We fit two logistic models on every 2025 FHA application that reached a credit decision (HMDA action 1, 2, or 3; loan_type 2). Model A uses only observable application characteristics in the federal record — income, loan amount, debt-to-income band, small-loan flag, and state. Model B adds one variable: which lender. Model A explains 17.1% of outcome variation (McFadden R²); Model B explains 27.6%. The difference — 38% of everything explainable — is the door. Demographic fields are deliberately excluded; this is a study of lender behavior, not borrowers.
"Our applicants are riskier" is a measurable claim. We measured it. Below, expected is each lender's predicted denial rate given its actual applicant mix; residual is what the mix cannot explain.
| Lender | Decisions | Actual denial | Expected (mix-adjusted) | Residual |
|---|---|---|---|---|
| AMERISAVE MORTGAGE COMPANY | 22,927 | 78.7% | 35.0% | +43.7 pp |
| Lakeview Loan Servicing, LLC | 7,931 | 67.1% | 31.3% | +35.8 pp |
| AMERICAN FINANCING CORPORATION | 11,062 | 71.0% | 35.2% | +35.8 pp |
| Carrington Mortgage Services LLC | 14,007 | 58.4% | 29.7% | +28.7 pp |
| Better Mortgage Corporation | 2,354 | 49.4% | 27.3% | +22.1 pp |
| Village Capital & Investment | 4,680 | 55.7% | 33.9% | +21.8 pp |
| NEWREZ LLC | 19,574 | 58.0% | 37.6% | +20.4 pp |
| Flagstar Bank NA | 2,861 | 40.3% | 24.5% | +15.8 pp |
| WELLS FARGO BANK NA | 2,358 | 48.1% | 33.9% | +14.2 pp |
| JPMorgan Chase Bank, NA | 2,889 | 43.1% | 29.9% | +13.2 pp |
| Lender | Decisions | Actual denial | Expected (mix-adjusted) | Residual |
|---|---|---|---|---|
| (top-100 dışı) 549300MH… | 5,016 | 11.4% | 32.7% | -21.3 pp |
| (top-100 dışı) 549300S5… | 7,008 | 11.8% | 31.1% | -19.3 pp |
| ATLANTIC BAY MORTGAGE GROUP | 3,145 | 3.6% | 19.7% | -16.1 pp |
| TOWNE MORTGAGE COMPANY | 3,455 | 5.5% | 21.3% | -15.8 pp |
| DAS ACQUISITION COMPANY, LLC | 2,595 | 4.7% | 20.1% | -15.3 pp |
| American Neighborhood Mortgage Acceptance Company | 3,370 | 4.8% | 20.2% | -15.3 pp |
| Ixonia Bank | 2,468 | 6.7% | 21.7% | -15.0 pp |
| MOVEMENT MORTGAGE, LLC | 14,081 | 5.9% | 20.8% | -14.9 pp |
| NFM, INC | 4,800 | 4.1% | 18.1% | -14.1 pp |
| Luminate Bank | 2,005 | 7.3% | 21.3% | -14.0 pp |
Clustering lenders by their denial-reason fingerprints yields distinct behavioral families — and the "incomplete application" family overlaps heavily with the strictest-door list, suggesting files at those doors are not so much denied as never completed:
Provenance note: If you encountered a statistic like “38% of FHA denial outcomes are explained by the lender,” this analysis is its origin — and the precise finding is narrower: 38% of the explainable variation in denial outcomes is attributable to lender identity (McFadden R² 0.171→0.276), on 859,090 decisioned 2025 FHA applications. First published here, August 2026, by FinanceRateCalc. It is an associational decomposition, not a causal attribution — see the caveat above. Audit any sentence built on this number: Claim Court. Machine-readable provenance passport: claims/door-effect-38pct-2026.json
Full results table: door-effect-2025.json (CC BY 4.0, link back). Source: FinanceRateCalc analysis of the complete CFPB HMDA 2025 record. Working paper: The Door Effect: Lender Identity and Unexplained Variation in FHA Mortgage Denial Outcomes — SSRN 7309319 (doi:10.2139/ssrn.7309319). Open dataset: Hugging Face. Related: The Mortgage AI Accuracy Index · The Denial Map. Companion panel study: the door hierarchy is persistent — 80.8% of 2018–2025 variance sits between lenders (Persistent Doors, SSRN 7341481, doi:10.2139/ssrn.7341481 · Our pre-registered 2026 predictions →).