Housing credit channel: methods note
Reference companion to the housing-credit-channel report, covering definitions, data provenance, delinquency measures, calibration and the price-to-income counterfactual.
Standalone for a reader who has not seen piece one. The HOAM measure definition and the exclusion-vs-bindingness framing are inherited from the first methods note and are pointed to, not re-derived. The composition data (cash share, first-time-buyer share, lock-in) is owned by this piece: piece one dropped that material from its body, and piece two carries it (§9). The 2012 counterfactual, parked as a forward reference in piece one's note §9, is earned here in full (§13).
1. What this piece claims, and what it does not
The piece makes one structural claim and names one limit:
- Structural claim (observed): the affordability extreme that coincided with forced-selling in 2006 has not produced forced-selling now, because the financing structure this cycle excludes entrants rather than imperiling owners, and both triggers the literature identifies for mortgage default (negative equity and illiquidity) are currently quiescent.
- Limit (interpretation, labeled): this rules out a 2008-shaped correction, not a correction. Demand-driven declines need no forced sellers.
The piece does not forecast. It names conditions and where they sit. Beat 7b names the variable to watch (the labor market) and its current reading; naming a variable is not predicting its path. The counterfactual in §13 is labeled counterfactual throughout and is mechanism-independent: it states how large a correction would have to be to close the gap, not how it would arrive.
2. The two default triggers: the Elul anchor
The piece's mechanism rests on a loan-level decomposition of default, not a stylised fact:
Elul, Souleles, Chomsisengphet, Glennon and Hunt, "What 'Triggers' Mortgage Default?" (Federal Reserve Bank of Philadelphia WP 10-13; American Economic Review 100:2, 2010, DOI 10.1257/aer.100.2.490): "both negative equity and illiquidity are significantly associated with mortgage default, with comparably sized marginal effects."
Data window: first-mortgage originations of 2005 and 2006 (LPS/McDash loan-level data, ~70% of US originations, merged with Equifax), performance followed through April 2009.
This citation does two things. It grounds the default chain (negative equity or illiquidity → default → foreclosure → supply flood → price collapse) in loan-level evidence rather than the piece's own assertion. And it names both triggers, so the labor-market leg (§12) is the same paper's second trigger, not a bolt-on. Beats 3–5 cover the equity/default leg; beat 7b covers the illiquidity leg. The piece's own anchor is satisfied, not undercut.
3. DRSFRMACBS vs MBA NDS: coverage composition, not a threshold difference
Two delinquency series appear in the piece. Both measure 30+ day past-due loans (plus nonaccrual). The level gap between them is coverage composition, not a threshold difference.
The 30+ day definition rests on documentation, not inference. The Federal Reserve's Charge-Off and Delinquency Rates release (updated 2026-05-19) defines delinquent loans as "those past due thirty days or more and still accruing interest, or in nonaccrual status" (federalreserve.gov/releases/chargeoff/chgallsa.htm). The FFIEC Call Report Schedule RC-N instructions confirm the single-family residential item captures loans "past due 30 days or more or in nonaccrual status." DRSFRMACBS is the series built from those Call Report items for all commercial banks.
The level gap today is composition. DRSFRMACBS (1.89%, Q1 2026) sits below MBA NDS (4.44%, Q1 2026) because the bank-held book skews prime: post-2008, originations shifted toward agency securitization and nonbank servicers, leaving the loans that remain on bank balance sheets cleaner than the broad market. MBA NDS covers all servicers (~75% of first-lien mortgages) including nonbanks and FHA/subprime that DRSFRMACBS omits, so it is the broader measure and leads the default section; DRSFRMACBS corroborates with the caveat above. Stated this way wherever both appear; never presented as a threshold difference.
4. MDSP: low, but not the series low (calibration correction)
The Mortgage Debt Service Payments-to-Disposable Personal Income ratio (MDSP) reads 5.88% in Q1 2026 against an 8.95% crisis peak (Q4 2007) and a 6.07% full-history median, at the 28th percentile and below the median.
It is not the series low. The trough was 4.76% in Q1 2021, the refinancing wave. The plan called 5.88% the "series low"; the script (scripts/article-credit-channel.ts, writing to credit-channel.json) caught the error before the article was drafted. The honest framing: owners' mortgage debt-service burden is low by historical standards and far below the crisis peak, but the trough was the 2021 refi wave. The "series low" claim is dropped everywhere.
5. Negative equity: level, the QoQ disclosure, the average buffer
CoreLogic (Cotality), Q4 2025 (released 2026-03-12):
| Measure | Today | Peak (Q4 2009) | Today / peak |
|---|---|---|---|
| Underwater share | 2.2% | 26% | 0.085 |
| Underwater count | 1.2M | 11.1M | 0.11 |
QoQ disclosure (the fragility-publishing move): the underwater share has been edging up off its trough in recent quarters. The article discloses this rise; it does not hide it. The precise prior-quarter level is not pinned here: CoreLogic's Q4 2025 release reports the rise directionally, but the exact Q3 2025 share is not confirmed from primary sources, so the article states the trend without asserting a specific quarter-on-quarter percentage. The level is still roughly one-twelfth of peak (2.2% vs 26%); the trend is stated, the conclusion unchanged.
The average homeowner equity buffer (~$295k) is supporting color, not load-bearing, and is held here rather than carried in the article body.
6. Homeowner vacancy: level and trend
The homeowner vacancy rate (Census, RHVRUSQ156N) reads 1.20% in Q2 2026 against a 2.90% crisis peak (Q1 2008) and a 1.50% full-history median, at the 30th percentile.
Trend disclosure (the open item): the series low was 0.70% in Q2 2023. Vacancy has since risen. At 1.20% it sits roughly 0.5 points off the trough. The piece discloses this rise wherever the number appears, on the same discipline as the negative-equity QoQ disclosure: selective trend-disclosure is an asymmetry a hostile reader will find, and the postcard sentence's credibility depends on publishing the fragility alongside the level. The level is still well below the crisis peak; the trend is stated, the conclusion unchanged.
7. Survivorship as structural point, not bias
A natural objection to today's low delinquency readings: the clean post-2008 origination cohort is a survivorship artefact, so the low levels prove nothing about the structure's durability.
The piece folds this in as the structural point rather than batting it away. The clean cohort is itself the mechanism: post-2008 underwriting plus the shift to agency securitization structurally excluded the high-risk originations that drove the 2006–08 wave. The underwriting change is regulatory fact, not interpretation: the Dodd-Frank Act (Pub. L. 111-203, §1411, codified at 15 U.S.C. §1639c) added an ability-to-repay requirement to TILA, effective January 2014, under which creditors must verify income and assets and assess a borrower's reasonable ability to repay; the Consumer Financial Protection Bureau's Qualified Mortgage rule (12 CFR §1026.43) defines the loan categories that satisfy it. These rules replaced the stated-income, low-documentation lending that characterized the 2005–06 cohort with documented-income, verifiable-capacity underwriting. The survivorship is not a measurement bias to dismiss; it is why owners are not imperiled. Stated this way, the objection converts into the strongest version of the piece's claim.
8. The exclusion-vs-bindingness clarification
The HOAM measure tracks how expensive a median home is for a median household. It does not track 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 objection "median household ≠ marginal buyer" converts to a clarification rather than a defense. HOAM measures the exclusion of entrants (how many would-be first-time buyers the market prices out), not the bindingness of the constraint on transactions. The composition data in §9 shows the exclusion directly. The measure is used as a summary of the affordability burden and as an index of exclusion, not as a claim about transaction-level distress.
Full framing lives in the first methods note §1 and §10; piece two inherits it without restating the derivation.
9. Lock-in: the FHFA NMDB gap
FHFA National Mortgage Database, distribution of the outstanding mortgage stock (not new originations), Q1 2026 (released 2026-06-26):
| Outstanding rate | Share of stock |
|---|---|
| Under 4% | 49.9% |
| Under 5% | 66.7% |
| Over 6% | 22.1% |
New-origination 30-year fixed: 6.66% (Freddie Mac PMMS, 2026-07-30).
Half the outstanding stock sits below 4% against a ~6.66% market rate. That is the lock-in gap, pinned to the stock distribution rather than inferred from the rate gap alone. Lock-in keeps existing owners in place (suppressing supply) and concentrates the affordability burden on entrants who face the full market rate, which is the mechanism by which today's structure excludes rather than imperils.
Composition data (owned by this piece): cash share ~30% of 2024 purchases (Redfin 32.6%; NAR 28–31%); first-time-buyer share 21% in the NAR 2025 Profile (transactions Jul 2024–Jun 2025), the lowest since the series began in 1981, against a ~40% pre-2008 norm. These show the exclusion directly and are owned by this piece; their provenance lives in the first methods note §10.
10. Months-supply vs raw inventory (the stock-growth caveat)
Two supply measures appear, in this order of authority:
- Months supply (leads): 4.6 (June 2026) vs ~11.0 at mid-2007 (NAR all-time high). This is normalized: sales at the current rate would clear the inventory in 4.6 months. It is the like-for-like comparison across the window and leads the supply section.
- Raw inventory (secondary, with caveat): 1.56M (June 2026) vs ~4.0M at mid-2007. Raw counts across 2007–2026 are not like-for-like: the US housing stock grew roughly 19% over the gap, so the same physical inventory represents a smaller share of the market today. The raw figure is cited only with the stock-growth caveat, or dropped under word pressure.
Months-supply carries the section alone if needed; the raw figure is corroborating.
11. Vintage map
The pinned figures span roughly nine months. "Current" is never used without a date. The beats resting on the oldest vintages age first.
| Vintage | Figure | Used in |
|---|---|---|
| Q4 2025 (released 2026-03) | CoreLogic negative equity 2.2% / 1.2M | §5 |
| Full-year 2025 (released 2026-01) | ATTOM foreclosure filings 367,460 | §3 corroboration |
| Q1 2026 | MBA NDS 4.44% / 0.64% foreclosure inventory | §3 |
| Q1 2026 | HOAM Q1 avg 41.2% (inherited) | throughout |
| Q1 2026 (released 2026-06) | FHFA NMDB outstanding-rate distribution | §9 |
| Q2 2026 | Homeowner vacancy 1.20% | §6 |
| June 2026 | UNRATE 4.2%; NAR months-supply 4.6; inventory 1.56M | §10, §12 |
| 2026-07-30 | New-origination mortgage 6.66% (Freddie PMMS) | §9 |
| Series-history | All percentile ranks, crisis peaks, medians | throughout |
Oldest-vintage claims: the negative-equity reading (Q4 2025) and the ATTOM filing count (full-year 2025). A Q1 2026 CoreLogic release will refresh the first; the MBA NDS (Q1 2026) is the freshest default-channel reading and carries the default section's lead.
12. The labor-market timing: sequence, not lead
UNRATE reads 4.2% (June 2026) against the 10.0% October 2009 peak, at the 23rd percentile of postwar history. This is the illiquidity trigger (Elul's second leg), and it is currently quiet.
Timing: one episode, not a general lead. In 2007–10, unemployment peaked in October 2009 and foreclosure inventory crested roughly a year later (Q4 2010, at 4.63%). The piece states this as a sequence in that episode: "in that one cycle, unemployment peaked about a year before foreclosure inventory did." It does not generalise to "unemployment leads foreclosure by about a year," which would assert a lead-lag relationship from a single observation. The sequence is used to handle the objection that low delinquencies are a lagging indicator: the mechanism by which a labor-market move would transmit to default takes time, so today's low delinquency reading reflects today's labor market with a lag, not a structural guarantee.
Equity cushions do not protect a borrower who loses their income. That is the substance of the labor-market leg: the structural protections (equity, underwriting, lock-in) are durable and turn slowly, but the illiquidity trigger is cyclical and can move in quarters. The labor market is therefore the variable to watch, and it is currently quiet.
13. The 2012 counterfactual: fully earned here
The 2012 trap counterfactual was parked in the first methods note §9 as a forward reference. Piece two earns it.
To return the price/income ratio to its 2012 level (1.087 on the 1987 base) from its 2024 level (1.51) would require either a roughly 28% fall in home prices or a roughly 39% rise in incomes, holding the other fixed. Both are calculated, counterfactual, and labeled as such throughout.
Mechanism-independent. The magnitude does not depend on the channel. A 28% fall is a 28% fall whether it arrives through forced-selling distress or through demand withdrawal. This clause makes the counterfactual robust to any assumption about how a correction arrives (which is the question beat 7 handles) by separating how large (this section) from how (the article's beat 7). The two questions are distinct and the counterfactual answers only the first.
The piece does not derive an income-growth-rate path to that 28%/39% (e.g. at 3–4%/yr, incomes reach the gap in roughly a decade). That derivation is held here as supporting arithmetic, not carried in the article, because it edges toward forecast and the article predicts nothing.
14. Data sources
| Series | Source | Frequency | Vintage used | Tier |
|---|---|---|---|---|
| DRSFRMACBS (30+ & nonaccrual, bank-held SFR) | Fed/FFIEC Call Report via FRED | Quarterly | Q1 2026 | 1 |
| MDSP (mortgage debt service % DPI) | BEA/Fed via FRED | Quarterly | Q1 2026 | 1 |
| RHVRUSQ156N (homeowner vacancy) | Census via FRED | Quarterly | Q2 2026 | 1 |
| UNRATE (unemployment) | BLS via FRED | Monthly | June 2026 | 1 |
| MBA NDS (30+ delinquency, foreclosure inventory) | Mortgage Bankers Association | Quarterly | Q1 2026 | 2 |
| CoreLogic negative equity | CoreLogic/Cotality | Quarterly | Q4 2025 | 2 |
| ATTOM foreclosure filings | ATTOM Data Solutions | Annual | full-year 2025 | 2 |
| NAR months-supply, inventory | National Association of Realtors | Monthly | June 2026 | 2 |
| FHFA NMDB outstanding-rate distribution | FHFA | Quarterly | Q1 2026 | 2 |
| Elul et al. default triggers | AER 100:2 / Philly Fed WP 10-13 (2010) | — | 2005-06 originations through Apr 2009 | 2 (causal) |
| HOAM, P&I burden, Case-Shiller, basis reversion, 2012 gap | inherited | various | see piece one's note | 1/2 |
Tier-1 series scripted in scripts/article-credit-channel.ts, written to notes/article-data/credit-channel.json. Tier-2 figures hand-pinned with URL and date in notes/article-data/credit-channel-citations.json. The fact ledger (notes/article-data/credit-channel-fact-ledger.md) reconciles both tiers and carries the vintage map and reconciliation rules.