The Archetype CPI

§ 1 · The problem

01One average, no actual households

Every quarter — and now every month — the Australian Bureau of Statistics publishes the Consumer Price Index. It is a genuinely good statistic for what it is: the price change of a basket weighted by aggregate national household spending. The trouble is that the aggregate basket belongs to no one. It blends the outright-owner retiree (who pays no rent and no mortgage) with the student handing over most of their income to a landlord, then reports one number for both.

The distortion is structural, not incidental, and housing is where it bites hardest:

• Rents carry a weight of just 6.6% in the 2025 CPI weighting pattern, because roughly two-thirds of households don't rent. A typical renter puts 25–35% of their spending toward rent. When rents surge, the CPI shrugs; the renter doesn't.
• Owner-occupier costs enter via the “net acquisitions” approach — the price of newly built dwellings (excluding land), weighted at 7.6%. Countries like the US instead impute a rent that owners notionally pay themselves. Both conventions flatten the owner/renter distinction into a single average, and both exclude the biggest cash cost of recent buyers: mortgage interest, which the CPI omits entirely.
• Neither convention prices the transition from renting to owning. Imputing owner costs from rents assumes rents and dwelling prices move together. They don't: gross rental yields have compressed for two decades, which is arithmetic proof that prices have outrun rents. The cost of becoming an owner — the deposit, priced off dwelling values — inflates on a different track from the cost of being a renter. §7–8 treat this properly.

This document specifies a transparent method for computing archetype-level inflation from published ABS data. Every weight, every source series, and every formula is stated. Anyone with the same inputs reproduces the same outputs.

§ 2 · Inputs

02Data: all of it public

The method consumes three ABS publications, nothing else. No scraped prices, no proprietary feeds — the point is that the ABS already publishes everything needed; it just averages it away at the last step.

1. Monthly CPI, groups and expenditure classes — price movements for the 11 CPI groups and the sub-indexes we need (rents, new dwelling purchase, electricity). This is the price signal; we use it unmodified.
2. Annual weight update of the CPI — the official all-households weighting pattern (2025 update, held until January 2027). Used as the reference column, and as the sanity check in §4.
3. Household Expenditure Survey — expenditure shares by household composition, tenure, and income source, from which the archetype weights in §3 are derived.
4. Total Value of Dwellings — the mean price of residential dwellings, Australia, quarterly. The asset-price series the CPI deliberately excludes; used only in §8, where the exclusion is the subject.
5. RBA Statistical Table F6 — the average interest rate on new owner-occupier housing loans funded each month, all institutions (series FLRHOFTA). The rate an actual new buyer got, fixed-rate era included.

Price relatives for the reference month (annual change to July 2026), as published:

Housing is split into its four sub-components because tenure is the whole game: a renter faces the rents index, an owner-buyer faces new dwellings, and everyone faces utilities. Utilities uses the electricity sub-index as proxy; “other housing” (rates, repairs) uses the housing group movement. Both proxies are flagged, not hidden.

§ 3 · Archetypes

03Five households you'd recognise

An archetype is a named, fixed basket: a weight vector over the 14 categories above, summing to 100. The weights are stylised from HES tenure and life-stage patterns — they are editorial choices, published in full precisely so they can be argued with. Disagree with a weight? §6 lets you change it and watch the number move.

Reading the ledger: the official column is the ABS 2025 weighting pattern. Note what averaging does — rent is 6.6% officially, but 28–35% for the three renting archetypes and zero for the two owners. Health is 6.7% officially, 15% for the self-funded retiree. No archetype is within cooee of the average on housing.

§ 4 · Method

04The formula, and its audit

The computation is deliberately boring: a fixed-weight (Laspeyres-type) aggregation, the same arithmetic the CPI itself uses, with the weight vector swapped.

Worked example: the share-house student, July 2026

Rent 35% × 3.6 = 1.26 pp. Food 20% × 3.2 = 0.64 pp. Utilities 4% × 6.1 = 0.24 pp. The remaining eleven categories contribute – pp between them. Total: – against the official 3.5%.

The audit line

A re-weighting method is only trustworthy if it reproduces the official number when fed the official weights. Running the formula with the ABS weight column returns – against the published 3.5%. The residual (– pp) comes from our two flagged proxies and from compositional effects the ABS captures at expenditure-class level. Any release where the audit residual exceeds ±0.3 pp is a methodology bug, and the release is held.

Honest limits of the arithmetic

• Fixed weights drift. Real households substitute away from what gets dear. Fixed baskets overstate inflation slightly over long horizons (the classic Laspeyres bias). We re-derive archetype weights each time the ABS updates its own, and say so in the changelog.
• Group indexes hide within-group variation. The student buys instant noodles, the family buys school shoes; both are “food” and “clothing”. Where a sub-index materially diverges (rents vs new dwellings), we split; where it doesn't demonstrably matter, we don't pretend precision we lack.
• Capital-city aggregate. All inputs are the weighted eight-capital-cities series. A Hobart archetype and a Sydney archetype differ mainly through the rent level, not the published rent index; regional variants are future work.

§ 5 · Results

05Who's actually paying, July 2026

Applying §4 to the five baskets, against the same month's headline:

The July 2026 spread is modest — about – percentage points — because this happens to be a month where rents, food and services are inflating at similar rates. That is itself information. Run the same baskets backwards through the quarterly series, though, and the divergence episodes identify themselves:

The back-series shows the divergence has structure, not noise. Inflation arrived in two waves aimed at different households. The 2022 wave was a homebuilder's crisis: new-dwelling costs peaked at +20.6% in the year to September 2022, driving the mortgaged family to 8.7% in December 2022 while the share-house student sat at 6.2% — a 2.5-point spread inside one official number (7.8%). Then the waves crossed: through 2023–24, as the headline fell to 4.1% and then 3.8%, the renting archetypes stayed roughly a point above it (student 4.9%, pensioner 5.1% in December 2023) as the rent surge worked through. Whoever you were, the official line was wrong about you — just in different directions at different times. And this chart still excludes mortgage interest, the thing that pushed measured living costs for employee households to 9.6% in mid-2023.

§ 6 · Your basket

06Build your own index

The archetypes are arguments, not verdicts. Start from the one nearest you, adjust the weights to your actual budget shares, and the same formula returns your inflation rate for the year to July 2026.

§ 7 · The washed-out event

07The Household Formation Index

The CPI prices the ongoing consumption of households that already exist. But the single largest price event in most people's financial lives is a one-off the CPI never sees as an event: forming a household. Moving out. Bond and rent in advance, a dwelling furnished from nothing — bed, fridge, washing machine, couch, curtains, the entire kit — connections, the move itself. Every component sits somewhere in the CPI, so officially the cost of this event doesn't exist; it is smeared across a dozen expenditure classes and weighted by how often established households replace a fridge, not by what it costs to acquire all of it at once, at today's prices, on the worst savings base of your life.

The Household Formation Index (HFI) reassembles the event. Two variants, because there are two doorways:

The owner variant's dominant component is the entry price of the dwelling itself, proxied here by the ABS new-dwellings index — a construction-cost measure that excludes land, which makes every owner-formation figure below a conservative floor. The bond is refundable in principle; it is priced here because it must be produced up front, at today's rent level.

The vintage table: when you formed is what you paid

Because the HFI is built from index levels, it answers the question the flow statistics wash out: how much more does the identical formation event cost than it did for the cohort just ahead of you? Setting the December 2019 (pre-COVID) cohort to 100:

The two cohorts are the same age, in the same labour market, often on the same street. The official statistics assign them the same inflation experience. The vintage table says the 2026 cohort pays – more for the owner-formation event than the 2019 cohort did — as a floor, before land. This is what “washed out” means operationally: a between-cohort price gap with no home in any published index, because each cohort only ever appears in the statistics after the event, as an established household.

§ 8 · The gap no index prices

08The renter→owner transition

Every price index — the CPI, the living cost indexes, and the archetype series above — prices the cost of staying what you are. A renter's index tracks rents; an owner's index tracks rates, repairs and (in the LCI) interest. None of them prices the doorway between the two tenures, and that doorway has its own inflation rate.

The usual defence of ignoring it leans on imputation: if owner-occupier costs can be proxied by rents (as in the US CPI's owners' equivalent rent), then rents stand in for housing generally and the tenure boundary dissolves. The defence fails on observable arithmetic. The gross rental yield — annual rent over dwelling price — has compressed from roughly 5% in the early 2000s to near 3% in the 2020s in the major capitals. Yield compression is the statement that prices and rents are different time series: dwelling prices have persistently outgrown the rents that are supposed to proxy them. An index built on that proxy silently understates the inflation faced by anyone trying to cross the boundary.

Proposed companion indicator: the Transition Cost Index

Rather than distort the archetype baskets, we propose publishing a separate indicator beside them, for the aspiring-buyer archetypes:

• Inputs are public: ABS Total Value of Dwellings for the mean dwelling price (wired into the charts below); RBA F6 for new-loan rates; state revenue offices for duty schedules; §5 for π_renter.
• It is an indicator, not a price index — it mixes an asset price with consumption prices, which is exactly why the ABS excludes it from the CPI, and exactly why it needs to exist somewhere else. Statistical purity for the CPI; decision-relevance here.
• It quantifies the saver's treadmill: when dwelling prices rise 8% while a renter's costs rise 3.5%, a deposit saved at the renter's real rate falls behind by design. That number deserves a monthly publication slot as much as any group index.

The effective rate: aspiring buyer vs debt-free retiree

The treadmill can be expressed as a single “effective” inflation rate, and it makes the magnitude problem explicit. Take a 30-year-old renter directing 20% of their outgoings at a deposit target that inflates with the mean price of residential dwellings (ABS Total Value of Dwellings — the actual asset, land included, not the CPI's construction-cost proxy): their effective rate is 0.8 × (renter-basket inflation) + 0.2 × (dwelling-price inflation). Put that person next to a debt-free retiree who owns outright — whose exposure to dwelling prices is precisely zero — and the two are not having small differences within one statistic; they are on different series:

The doorway in dollars

Indexes abstract; the doorway itself is denominated in dollars. Combine the mean dwelling price with the average rate on new owner-occupier loans actually funded each month (RBA F6) and the first-home buyer's entry has two moving parts: the deposit, set by the price level, and the first-year repayments, set by price and rate together on a 30-year, 80%-LVR loan:

Reading the vintage rows: price, rate, deposit and repayment are as at purchase. Variable rates later reprice every cohort — what a vintage locks permanently is the principal, and the deposit already paid. The 2021 rows are the cruellest: buyers who cleared the deposit hurdle at 2.5% rates kept their small principal when rates tripled; savers who were six months short watched the deposit target and the rate rise together.

§ 9 · Prior art

09What the ABS already does, and where this differs

The Selected Living Cost Indexes are the official cousin of this work: four household types (employee, age pensioner, other government transfer recipient, self-funded retiree), an outlays approach that includes mortgage interest and excludes new dwelling purchase. They are good, and under-read. The differences here are deliberate:

• Granularity of archetype. “Employee households” spans the share-house student and the partner-and-two-kids mortgagor — the two ends of §5's chart. Income source is the SLCI's cut; tenure and life stage are ours, because housing tenure is where the dispersion lives.
• Cadence. SLCIs are quarterly; the CPI is now monthly. Building on monthly group indexes gives archetype readings twelve times a year.
• Open weights. SLCI weights are published in aggregate but the full derivation isn't reproducible from public files. Here the entire weight matrix is in the page source, versioned, with a changelog.
• The formation index and transition indicator (§7–8) have no official counterpart at all — the SLCIs, like the CPI, only ever observe a household after it exists.

§ 10 · Governance

10Versioning, licence, corrections

• Versioning. Weight matrices are semantically versioned. Changing a weight is a minor release with a stated rationale; changing the category structure is a major release. Past releases stay archived so any published number remains reproducible.
• Corrections. Errors are corrected in place with a dated note, never silently. The §4 audit line is recomputed and shown on every release.
• Licence. ABS source data is CC BY 4.0; this methodology and its weight matrices are published under the same licence. Take it, fork it, argue with it — attribution is the only ask.
• Status. Release 0.2 is a draft for comment: five archetypes, a quarterly back-series to 2016 for every archetype, and the Household Formation Index with its vintage table (0.2 added the back-series and HFI; 0.1 was the single-month snapshot). On the roadmap: monthly automation from the ABS API, regional variants, an SLCI-style mortgage-interest extension for the mortgaged archetypes, and the TCI with live dwelling-price inputs in place of the ex-land construction floor.