Analysis · HMDA 2025
FHA looks cheaper until you add the insurance
On 2025 purchase loans, FHA borrowers got a lower interest rate than conventional ones and paid roughly twice as much at closing. Neither number captures what actually makes FHA expensive.
What the register shows
Purchase loans only — refinances have a different mix of borrowers and would muddle the comparison.
| Loan type | Loans | Median rate | Median fees | Median LTV | Median income |
|---|---|---|---|---|---|
| Conventional | 2,574,567 | 6.625% | $5,451 | 79.8% | $129 |
| FHA | 634,821 | 6.250% | $11,271 | 97.3% | $94 |
| VA | 330,071 | 6.125% | $6,456 | 100.0% | $112 |
| USDA | 37,158 | 6.250% | $6,374 | 100.0% | $77 |
FHA borrowers got a rate 0.375 percentage points lower than conventional ones and paid $5,820 more at closing. Most of that gap is not a lender markup: FHA charges an upfront insurance premium of 1.75% of the loan, which on the median FHA loan of $305,000 comes to about $5,338 — and it lands inside the reported fees.
The cost that appears nowhere
FHA also charges an annual premium of 0.55% of the balance, collected monthly. It is not interest, so it is absent from the rate. It is not a closing cost, so it is absent from the fees. On the median FHA loan it is about $140 a month.
Cost over 7 years on a $305,000 loan
As the filings show it
Conventional $140,839 · FHA $138,621
With the annual premium added
Conventional $140,839 · FHA $150,364
The premium adds roughly $11,743 over 7 years — enough to move FHA from looking competitive to costing about $9,525 more than conventional on the same loan amount.
Why FHA borrowers take it anyway
None of this makes FHA a bad deal, and the comparison above is not the one most borrowers actually face.
Look at the LTV and income columns. FHA borrowers put down a fraction of what conventional borrowers did and earn substantially less. For many of them the alternative was not a cheaper conventional loan — it was no loan at all, or a conventional one priced for a thin credit file and a small deposit, which would have carried both a worse rate and private mortgage insurance of its own.
FHA exists precisely to price that borrower into the market. The right question is not “is FHA more expensive than conventional” but “is this FHA loan more expensive than the conventional loan I could actually get” — and the answer depends on a credit score that this data does not contain.
What is genuinely worth knowing
The annual premium usually does not go away. On most FHA loans written since 2013 with a small deposit, it is charged for the life of the loan. Conventional private mortgage insurance, by contrast, must be cancelled once the balance falls far enough. That difference compounds quietly for years and is the single strongest argument for refinancing out of FHA once equity allows.
It is invisible in every number a borrower is shown. Not in the rate, not in the fees, not in the APR — because the APR is computed on the payment stream the note requires, and the premium is collected alongside it. A borrower comparing disclosures sees none of it.
Refinancing is the exit, and it has a cost. Moving from FHA to conventional means paying closing costs again. Whether that is worth it depends on the same horizon arithmetic as everything else on this site — the comparison tool is built for exactly that question.
Caveats
The insurance figures are current programme rules, not observations. The upfront and annual premiums come from FHA’s published schedule and are applied here to the median loan. Individual premiums vary with deposit size and term.
No credit scores. The single largest driver of what a borrower could alternatively have got on a conventional loan is not in this data.
Conventional PMI is not modelled. A conventional borrower at high LTV would pay mortgage insurance too, until it cancels. The comparison above is therefore generous to conventional at the high-LTV end.
Medians on different populations. FHA and conventional borrowers are not the same people buying the same houses, which is the point of the first section.
Figures from the register are recomputed on each build; the methodology sets out thresholds and sanitation.