Mortgage denial statistics disagree with each other more often than they should, and the reason is usually definitional rather than computational. Two sources can process the same federal file correctly and report figures several points apart because one counted approved-but-not-accepted applications and the other did not, or because one reported a median across lenders and the other a volume-weighted share.
These are the definitions behind every figure published here. They are also emitted as a schema.org DefinedTermSet in this page's markup, so a machine reading it gets the same definitions a person does — without inferring them from prose.
An FHA mortgage application that reached a credit decision by the reporting institution: HMDA action_taken 1 (originated), 2 (approved but not accepted) or 3 (denied). Applications withdrawn by the applicant, closed for incompleteness, purchased by another institution, or on the preapproval track are outside this universe. The denominator choice is the single largest source of divergence between published mortgage denial statistics.
Where this is used →Denials divided by decisioned applications: action_taken 3 over actions 1, 2 and 3. In the complete 2025 federal record the national FHA denial rate was 22.1% of 1,187,606 decisioned applications, excluding reverse mortgages.
Where this is used →Observed denials divided by expected denials, where expected is computed by applying market-wide denial rates for each borrower-profile cell to a lender's own application mix. Indirect standardization, the technique used for standardized mortality ratios in public health, applied to credit. A ratio of 1.00 means an institution denies exactly as often as the market would facing the same book.
Where this is used →A bucket defined by the Cartesian product of state, loan-amount band, income band, debt-to-income band and combined loan-to-value band, used to standardize denial rates for applicant composition. HMDA contains no credit scores, so cells control the profile dimensions the federal record contains rather than the full underwriting information set.
Where this is used →The ratio of the denial rate on FHA applications under $150,000 to the rate on applications of $250,000 or more, within the same geography. In 2025 it exceeded 1.0 in every state measured and peaked at 4.45x in Idaho, where 53.4% of small applications were denied against 12.0% of large ones.
Where this is used →The distribution of cited denial reasons for a single lender, expressed as shares of that lender's formal denials across eight HMDA categories. Reason shares differ far more between institutions than aggregate denial rates do, which makes the fingerprint more diagnostic than the rate alone.
Where this is used →A four-step framework for a borrower after a mortgage denial: decode the stated reason against the lender's own record, understand the year the application was decided, compare the door to its peers, and decide when and where to reapply. It is not a geographic map.
Where this is used →The finding that "incomplete application" is cited on a median 1.8% of denials across the 100 largest FHA lenders but 75.2% at one servicer-rooted institution, with a small cluster of similar outliers. First published 22 July 2026.
Where this is used →The finding that among builder-owned FHA lenders, collateral appears in a median 0.1% of cited denial reasons against 11.4% elsewhere, with the weight shifted onto debt-to-income, and that these lenders deny at a higher median rate (21.8%) than their peers (12.8%) rather than a lower one. First published 24 July 2026.
Where this is used →A result an unaffiliated third party has actually recomputed from the primary source and reported. Distinct from reproducible, which means only that the methodology is specified well enough that they could. No figure published by FinanceRateCalc has been independently reproduced.
Where this is used →A divergence between two implementations traceable to a rule the publisher never stated. Under the published divergence protocol this is assigned to the publisher rather than the replicator, and the remedy is to publish the missing rule.
Where this is used →Each artifact is derived from the same public federal file and points back to the others, so anyone arriving at one can reach the rest. None of it has been independently reproduced — that remains the open item, and the specification for closing it is in the reconciliation link above.
Every headline figure as structured JSON, each with its own verification status: claims.json →