Housing affordability: methods note
Reference companion to the mortgage-spreads report. The rebase table and its base-year sensitivity, the correlated-measures disclosure, the metro-dispersion accounting, the buydown detail, the annual-vs-monthly choice, and the provenance of every series. The report states results; this note defends them.
1. What the report measures, and what it does not
The report's central object is the monthly cost of owning a median home as a share of median household gross income. Two implementations are used:
- P&I burden: principal and interest only, 20% down payment, 30-year fixed at the prevailing Freddie Mac rate, applied to the Case-Shiller-anchored median price. Computed annually back to 1987.
- HOAM: the Federal Reserve Bank of Atlanta's Housing Affordability Index. Per its published methodology, HOAM assumes a 10% down payment (not 20%), a 30-year fixed rate from Freddie Mac, and adds property taxes, property insurance, and private mortgage insurance (PMI, at 0.558% of the mortgage amount) to the numerator. The measure and its inputs are described at the Atlanta Fed's Home Ownership Affordability Monitor page.
The measure answers one question: how expensive is the typical home for the typical household? It does not answer who is actually transacting. Roughly 30% of 2024 purchases were all-cash; existing owners carry realized equity; the lock-in effect self-selects the transacting pool. The measure tracks the exclusion of marginal buyers, not the bindingness of the constraint on transactions. The composition data lives here; the report uses the measure only as a summary of the affordability burden, not as a claim about who faces it.
2. The price-to-income rebase, and why the base year matters
The report's structural claim (that today stacks an index-record price on a 6%-plus mortgage) rests on dividing the S&P CoreLogic Case-Shiller US National Home Price Index by the Census median household income, both rebased to 100 at a shared start year. The headline 2024 gap is sensitive to that choice:
| Base year | 2024 gap (index/income, rebased to 100) |
|---|---|
| 1987 | 1.51 |
| 2000 | 1.54 |
| 2012 | 1.39 |
Anchoring to 2000 (which brackets the bubble on both sides) gives 2024 the cleanest win over the 2006 peak: a 1.54 gap against 1.50 at the bubble top. Anchoring to 1987 (the first year Case-Shiller and median-income data overlap) gives 1.51: a knife-edge, but a real one. Anchoring to 2012 (the post-crash trough) understates today's gap because 2012 is itself a series low; anyone rebasing to 2012 and claiming relief is choosing the friendliest start. The report asserts the structural combination (an index-record price stacked on a 6-per-cent-plus mortgage) without resting on the knife-edge margin; the sensitivity table lives here.
3. Three measures, same inputs: not independent replications
Three affordability readings agree on direction, less so on margin:
| Measure | 2006 | 2024 | Inputs |
|---|---|---|---|
| Price/income rebase (1987 anchor) | 1.50 | 1.51 | Case-Shiller, median HH income |
| P&I burden (20% down, 30y fixed) | 29.2% | 30.4% | + 30-year mortgage rate |
| Atlanta Fed HOAM (10% down; adds PMI, tax, insurance) | 42.3% | 43.3% | + PMI, property tax, insurance |
These are not independent replications. All three divide a price index by median household income; two also divide by the same mortgage rate; the third adds the same tax-and-insurance load. They confirm the arithmetic (when three measures built from overlapping inputs move together, the shared signal is real), but they should not be sold as three separate lines of evidence. The report uses them to confirm direction; it does not multiply the margins.
4. The mortgage basis: what it is, what widened, and the 10y-vs-30y choice
The 30-year mortgage rate decomposes into a risk-free leg plus a mortgage basis: the spread originators and the MBS market add over the duration-matched Treasury yield. What widened in 2022–23 was not originator margin, which was compressing over the window. It was the MBS-Treasury option-adjusted spread (OAS), pushed out by three forces:
- Federal Reserve balance-sheet runoff removed the marginal buyer of agency MBS.
- Post-SVB, banks had no appetite for the paper.
- Rate volatility stayed elevated, raising the option cost embedded in MBS.
The OAS is the historically faster-reverting piece (it normalizes as the MBS market finds private buyers and as rate vol settles), which is why the basis was identified as the component of the 2024 mortgage rate most likely to unwind without the Federal Reserve doing anything.
Why the 10-year, not the 30-year, Treasury. Mortgage duration runs roughly five to seven years; the 30-year Treasury over-matches duration and compresses the basis. The convention is the 10-year. Recomputing on DGS30 gives a 2024 basis of 2.32pp at the 89th percentile, close to the 10-year figure, but the 30-year historical distribution is contaminated by the Treasury's suspension of 30-year issuance from February 2002 to February 2006, during which FRED publishes fitted rather than market yields. Excluding that window moves the 30-year-basis median from 1.32pp to 1.38pp. The 10-year basis has no such gap and is used throughout. The basis series runs monthly from 1971 (645 observations through 2024); median 1.656pp.
5. Metro dispersion, and why the national figure sits above the metro median
Across 671 metros in March 2026, housing cost as a share of median household income:
| Percentile | Cost share |
|---|---|
| 10th (most affordable) | 23.5% |
| Median (unweighted metro) | 35.2% |
| 90th (least affordable) | 53.1% |
That is a 2.3-fold spread. The national figure (41.2% for Q1 2026) sits above the unweighted metro median, which looks incoherent until the weighting is named: the national HOAM is computed on national medians, while 35.2% is the middle of 671 metros unweighted by population. Population concentrates in the expensive metros. The typical metro is more affordable than the typical American's housing market. The min/max endpoints (Madisonville KY at 12% to Jackson WY at 236%) are unweighted tail outliers and are not cited in the report; p10–p90 is the honest range.
6. Builder buydowns: evidence versus forecast application
For roughly two years, builders have subsidized buyers' rates below the headline 30-year through buydowns (permanent and temporary). The payment series used in the report does not see those effective rates; it sees the headline. This breaks the transmission from "mortgage rate falls → new-home demand rises" in a way that does not apply to existing-home volume, where the headline rate is the rate the borrower pays.
That distinction scopes what the report excludes, and it is narrower than "new homes are unreliable." Two things, kept separate:
New-home sales as corroborating evidence: the report uses them, and the buydown caveat makes them a stronger test, not a weaker one. New single-family sales fell roughly 9 percent over the retracement window despite builder subsidies actively propping that segment. Buydowns were the one place relief reached the borrower, and activity still declined. Stated that way, the caveat stops undercutting the number and becomes the sharpest part of the volume evidence.
New-home demand as a forecast application: the report excludes it. Because the headline mortgage rate does not govern builder-subsidized demand, keying a homebuilder-demand forecast to the headline rate would be a category error. The report confines its forecast implication to existing-home volume and origination, where the transmission from headline rate to borrower is intact.
(Specific buydown rate points are not pinned here; the point is directional.)
7. Annual vs monthly: which to cite when
Annual averages smooth peaks. The choice depends on the point:
- When the point is the extreme, cite the monthly reading: October 2023 HOAM of 45.3% payment share, the worst month in the series. The 2024 annual average (43.3%) understates it.
- When the point is the central tendency over a year (the basis-reversion comparison), cite the annual average: HOAM moved from a 2024 average of 43.3% to a Q1-2026 average of 41.2%.
- Median household income is annual (Census, September release). 2025 is not yet published at the time of writing. Series that update monthly (mortgage rates, the HOAM index, Treasury yields) are carried through July 2026 where it matters. The 2024 vintage therefore governs every annual comparison, and that limitation is disclosed in the report rather than hidden.
Two post-2019 deterioration figures appear in the report, and both are correct: they belong to different measures. The HOAM index worsened by 14.3 points from 2019 to 2024 (29.03 → 43.32); this is the figure in the reversion paragraph, where the comparison is HOAM-to-HOAM. The P&I burden series worsened by 12.7 points over the same window (17.7 → 30.4); this is the figure in the counterfactual, where the comparison is P&I-to-P&I. The two series share inputs but differ in numerator (HOAM adds taxes and insurance), so their levels and their swings differ. Each comparison is internally consistent; the two numbers are not in conflict.
8. Case-Shiller as a repeat-sales anchor
Case-Shiller is a repeat-sales index. It tracks the price trajectory of the same homes reselling over time, not the literal price of a representative house. The $85,000 figure cited for 1987 (in 2024 dollars) is an index-anchored illustration obtained by scaling the 2024 NAR median existing-home price ($410,000) back along the index, used only to make the 1987 mortgage-rate burden intuitively comparable. It is not a Census-measured 1987 price.
9. The 2012 labelling trap (forward reference)
Two distinct quantities share the word "2012":
- The 2024 gap measured from a 2012 start: 1.39 (row in the rebase table). This is today's price/income position if you treat 2012 as year zero.
- The gap measured at 2012, on the 1987 base: 1.09. This is a series low. It shows how affordable 2012 itself was relative to the 1987 anchor.
The close-arithmetic that turns on this distinction (returning to "the 2012 level" requires a 28% price fall or a 39% income rise) belongs to the follow-up piece on how affordability gaps close and is not used in this report. It is parked here only to flag the labelling trap: the two quantities share the word "2012" and nothing else.
10. Data sources and vintages
| Series | Source | Frequency | Vintage used |
|---|---|---|---|
| Case-Shiller US National (CSUSHPISA) | S&P / FRED | Monthly | through 2024 annual |
| Median household income (MEHOINUSA646N) | Census / FRED | Annual | 2024 (latest published) |
| 30-year fixed mortgage (MORTGAGE30US) | Freddie Mac / FRED | Weekly → monthly avg | through 2026-07 |
| 10-year Treasury (DGS10) | Fed Treasury / FRED | Daily → monthly avg | through 2026-07 |
| 30-year Treasury (DGS30) | Fed Treasury / FRED | Daily | cross-check only (see §4) |
| HOAM index | Federal Reserve Bank of Atlanta | Monthly | through 2026-03 |
| HOAM metro dispersion | FRB Atlanta | Monthly, 671 metros | 2026-03 |
| New single-family sales (HSN1F) | Census / FRED | Monthly | through 2026-06 |
| Housing starts (HOUST) | Census / FRED | Monthly | through 2026-06 |
| Existing-home sales (EXHOSLUSM495S) | NAR / FRED | Monthly | 2025-06 onward (FRED truncation) |
| First-time buyer share | NAR Profile of Home Buyers and Sellers | Annual | 2024 (24%), 2025 (21%) |
| Cash share | Redfin, NAR | Monthly/annual | 2024 |
Why the 2024 vintage governs annual series. Median household income is the binding constraint: Census publishes it annually in September, and the 2025 figure is not out at the time of writing. Every annual affordability comparison therefore uses 2024 income. Monthly series are not so constrained and are carried to their latest observation. The report states this rather than letting the reader discover it.
11. The spread-normalization consensus: sourcing the report's foil
The report opens on a reversal of the widely-held 2024–25 view that mortgage spreads were abnormally wide, would normalize, and that this normalization was housing's rate-side unlock. That foil is load-bearing for the opening, so it is documented here rather than left as assertion. Three representative 2024–25 sources state the view explicitly:
HousingWire, 2025 mortgage-rate forecast (Nov 2024): "We expect stability on the underlying bond markets to allow mortgage spreads to continue to ease a bit lower in 2025... A slowly decreasing spread gives us a little optimism that mortgage rates will touch the low end of the range." Spread compression is named as the mechanism delivering rate relief.
HousingWire, "Mortgage spreads are almost back to normal" (Jul 2025): "Mortgage spreads have improved significantly since 2023, leaving us just 0.49% from normal levels... we can achieve near 6% mortgage rates without the 10-year yield dropping below 4%, provided spreads continue to improve." Spread normalization is framed as the route to affordable rates independent of the Treasury leg, precisely the transmission the report argues has fired without producing the relief.
Fannie Mae, Economic & Housing Outlook (Sep 2025): forecast the 30-year mortgage to end 2025 at 6.4% and 2026 at 5.9%, a path that assumed further easing in the spread component alongside a stable-to-falling Treasury yield.
The report's phrasing ("the consensus," "the working assumption in housing-market commentary," "most analysts were watching") refers to this set of forecasts and commentary. The reversal the report reports is against this documented expectation, not a strawman.