Housing lock-in decay: methods note
Methods behind the mortgage lock-in study: national loan counts, payoff comparisons, amortization assumptions, runoff scenarios and source vintages.
Read the research article for the findings and their implications.
1. The analysis measures runoff, not a release date
The article separates two processes. A narrower gap between outstanding mortgage rates and new-origination rates can increase refinancing and sales. Mortgage terminations also continue while that gap remains open through voluntary payoff, adverse termination and maturity. The analysis measures the second process over an observed interval and presents arithmetic on the existing stock under stated constant-runoff assumptions. It does not forecast mortgage rates, sales, listings or net inventory.
A mortgage payoff is not necessarily a home sale: voluntary payoff combines refinancing and property sales. A sale is not necessarily an addition to net inventory either, because a seller may buy another home. The national runoff calculation therefore informs a transaction-volume baseline but does not map one-for-one into transactions or available inventory.
2. FHFA supplies the national stock calculation
Observed data. The Federal Housing Finance Agency’s National Mortgage Database is a nationally representative sample of closed-end first-lien residential mortgages. Its 2026 Q1 aggregate release reports total active loans and their interest-rate distribution, with VALUE1 weighted by loan count and VALUE2 weighted by unpaid principal balance (UPB). The public release provides rate buckets, not a rate-by-vintage cross-tab.
Calculation. The estimated sub-4% mortgage count equals the published total loan count multiplied by the combined count-weighted shares below 3% and from 3.00% to 3.99%. That produces 33.104m loans in 2022 Q1 and 25.670m in 2026 Q1. From 2022 Q1 to 2026 Q1, the estimated bucket therefore declined by 7.434m, or 22.46%, and 77.54% of the peak count remained. Compound annual net runoff over that interval was 6.16%, rounded to 6.2% in the article.
The same method applied to UPB produces an estimated $8.457tn in 2022 Q1 and $6.045tn in 2026 Q1. The resulting 28.52% net decline is equivalent to 8.05% compound annual UPB runoff, rounded to 8.1% in the article.
These are aggregate-bucket calculations, not the survival history of a tracked cohort. FHFA’s totals and published shares are rounded. The change is net runoff after any new loans enter the bucket, so gross exits would be larger if inflows occurred.
3. Scheduled amortization is removed multiplicatively
UPB falls when borrowers make scheduled principal payments even if no mortgage terminates. The 8.1% UPB runoff therefore cannot serve as a household lock-in clock.
Assumption arithmetic. The scheduled-amortization proxy begins with 22 Freddie Mac 30-year TBA-eligible cohorts whose weighted-average current mortgage rate was below 4% in the March 2022 factor. The cohorts represented 84.64% of the selected UPB and had a 3.13% weighted-average note rate, an 18.2-month weighted-average loan age and a 337.3-month weighted-average remaining term. Each cohort’s remaining balance is projected for 48 monthly payments using its observed note rate and remaining term, with exits and curtailments set to zero and the starting mix held fixed. Aggregate scheduled paydown is 9.58% over 48 months, equivalent to 2.49% a year.
The decomposition is multiplicative because scheduled amortization and exits act on a declining balance:
observed FHFA UPB retention = scheduled-only Freddie retention × implied exit retention
Dividing FHFA’s four-year UPB retention by the scheduled-only retention and annualizing the result gives the implied exit-driven UPB runoff used below.
The count runoff and exit-driven UPB runoff are separately constructed on the same FHFA sub-4% bucket from 2022 Q1 to 2026 Q1, while the UPB construction imports the Freddie amortization adjustment. Their 0.45 percentage-point difference over that interval lies inside the coverage and construction mismatch and cannot show whether larger or smaller loans exit faster.
4. The Freddie mix can lower the exit-runoff estimate
The 2.49% scheduled-amortization proxy covers qualifying Freddie 30-year TBA cohorts. FHFA All Mortgages also includes 15-year, FHA/VA, jumbo and non-agency loans, and the term and product shares within the national sub-4% bucket are not pinned.
Sensitivity calculation. At the Freddie proxy’s 3.1329% weighted-average note rate, a new 15-year level-payment mortgage amortizes 5.31% of principal in its first year, compared with 2.04% for a new 30-year mortgage. This does not estimate the national term mix. The calculation gives 5.71% implied exit-driven UPB runoff from 2022 Q1 to 2026 Q1 after removing the 2.49% amortization rate derived from March 2022 Freddie 30-year TBA cohorts; because the national book also contains 15-year, FHA/VA, jumbo and non-agency loans, higher true amortization puts exit runoff below 5.71%.
Sensitivity calculation. If a constant annual UPB runoff of 8.05% is assumed, the arithmetic balance half-life is 8.26 years. That is not a lock-in clock, because the UPB measure includes 2.49% scheduled amortization on mortgages whose households may remain locked in. The article omits this balance half-life.
5. Freddie measures a narrower form of termination
Observed data and calculation. Freddie Mac’s Daily Prepayment Report groups 30-year TBA-eligible securities by coupon and origination year, omits cohorts below $500m in current UPB, and reports monthly single-month mortality. It defines voluntary payoff as full payoff through refinancing or a property sale; scheduled payments, curtailments and involuntary prepayments are excluded. Calendar-year payoff is calculated as one minus the product of monthly retention rates.
For low-rate Freddie cohorts originated in 2020 and 2021, calculated voluntary payoff was 3.64% in 2023, 3.78% in 2024 and 4.06% in 2025. These are UPB-weighted gross voluntary payoff rates for a qualifying agency subset. They differ from FHFA’s count-weighted national net runoff across every form of bucket exit after inflows. The FHFA-Freddie gap cannot be assigned to involuntary exits or any other cause because coverage, weighting and net-versus-gross construction all differ.
Within the 2022 Freddie vintage, calculated voluntary payoff for cohorts below 4% was 3.00% in 2023, 3.85% in 2024 and 4.22% in 2025. For cohorts at 5.5% or above, it was 4.74%, 6.62% and 8.26% over the same complete calendar years. Holding origination year and broad loan age constant reduces vintage confounding, but the cohorts are not otherwise identical. The comparison is descriptive and does not estimate a causal rate effect.
6. Tenure and mobility are directional proxies
Observed survey evidence. The National Association of Realtors’ 2025 Profile covers primary-residence transactions from July 2024 through June 2025 and reports an 11-year median seller tenure, against six years during 2000–08. The statistic is selected on households that sold; it is not a survival curve for all owners and must not be inverted into an annual mortgage exit rate.
Observed counts and calculation. Census Current Population Survey Table A-4 implies that 4.11% of people living in owner-occupied units in 2023 had moved in the preceding year, down from 9.05% in 2000. CPS classifies people by current tenure, not mortgages by termination, and its mobility question changed in 2004. The series is a directional indicator of mobility, not a mortgage hazard.
Neither proxy calibrates the article’s runoff assumptions. They show why a timeless turnover rate would be difficult to defend.
7. Constant-runoff sensitivities are arithmetic
Sensitivity calculation. If a constant annual loan-count runoff of 3% is assumed, 85.9% of the starting loan count remains after five years, 73.7% remains after ten years and 63.3% remains after 15 years; the arithmetic loan-count half-life is roughly 23 years.
Sensitivity calculation. If a constant annual loan-count runoff of 4% is assumed, 81.5% of the starting loan count remains after five years, 66.5% remains after ten years and 54.2% remains after 15 years; the arithmetic loan-count half-life is roughly 17 years.
Sensitivity calculation. If a constant annual loan-count runoff of 6% is assumed, 73.4% of the starting loan count remains after five years, 53.9% remains after ten years and 39.5% remains after 15 years; the arithmetic loan-count half-life is roughly 11 years.
Each sentence compounds a fixed rate against a fixed starting stock. None is a dated forecast. If the rate gap changes, the rate-sensitive payoff process changes and the constant-runoff arithmetic no longer describes the whole path.
8. Vintage map
| Evidence | Vintage used | Publication or access lag | Role |
|---|---|---|---|
| FHFA NMDB outstanding stock | 2026 Q1 | Released 26 June 2026 | National loan-count and UPB result |
| Freddie Daily Prepayment Report | August 2026 factor in workbook dated 5 August 2026; calculations use complete calendar years through 2025 | Weekly or monthly report | Voluntary-payoff cross-check and same-vintage comparison |
| NAR seller tenure | Transactions from July 2024 through June 2025; report released November 2025 | Roughly five months after the transaction window | Directional tenure evidence |
| Census CPS mobility | 2023; historical workbook corrected December 2024 | Roughly one-year publication lag | Directional mobility evidence |
| ACS owner age | 2024 | Latest annual one-year table before the September 2026 release cycle | Composition only; not used in the article |
| Existing-home sales | June 2026 in the house data file | Monthly | Gate check only; not used in the runoff calibration |
| Freddie PMMS market rate | 30 July 2026 | Weekly | New-origination comparison only |
The article names these vintages separately and does not describe the collection as current.
9. Sources and reproducibility
| Input | Source | Local artifact |
|---|---|---|
| Outstanding mortgage counts, UPB and rate buckets | FHFA National Mortgage Database | notes/article-data/lock-in-decay.json |
| NMDB definitions and weighting | FHFA aggregate statistics technical notes | notes/article-data/lock-in-decay-citations.json |
| Cohort factors and voluntary payoff | Freddie Mac Daily Prepayment Report guide | notes/article-data/lock-in-cpr.json |
| Seller tenure | NAR 2025 Profile highlights | notes/article-data/lock-in-decay-citations.json |
| Mobility | US Census Bureau, CPS Table A-4 | notes/article-data/lock-in-mobility.json |
scripts/article-lock-in-cpr.py processes Freddie cohort data, scripts/article-lock-in-mobility.py processes CPS mobility data, and scripts/article-lock-in-decay.ts combines those outputs with the FHFA stock data. Generated values and full-precision calculations live in notes/article-data/lock-in-decay.json; the editorial fact ledger is notes/article-data/lock-in-decay-fact-ledger.md.