparity finding. They will stop a submission, or make a record impossible to trace, which is a different and more immediate cost. The distinction is worth being explicit about, because effort spent perfecting them is not effort spent on exam risk — and effort spent ignoring them is how a filing season goes long.

Universal Loan Identifier — field 3

Feeds: No disparity test. It is the key by which a LAR row is matched back to a loan file, so it is what makes any comparative file review possible at all.
Bad value: Duplicated across records, or regenerated between submission and resubmission so the two cannot be reconciled.
Consequence: A duplicate ULI is a hard submission failure. A ULI that changes between filings means an examiner cannot follow a record across years, which reads as a recordkeeping problem under C3.

Legal Entity Identifier — field 2

Feeds: No disparity test. It attributes the record to the filing institution.
Bad value: A parent LEI on a subsidiary’s records, or an expired registration.
Consequence: Misattributes lending between affiliates, which matters because S5 compares an institution’s prohibited basis percentages against those of its subsidiaries. Wrong LEI, wrong comparison.

Preapproval — field 7

Feeds: Population definition rather than a test of its own, and it governs whether Action Taken codes 7 and 8 are valid on the record.
Bad value: Preapproval requested recorded on a purpose for which the programme does not exist.
Consequence: A validity failure, and preapproval requests mixed into the application population change denominators in U1.

Type of Purchaser — field 58

Feeds: No fair lending disparity test directly. It describes secondary market disposition.
Bad value: Not applicable reported for a loan sold in the same calendar year.
Consequence: Filing accuracy. Worth noting only because a purchaser-driven pricing policy is a legitimate explanation for a rate pattern, and this is the field that evidences it.

Loan Term, Multifamily Affordable Units, Manufactured Home property fields, Mortgage Loan Originator NMLSR Identifier — fields 82, 92, 89–90, 95

Feeds: No disparity test. Loan term is a pricing covariate in practice; the NMLSR identifier is how a pattern gets attributed to an individual originator once one is found.
Bad value: Term in years where months are required; the manufactured home fields populated on a site-built property; a blank NMLSR identifier on a broker-originated loan.
Consequence: Mostly edit-check exposure. The exception is the NMLSR identifier: without it, a pricing disparity concentrated in one originator’s book cannot be localised, and an institution-wide finding is harder to answer than an individual one.

Administrative and free-text fields

Grouped rather than enumerated, because none of them feeds an examiner test and the failure mode is the same for all: a validity or conditional-logic failure that costs a resubmission.

Fields	What they are	Failure mode
1	Record Identifier, always 2 on a LAR row	Anything other than 2 is a syntactical failure that stops the file
24, 30	Free-form text for other Hispanic or Latino ethnicity	Populated without the code that requires it, or omitted when the code is present
38–40, 46–48	Free-form text for enrolled or principal tribe, other Asian, other Pacific Islander	Same conditional mismatch, six times over
65, 67	Conditional free-form text for credit scoring model code 8	Model reported as other with no name given
72	Conditional free-form text for denial reason code 9	Other selected with the reason left unstated — the commonest of these, and the one that also costs you U2
101, 107	Conditional free-form text for AUS code 5 and AUS result code 16	Other selected without the system or result named

The pattern across all of them is one rule: a free-form field is required exactly when its partner code says other, and prohibited otherwise. Treating them as optional notes rather than as conditional requirements is what produces the resubmission.

Methodology and sources

How this map was built. Two published documents, read directly and parsed rather than summarised, on 8 September 2026.

The field numbers, names and code values come from the FFIEC Filing Instructions Guide for HMDA data collected in 2026. All 110 LAR fields were parsed from the field order table, and every one of them is accounted for on this page — the analytically significant fields individually, the conditional free-text fields grouped. One caution for anyone repeating the extraction: that page carries two numbered tables, and the LAR field order is the second one. The first is the transmittal sheet.

The examiner tests come from the Interagency Fair Lending Examination Procedures (FFIEC, August 2009, the current edition, which replaced the March 1994 procedures). Every mapping on this page is tied to a numbered risk factor or a named step in Part I of that document, and the risk factor identifiers used here were verified against the full document rather than assumed: the ranges are C1–C7 for compliance programme, M1–M7 for marketing, O1–O5 for overt indicators, P1–P7 for pricing, R1–R12 for redlining, S1–S8 for steering and U1–U9 for underwriting.

What is ours and what is not. The field definitions are the FFIEC’s. The risk factors are the FFIEC’s, and they are quoted rather than paraphrased wherever the exact wording carries the point. The join between them — which field feeds which factor, what a defective value looks like in practice, and what it costs downstream — is ours, and it is editorial judgement rather than a published mapping. Where a field plausibly feeds a test but the procedures do not say so, this page says the field is a control or a population definition rather than claiming a test it does not name. No figure, threshold or code value on this page comes from anywhere but the two documents above.

Filing Instructions Guide for HMDA data collected in 2026. Federal Financial Institutions Examination Council and Consumer Financial Protection Bureau. ffiec.cfpb.gov/documentation/fig/2026/overview. Field order, names, valid values and code descriptions read on 8 September 2026.
Interagency Fair Lending Examination Procedures. Federal Financial Institutions Examination Council, August 2009. Read from the Federal Reserve’s Consumer Compliance Handbook reprint (page footers stamped 11/09) on 8 September 2026. Risk factors and examination steps quoted from Part I.

Related reference pages on this site, each sourced the same way: what regression can and cannot prove, comparative file review versus matched-pair testing, BISG proxy methodology and its limits, REMA versus assessment area, and how to fix the ten most common HMDA edit check errors, which covers the edit codes behind the filing failures described above.

How should you cite this page?

Methodology. Every field number, name and code value on this page is read from the FFIEC Filing Instructions Guide for HMDA data collected in 2026, and every examiner test is tied to a numbered risk factor or named step in the Interagency Fair Lending Examination Procedures (FFIEC, August 2009). Both were read on 8 September 2026 and both are named and linked in methodology and sources above. Nothing here is derived from a secondary summary, and no threshold, code or figure appears that is not in one of those two documents. Last updated September 8, 2026.

Every field heading carries a stable id, so you can link to one field rather than to the page — for example #field-census-tract or #field-rate-spread.

Plain text

RATA Associates. "What Your HMDA LAR Looks Like to an Examiner: A Field-by-Field Risk Map." Published September 8, 2026; updated September 8, 2026. https://rataassociates.com/hmda-lar-examiner-risk-map/

HTML link

<a href="https://rataassociates.com/hmda-lar-examiner-risk-map/">What Your HMDA LAR Looks Like to an Examiner: A Field-by-Field Risk Map</a> &mdash; RATA Associates

You may republish these figures with attribution and a link to https://rataassociates.com/hmda-lar-examiner-risk-map/. If you believe a mapping or a quotation here is wrong, tell us — we would rather fix it than be cited incorrectly.