HMDA 2025 · Harris County, Texas · 531 loans, one profile
The rate is not the price.
Same county. Same year. Same conventional 30-year purchase loan, same LTV band, same DTI band, same loan size — borrowers an underwriter could not tell apart. The one at the 90th percentile paid 3.71 times what the one at the 10th paid in closing costs.
Nobody publishes that spread, so nobody shops it. Everything below is computed live from public rules on one real profile, and every figure says where it came from.
3.71×
$9,842
between the 10th and the 90th percentile, on the same mortgage
n = 531 · 523 report a cost figure
No social security number. No data sold. Nobody calls you.
Total loan costs paid, by percentile
Counted on the regulator’s own microdata, not on a survey.
Conventional · first lien · 30 years · principal residence · 1 unit · site built · LTV 79–81 · DTI 36–43 · loan $300k–$500k
Closing costs charged to the borrower
60.1% of the cohort bought discount points, median $4,000 — so comparing two rates without comparing their points compares two different things.
What this number covers
- A. Origination Charges — the lender’s own fees, discount points included
- B. Services You Cannot Shop For — appraisal, credit report, flood certification, FHA upfront premium when the borrower pays it
- C. Services You Can Shop For — settlement agent, lender’s title policy
What it leaves out, and we compute separately
- E. Taxes and Other Government Fees — recording, transfer taxes. In Philadelphia these run 4.578%: on a $500,000 house, $22,890.
- F. Prepaids — prepaid interest, premiums, taxes
- G. Initial Escrow Payment at Closing
- H. Other — the owner’s title policy, which is a different policy from the lender’s
Of the 531 loans in the cohort, 523 report a cost figure; the percentiles are counted on those 523. This is history from one county, not a quote, and not your county until we have counted it.
The file this page runs on
One buyer, loaded and computed.
Forty is the median age of a first-time buyer in the United States and an all-time high — the person using a comparison site is not twenty-five. A 691 score with 10% down is the band where the spread between lenders is widest and where every existing comparison site serves worst: they calibrate their storefront for 780 and 20% down.
Buyer
María
- Age
- 40
- County
- Tarrant County, TXFIPS 48439
- Score
- 691Classic FICO
- Household
- 4
Property
$434,100
- Pre-approval she arrived with
- FHA6.5% · 30 yr
- Down payment
- $15,1943.5%
- LTV
- 96.50%
- Years she expects to stay
- Assumed7 years
Rate anchor
6.766%
- OBMMIC30YF
- Modeled2026-09-08
- Property tax rate
- Modeled1.245%ACS 2024 1-year
- Homeowners insurance
- Modeled$2,251ACS 2024 1-year
No published index observes her exact LTV × FICO cell, so the general 30-year conforming index is used. Between the general index and the LTV>80 / FICO<680 cell the observed difference was 16.5 basis points, so this anchor runs low for her band. We say so instead of quietly picking the number that flatters the page.
Her pre-approval letter says 6.5%. That is not a market figure — it is what somebody told her — and this product exists to show that it is not the price either.
The contrast, computed live
The monthly payment says one thing. The clock says another.
She qualifies for both. Same house, same day, same 30-year term. What separates them is not the rate and not really the payment — it is how long each mortgage insurance lasts, and that is the one thing no comparison site puts on the screen.
FHA, 3.5% down
LTV 96.50%The pre-approval she already has
Full monthly payment, first year
$3,598.14
Mortgage insurance, first year
$191.08
Falls every year
How long it lasts
Never cancels
360 months
- How the premium behaves
- Falls every year
- Mortgage insurance over the whole loan
- $45,589
- Total cost over the 7 years she plans to stay
- $316,841
The FHA premium and its duration come straight from HUD’s published table, applied to her loan. Nothing here is modeled and nothing is a proxy: the government prices its own insurance in public, which is the one part of this whole comparison that is not opaque.
Conventional 97
LTV 97.00%Fixed only, one unit, first-time buyer, through DU — she meets all of it
Full monthly payment, first year
$3,752.51
Mortgage insurance, first year
$378.97
Flat until it ends
How long it lasts
Ends by law
month 148, automatically · January 2039
Cancellable in writing from month 137
- How the premium behaves
- Flat until it ends
- Mortgage insurance over the whole loan
- $56,087
- Total cost over the 7 years she plans to stay
- $332,971
Private mortgage insurance modeled on Enact Mortgage Insurance’s rate card — the only public one left. The other five insurers quote solely through the lender’s pricing engine, and the borrower does not even choose which one prices the loan. This is a declared proxy, not a quote.
Principal, interest, mortgage insurance, property tax and homeowners insurance. The two programs land within a few percent of each other here — which is exactly why the monthly payment is the wrong place to decide.
Two different machines: FHA charges basis points on the average balance of each year, so the premium falls. Private MI is a fixed rate on the original loan amount and never moves.
Ordering rule: total cost at the tenure horizon the buyer declared (7 years), lowest first. It is the criterion HUD blessed in writing as a neutral display in 1996. Nothing on this page is ranked by anything else, and nothing is hidden or paged.
What these totals rest on, so it can be checked: mortgage insurance duration and FHA premiums come from the rule and are exact; the private premium is modeled on the one public rate card; the rate anchor is an observed index for a broader band than hers; property tax and homeowners insurance are state-level estimates from the Census; and the seven-year horizon is an assumption until she tells us hers. Every one of those is labelled where it appears. And what these two totals do NOT carry: the lender’s own closing costs. Those are the line that varies 3.71x between lenders, they are the same question on both programmes, and they are priced against the regulator’s own file on the comparison page — not guessed at here.
What the engine returns
$16,130
FHA · 7 years
Over seven years FHA is cheaper for her, and it stays cheaper at every horizon we ran. That is not the answer anyone expects, and it is not the answer for the next person.
Her private premium costs more per month and more in total than FHA’s, because at her score the insurer prices her near the top of its card. That is not a fact about the two programs. It is a fact about her, and it reverses further up the score scale.
FHA duration depends only on LTV and term — never on the score. Above 95% LTV the annual MIP runs the full term (HUD Handbook 4000.1, Appendix 1.0). To stop paying it she has to refinance out of FHA entirely.
The Homeowners Protection Act forces the servicer to terminate private MI when the scheduled balance hits 78% of the original value, and to cancel it on written request at 80% (12 U.S.C. 4901–4902). The thresholds run on the ORIGINAL value: the house going up in price does not move this date.
The gap, year by year
Conventional minus FHA, in total dollars, at every horizon from one year to thirty. The gap grows for twelve years, and then the private mortgage insurance dies and the gap dies with it. By year thirty the two loans have cost almost exactly the same. Compare them at the wrong horizon and you get the wrong loan.
Conventional costs more →
Years in the house
And now change one number.
Same house, same day, same down payment. Only the credit score moves. The private insurer reprices, the loan-level fee reprices, and somewhere between 720 and 760 the answer flips.
| Score | Private MI, month 1 | Rate | Cheaper over 7 years |
|---|---|---|---|
| 660 | $494.77 | 6.931% | FHA$26,317 |
| 691 | $378.97 | 6.766% — general index, not this band | FHA$16,130 |
| 700 | $301.77 | 6.766% — general index, not this band | FHA$8,593 |
| 720 | $259.66 | 6.766% — general index, not this band | FHA$4,529 |
| 740 | $200.01 | 6.766% — general index, not this band | Conventional$1,534 |
| 760 | $157.90 | 6.766% — general index, not this band | Conventional$6,124 |
Nine points of credit score — 691 to 700 — is worth more to her than a lot of decimal places of rate, and no comparison site tells her that, because none of them models mortgage insurance with a duration.
The rate only moves on the 660 row, and not because the market charges the same from 691 to 760. It is that the only band with its own observed index is the one below 680; every other row falls back to the general 30-year conforming index. The flat column is a hole in the public data, not a finding about pricing.
691 → 700
Declared proxy$11,425
$77.20 × 148 months
Three things nobody else does
A price with no clock attached is half a price.
One
We show how long each component lasts, not just what it costs this month.
She has $43,000 saved. The FHA duration cliff is not at «10% down» — it is at 90.00% LTV measured on the base loan amount, and $43,000 leaves her just over the line. Four hundred and ten dollars over.
Everything she has saved
$43,000 down
$43,000
- LTV on the base loan amount
- 90.09%
- Annual MIP runs for
- 360 months
- Mortgage insurance she will pay
- $38,693
Four hundred and ten dollars more
exactly 90.00% LTV
$43,410
- LTV on the base loan amount
- 90.00%
- Annual MIP runs for
- 132 months
- Mortgage insurance she will pay
- $19,922
What the $410 buys
From the rule+ $410
more at closing
− $18,771
less mortgage insurance over the loan
24 CFR 203.284 measures the FHA LTV on the original principal obligation EXCLUDING the financed upfront premium, and Appendix 1.0 of Handbook 4000.1 cuts the duration at 90.00%. Almost every mortgage calculator on the internet adds the upfront premium into the LTV, which pushes the borrower into a worse band and then charges the premium on a bigger base — the same mistake billed twice.
Two
We geolocate the property tax and the insurance, because they are not a rounding error.
On her house the county takes more every month than the mortgage insurance does. Every national calculator uses one country-wide average for this line, and the country-wide average is wrong everywhere.
Her house, per month
Modeled- Property tax
- $450.38
- Mortgage insurance
- $191.08
- Homeowners insurance
- $187.58
2.36×
The county takes this many times what the mortgage insurer takes
The same $434,100 house, moved between states
Property tax per month on her house
- IL$649
- NJ$606
- VT$506
- TX$450
- AZ$154
- AL$135
- HI$111
Effective property tax rate, computed from the Census ACS with the NAHB method. Illinois to Hawaii is a 5.8x spread — larger than any rate difference anyone will ever quote her, and it is the line nobody shops.
These are STATE effective rates over the existing housing stock. The tax is actually set by the county and the school district, and in Texas the first escrow is often built from the seller’s bill while the buyer’s arrives reassessed at the purchase price. So this figure is the right order of magnitude and can still run low for one specific house. County-level ingestion uses the same Census tables and is the next thing we build.
Three
We warn about the year-two escrow hit, and we show the mechanism.
This is the single most common surprise in the regulator’s own complaint file, and it is the only one that can be seen coming before signing, because it is arithmetic.
At closing, before month one
From the rule$4,129
Block G of the Loan Estimate. Real cash, and no closing-cost estimator on the internet shows it.
- Of that, cushion
- $1,276 · 2
- Monthly payment, year one
- $637.96
Two months. The law caps it at one sixth of the annual disbursements and not a dollar more.
Then the servicer re-runs the account
month 14 · November 2027The shortage is spread over twelve months
A shortage is not a fee. It is the gap between what they estimated the bills would be and what the bills actually were. §1024.17(f)(3).
And the base deposit goes up
Because next year’s bill is bigger than last year’s. This one is permanent.
Both land in the same month, and that is the surprise — but only the second one is guaranteed. The first appears when the first-year estimate was too low, which in Texas is close to routine.
How big is it? That depends, and we will not invent it.
The mechanism is in the regulation. The magnitude is not: there is no primary source for what a typical year-two shortage costs. So we run both ends and label both.
If the first-year estimate was right
Modeled+ $4.50
per month
- The shortage is spread over twelve months
- $0.00
- And the base deposit goes up
- $4.50
Property tax flat, insurance up 2.4% — the bottom of the range the NAIC publishes. No shortage at all: only the base deposit moves.
If the county reassesses at the purchase price · 18%
No source+ $152.38
per month
- The shortage is spread over twelve months
- $66.81
- And the base deposit goes up
- $85.57
The normal Texas outcome when the first escrow was built from the seller’s tax bill. Now both increases fire at once. The 18% is an illustration, not a measurement — no source quantifies it.
A widely repeated figure puts this jump at about $180 a month. We could not source it. What we can do is show which assumption produces which number, and let the reader see that the whole story lives in one input nobody asks about: whether the tax bill in the first escrow was the seller’s or hers.
12 CFR 1024.17: the servicer analyses the account before opening it and at the end of every escrow account computation year. Aggregate accounting is mandatory, the cushion cannot exceed one sixth of estimated annual disbursements, and a shortage of one month or more may be collected over twelve months or longer.
The promise
No social security number. No data sold. Nobody calls you.
This is not a courtesy paragraph at the bottom of a lead-generation form. It is the architecture, and it is the reason the numbers above could be built the way they were.
We never ask for a social security number.
Six pieces of information turn an inquiry into an «application» under 12 CFR 1026.2(a)(3)(ii). We hand you five and keep the sixth. Without it there is no application, no obligation to issue a Loan Estimate, no hard credit pull, and no trigger lead sold to twenty callers. It is a legal decision wearing a UX costume.
We do not sell your information, ever.
Congress cut off the sale of mortgage trigger leads by the credit bureaus in March 2026, but left an exception when the consumer consents. Asking you for that consent would be reproducing the problem with your permission. We do not ask.
No lender calls you, and we never introduce you to one.
No warm handoff, no live transfer, no «best match», no email afterwards promoting anyone, and no introduction even if you ask for one — under Regulation Z an introduction that can affirmatively influence which lender you pick is a referral. Those are the exact behaviours the CFPB describes as steering, and a platform that does them is running a referral-fee scheme with triple damages per borrower.
What we do give you is the document that actually binds.
A Loan Estimate is free, legally due within three business days, and cannot be charged for beyond the credit report. Multiple mortgage inquiries inside a 45-day window count as one on your credit report. We build the package, hand you the script, and normalise the estimates you bring back side by side.
A disclaimer does not cure a violation — HUD said so in 1999 and it is still true. Neutrality here is structural: the ordering rule is printed under every table, nothing is hidden or paged, everyone gets the same typography, and what a lender charged is labelled history rather than dressed up as an offer.
Run it on your own numbers.
Occupancy first, county second, then the eighteen other fields that actually move the price and that nobody asks for. It takes a few minutes, it costs nothing, and it ends with a package you can take to real lenders.
No account. No social security number. No phone calls.