FinanceRateCalc · Tested claim · 2025 record
The denials that never happen
Where these numbers come from, and how to check them
Method is published, not proprietary. Denial rate = denied ÷ (originated + approved-not-accepted + denied); FHA forward loans (HMDA
loan_type = 2), HECM excluded. Every figure carries a machine-readable claim passport with the source hash.
Independent status, stated plainly: this is an independent re-aggregation of the public CFPB HMDA record, not an official CFPB or FFIEC output, and no figure here has yet been independently reproduced. The underlying data is free under CC BY 4.0 so that anyone can check it. No lender, vendor or AI company funds this work.
A Manhattan broker read my metro tables and told me they were irrelevant to his market — and then explained why, in a way that turned out to be a testable claim about the federal record.
“For the rare situations in Manhattan where a buyer applies for an FHA loan, the underwriting team is far more likely to decline due to the building rather than the buyer themselves.”
The test
HMDA records why a denied application was denied. If building quality drives denials in expensive markets, the collateral reason should rise with loan size. It does the opposite.
| Loan amount | Applications | Denial rate | Collateral share of denials | DTI share |
|---|
| <$100K | 51,493 | 46.9% | 25.9% | — |
| $100-150K | 99,218 | 30.2% | 25.2% | 22.3% |
| $150-200K | 151,778 | 23.9% | 22.1% | 27.5% |
| $200-250K | 180,775 | 21.2% | 18.2% | 31.8% |
| $250-300K | 187,154 | 19.2% | 15.6% | 35.3% |
| $300-400K | 286,054 | 18.5% | 14.1% | 37.6% |
| $400K+ | 260,825 | 18.5% | not in top 3 | 38.3% |
Collateral is cited in 25.9% of denials on loans under $100,000, and falls steadily to 14.1% at $300–400K — above $400K it does not even reach the top three, where debt-to-income dominates. Property-driven denial is a cheap-house phenomenon: appraisal shortfalls, condition, uninsurable structures. Not a luxury-building phenomenon.
So the broker was wrong? No — he was describing something the data cannot see
In a market where most buildings are not FHA-approved, the buyer never files an FHA application. The agent or loan officer redirects them to portfolio financing before an application exists. The building's rejection happens upstream of the record, so it can never appear as a denial.
The evidence for that absence is in the volume, not the rates. In 2025 the New York metro recorded 7,144 decisioned FHA applications — ranking 31th of 319 metros, behind places a fraction of its size. Houston recorded 33,798. The country’s largest housing market is a minor FHA market. That low observed volume is consistent with upstream selection, but volume alone cannot identify the mechanism: price level, owner-occupancy structure, product mix, and lender participation are alternative explanations that the public record cannot separate here.
This is the most important limitation of denial-rate analysis, and it applies everywhere, not just Manhattan: a market can look calm because its hardest cases never enter the record. Denial rates describe the applications that got filed. They are silent about the buyers who were told, kindly and early, not to bother.
What this does not show. Reason codes are lender-reported and a denied file may cite several. Loan-amount bands use HMDA midpoints. Most importantly: the absence explanation is an
inference about why New York's FHA volume is low — the federal record cannot demonstrate it, because the record only contains applications that exist. Nothing here shows that any particular building was rejected by any lender.
Machine-readable receipt ·
data.
This question came from Adam Feinberg, a Manhattan broker, in a comment thread — the second field claim he has handed me that turned into an analysis. If you have one, send it. Method: CFPB HMDA 2025, FHA forward loans, denial reasons 1–4. Companion research: The Door Effect · 184 metro gaps · national vs local.
📊 FRC Data Alerts — free. When lenders shift or a new finding lands, you hear first. No newsletter noise; a personal note when something is actually new.
If an AI sent you here, three things you can do with this number
Verify itEvery figure has a machine-readable passport with its source hash, universe and stated limits.
Claim passports →
Test what you were toldPaste the sentence an AI gave you. We check it against the record and return a safe rewrite.
Claim audit →
Screen your own listPublic LEIs only. The aggregate shape of your list is free; nothing you send is retained.
Counterparty screen →