Mortgage Denial Explorer
Denial rates by race and ethnicity for 2.8 million home-purchase applications in the 2025 HMDA data: pick a state, an income band, a debt-to-income band and a loan type, compare raw and adjusted gaps, read the lenders' stated reasons, and ask the paper's model for a predicted denial probability.
How to read this
Each bar is the share of 2025 home-purchase applications that were denied, among applications that reached a decision (the paper's complete cases). Pick a state on the map and narrow by income, debt-to-income ratio and loan type. The raw gap is a group's denial rate minus the White rate; the adjusted gap compares each group with what a model fitted to White applicants predicts for applications like theirs. Neither is proof of discrimination: the public data have no credit score. Cells with fewer than 30 applications are not shown. The opening view is Figure 1 of the paper.
Denial rates
Raw and adjusted gaps against White applicants
Adjusted gap: the group's excess denial rate (actual minus the White benchmark, a logit of denial on income, debt-to-income, combined loan-to-value, loan amount, property value, loan type, age, co-applicant, sex, jumbo status and state, fitted to White applicants only) minus the White excess in the same selection. The intervals are ±1.96 binomial standard errors; they ignore clustering by lender and the uncertainty in the benchmark, so they are too narrow.
The paper's model ladder (all states)
Adjusted gaps in percentage points from the paper's logistic regressions, with lender-clustered 95% intervals: each model adds controls to the one before (Table 3 of the paper).
Why lenders said no
What the model predicts
Set an application's recorded characteristics and see the denial probability that model (3) of the paper assigns it for each group. The differences between bars are the model's group shifts, not the effect of race: anything the public data omit, above all the credit score, is folded into them.
Data: FFIEC/CFPB, HMDA Snapshot National Loan-Level Dataset, 2025 (first-lien home-purchase applications for a principal residence, 1–4 unit site-built, decided as originated, approved but not accepted, or denied; complete cases). Group definitions follow the paper: Hispanic of any race, then non-Hispanic single races. The view is saved in the address bar, so a link reopens it.