One finding, measured, not assumed.
We took all 859,090 FHA applications that reached a credit decision in the 2025 federal record and asked a blunt question: how much of the outcome is the applicant's file — income, loan size, debt ratio, location — and how much is simply which lender they walked into? Two statistical models, one difference. The answer: 38% of everything explainable about a denial is the door, not the file.
The extremes make it concrete. One major lender's applicant mix predicts a 35% denial rate; its actual rate is 78.7% — a +43.7-point gap no file can explain. On the other end, several lenders deny 13–21 points less than their mix predicts. And the strictest doors share a fingerprint: an outsized share of their denials are coded "application incomplete" — files at those doors aren't so much denied as never finished. (Honest caveat, always: HMDA contains no credit scores, so the residual means "not explained by observable federal-record characteristics" — consistent with overlays, channel, documentation practices. Full method and tables: financeratecalc.com/door-effect.html)
One program, three numbers. You'll see the 2025 "FHA denial rate" quoted as 12.7%, 22.1%, or 38.1% — all technically defensible: purchase-only, all decisioned applications, refinance-only. The denominator you choose is the story you tell. We publish ours in every table: decisioned applications, actions 1-2-3, stated up front. Ask any number you meet: denied out of what?
P.S. Our data now speaks agent. AI tools can query the federal record directly — npx frc-mcp, or the no-install remote endpoint at financeratecalc.com/mcp-server.html. Same red line, written into the code: historical aggregates only, never individual predictions.
— The Denial Dispatch, by FinanceRateCalc