Debt-to-income first, credit history second, incomplete files third. And the withdrawn applications that denials alone miss.
This study was prepared November 1, 2026, from 2025 Home Mortgage Disclosure Act (HMDA) data. It answers one question with original computations from public data: why are VA purchase loans actually denied?
The lead number
In calendar year 2025, 4,278 VA purchase loan denials listed “credit application incomplete” as the primary denial reason. That is 14.08% of the 30,378 VA purchase denials that carried a stated primary reason, more than double the incomplete share for FHA (5.17%), conventional (6.19%), or USDA (6.94%) loans. Incomplete files are not a footnote in the VA denial story. They are the third largest reason VA purchase applicants are denied, and the single most distinctive thing about VA denials compared with other loan programs.
Executive summary
Every number below comes from the 2025 Home Mortgage Disclosure Act (HMDA) loan-level data, purchase loans only, computed October 31, 2026. Denominators are stated with every percentage, because the denominator is what makes a percentage honest.
- VA purchase loans are denied least often. Of all VA purchase applicants, 6.61% were denied (Rate A: denied divided by all applications). Of applications that reached a yes or no decision, 8.42% were denied (Rate B: denied divided by originated plus denied). On both definitions, VA sits below FHA (10.60% / 13.06%), USDA (10.20% / 12.94%), and conventional (12.04% / 15.55%). In plain terms, about 1 in 15 VA purchase applicants was denied, or about 1 in 12 of the applications that reached a decision.
- Debt-to-income ratio is the top cited primary denial reason for VA, at 29.25% of VA denials with a stated primary reason (8,885 of 30,378). It is the top reason for every loan type, but its share is lowest for VA and highest for FHA (41.42%).
- Credit history is second for VA, at 23.02% of stated VA denials (6,994 of 30,378). “Credit history” is a bucket, not a diagnosis: under the adverse-action mapping it covers late payments, collections, judgments, foreclosure, repossession, bankruptcy, thin files, and no file. HMDA cannot separate those, so this study does not claim which credit event drove any denial.
- Incomplete applications are third for VA and are the VA outlier. See the lead number above. Ohio is the extreme case: 17.8% of Ohio’s 844 stated VA denials cited incompleteness, the highest share of the ten largest VA states.
- Withdrawals are bigger than denials, and they are not denials. 85,628 VA purchase applications were withdrawn in 2025, which is 18.60% of all 460,346 VA purchase applications and about 2.8 times the 30,421 denials. Any account of VA access that counts only denials misses the larger withdrawn pool. Withdrawals may reflect borrower behavior as much as lender decisions; this study does not treat them as lender rejections.
- Cash to close is a smaller cited VA driver than the old draft suggested. Insufficient cash was the primary reason on only 5.04% of stated VA denials (1,532 of 30,378), and it appears as any of the up to four cited reasons on only 8.28% of stated VA denials, versus 13.53% for FHA and 16.45% for conventional. Zero down removes the down payment, but the data says cash is rarely the leading reason a VA file is denied. It is demoted from a headline finding to a supporting one in this revision.
- The score floors veterans meet are lender overlays, not VA rules. The VA sets no minimum credit score (Pamphlet 26-7, Chapter 4). Published lender floors on file range from Rocket Mortgage’s 580 baseline to Veterans United’s 620 in most cases, with Rocket requiring a 640 or higher median score for manual underwriting. The spread between those published floors is itself a barrier: the same veteran can be denied by one lender’s overlay and approved under another’s.
What this study does not do is equally important, and it is stated in the body of this article, not buried in a footnote. It does not claim any credit-score-specific denial rate, because public HMDA redacts credit scores. It does not claim a count of eligible veterans who never apply, because no public dataset yields that count. It does not claim a trend against 2024, because the 2024 figures in circulation use a different denominator and a different year. It does not blend calendar-year HMDA data with fiscal-year VA program totals into any single rate.
Finding 1: VA purchase loans are denied less often than FHA, USDA, and conventional loans, on either definition
Table 1. Denial rates by loan type, purchase loans, 2025
| Loan type | Applications (actions 1-5) | Originated | Denied | Withdrawn | Denial rate A: denied / applications | Denial rate B: denied / (originated + denied) |
|---|---|---|---|---|---|---|
| VA | 460,346 | 331,042 | 30,421 | 85,628 | 6.61% | 8.42% |
| FHA | 907,898 | 640,557 | 96,258 | 141,198 | 10.60% | 13.06% |
| USDA | 55,980 | 38,423 | 5,709 | 9,460 | 10.20% | 12.94% |
| Conventional | 3,950,124 | 2,582,011 | 475,430 | 579,912 | 12.04% | 15.55% |
Source: author’s computation from FFIEC/CFPB HMDA 2025 loan-level data, retrieved October 31, 2026. Full counts, including approved but not accepted and closed for incompleteness, are in the accompanying tables file.
Why two rates? Rate A answers “of everyone who applied, what share was denied?” Rate B answers “of applications that reached a yes or no decision, what share was denied?” Rate B is always higher because it removes withdrawals and other non-decisions from the base. This article reports both every time, because published secondary analyses appear to use the originated-plus-denied base, and a reader comparing this study with those analyses needs to know which base each figure uses.
The comparison holds on both bases: VA purchase applicants are denied roughly half as often as conventional applicants (6.61% versus 12.04% on Rate A; 8.42% versus 15.55% on Rate B), and below FHA and USDA on both bases as well.
If your VA loan was denied, the first question to ask is not “what did the VA decide?” The VA does not approve or deny individual loans. The question is “what did this lender decide, and would another lender decide differently?” That distinction is the subject of my denial playbook.
Finding 2: The reason mix: debt-to-income first, credit history second, incomplete files third
Table 2. Primary denial reason mix, purchase loans, 2025 (percent of denials WITH a stated primary reason)
| Primary reason | VA (of 30,378) | FHA (of 95,933) | Conventional (of 464,862) | USDA (of 5,530) |
|---|---|---|---|---|
| Debt-to-income ratio | 29.25% (8,885) | 41.42% | 32.14% | 31.92% |
| Credit history | 23.02% (6,994) | 17.86% | 30.56% | 17.65% |
| Credit application incomplete | 14.08% (4,278) | 5.17% | 6.19% | 6.94% |
| Collateral | 10.74% (3,264) | 11.10% | 11.56% | 12.19% |
| Other | 9.96% (3,027) | 7.26% | 7.49% | 18.12% |
| Unverifiable information | 5.05% (1,535) | 6.41% | 6.96% | 5.53% |
| Insufficient cash (down payment, closing costs) | 5.04% (1,532) | 7.19% | 3.64% | 3.56% |
| Employment history | 2.75% (836) | 3.52% | 1.41% | 3.82% |
| Mortgage insurance denied | 0.09% (27) | 0.06% | 0.03% | 0.27% |
Source: author’s computation from HMDA 2025, denial_reason-1. Shares sum to 100% within each column (rounding). Nearly all VA denials carried a stated primary reason (30,378 of 30,421; only 43 not stated, 0.14%).
Three things stand out in this table. First, debt-to-income leads everywhere, but VA’s DTI share (29.25%) is the lowest of the four programs, while FHA’s (41.42%) is the highest by a wide margin. Second, credit history is the largest secondary driver for conventional loans (30.56%, the highest of any program) and second for VA (23.02%). Third, collateral is roughly even across programs (10.74% to 12.19%): property value and property problems are a shared barrier, not a VA-specific one.
A note on the DTI finding, because it is easy to misread. The VA’s 41% debt-to-income figure is a checkpoint that triggers closer scrutiny, not a hard limit (Pamphlet 26-7, Chapter 4). A file above 41% is not automatically denied; it is examined more closely, and compensating factors can carry it. Residual income, the money left each month after housing costs, debts, and taxes measured against VA guideline tables by region and family size, can carry a file that debt-to-income alone would sink, and weak residual income can sink a file with an acceptable ratio. Borrowers almost never arrive asking about residual income, but it is often the number that decides their file. My residual-income guide explains the tables and the math.
When an automated underwriting system refers a file instead of approving it, that is not a denial. It is a routing to manual underwriting, where a human underwriter reviews the full file, including compensating factors. That path is explained in the manual underwriting guide.
Finding 3: Incomplete applications are a barrier in their own right, and they are distinctively a VA problem
The incomplete finding deserves its own section because it is the finding most likely to change how a veteran acts. A denial coded “credit application incomplete” is not a finding that the veteran was unqualified. It is a lender report that the application was incomplete. HMDA does not record which document or item was missing, so this study does not claim what was missing in any individual case. But the pattern across 4,278 VA denials is too large to treat as paperwork noise: 14.08% of stated VA denials, against 5.17% for FHA, 6.19% for conventional, and 6.94% for USDA. On the any-mention measure (the code appearing in any of the up to four cited reasons), VA incompleteness is 15.05%, again the outlier.
Two distinctions keep this finding honest. First, a denial coded incomplete is different from an application “closed for incompleteness” (HMDA action 5), which is a separate outcome, not a denial. VA purchase applications closed for incompleteness numbered 5,441 in 2025. The two groups must never be added together: one is a denial outcome, the other is a non-decision outcome. Second, an incomplete-coded denial is different from an adverse credit or income finding. A veteran denied for incompleteness was not necessarily denied on the merits of the file, and the practical next step is different: ask the lender exactly what was outstanding, supply it, and ask whether the file can be reconsidered or re-submitted, rather than assuming the answer was no on credit or income.
Why VA files would show incompleteness at more than double the rate of other programs is not something HMDA can answer, and this study does not guess. Possibilities a reader might raise, such as entitlement documentation, disability rating paperwork, or the extra documents a VA file requires, are hypotheses, not findings. They are listed here only so a journalist reading this study does not mistake the gap for an answer. What HMDA does establish is the size of the gap and where it is largest: by state, VA incompleteness runs from 11.1% (Florida) to 17.8% (Ohio) of stated VA denials, with Ohio highest of the ten largest VA states (see Finding 7).
Finding 4: Withdrawals: the larger pool that denials alone miss
Table 3. Denials versus withdrawals, purchase loans, 2025
| Loan type | Applications | Denied | Withdrawn | Withdrawn as share of applications | Withdrawn per denial |
|---|---|---|---|---|---|
| VA | 460,346 | 30,421 | 85,628 | 18.60% | 2.81 |
| FHA | 907,898 | 96,258 | 141,198 | 15.55% | 1.47 |
| Conventional | 3,950,124 | 475,430 | 579,912 | 14.68% | 1.22 |
| USDA | 55,980 | 5,709 | 9,460 | 16.90% | 1.66 |
Source: author’s computation from HMDA 2025. Withdrawn is HMDA action 4. Withdrawn counts are included in the Rate A application denominator and excluded from the Rate B denominator.
Denials and withdrawals are different outcomes and this study never merges them. A denial is a lender decision. A withdrawal is an application the applicant withdrew (or that was withdrawn) before a decision. The VA withdrawal pool is large in absolute terms (85,628), large as a share of applications (18.60%), and large relative to denials (2.81 withdrawn for every denial, versus 1.22 for conventional). Any story about VA access that reports only the 6.61% denial rate understates how many VA purchase applicants did not reach an origination: 30,421 denied plus 85,628 withdrawn is 116,049 VA purchase applicants who did not close, out of 460,346 who applied.
This study does not claim why those applicants withdrew. Withdrawal can reflect a borrower who found another house, changed plans, chose another lender, or walked away from conditions, as well as a borrower who was effectively discouraged. HMDA does not record the borrower’s reason, so no share of withdrawals is attributed to lender behavior, overlays, or discouragement here. The finding is the size of the pool and the discipline it imposes: do not present a denial rate as a completion rate, and do not present withdrawals as rejections.
Finding 5: Cash to close, demoted: real, but rarely the leading VA reason
The old draft made cash a headline barrier. The HMDA data does not support that placement for VA, so this revision demotes it. Insufficient cash (down payment, closing costs) was the primary reason on 5.04% of stated VA denials (1,532 of 30,378), behind DTI, credit history, incompleteness, collateral, other, and unverifiable information. As any of the up to four cited reasons, cash appears on 8.28% of stated VA denials, versus 13.53% for FHA and 16.45% for conventional. Cash is a more frequent contributing reason outside VA than inside it.
None of that means cash does not matter to veterans. With full entitlement, a VA loan requires no down payment, which removes the single largest cash barrier other buyers face. But closing costs, prepaid items, and the VA funding fee (which can usually be rolled into the loan, but still affects the file) remain, and a veteran can be short on those even with zero down. My closing-costs guide walks through the zero-cash-at-closing paths that actually exist, including seller-paid strategies.
The demotion is about what the denial data shows, not about whether cash is hard. If cash were the leading VA barrier, it would lead the reason table. It ranks seventh as a primary reason. A study that kept it as a headline finding would be writing from assumption, not from the data it computed.
Finding 6: The overlay wall, measured from published floors only
VA Pamphlet 26-7, Chapter 4, states the rule plainly: “VA does not have a minimum credit score requirement.” Every score floor a veteran encounters is a lender overlay, a rule the lender adds on top of the VA program. The old draft claimed, without a source, that most lenders typically set their floor around 620. That claim is withdrawn in this revision: it is unsupported, and it is contradicted by the verified published floor below it.
Table 4. Published credit score floors: VA rule versus lender overlays (verifiable published statements only)
| Source | Floor | What it covers | Source and date |
|---|---|---|---|
| VA program rule | No minimum credit score | All VA loans; the VA itself sets no floor | Pamphlet 26-7, Chapter 4, via KnowVA |
| Rocket Mortgage | 580 baseline | VA loans generally, per Rocket’s published VA credit score page | rocketmortgage.com/learn/va-loan-credit-score, updated September 5, 2026, read October 31, 2026 |
| Rocket Mortgage, manual underwriting | 640 or higher median score, with debt-to-income no higher than 45% | Files manually underwritten by Rocket | Same Rocket page, same dates |
| Veterans United | 620 in most cases | VA loans at Veterans United, per its own published page | veteransunited.com, bad-credit page, as cited in the prior draft |
| Specialty and broker channels | No single floor cited here | Lenders reached through brokers may apply different overlays; no single published floor is claimed in this table | See scope note below |
Scope note, stated plainly: this table is not a market survey. It lists only floors that named lenders publish themselves, plus the VA rule. It does not claim a typical, average, or most common lender floor, because no verified survey of lender floors was pulled for this revision. A broader lender-floor survey, including broker-channel floors below 580, is listed under “Data still to add before publication” and must not be estimated in the meantime. A veteran with a score below 620 is therefore not failing a VA requirement. That veteran is failing one lender’s overlay, and other lenders, including brokers who shop a file across many lenders’ guidelines, may apply a different one.
This is the barrier most likely to be misdiagnosed, because the denial conversation rarely names it. A borrower told “you need a 620” hears a VA rule. It is not one. My credit-score hub explains what the VA actually requires and what lenders add, score band by score band.
For veterans with damaged credit generally, the bad-credit hub is the starting point.
Two related guides for files the overlay wall affects most: the bankruptcy guide and the foreclosure guide. HMDA cannot say what share of VA denials followed a bankruptcy or foreclosure, because it never records whether an applicant had one, so neither guide’s topic is given a denial share in this study. The collections guide is here.
Finding 7: Where VA denials cluster: ranked states, with Ohio in focus
Table 5. Featured states ranked by VA denial rate A, purchase loans, 2025
| Rank (lowest VA denial rate first) | State | VA applications | VA denial rate A | FHA denial rate A | Conventional denial rate A | VA incomplete share of stated VA denials |
|---|---|---|---|---|---|---|
| 1 | Virginia | 27,001 | 4.76% | 10.24% | 8.93% | 15.5% |
| 2 | North Carolina | 26,007 | 5.08% | 10.03% | 11.87% | 15.8% |
| 3 | Tennessee | 14,411 | 6.00% | 8.74% | 12.59% | 14.5% |
| 4 | Arizona | 13,814 | 6.10% | 9.10% | 9.01% | 13.2% |
| 5 | South Carolina | 16,101 | 6.45% | 11.34% | 16.78% | 14.1% |
| 6 | Ohio | 12,868 | 6.57% | 9.25% | 8.76% | 17.8% |
| 7 | California | 21,675 | 7.07% | 9.52% | 8.95% | 12.1% |
| 8 | Georgia | 22,915 | 7.14% | 10.87% | 12.87% | 11.4% |
| 9 | Texas | 62,355 | 7.36% | 10.75% | 17.12% | 11.5% |
| 10 | Florida | 44,067 | 7.82% | 13.91% | 13.61% | 11.1% |
Source: author’s computation from HMDA 2025. Featured set: the ten states with the most VA purchase applications in the computed data itself, which includes Ohio (ranked 10th by volume) and the largest VA states. National VA denial rate A for reference: 6.61%. Full state table (all 54 reported state codes) is in tables/denial_rates_states.csv. VA reason mix by featured state, including DTI and credit history shares, is in the October 31 study tables.
Three patterns in this ranking are worth a journalist’s attention. First, VA denial rates vary less by state (4.76% to 7.82%) than conventional rates do (Ohio 8.76% versus Texas 17.12%). Second, Florida is the highest-denial large state for VA (7.82%) and for FHA (13.91%). Third, Ohio’s ordering differs from the national ordering: in Ohio, VA (6.57%) sits below USDA (7.81%), conventional (8.76%), and FHA (9.25%), while nationally conventional (12.04%) exceeds FHA (10.60%). USDA state rates rest on small application counts and should be read with that caveat.
Ohio spotlight (purchase loans, 2025)
| Loan type | Applications | Originated | Denied | Withdrawn | Denial rate A | Denial rate B |
|---|---|---|---|---|---|---|
| VA | 12,868 | 9,161 | 845 | 2,516 | 6.57% | 8.44% |
| FHA | 30,713 | 21,417 | 2,842 | 5,397 | 9.25% | 11.72% |
| Conventional | 145,275 | 101,030 | 12,724 | 22,531 | 8.76% | 11.19% |
| USDA | 2,023 | 1,407 | 158 | 384 | 7.81% | 10.10% |
Ohio VA primary reason mix (844 stated of 845 denied): debt-to-income 27.5%, credit history 24.4%, credit application incomplete 17.8%, collateral 10.5%, other 10.1%, insufficient cash 3.6%, unverifiable information 3.2%, employment history 3.0%, mortgage insurance denied 0.0%. Ohio’s 17.8% incomplete share is the highest of the ten featured states, and it is the finding that most directly suggests a local action: for Ohio VA applicants, the file being complete may matter as much as any single underwriting factor. HMDA does not say what was missing, so that suggestion stops at the data’s edge.
What this study does not measure (limitations, in the body of the article)
- No credit scores. Public HMDA does not contain credit scores: lenders report the score relied on, but the CFPB redacts it from the public release. This study makes no score-specific denial claim. Nothing here says what percent of sub-620, 580, or any other score band of VA applicants is denied, and no such figure can be derived from public HMDA.
- “Credit history” is a bucket, not a diagnosis. Code 3 combines late payments, collections, judgments, foreclosure, repossession, bankruptcy, thin files, and no file. HMDA cannot separate these, so this study cannot say which credit event drove any denial, and it cannot give post-bankruptcy or post-foreclosure approval rates.
- No AUS outcomes. The file records which automated underwriting system was used, not the AUS finding. A file routed to manual underwriting after an AUS Refer cannot be distinguished from an AUS approval in public HMDA, so manual versus automated paths cannot be separated here.
- No eligible-nonuser count. This study counts applications and outcomes. It does not count eligible veterans who never applied, and it does not estimate that count. Subtracting VA loan totals from veteran population counts does not yield eligible non-users, because eligibility, intent, and timing differ across those universes.
- Calendar-year HMDA only. HMDA is calendar-year data. It is not blended with fiscal-year VA program data into any single rate in this study. VA program totals, where mentioned in the prior draft, are dropped from this revision’s findings for exactly that reason.
- Reasons are lender-reported and optional. Regulation C permits, but does not require, reporting denial reasons, and a denial may carry up to four reasons. Reason shares are therefore shares of denials with a stated reason, not of all denials, and they reflect what lenders chose to report. USDA reason detail deserves extra caution: 3.14% of USDA denials carried no stated primary reason, versus 0.14% for VA.
- No causal claims. Differences across loan types and states are descriptive. They do not by themselves prove that any lender, program rule, or borrower characteristic caused a denial. In particular, this study does not claim any share of denials was caused by overlays: HMDA records lender decisions, not the guideline that produced them.
- Coverage is not literally every lender. Small-lender partial exemptions mean some smaller institutions report fewer fields. A small number of denied rows carry an exempt or not-stated primary reason.
- Vintage note. Figures are API-served 2025 data as retrieved October 31, 2026. The API endpoint does not expose whether the vintage served is the frozen Snapshot National Loan-Level Dataset or the Dynamic dataset, so a frozen-Snapshot tabulation could differ slightly. A frozen-Snapshot re-run is listed under “Data still to add before publication.”
- No trend against 2024. This study computes one year, 2025, only. No 2024 primary computation was run, so no year-over-year trend is claimed.
About the 2024 figures you may have seen
A secondary analysis (Vetted VA / Polygon Research, 2025 VA Loan White Paper, analyzing 2024 HMDA Modified LAR data) reported purchase denial rates of VA 8.4%, conventional 15.6%, FHA 12.7%, and USDA 12.0%. Those figures track this study’s Rate B definition (denied divided by originated plus denied) far more closely than Rate A, which suggests the secondary analysis used an originated-plus-denied denominator, although its exact method is not published in the material relied on. The years differ (2024 versus 2025), so no trend is claimed here, and the closeness of the two sets is consistent with a shared denominator definition, not proof of it. This study’s 2025 figures are its own computations from primary data and are not copied from the secondary analysis. This secondary comparison is labelled as such wherever used.
Affordability: demoted, not a finding
The prior draft’s fifth finding, that a veteran earning the local median income could afford only 7.7% of homes listed for sale with a zero-down VA loan in an October 2026 analysis of 196 metros, is demoted in this revision to this paragraph. It is Movoto’s analysis, built from Census income data and Movoto listings with Movoto’s own assumptions, and it is cited here only as Movoto’s analysis, not as a government finding and not as a finding of this study. Home prices are a real barrier the loan program cannot fix, but this study computed no affordability measure, and a secondary affordability claim does not belong beside primary HMDA findings. If affordability is added to the published version, it should be added as a separately labeled section with its own methodology, not merged into the denial findings. No Movoto link is included, because no verified Movoto source link was supplied for this article.
The racial gap: removed from the findings
The prior draft reported, from a secondary analysis, that minority VA applicants were denied at 10.3% with a 3.6 point gap to non-Hispanic white VA applicants. That figure is removed from this revision’s findings. It came from a competitor-sourced secondary analysis, not from a computation in this study, and this study pulled no race or ethnicity cut of 2025 HMDA data. A 2025 race and ethnicity cut is listed under “Data still to add before publication.” Until it is computed from primary data and its denominators are stated, no racial gap figure appears in this article’s findings, and no disparity claim is made here.
Methodology and reproducibility
Source. FFIEC/CFPB Home Mortgage Disclosure Act (HMDA) loan-level data, calendar year 2025, retrieved October 31, 2026, through the documented FFIEC Data Browser API:
- Nationwide CSV endpoint
- One call per loan type, each with exactly two filters (the endpoint enforces a maximum of two filter criteria): years=2025, loan_types=<1, 2, 3, or 4>, loan_purposes=1 (home purchase). All actions were returned and filtered locally; no action filter was applied at the API level.
- The API serves the published HMDA loan-level dataset. The endpoint does not expose whether the vintage served is the frozen Snapshot National Loan-Level Dataset or the Dynamic dataset, so figures are reported as API-served 2025 data as retrieved October 31, 2026. They may differ slightly from a frozen-Snapshot tabulation.
Universe and definitions (fixed before computing):
- Purchase loans only (loan_purpose = 1). Refinances, home-improvement loans, and other purposes are excluded from every figure in this article.
- Application universe = action_taken in {1 originated, 2 approved but not accepted, 3 denied, 4 withdrawn, 5 closed for incompleteness}. Excluded and reported separately: 6 = loan purchased by the institution (not an application) and 7/8 = preapproval requests.
- Denial rate A (headline) = denied / applications (actions 1-5).
- Denial rate B (alternate, shown alongside) = denied / (originated + denied).
- Primary denial reason = denial_reason-1. Reason shares use as denominator the denials WITH a stated primary reason (denial_reason-1 in codes 1-9). Rows coded 1111 (exempt/not stated) or blank are counted separately as “not stated” and excluded from reason shares. Lenders may report up to four reasons per denial (denial_reason-1 through -4); the primary table uses reason 1 only, and the any-mention table counts a code if it appears in any of the four fields (those shares sum above 100% by construction).
- Reason codes follow Regulation C, Appendix A: 1 = debt-to-income ratio, 2 = employment history, 3 = credit history, 4 = collateral, 5 = insufficient cash (down payment, closing costs), 6 = unverifiable information, 7 = credit application incomplete, 8 = mortgage insurance denied, 9 = other.
- State tables use the property state (state_code) as reported. The featured state set is the top 10 states by VA purchase applications in the computed data itself.
- No sampling, weighting, or estimation was used. Every figure is a direct count or ratio of counts from the downloaded loan-level rows.
Loan-type codes: 1 = conventional, 2 = FHA, 3 = VA, 4 = FSA/RHS (USDA).
Validation. The Ohio subset of the nationwide VA file matches a separately pulled Ohio VA file exactly (12,868 applications, 9,161 originated, 845 denied), supporting the internal consistency of the pulls.
Computation files. The October 31 study’s compute script and raw downloads are retained in the study workspace (compute.py, data/, tables/). The tables accompanying this revision are copies of that study’s national tables plus the ranked featured-state table built from its state table. A reader with the same API endpoint and the filters above can reproduce every rate in this article from the stated sources.
Reproduction check (November 2, 2026)
Every headline figure in this article was recomputed from the derived tables below on November 2, 2026, and matches:
- VA denial rate A: 30,421 denied / 460,346 applications = 6.61%.
- VA withdrawals: 85,628 / 460,346 = 18.60% of applications; 2.81 withdrawals per denial.
- VA denial reasons (share of the 30,378 denials with a stated primary reason): debt-to-income 8,885 = 29.25%; credit history 6,994 = 23.02%; incomplete 4,278 = 14.08%; collateral 3,264 = 10.74%.
- Ohio VA: 845 denied / 12,868 applications = 6.57%; incomplete 150 / 844 stated = 17.8%.
Download the derived tables (CSV; public HMDA counts only, no borrower-level data): Denial rates by loan type, Primary denial reasons, Denial reasons, any mention, Featured states ranked. The national tables are direct counts from the FFIEC Data Browser API pulls described above; the featured-state table is ranked from the state table by VA denial rate A.
Appendix: Data still to add before publication (not findings)
Nothing in this list is claimed as a finding anywhere in this article. Each item must be pulled from its primary source and verified before it is added to a published version.
- Frozen-Snapshot re-run: re-tabulate the 2025 purchase figures from the frozen Snapshot National Loan-Level Dataset and report any differences from the API-served figures used here.
- 2025 race and ethnicity cut: compute VA denial rates and reason mix by race and ethnicity from primary HMDA data, with denominators stated, before any disparity figure is published.
- VA program totals from primary sources: VBA Annual Benefits Report figures or a VA press release, pulled and cited directly, if program totals are to appear at all, and kept in a separate section that is never blended into HMDA rates.
- Broader lender-floor survey: published score floors from additional named lenders, including broker-channel floors, each cited to the lender’s own page, before any “typical lender floor” claim is made.
- Any-mention reason mix by state: computed from the same HMDA pull, labeled as any-mention (shares sum above 100%), if state-level secondary drivers are to be reported.
- Refinance cut, if wanted: computed separately from a separate pull, labeled as refinances, and never mixed into purchase rates.
What was fixed or demoted from the October 31 barriers draft
- Retitled. The old title, “The Veterans the VA Loan Leaves Behind,” and its subtitle promised a count of eligible veterans who never buy. That count cannot be computed from public data, so the title now promises only what the data supports: why VA purchase loans are denied, from 2025 HMDA data.
- Finding 1 rebuilt. The old Finding 1 table used 2024 secondary figures (VA 8.4%, conventional 15.6%, FHA 12.7%, USDA 12.0%) without stating its denominator. It is replaced with 2025 primary computations under both denial-rate definitions, and the 2024 secondary figures appear only in the labeled comparison note with their denominator caveat.
- Reason mix added. The old draft admitted the national VA reason mix had not been computed. It is now the core of the article (Findings 2 and 3), computed from primary data.
- Withdrawals added. The old draft never separated withdrawals from denials. Finding 4 keeps them strictly distinct and reports the 85,628 VA withdrawals alongside the 30,421 denials.
- Incomplete applications promoted to a full finding. The old draft had no incomplete finding. It is now Finding 3 and the lead number, including Ohio’s 17.8%.
- Cash demoted. The old draft’s cash barrier is demoted from a headline finding to Finding 5, on the evidence that cash ranks seventh as a primary VA denial reason.
- “Most lenders typically around 620” withdrawn. The unsupported claim is removed and replaced with the verified published floors in Table 4 (Rocket 580 baseline and 640 manual underwriting; Veterans United 620 in most cases; VA no minimum).
- Movoto affordability demoted. The old Finding 5 is demoted to a labeled paragraph under limitations, identified as Movoto’s analysis with its own assumptions, not a finding of this study.
- Competitor-sourced racial gap removed. The old draft’s minority denial figures are removed from the findings and moved to “Data still to add,” pending a primary-data race and ethnicity cut.
- Fiscal-year VA totals dropped from the findings. The old draft’s VBA FY2025 and FY2024 totals and its $0 down share are removed from the findings to avoid blending calendar-year HMDA data with fiscal-year program data. VA program rules (no down payment with full entitlement, no VA minimum score, 41% DTI checkpoint, residual income) remain, cited as rules, not statistics.
- “Five barriers” framing replaced. The old draft’s five-barrier summary is replaced by findings that follow the data’s own ranking: DTI, credit history, incompleteness, withdrawals as a separate pool, cash demoted, overlays measured from published floors, and states ranked.
Sources
Primary data and definitions:
- FFIEC/CFPB HMDA Data Browser API documentation, nationwide CSV endpoint at ffiec.cfpb.gov/v2/data-browser-api/view/nationwide/csv, retrieved October 31, 2026 (see Methodology for endpoint, filters, and definitions).
- Regulation C (12 CFR 1003), Appendix A, denial reason codes; Regulation B adverse-action mapping for code 3 (credit history bucket).
- VA Pamphlet 26-7, Chapter 4 (Credit Underwriting), via KnowVA, cited in text without a KnowVA hyperlink, because no working Chapter 4 KnowVA URL was verified for this article: no minimum credit score; 41% debt-to-income checkpoint; residual income tables; AUS findings as risk classifications, not approvals or denials.
Published lender statements (Table 4):
- Rocket Mortgage, VA loan credit score page (updated September 5, 2026, read October 31, 2026): 580 baseline; 640 or higher median score with debt-to-income no higher than 45% for manual underwriting.
- Veterans United, bad-credit page (620 in most cases, as cited in the prior draft).
Secondary comparison only (labelled as such wherever used):
- Vetted VA / Polygon Research, 2025 VA Loan White Paper (June 2025), analysis of 2024 HMDA Modified LAR data (see the comparison note; years and denominator differ, no trend claimed).
Secondary analysis demoted to a labeled paragraph (not a finding):
- Movoto affordability analysis (October 2026), 196 metros, median-income veteran affordable share 7.7% of listings with a zero-down VA loan at 7.32%, as described in the prior draft. Assumptions are Movoto’s; not recomputed here. No Movoto hyperlink is included, because no verified Movoto source link was supplied.
This article is educational only and is not a commitment to lend, a guarantee of approval, or an offer of specific loan terms. Individual results depend on the full file, including credit history, income, debts, residual income, property, and the lender’s guidelines, which change. VA program rules are cited from Pamphlet 26-7; lender overlays vary by lender and are identified as overlays wherever they appear.
