method string | global_rate_pct float64 | lender_count int64 | rows list |
|---|---|---|---|
Indirect standardization. Cells = state x loan-amount band x income band x DTI band x CLTV band. Market cell rates computed on all decisioned FHA applications (actions 1,2,3; denial=3), cells with n>=25; smaller cells fall back to the national rate. Expected denials = sum of cell rates over each lender's own applicatio... | 21.84 | 277 | [
{
"lei": "549300HW662MN1WU8550",
"apps": 112470,
"observed_denials": 24022,
"observed_rate_pct": 21.4,
"expected_denials": 22531.7,
"expected_rate_pct": 20,
"adjusted_ratio": 1.07
},
{
"lei": "549300FGXN1K3HLB1R50",
"apps": 96051,
"observed_denials": 28619,
"observed_... |
Peer-Adjusted FHA Denial Rates — 2025
Every published mortgage denial statistic carries the same caveat: rates partly reflect who applies where. This dataset quantifies that caveat at the lender level, apparently for the first time in public.
Method
Indirect standardization (the technique used for standardized mortality ratios in public health), applied to credit:
- All 1,217,297 decisioned FHA applications in the complete 2025 CFPB HMDA record are assigned to borrower-profile cells: state × loan-amount band × income band × DTI band × CLTV band → 40,459 cells.
- The market-wide denial rate is computed for each cell (cells with n<25 fall back to the national rate).
- For each lender, those market rates are applied to that lender's own application mix → expected denials.
peer_adjusted_ratio = observed_denials / expected_denials.
Reading it: 1.00 = denies exactly as the market would, facing the same book. 2.13 = denies 2.13× as often as peers facing comparable applications. 0.14 = far less often.
Headline findings
- Applicant mix is real: expected rates range 12.8%–36.9% across the 100 largest lenders (a 2.9× range)
- The observed range is 1.8%–78.7% (44×)
- Rank correlation between raw and adjusted measures: 0.93 — strict doors stay strict after the borrower profile is held constant
- Sharpest case: Lakeview Loan Servicing 2.30× vs Lakeview Community Capital 0.08× — same corporate parent, 28-fold difference
Analysis: https://financeratecalc.com/adjusted-denial-rates.html
Files
fha_peer_adjusted_denial_2025.csv— one row per lender (277 lenders with ≥500 decisioned applications)adjusted_denial_2025.json— same data plus full method string
Limitations (please read)
HMDA contains no credit scores. The adjustment therefore controls for the profile dimensions the federal record does contain — not for every underwriting-relevant factor. Residual differences may reflect unobserved applicant characteristics as well as lender behavior. Ratios are historical observations, not evidence of wrongdoing, not recommendations, and not predictions about any individual application.
Citation
Primary data: CFPB HMDA 2025 Snapshot. Derived measure: Yetiş, Z. (2026), Peer-Adjusted FHA Denial Rates, FinanceRateCalc. https://financeratecalc.com/adjusted-denial-rates.html — companion working paper: SSRN abstract 7156938.
CC BY 4.0 · Not a lender · No ads · A denial is a data point, not a verdict on the applicant.
- Downloads last month
- 36