Our August call was a hike. The RBA held.

On 29 July 2026 this framework published a HIKE call for the 11 August 2026 meeting. The Reserve Bank’s Monetary Policy Board held at 4.35 per cent, and the RBA’s daily series has shown 4.35 per cent on every day since. The call was wrong. This was the first meeting the framework has been scored against, so its live record is one call and one miss.

There is a complication we are not going to use in our favour. Two readings were published for August. The 11 July 2026 edition read HOLD, and the pipeline’s ledger, which keeps the first call registered for each meeting, recorded HOLD on 4 July 2026 with score shares of 0.574 hold and 0.406 hike. By that ledger’s rule, August counts as a hit. We are scoring it on the HIKE call, because that was the last reading we published before the meeting and the one that stood when the Board decided.

The market read it correctly. ASX 30 Day Interbank Cash Rate Futures on 28 July 2026 implied a 78 per cent chance of no change at the August meeting. The 29 July edition printed that figure, disagreed with it, and said the disagreement would be settled on 11 August. It was, in the market’s favour.

Two statements in that edition did not hold up, independent of the outcome. It said the call would be reviewed once the 4 August household spending indicator was ingested, and that a weak spending print could return the reading to HOLD. The data pipeline did not run between 29 July and 11 September, so no review took place. More fundamentally, that indicator is not an input to the score: the domestic-activity driver is built from the quarterly National Accounts (GDP, domestic final demand, household consumption and business investment). The sentence overstated what the model could see, and we should have checked it before publishing.

What the model got wrong

Our data set does not contain the Board’s reasoning, and we are not going to reconstruct it from commentary. What we can do is test each part of the August case against the data that has arrived since.

• The margin was inside uncertainty we had already disclosed. The August reading cleared an assumed boundary by 0.03, and that edition’s own objections section said such a margin was not a strong claim. The boundary is a design choice, not an estimate of where the Board acts. A hold at that reading is one piece of evidence that the Board’s threshold, if it has a stable one, sits higher than ours, or that it gave weight to inputs the score does not contain, such as staff forecasts and business liaison. One meeting cannot separate those explanations.
• The inflation-level driver leaned on the least reliable series. The August call rested on a June jump in services inflation to 4.0 per cent and non-tradables inflation to 4.9 per cent. In July services fell to 3.7 per cent from 4.0 and non-tradables fell to 4.4 per cent from 4.9, and the persistent-inflation composite that gives them extra weight fell to 3.7 per cent from 3.9. The trimmed mean, the measure the RBA prefers, was 3.6 per cent in June and 3.6 per cent in July. The August edition’s third objection warned that the unadjusted services series was noisy. The objection was right, and the call was published anyway.
• The momentum argument was not wrong. The August edition said a July trimmed-mean monthly rise below about 0.19 per cent would remove the momentum case. The July rise was 0.5 per cent, after 0.4 and 0.3 per cent in May and June, and three-month annualised trimmed-mean inflation is now 4.91 per cent against an annual rate of 3.6. That part of the case has strengthened since the meeting.

What we changed, and what we left alone

The ten weights and the ±0.306 hold band are unchanged. They were fixed before the August outcome, and adjusting them after a single miss would fit the model to one observation and make the next call worthless as a test. What has changed is how a call this close to the line is reported: this edition leads with the margin and with the August result, and treats a reading a few hundredths from the boundary as a narrow call rather than a clean one.

The September call, and how much weight it can bear

On data available at 11 September 2026 at 9:04 pm AEST, the pressure score is +0.346. The rule is HIKE above +0.306, CUT below −0.306, and HOLD in between, so the reading for 29 September 2026 is HIKE: a 0.25 percentage point rise to 4.60 per cent.

Inside the model the reading is stable. Moving any single driver weight up or down by 25 per cent, 20 tests in all, leaves 20 of 20 at HIKE, with scores from +0.314 to +0.374. Removing one driver at a time changes the call in 1 of 10 cases: without the inflation-level driver the score falls to +0.186 and the reading becomes HOLD. The call is therefore mostly a statement about one number, the gap between persistent inflation and the 2.5 per cent target midpoint, which contributes +0.206 of the +0.346 total (59.6 per cent).

Outside the model the evidence is thinner. The August reading also cleared the boundary, by 0.03, and the Board held. A margin of 0.040 is not materially different. By the framework’s rule the call is HIKE; by its one-meeting record, readings this close to the line have not predicted what the Board does. Our reading is that the July data keep a rise on 29 September a live option, and that the rule falls narrowly on that side of the line.

The score shares, 0.553 hike, 0.427 hold and 0.020 cut, rescale the same score into three numbers. They look like probabilities and are not. They have never been calibrated against Board decisions, and with one completed meeting they cannot be.

What changed between the June and July data

Eight releases on the September input list have landed since the August edition was published:

1. 30 July 2026
   International Trade Price Indexes, AustraliaJune 2026 · external context only
2. 4 August 2026
   Monthly Household Spending IndicatorJune 2026 · listed in the release calendar; not an input to the score
3. 13 August 2026
   Average Weekly Earnings, AustraliaMay 2026 · wages
4. 19 August 2026
   Wage Price Index, AustraliaJune 2026 · wages
5. 20 August 2026
   Labour Force, AustraliaJuly 2026 · labour
6. 26 August 2026
   Consumer Price Index, AustraliaJuly 2026 · inflation
7. 27 August 2026
   Monthly Household Spending IndicatorJuly 2026 · listed in the release calendar; not an input to the score
8. 2 September 2026
   Australian National Accounts: National Income, Expenditure and ProductJune 2026 · domestic activity, GDP and consumption

Inflation: the domestic measures gave back June’s jump

In the monthly CPI indicator for July 2026, released on 26 August 2026, headline inflation fell to 3.5 per cent from 3.8. This time the fall came from domestically generated prices. Non-tradables inflation fell to 4.4 per cent from 4.9 and services fell to 3.7 per cent from 4.0, while tradables inflation rose to 1.7 per cent from 1.5. That is close to the reverse of the June pattern the August call was built on.

The underlying measures did not improve. The trimmed mean held at 3.6 per cent and the weighted median held at 3.6 per cent. The monthly pace is where the hawkish evidence sits: the trimmed mean rose 0.5 per cent in July, and on the framework’s persistent composite the last three months annualise to 4.3 per cent, above the composite’s annual rate of 3.74 per cent.

Jobs: softer at the margin, not weak

The July 2026 Labour Force release on 20 August 2026 put unemployment at 4.46 per cent, from 4.43 per cent in June and 0.38 percentage points higher than six months earlier. Employment fell by 15,827 after a rise of 80,249 in June, hours worked fell 0.62 per cent in the month, and participation fell to 66.9 per cent from 67.0. Underemployment fell to 6.36 per cent from 6.45 and underutilisation fell to 10.82 per cent from 10.88. Job vacancies for the June quarter were 329.5 thousand, down 2.1 per cent on the quarter.

Read together, that is a labour market cooling slowly from a tight position. Unemployment remains below the 4.6 per cent NAIRU estimate the framework uses as a reference, and three-month annualised employment growth is still 2.82 per cent.

Wages and costs: steady pay, rising unit costs

The Wage Price Index for the June quarter 2026, released on 19 August 2026, held at 3.2 per cent over the year. Private-sector wage growth fell to 3.1 per cent from 3.2 and public-sector wage growth was 3.4 per cent. Average weekly ordinary-time earnings grew 3.67 per cent in the year to May 2026. The National Accounts added the less comfortable half: nominal unit labour costs rose 3.58 per cent over the year and 1.2 per cent in the quarter, while productivity (GDP per hour worked) fell 0.2 per cent. Pay growth below the framework’s 3.5 per cent benchmark is not adding pressure; labour cost per unit of output is.

Activity: growth slowed in the June quarter

The June-quarter National Accounts on 2 September 2026 showed real GDP up 0.4 per cent in the quarter, while annual growth fell to 2.14 per cent from 2.50. Quarterly growth in domestic final demand fell to 0.3 per cent from 0.9, household consumption grew 0.4 per cent, and private business investment fell 0.5 per cent after 6.6 per cent growth the quarter before.

Markets: yields have risen since the hold

The 2-year Australian government bond yield was 4.581 per cent on 11 August 2026 and 4.835 per cent on 9 September 2026, a rise of 0.25 percentage points since the August decision. The 10-year yield was 5.202 per cent. Consumer inflation expectations, shown for context and excluded from the score, fell to 4.9 per cent from 5.5 in the latest quarterly reading.

Driver by driver: where the pressure comes from

The ten drivers sit in four official-data blocks. Inflation pressure scores +0.568 (hike pressure); Labour market and wages scores +0.078 (neutral / hold); Domestic activity, credit and financial conditions scores +0.327 (hike pressure); External and risk conditions scores +0.060 (neutral / hold). Each contribution below is the driver’s score multiplied by its fixed weight; together they sum to the composite of +0.346.

Pushing toward a rise

• Inflation level (+0.206). The persistent-inflation composite is 3.74 per cent, 1.24 percentage points above the target midpoint. It weights the trimmed mean at 50 per cent, the weighted median at 20 and services and non-tradables at 15 each, which is why it sits above the trimmed mean.
• Inflation momentum (+0.047). The last three monthly composite readings annualise to 4.3 per cent, above the annual rate. On the trimmed mean alone, which is seasonally adjusted and free of this framework’s weighting, the three-month pace is 4.91 per cent.
• Credit growth (+0.040). Housing credit grew 7.55 per cent and business credit 10.46 per cent over the year to July 2026. These are transmission indicators, and they show credit still expanding at those rates after three rate rises this year.
• 2-year bond proxy (+0.035). After a stated 0.05 point term adjustment the proxy is 4.785 per cent, above the 4.35 per cent cash rate.
• Taylor-rule benchmark (outside the score). An adapted Taylor rule with a 1.0 per cent neutral real rate puts a benchmark cash rate at 5.49 per cent, 1.14 percentage points above the actual rate. It is reported as a cross-check and carries no weight.

Neutral, with one driver pulling the other way

The model tags six drivers as neutral or hold. Most of them are close to zero rather than arguing for patience:

• Labour market (+0.016). Employment growth scores at +1.00 and the employment-to-population ratio at +0.37; hours worked (−0.91) and underutilisation (−0.54) score against them. The net is near zero.
• Wages and labour costs (+0.009). Headline and private WPI score slightly below neutral; unit labour costs and weak productivity score above it.
• Inflation expectations (+0.003). Market economists expect 2.5 per cent inflation in two years and the 10-year break-even rate is 2.27 per cent, both near the target midpoint; business expectations of 3.59 per cent are higher.
• Australian dollar and commodities (+0.001 and +0.001). The trade-weighted index is down 0.3 per cent over six months and the RBA commodity price index is up 1.5 per cent. Together they carry 3 per cent of the weight.
• Domestic activity (−0.013). The only driver below zero, reflecting the slower June quarter: domestic final demand 0.3 per cent and business investment −0.5 per cent.

The hold case inside this model is therefore the absence of a strong offset rather than a set of drivers pointing the other way. Of ten drivers, one is negative, worth −0.013 in total. That is also where the framework is weakest: a Board that weighed the slower June quarter, softer hours and July’s fall in services inflation more heavily than this framework does would not be reading the data unreasonably.

Market pricing and the model

This build has no current futures pricing for the September meeting. ASX and FENICS-derived market series are excluded from the public build pending redistribution rights. The most recent ASX RBA Rate Tracker settlement in our files is 28 July 2026, which priced the August meeting at 78 per cent no change. That reading is scored in the August section above; showing a July settlement as September pricing would be wrong. Current market-implied probabilities are published on the ASX RBA Rate Tracker.

The market data we do hold points the same way as the model. The 2-year yield has risen 0.25 percentage points since the August decision, and the bond-market directional proxy leans HIKE on score shares of 0.88 hike and 0.10 hold. One internal benchmark disagrees: the simpler rules-based reaction function in the same pipeline leans HOLD, on shares of 0.513 hold and 0.467 hike.

Market pricing source: ASX RBA Rate Tracker. Benchmark leans are computed by the same pipeline and carry the statuses shown in the data file.

Sixteen scenarios test both sides of the decision

The same weights and thresholds are applied to every stress test: 7 HIKE, 6 HOLD and 3 CUT. The defined state today’s data sits closest to is Domestic activity reacceleration, a HIKE state, with a similarity index of 81.9; the nearest cut state, Household balance-sheet squeeze, ranks 14th. The similarity index is 55 per cent distance from current levels and 45 per cent agreement with the direction of travel. It measures resemblance, not likelihood, and the framework marks the closest state’s evidence as partial.

What could flip the call before 29 September

The reading is conditional on data available at 11 September 2026 at 9:04 pm AEST. ABS release dates were re-verified against the ABS product pages on 11 September 2026 and remain subject to change. Two scheduled ABS releases fall between now and the decision:

1. 24 September 2026
   Labour Force, AustraliaAugust 2026 · feeds the labour driver
2. 29 September 2026
   Monthly Household Spending IndicatorAugust 2026 · listed in our release calendar but not an input to the score; lands at 11:30am on decision day, two hours before the 1:30pm model lock

Only the Labour Force release on 24 September 2026 can move a scored driver. The arithmetic is set out now so it cannot be revised later. The composite clears the boundary by 0.040. The labour driver scores +0.091 with a weight of 0.18; with nothing else changing, a labour score of −0.129 or lower returns the reading to HOLD. That is a fall of 0.220 in one release. Because the labour score averages seven components, a swing of that size would need weakness across several of them at once rather than one soft employment number.

The two daily market inputs update on every data run. The 2-year proxy would need to fall from +0.87 to −0.12 to flip the call alone, which would require the 2-year yield to fall to around the cash rate or below. The Australian dollar driver cannot flip the call alone at any value, because its 2 per cent weight caps its effect below the margin. The inflation level and momentum drivers cannot change before the meeting: the next monthly CPI indicator is released on 30 September 2026, the day after the decision.

The Board also works from material this score does not contain, including staff forecasts and business liaison. There is no Statement on Monetary Policy at this meeting; the next forecast round is published with the 3 November 2026 decision. If the Board holds on 29 September, the most likely data-based reasons are the ones set out above: a trimmed mean that has not risen in annual terms, domestic services inflation that fell back in July, and slower growth in the June quarter.

Does the RBA really look at these ten inputs?

Broadly, yes. Mechanically, no. The RBA has no published rule saying these ten variables are required or that they receive fixed weights. Its mandate is price stability and full employment, and staff brief the Board on Australian and global conditions, financial markets, forecasts and risks. The score compresses part of that information set into a single number, and the August outcome is a reminder of how much it leaves out.

Method, reproducibility and limits

The reproduction panel beneath the scenario library is the audit trail: every scenario score recomputes from the published weights, and every input traces to a hash-verified official file. 31 of 31 integrity checks pass on this run. Those checks establish data lineage and arithmetic consistency. They do not establish forecast accuracy, which only the record of meeting outcomes can.

Formally, each driver i is mapped to a bounded score s_i ∈ [−1, +1] from its official observations, and the composite is the weighted sum S = Σ w_i·s_i with fixed weights that sum to one (inflation block 45%, labour and wages 33%, activity 10%, financial and external conditions 12%). The decision rule is: HIKE if S > +0.31, CUT if S < -0.31, otherwise HOLD. Today S = +0.35.

The framework is not a statistically calibrated probability model or a causal estimate of RBA behaviour. Its weights, transformations and ±0.306 hold band are design assumptions, stated in advance and held fixed between meetings, including after the August miss. TimesFM is a separate appendix experiment and does not enter this score.

Where this method sits in the literature

Scoring official data into a policy-pressure reading is a monetary-policy reaction function — a well-studied idea, not an invention of this site. Situating the framework honestly: it is a fixed-weight, preregistered variant of that tradition, far simpler than the models central banks actually run.

• Taylor, J.B. (1993), “Discretion versus policy rules in practice”, Carnegie-Rochester Conference Series on Public Policy — the origin of mechanical benchmark rules; our adapted Taylor comparison descends from this.
• de Brouwer, G. and Gilbert, J. (2005), “Monetary Policy Reaction Functions in Australia”, Economic Record — estimated RBA reaction functions; unlike that work, our weights are asserted, not estimated, which is why we label the score uncalibrated.
• Orphanides, A. (2001), “Monetary Policy Rules Based on Real-Time Data”, American Economic Review — why rules evaluated on final-vintage data flatter themselves; the reason our historical checks are labelled final-vintage and our live ledger starts at zero.
• Ballantyne et al. (2019), “MARTIN Has Its Place: A Macroeconometric Model of the Australian Economy”, RBA RDP 2019-07 — the full-system model the RBA actually maintains; a ten-driver score is not a substitute for it and does not try to be.
• CAMA RBA Shadow Board, Australian National University — the established precedent for publishing independent, before-the-meeting RBA calls and living with the outcomes.

Conclusion: a narrow call from a framework that has just been wrong

The August test went against us. The framework read HIKE on a margin of 0.03, the futures market priced no change, and the Board held. Part of that case, a June jump in services and non-tradables inflation, reversed in July. Another part, the pace of underlying inflation, has strengthened: the trimmed mean rose 0.5 per cent in July and is running at 4.91 per cent annualised over three months.

On the July data the rule reads HIKE for 29 September 2026, by 0.040. Inside the model that result survives every weight test; outside it, the only evidence we have is a hold at a similar margin six weeks ago. We are publishing the call because the rule produces it, and because a call made in advance is only worth recording if it is published when it is uncomfortable. It is a narrow reading, and an analyst looking at the same releases could reasonably expect a hold.

The last scheduled input is the Labour Force release on 24 September 2026. The decision is announced at 2:30pm AEST on 29 September 2026, and this call will be scored against it in the next edition, whichever way it goes.

Common questions about the September RBA decision

Primary sources

• Reserve Bank of Australia - Cash rate target overview
• Reserve Bank of Australia - Upcoming Monetary Policy Board meetings
• Australian Bureau of Statistics - Consumer Price Index, Australia (July 2026)
• Australian Bureau of Statistics - Labour Force, Australia (July 2026)
• Australian Bureau of Statistics - Wage Price Index, Australia (June quarter 2026)
• Australian Bureau of Statistics - Australian National Accounts (June quarter 2026)
• Australian Bureau of Statistics - Average Weekly Earnings, Australia (May 2026)
• Australian Bureau of Statistics - Job Vacancies, Australia (June quarter 2026)
• Reserve Bank of Australia - Monetary policy objectives and decision process
• Reserve Bank of Australia - Statistical tables (F1 cash rate, F2 bond yields, D2 credit, F11 exchange rates)
• ASX - RBA Rate Tracker (30 Day Interbank Cash Rate Futures)

Previous editions, retained unedited

• Will the RBA Raise Rates in August 2026? Our Model Just Flipped to Hike (29 July 2026). The HIKE call for 11 August scored above.
• Three hikes, then a pause: the public data lean HOLD for the RBA’s August decision (11 July 2026). The original HOLD reading for August.

Publication and correction policy

Status: independent personal research; not commissioned, sponsored or externally peer reviewed. A practising economist has not yet reviewed the framework or this article, and this disclosure stays until one does. No employer, RBA, ABS or Australian Government affiliation or endorsement is claimed.

Author and contact: written by Marcello Babbili and published by Not A Tech Guy. Questions, corrections and methodological challenges are welcome: notatechguy@agentmail.to. Substantive challenges that change a published number will be credited in the correction note.

Versioning: this page identifies the observation dates, generation time and model version. Material corrections are made in place with the modified timestamp updated. Earlier meeting calls are kept rather than rewritten after the outcome: the July and August editions remain online unchanged, including the August call scored as a miss above.

Scheduled review: this reading incorporates the July 2026 Labour Force, July 2026 CPI indicator, June quarter 2026 Wage Price Index and June quarter 2026 National Accounts. It is due to be re-run after the 24 September 2026 Labour Force release and scored against the outcome after 29 September 2026. The August edition carried a similar review promise that was not kept; if this one is missed, the next edition will say so.