“Mortgage stress” arrives as one averaged headline — 1.447 million mortgage holders, 26.8%, “At Risk.” This brief decomposes it: a modelled estimate is not an observed count, and an average is not a distribution.
Roy Morgan's modelled “At Risk” estimate covers 26.8% of mortgage holders — 1,447,000 — for the three months to March 2026. The RBA's observed data, to end-2025, records around 1% of variable-rate owner-occupiers with a cash-flow shortfall. A modelled estimate of burden; an observed count of distress — two different measures.
It is a sentence built to alarm: more than one in four Australian mortgage holders, 1.4 million households, in “mortgage stress.” It sounds like a count of people in trouble — borrowers falling behind, the system under strain. It is, more precisely, something narrower: a modelled estimate of repayment burden.
The figure is Roy Morgan's. Its modelled “At Risk” measure placed 1,447,000 mortgage holders — 26.8% — above an income-scaled repayment-burden threshold in the three months to March 2026. It is a real, carefully constructed figure, published by a respected research firm. It is also not a count of borrowers actually behind on the mortgage. That is a different measure, and a different — far smaller — number.
This brief places the measures on one sheet. The point is not that the headline is wrong. It is that “1.4 million in mortgage stress” answers a specific question — how many borrowers carry a heavy modelled repayment burden? — and a reader who takes it as how many are behind on the loan? has reached for the wrong number.
The headline figure and the arrears figure are not two readings of one thing. One is a modelled estimate of repayment burden; the other is an observed count of borrowers behind. This brief draws them on two separate planes — because they are not on one axis.
| Measure | What it is | Latest verified figure | Period & source |
|---|---|---|---|
| Mortgage stress | a modelled estimate — repayment burden above an income-scaled threshold | 1,447,000 / 26.8% “At Risk” | Roy Morgan · 3 mo to March 2026 |
| Arrears | observed — borrowers with a cash-flow shortfall, or behind on payments | ~1% shortfall; <1% 90+ days | RBA Financial Stability Review · to end-2025 |
| Arrears (alternate series) | observed — 30+ days in arrears | 1.68% of mortgages | Cotality · Q1 2025 |
| Default / possession | the outcome | no current national figure verified — not quantified | — |
Roy Morgan's modelled “At Risk” share is 26.8% of mortgage holders. The RBA's observed cash-flow-shortfall measure is around 1%; 90-plus-day arrears are below 1%. The two sit roughly twenty-five-fold apart — and that gap is not an error in either. It is the distance between a model of burden and a count of distress. A modelled repayment-burden ratio is not a count of borrowers actually behind; the two were never the same number.
Both belong in the picture. Roy Morgan's estimate is correct as a modelled estimate; the RBA's data is correct as observed data. The brief alleges no error by either — it simply draws them on two planes, so neither is read as the other.
Roy Morgan's measure flags a borrower whose modelled repayment burden crosses an income-scaled threshold. The RBA's measure counts only borrowers actually behind. The designs answer different questions, so they return different numbers.
Borrowers “paying more than a certain proportion of their after-tax household income (25% to 45% depending on income and spending) into their home loan, based on the appropriate Standard Variable Rate reported by the RBA and the amount they initially borrowed.”
A borrower can clear that threshold — carry a burden the model flags as “At Risk” — and still be paying every instalment, in full, on time. The model does not claim otherwise; it measures burden, not arrears. Roy Morgan's deeper band, “Extremely At Risk” — 1,020,000 mortgage holders, 18.9% — uses the amount now outstanding rather than the amount initially borrowed: a different denominator. The two bands are not interchangeable, and the brief never averages them.
Roy Morgan itself notes that the dominant driver of the “At Risk” level is household income and employment, not the interest rate. That is why the modelled 26.8% and the observed ~1% sit roughly twenty-five-fold apart — not because either is wrong, but because one estimates a burden and the other counts who is behind. And beyond arrears sits a third measure entirely: default, the outcome — for which no clean current national figure was verified, so AU-03 names it and leaves it unquantified.
Beneath the headline, repayment difficulty is not spread evenly across borrowers. It concentrates — and the axis it concentrates on is leverage: how much was borrowed against the property, and how thin the buffer behind it.
The RBA's Financial Stability Review and its arrears research establish the distribution directly. Observed arrears run far higher for borrowers carrying the most leverage, and close to nothing for the well-buffered majority. The table below sets out that distribution — all of it one measure, observed arrears, so the cohorts are comparable to one another.
| Borrower cohort | Observed arrears | Reference |
|---|---|---|
| High-LVR · loan-to-value ratio ≥ 80% | ~2.5% | peak, 2024 — well above the whole-of-book rate |
| High loan-to-income · LTI > 4 | ~1.5% | RBA arrears research |
| Whole-of-book · 90+ days in arrears | <1% | RBA FSR Mar 2026 — declined over the past year to ~pre-pandemic |
Read the other way, the same data is a resilience picture. The median prepayment buffer is larger than before the pandemic for every income quartile; fewer than 1% of households are in negative equity; whole-of-book 90-plus-day arrears are below 1% and have declined over the past year. The well-buffered majority — most borrowers, across every income band — carries little of the strain. The exposure is real, and it is concentrated in a high-leverage, thin-buffer minority.
One axis this brief does not use is loan vintage. No publisher splits mortgage stress or arrears by the year a loan was written, so AU-03 frames the cohort by leverage — the verified proxy. Recent high-LVR lending, including first-home-buyer lending after the October 2025 expansion of the 5%-deposit scheme, clusters at the top of the LVR range, which places recent buyers disproportionately in the exposed group. But that is a statement about leverage, supported by LVR-keyed data — not a published purchase-year table.
AU-03 follows AU-01 and AU-02 inside The Australian Property Decode. The first brief decomposed how Australia measures the price of a house — three private indices, three methods. The second decomposed how it measures the rent on one — a flow measure and a stock measure. The third decomposes the borrower carrying the loan — a modelled estimate, an observed count, and a distribution.
The series asks one question of each published property figure: which population does this number actually describe? For the house price it was which sample of sales; for the rent, which sample of tenants; for mortgage stress, which kind of measure — and which borrowers, on which side of a leverage line. The headline rarely says. The brief does.
General information only. The scenario below is a modelled illustration, built to make the data concrete — it is not advice, and it describes no real person, household, or transaction.
Picture a modelled household with a mortgage, reading that 26.8% of mortgage holders are “At Risk” — more than one in four. They read it as a one-in-four chance that they, specifically, are in trouble. The number has set the mood at the kitchen table, and shaped how they think about the year ahead.
But 26.8% is the modelled “At Risk” estimate — the share of mortgage holders whose modelled repayments cross an income-scaled burden threshold. It is not a one-in-four probability of falling behind on the loan. The measure that counts borrowers actually behind — the RBA's observed data — sits near 1%. The two answer different questions, and this household has been reading the answer to the one it did not ask.
Whether this modelled household sits in the genuinely exposed group is not settled by the headline average at all. It turns on leverage and buffer — how high their loan-to-value ratio is, how much of a prepayment buffer they hold. The RBA's data shows the exposure concentrated there: arrears reached around 2.5% at their 2024 peak for high-LVR borrowers, against a whole-of-book rate below 1%, while prepayment buffers across every income quartile are larger than before the pandemic.
The composite is illustrative — a modelled household, not a surveyed one. Its only purpose is to make the decomposition concrete: the headline is a modelled estimate laid across an uneven distribution. The brief offers no view on this household's position, and no suggestion of what they should do — only on which number the headline is, and which it is not.
If you read one thing: “1.4 million in mortgage stress” is a modelled estimate of repayment burden — not a count of people behind on the loan. That observed number is around one percent, and the exposure is concentrated by leverage.
Australia's mortgage-stress headline is not a count of households behind on the loan. It is a modelled estimate of repayment burden — Roy Morgan's “At Risk” measure, 1,447,000 / 26.8% in the three months to March 2026. The observed arrears figure — borrowers actually behind — is far smaller: around 1% with a cash-flow shortfall, fewer than 1% 90 or more days in arrears, by the RBA's data to end-2025. Both numbers are real; they measure different things. Mirror Brief AU-03 makes one claim: read whether a mortgage-stress number is a modelled estimate or an observed count before reading it as “how many are in trouble” — and remember the headline is an average laid across a distribution concentrated, by leverage, in a minority. Every figure here is verbatim from a publisher's own release. The brief alleges nothing against Roy Morgan, the RBA, or Cotality — each measure is correct for what it measures.
Mirror format — RAOSCAFF anchors on the publishers' own released figures (Roy Morgan, the RBA Financial Stability Review, Cotality), places the modelled stress measure beside the observed arrears measure, and decomposes the design difference behind the gap between them. No primary data collection, no analyst estimate, no extrapolation.
Every figure traces to a Roy Morgan release, the RBA's March 2026 Financial Stability Review, the RBA's arrears research, or Cotality's mortgage-arrears article, fetched live on 21 May 2026 by Phase 0. Each figure carries its reference period in the same sentence — Roy Morgan is the three months to March 2026; the RBA FSR is data to end-2025; Cotality is Q1 2025; the cash rate is 5 May 2026 — and periods are never blended. Full source list in the companion FACTS.md.
FACTS.md is the source-of-truth file; every figure in this report traces to it. The hero is a two-measure panel built from the publishers' verified figures, with the modelled and observed measures drawn on separate planes so neither is read as the other. Phase 0 returned GREEN with one framing constraint: because no publisher splits mortgage stress by loan vintage, the cohort axis is leverage — loan-to-value ratio, loan-to-income, buffer — the verified proxy, and AU-03 claims no purchase-year table.
Default and possession are not quantified — no clean current national figure was verified. The cohort is framed by leverage, not loan vintage. Roy Morgan's “~1.6 million” figure is a projection of a post-further-rate-rise scenario and is not used as a current number; the current figure is 1,447,000 / 26.8%. The brief makes no forecast of stress, arrears, or interest rates, and does not adjudicate which measure is most useful — each answers a different question.
Predict-not-recommend. Defamation-disciplined: the brief critiques what each measure measures — never the integrity, competence, or honesty of any publisher or person; Roy Morgan's modelled estimate is stated to be correct as a modelled estimate, the RBA's observed data correct as observed data. Roy Morgan, the Reserve Bank of Australia, and Cotality are cited as the authoritative publishers of their respective series. The brief adds only the cross-measure comparison no single publisher prints.