August 2026 RBA forecasts
Australian economic
scenario analysis.
Explore shocks to the RBA outlook.
Loading the forecasts and model responses…
About the models
02 / Explore the outcome
The path ahead
Choose variables
View the numbers
What’s behind the scenarios?
Scenarios add model responses to the RBA baseline. Common policy, commodity-price, exchange-rate and foreign-demand shocks show both models.
Uncertainty and error bands
The RBA publishes 70% and 90% historical forecast-error bands for GDP growth, unemployment, headline inflation and trimmed-mean inflation. These describe uncertainty around the RBA forecast, not uncertainty about the effects of an individual shock.
This preview shows central scenario paths only. The pale area between the two lines shows the change from baseline; it is not an error band. No confidence intervals have been calculated for the reconstructed MARTIN or saved DSGE responses used here. The RBA’s April 2025 model comparison explicitly says its graphs omit IRF uncertainty. The 2018 paper includes 68% bands for separate SVAR estimates; those cannot be reused as DSGE bands.
A possible extension would retain the RBA baseline bands as context. Moving those bands with the scenario would assume unchanged forecast uncertainty and a known shock response; it would need to be labelled an illustrative fixed-width approximation. Proper scenario intervals require jointly simulated forecast and model-response uncertainty, preserving correlations across variables, horizons and shocks.
RBA August 2026 forecast charts · RBA forecast-uncertainty methodology · RBA model uncertainty · 2018 model paper
Timing, units and the approximation
Selecting a shock automatically shows its associated forecast variable, or a model-response chart where no RBA forecast is available. These extra charts show deviations from the model baseline; MARTIN percentage deviations use the initial saved baseline as the denominator. They include the combined effects of selected shocks.
September 2026 is the first forward quarter after the August forecast release. March and June 2026 show the latest available actuals. Where an actual is unavailable, history is left blank or the June RBA forecast is retained and identified in the data table. September onward retains the August RBA forecast vintage; it is not rebased to later data. The RBA publishes June and December endpoints. Forecast September and March values are linear interpolations, not unpublished RBA forecasts.
GDP, consumption, investment and other activity forecasts show year-ended growth. Their responses are converted from level deviations using the difference between the current and four-quarter-lagged response. Inflation responses use four-quarter changes. The 2–3% band and 2.5% midpoint are headline CPI targets, shown as reference values on trimmed-mean charts. Index and oil-price forecasts receive percentage changes to their levels. Prior to the shock, all deviations are zero.
Chart scales are fixed using the baseline and both directions of the example shocks across both models. Larger combined shocks can extend beyond the chart; a notice appears and full values remain in the table and download. This is a local, additive approximation. It does not re-estimate the models, reproduce the RBA’s judgemental forecasts, enforce all accounting identities after the overlay, or model nonlinear interactions and interest-rate bounds. Large shocks are less reliable. Model vintages and policy-shock persistence differ.
Cross-model comparisons match the initial cash-rate change, commodity-price change nominal trade-weighted exchange-rate change, or initial trading-partner GDP response. Exchange-rate shock mechanisms differ; their later paths need not match. A mix containing model-specific shocks shows only the model that supports the entire scenario. Remove those shocks before adding shocks exclusive to the other model. The MARTIN one-year cash-rate example holds the rate above its own baseline for four quarters and holds government spending and the neutral rate fixed, as in the replication exercise. Adding other shocks adds their separate responses; it does not enforce that same total cash-rate path in the combined scenario.
Model coverage and verification
MARTIN: all 217 equations and 86 documented equation innovations extracted from the public 2019 workfile. Responses are local derivatives around its saved 2019 baseline. The source contains simulated substitutes for proprietary series. The reconstructed one-year rate shock differs from published GDP responses by at most 0.026 percentage points of the baseline level, consistent with the difference in data and linearisation. Native EViews execution has not been performed.
DINGO: baseline solution released with Fink & Hambur (2026), extending the Gibbs–Hambur–Nodari model. Uses the saved posterior-mean parameters and baseline price rigidity, including separate mining-investment dynamics. All 152 endogenous variables and 60 exogenous entries are extracted: 29 structural shocks (two inactive) and 31 measurement errors. This is not the current internal RBA version.
The solution matrices reproduce the released baseline import-cost, consumption and monetary-policy scenario spreadsheets with a maximum absolute difference below 0.000001. Of 348 saved IRF series, 347 agree; one stored exchange-rate shock cash-rate series is inconsistent, so responses are computed from the decision rules. No fresh estimation is claimed.
Responses without a defensible forecast mapping are marked unavailable. In particular, the DSGE model has no unemployment equation and no separate headline-CPI inflation measure. Its inflation observable is trimmed mean (RDP 2018-04, Appendix A). Mapping qualifications appear beneath the relevant charts.
Sources and downloads
Historical observations: RBA statistical tables and ABS population. Retrieved 27 September 2026. Monthly rates are averaged over the quarter; GDP per capita is the published ABS measure.
- RBA August 2026 detailed forecasts, finalised 5 August.
- Official forecast archive workbook
- Ballantyne et al. (2019), MARTIN Has Its Place · Replication files
- Fink & Hambur (2026), Shifts in Australian Price-setting Behaviour · Replication files — baseline solution used here
- Gibbs, Hambur & Nodari (2018), DSGE Reno · Replication files
- RBA (2025), model comparison and DINGO lineage
- Forecasts and mapped responses (JSON)
- Every MARTIN response (JSON, 8 MB)
- Every DSGE response (JSON, 2 MB)
Sources downloaded 25 September 2026. This is an independent academic tool, not an RBA product.
Full impulse-response explorer
Inspect any extracted model variable against any source shock over 40 quarters. These responses use native model units and do not include the forecast baseline. This explorer is separate from the scenario controls above.