← Tools The Lock-In Ledger Outstanding first-lien mortgages by note rate 2013 Q1 – 2026 Q1 Calibrated to FHFA NMDB

How many people still hold a frozen mortgage — and how many are paying today's price?

The national chart hides the two things that actually decide the answer: where the house is, and what it cost when it was bought. Pick both.

CROSSOVER QUARTER

Share of outstanding mortgages by note rate

Note rate 2022 Q1 Latest Change Latest share
What is measured, what is estimated

This tool answers a question the public data does not answer directly. Here is exactly where each number comes from.

Published
The national rate mix, quarterly. FHFA's National Mortgage Database — a 5% sample of every closed-end first-lien residential mortgage in the country — publishes the share of outstanding loans in each rate band every quarter back to 2013 Q1. Reported figures used as calibration targets: 2022 Q1 (24.6 / 40.5 / 20.2 / 7.4 / 7.3), 2024 Q3, 2025 Q2, and 2025 Q3 (20.0 / 31.5 / 17.1 / 10.2 / 21.2).
Published, one dimension
State-level. FHFA publishes NMDB outstanding-mortgage statistics by state, and Redfin's analysis of that file gives the share of mortgages under 5% in every state as of 2022 Q1 — from Utah at 93.2% down to West Virginia at 76.7%. Each state's series here is fitted so that it reproduces that state's published position relative to the national figure, then run forward on the same engine. Fitted values land within roughly half a point of the published ones.
Estimated
Purchase-price bands. No public source crosses loan size or purchase price with note rate, by state, by quarter. FHFA's outstanding-mortgage file has no loan-amount dimension at all. Every price-band figure here is model output, not observation.

How the model works

It is a vintage-cohort simulation, not a curve fit. Every quarter from 1998 to 2026, loans are originated at that quarter's prevailing rate (spread across seven rate strata to reflect FHA/VA/jumbo/term mix). Each existing loan then faces three competing hazards each quarter:

  • Refinance — a logistic function of how far the loan's note rate sits above the current market rate, with a wave multiplier when rates hit a new low, and a 52% "inert" segment that rarely refinances at any incentive.
  • Turnover — the borrower sells. Suppressed by the lock-in term 1/(1+K·gap), where gap is how far below market the note rate sits. This is the mechanism the whole chart is about.
  • Payoff — the loan leaves the pool entirely.

Refinancing borrowers re-enter at the current rate and keep their purchase-price band. Selling borrowers re-enter at the current rate and at the current price distribution — which is why a home bought for $200k in 2012 and resold in 2024 leaves the lowest band.

How well it fits

The model is also constrained to reproduce the one date the whole chart turns on: nationally, loans at 6% or more overtook sub-3% loans in 2025 Q3, exactly as FHFA's published figures show.

As an out-of-sample check, the model was never fitted to 2026 Q1. It puts 68.3% of loans below 5% that quarter against FHFA's actual 66.7% — close, but a reminder that this is an estimate and the second decimal place is not real.

Where the price bands come from

Purchase prices in each state and quarter are modelled as lognormal around the state's median sale price, which is carried backwards from its 2026 level using the national median-price path scaled by a state amplitude (higher for boom-and-bust states such as Nevada and Florida, lower for steady ones such as Texas and Iowa). Larger loans refinance more readily — the payoff on a $900k balance clears closing costs at a much smaller rate move than on a $150k balance — so the refinance hazard is scaled by 1+0.22·ln(loan/median), and turnover is scaled mildly the other way.

The single most important caveat

Most of the spread you see between price bands is vintage composition, not borrower behaviour. A $1M–$2M purchase was rare in 2012 and common in 2025, so that band is packed with recent, expensive loans. A $250k purchase was ordinary in 2014 and unusual now, so that band is packed with old loans that had a decade to refinance. The bands differ mostly because they were bought at different times. That is a real and useful fact about the market, but it is not evidence that expensive borrowers behave differently.

About the chart that prompted this

The widely shared version of this chart puts the sub-3% peak at 33.3% and the 2025 Q3 figure at 22.3%. FHFA's published numbers for all outstanding mortgages are 24.6% and 20.0%. The gap is almost certainly a denominator difference — the higher figures are consistent with restricting to Fannie Mae and Freddie Mac acquisitions, which refinanced far more heavily in 2020–21 than FHA, VA and portfolio loans did. The direction of the story is unaffected; the levels are not comparable.