Levels, Overlap and a Short Baseline: The Limits of the Inflation Composite
A rigorous reading of Inflation (21220) needs its construction understood as well as its reading. The certified composite for the June quarter 2026 is +2.1049σ, ranked 100.0 against 22 quarters. Three properties of the node’s design determine what that figure can and cannot support: four of its five inputs are price levels standardised against a fixed average; those four move almost as one; and the fifth rests on a 20-quarter baseline that also bounds the ranking population.
Levels, not rates
Headline CPI, Housing CPI, Trimmed Mean CPI and PPI House Construction enter the composite as index levels, each standardised against its mean and standard deviation over March 2011 to December 2025. A Z-score on an index level measures cumulative price change relative to the baseline’s average level, not the current rate of inflation.
The consequence can be quantified. Headline CPI’s baseline standard deviation is 9.2647 index points. At its June quarter 2026 level of 102.31, a year of 2.5 per cent inflation, the midpoint of the Reserve Bank’s target band, adds about 2.56 points, or roughly +0.28σ a year. An index rising steadily at the target rate therefore drifts further above its baseline mean every year. On these four inputs, a reading beyond +2σ indicates a price level well above the baseline average; it does not on its own indicate inflation above target, and a return below +2σ would require prices to fall or the baseline to be re-anchored.
Only the expectations input is a rate. The pace of inflation is read from annual growth rates, which the node page charts alongside the composite but which are not composite inputs.
Overlap: four inputs, close to one signal
Across the 60 baseline quarters, the Z-scores of the four price inputs are almost perfectly correlated: Headline CPI with Trimmed Mean CPI 0.998, Headline with Housing CPI 0.994, Housing with Trimmed Mean 0.995, Headline with PPI House Construction 0.991 and Housing with PPI House Construction 0.981. Part of this is construction: the housing group sits inside the headline basket, and the trimmed mean is calculated from the same basket. Part is shared trend: all four indices rose over the period.
The expectations input behaves differently. Over its 20 baseline quarters its Z-score correlates at −0.168 with Headline CPI, −0.154 with Housing CPI and −0.018 with PPI House Construction.
The composite therefore gives 80 per cent of its weight to what is, over the baseline, very nearly a single price-level signal, and 20 per cent to a largely independent expectations signal. Agreement among the four price inputs is not four independent confirmations. The independent information in the reading is closer to two signals than five.
Pace and relative growth are read from the annual growth rates: 6.80 per cent for housing costs, 3.94 per cent headline, 3.65 per cent trimmed mean and 5.92 per cent house construction producer prices over the year to June 2026.
The short baseline and the ranking population
The expectations series begins in March 2021, so its baseline holds 20 quarters, below the platform’s minimum observation count; this is disclosed on every certification of the node. Its baseline mean of 4.7271 per cent was set entirely within the post-2021 inflation period. Its Z-score is measured against recent experience rather than a full cycle: it read +3.1355σ in the June quarter 2022 and −1.7664σ in the March quarter 2025.
The same series bounds the ranking population. The empirical position ranks the reading only against quarters in which all five inputs are present, which begins in March 2021: 22 quarters to June 2026, above the floor of 20 at which an empirical position replaces a model-based one. No input carries a recorded series break. A position of 100.0 against 22 quarters places the June quarter 2026 above every quarter since early 2021, a window that begins as inflation was starting to rise. It does not rank the reading against the low-inflation decade before it.
Further boundaries
Basket reweighting. The ABS periodically reweights the CPI basket. A change in housing’s share alters the relationship between Housing CPI and Headline CPI independently of price movement, and the node does not flag reweighting separately.
Temporal resolution. The node is quarterly. The ABS also publishes a monthly trimmed mean annual rate, but its history is too short to support a fifteen-year baseline, so it is used for current reading only and sits outside the composite.
Fixed anchor. Every baseline ends in December 2025. Readings after that date are measured against a distribution that does not include them, which is the intended design, and which also means the level drift described above accumulates until the anchor is formally revised.
Expert-level implications
Credit risk modelling: do not treat agreement among the four price inputs as independent confirmation. Treat the composite as one price-level signal plus one expectations signal, and model inflation pace from growth rates.
Institutional research: the composite’s level drift makes it a measure of accumulated repricing. For turning points, monitor the annual growth rates and the expectations input; the four price-level inputs will not fall in a period of positive but slowing inflation, so a decline in the composite would come first from the expectations input.
Multi-node analysis: read Inflation (21220) alongside Interest Rates (21210), certified at +0.9807σ for the June quarter 2026, and Construction Costs & Supply Chain (21260), certified at +1.0353σ for the June quarter 2026. The latter combines construction input-cost escalation, measured as an annual growth rate, with dwelling approvals and construction insolvency, so it carries pace and activity information that the price-level inputs of Inflation (21220) do not.
