Denominator diagnosis ยท worked example

12.23% and 7.89%: of what?

A consumer finance page, Signs Your Mortgage Will Be Denied in Underwriting (LendingTree, marked “Updated Oct 20, 2026”), gives national denial rates of 7.53% for conventional loans, 12.23% for FHA loans and 7.89% for VA loans, from its own analysis of Home Mortgage Disclosure Act data, for 2025. It does not say which applications are counted. The link it gives for the data opens a CFPB report on 2023 mortgage activity. Answer engines repeat these figures; one gave 12.23% as the “overall” FHA denial rate. Page read on 15 November 2026.

Finding. Neither figure is reproduced by any definition computed from the 2025 federal file. Among decisioned applications of every purpose, the 2025 rates are 22.08% for FHA and 16.15% for VA. For home purchase alone they are 12.78% and 8.24%. The quoted figures sit near the purchase rates and below them. The conventional figure was not searched: purchase and refinance rates for conventional loans are not published here yet.

How the search works

For each program the 2025 HMDA file is counted by loan purpose (8), scope (all rows; or first-lien, owner-occupied, site-built, 1–4 unit) and base (denied over decisioned applications; over all applications acted on; over denied plus originated). That gives 48 denial rates per program, each with its counts, on the denominator map. A quoted figure within 0.15 points of a cell is a match.

FHA: quoted 12.23%

48 definitions computed from the 2025 file. The ten closest, nearest first. None within 0.15 points. The closest is 12.70% (home purchase, decisioned), 0.47 points away.

Loan purposeScopeBaseDenied / denominatorRateDistance
home purchasefirst-lien, owner-occupied, site-built, 1–4 unitsdecisioned90,590 / 713,09412.70%0.47
home purchaseall rowsdecisioned96,186 / 752,77012.78%0.55
home purchasefirst-lien, owner-occupied, site-built, 1–4 unitsdenied + originated90,590 / 697,24412.99%0.76
home purchaseall rowsdenied + originated96,186 / 735,34513.08%0.85
home purchaseall rowsall acted on96,186 / 906,13310.61%1.62
home purchasefirst-lien, owner-occupied, site-built, 1–4 unitsall acted on90,590 / 855,47810.59%1.64
all purposesfirst-lien, owner-occupied, site-built, 1–4 unitsall acted on247,723 / 1,479,06216.75%4.52
all purposesall rowsall acted on262,250 / 1,563,37016.77%4.54
refinancefirst-lien, owner-occupied, site-built, 1–4 unitsall acted on44,658 / 238,08418.76%6.53
refinanceall rowsall acted on46,808 / 249,35518.77%6.54

VA: quoted 7.89%

48 definitions computed from the 2025 file. The ten closest, nearest first. None within 0.15 points. The closest is 8.06% (home purchase, decisioned), 0.17 points away.

Loan purposeScopeBaseDenied / denominatorRateDistance
home purchasefirst-lien, owner-occupied, site-built, 1–4 unitsdecisioned28,831 / 357,5378.06%0.17
home purchaseall rowsdecisioned30,421 / 369,2758.24%0.35
home purchasefirst-lien, owner-occupied, site-built, 1–4 unitsdenied + originated28,831 / 350,0438.24%0.35
home purchaseall rowsdenied + originated30,421 / 361,4618.42%0.53
home purchaseall rowsall acted on30,421 / 460,3436.61%1.28
home purchasefirst-lien, owner-occupied, site-built, 1–4 unitsall acted on28,831 / 445,4446.47%1.42
refinancefirst-lien, owner-occupied, site-built, 1–4 unitsall acted on25,284 / 263,5789.59%1.70
refinanceall rowsall acted on26,299 / 273,6069.61%1.72
all purposesfirst-lien, owner-occupied, site-built, 1–4 unitsall acted on110,671 / 945,49811.71%3.82
all purposesall rowsall acted on115,026 / 977,62111.77%3.88

What could explain the gap

These are possibilities, not findings. The publisher may have applied a filter this map does not have (a loan-amount limit, a subset of lenders, conforming loans only, a different treatment of preapprovals or purchased loans). The figures may come from another year’s file: the link on the page points to a report on 2023. Or the method may be one the public file cannot reproduce. Only the publisher can say which. A statement of the denominator on that page would settle it in one line.

What this does not say

It does not say the figures are wrong, and it says nothing about the publisher’s intent. It says that, as published, they cannot be traced to the 2025 federal file under any of the definitions listed here, and that a reader cannot tell what they measure. It says nothing about any applicant, and it does not compare programs or lenders.

Counts and rates: data/purpose/fha-2025.json, data/purpose/va-2025.json. Receipts for the purchase and all-purpose rates are on the denominator map. To run the same search on any table: place a number. A five-claim example: sample Evidence Brief. Generated by scripts/build_diagnosis_page.py.

FinanceRateCalc · Measured, not assumed. · Corrections · If you publish one of these figures and can state its denominator, write to the address on About and this page will say so.