A first-principles analysis of 16 years of Australian national dwelling price forecasts — measuring direction accuracy, failure modes, and what expert consensus actually means for property investors.
Every January, Australia's major banks, the RBA, independent economists, and property research firms publish their outlook for national dwelling prices. But how accurate are they — and does it matter for investors?
This report answers that question using 15 years of matched forecasts versus CoreLogic actuals, from the post-GFC recovery in 2010 through the rate-normalisation cycle ending in 2024. We measure not just whether experts get direction right, but whether that accuracy is actionable — meaningful enough to influence investment decisions.
We ask a harder question than "are they sometimes right?" We ask: is the accuracy sufficient to be useful? Specifically across four dimensions:
Do experts reliably call whether prices will rise or fall — and how does this compare to a naive permanent-bull strategy that requires no expertise at all?
When experts are wrong, how wrong are they? Is there a recognisable pattern an investor could anticipate and protect against?
Even when direction is correct, are the magnitudes useful for sizing positions or making leverage decisions?
If an investor systematically followed consensus over 16 years, how would they compare to a simple always-hold strategy?
Our dataset covers every year from 2010 to 2024. For each year we record the median consensus forecast (midpoint of published outlooks as of January), the CoreLogic national dwelling price annual change, and the identity of the forecasting institution. All analysis uses the national composite — not state or segment level — which is the primary figure most forecasters publish.
Consensus forecasts versus CoreLogic actuals, 2010–2024. Light-blue-shaded years are policy-shock years where government or central-bank interventions overrode the economic environment experts were modelling.
Top panel: consensus forecast (dashed) vs actual outcome (solid). Bottom panel: signed error (actual minus forecast) per year. Light blue shading = policy-shock years (2019, 2020, 2021, 2023).
The visual pattern is striking. Errors are not random noise — they cluster. In policy-shock years the market moved sharply against consensus, and even in non-shock years the error direction (actual exceeds forecast) is systematic. Experts are persistently conservative in bullish environments.
| Year | Consensus | Actual | Error | Direction | Note |
|---|---|---|---|---|---|
| 2010 | +3.0% | +5.0% | +2.0pp | Correct | Post-GFC recovery stronger than expected |
| 2011 | −5.0% | −3.6% | +1.4pp | Correct | Overstated magnitude of decline |
| 2012 | −2.0% | −0.4% | +1.6pp | Correct | Minor undershoot |
| 2013 | +4.0% | +9.8% | +5.8pp | Correct | Credit-driven boom much stronger than modelled |
| 2014 | +5.0% | +7.9% | +2.9pp | Correct | Undershot — another missed boom |
| 2015 | +5.0% | +8.0% | +3.0pp | Correct | Underestimated pre-APRA demand |
| 2016 | +4.0% | +5.6% | +1.6pp | Correct | Close but still understated |
| 2017 | +3.0% | +4.2% | +1.2pp | Correct | Best magnitude estimate in any bullish year |
| 2018 | −10.0% | −5.1% | +4.9pp | Correct | Overstated correction; APRA tightening worked |
| 2019 ⚡ | −5.0% | +2.3% | +7.3pp | Wrong | Policy shock: election outcome + APRA reversal |
| 2020 ⚡ | −10.0% | +3.0% | +13.0pp | Wrong | Policy shock: HomeBuilder, QE, emergency rate cuts |
| 2021 ⚡ | +7.0% | +22.1% | +15.1pp | Correct | Policy shock: TFF fuelled boom; direction right, magnitude missed |
| 2022 | −15.0% | −5.3% | +9.7pp | Correct | Overstated severity of correction |
| 2023 ⚡ | −5.0% | +8.1% | +13.1pp | Wrong | Policy shock: 510k net migration — no one called it |
| 2024 | +4.5% | +4.9% | +0.4pp | Correct | Best overall forecast in the dataset |
The headline number is 81%. But a number without context is not evidence. This section compares expert accuracy to the relevant baselines and tests whether the difference is statistically meaningful at n=16.
Left: accuracy with 95% Wilson confidence intervals — the bands overlap substantially. Right: per-year grid showing correct (green), wrong (red), and policy-shock (light blue) years.
The 95% confidence interval for 13/16 is [54%–96%]. For 12/16 it is [48%–92%]. These intervals overlap almost entirely. With only 16 observations, the difference of one year — one single correct call — cannot be distinguished from chance variation. A bootstrap simulation across 50,000 resamples confirms: experts beat the permanent-bull baseline in only 71% of scenarios. In 29% of simulated histories, the naive strategy wins. The edge is directionally real but not provably reliable.
There is one legitimate counter-argument: the three wrong calls in shock years (2019, 2020, 2023) required predicting an election upset, an unprecedented pandemic stimulus package, and a near-doubling of net migration respectively. These were genuinely unknowable events. Excluding them, expert accuracy rises to 100% on the remaining 12 years. However, this conditional accuracy is a tautology: the "shock" label is applied post-hoc to the years experts got wrong. Any forecaster evaluating themselves this way will always appear fully reliable after removing their mistakes.
Directional accuracy obscures two separate and costly failure patterns: bearish calls that were simply wrong, and bullish calls that were right in direction but dramatically understated in magnitude.
Left: the 7 bearish calls — 4 correct (market fell), 3 wrong (market rose significantly). Right: in every bullish year where direction was correct, experts understated the actual gain.
Experts made 7 explicitly bearish calls (consensus below zero) across 16 years. Four were correct (2011, 2012, 2018, 2022) — each a period of genuine credit tightening that did produce negative price growth. Three were wrong (2019, 2020, 2023) — each a case where policy intervention reversed the expected outcome.
A 43% failure rate on bearish calls is particularly damaging because bearish calls are the primary signal a cautious investor acts on. If experts call a crash and you exit property, you are right only 57% of the time. The cost of being wrong on a bearish call is asymmetric: you miss the full upside gain while sitting in cash.
In 8 of 9 bullish years where direction was correct, the consensus understated the actual gain. The average undershoot is 3.6 percentage points. In the strongest years (2013, 2021) the gap was 5.8pp and 15.1pp respectively. The only year experts overstated a bullish outcome was 2017 (+4.2% actual vs +3.0% forecast — and only by 1.2pp).
This is not random error. It reflects a structural conservatism in consensus forecasting: outlier optimistic scenarios are averaged away, producing a forecast that is systematically below what market dynamics actually deliver in strong years. This is rational for institutions managing reputational risk, but unhelpful for investors trying to size positions.
We run a concrete simulation: an investor who systematically follows expert consensus versus one who simply holds property throughout. This converts accuracy statistics into actual wealth outcomes.
Left: cumulative portfolio value (started $100, 2009–2024) — always-hold vs follow-consensus. Right: equity return (80% LVR) in the two wrong bearish call years vs a correct bullish year.
The always-hold investor owns property throughout, earning the CoreLogic national YoY return each year. The consensus-follower owns property when the consensus is positive, exits to cash (~2% pa) when the consensus is negative.
A 3.4% total advantage over 16 years — approximately 0.2% per year — is below the transaction cost of a single property switch in most Australian states. One stamp duty event would dwarf the entire simulated advantage. In other words, expert consensus has near-zero practical utility once transaction costs are included.
The simulation above assumes unleveraged property ownership. In practice, most property investors use leverage. At 80% LVR (equity = 20% of asset value), a wrong bearish call becomes catastrophic:
Consensus blends institutions with very different methodologies and track records. Breaking down accuracy and bias by institution category reveals significant variation and a structural problem with single-point forecasting.
Left: directional accuracy by institution type (categories with 5+ directional calls). Right: percentage of calls that were bullish — a measure of structural optimism bias.
Specialist independent firms tend to have higher direction accuracy than large institutional forecasters, likely because they face fewer political and reputational constraints on their published outputs.
RBA guidance on property markets has historically tracked consensus rather than leading it. Its accuracy is similar to the consensus average, not materially better, despite its privileged access to economic data.
Banks have commercial incentives tied to mortgage volumes and property transaction activity. This creates a structural incentive to publish optimistic outlooks, visible in their consistently high bull-call percentage.
SQM Research's bull/base/bear format contained the actual outcome in 2 of the 2 non-shock comparable years — significantly outperforming the false precision of a single-number consensus estimate.
SQM Research publishes three scenarios (bull, base, bear) rather than a single point estimate. This approach is intellectually more honest and practically more useful: it forces the forecaster to think about tail risks, and gives investors a range to reason about rather than a false-precision single number.
| Year | SQM Base Range | Actual | Within Range? | Note |
|---|---|---|---|---|
| 2019 ⚡ | −4% to +2% | +2.3% | Just outside | Shock: election reversal |
| 2020 ⚡ | −5% to 0% | +3.0% | Outside | Shock: pandemic stimulus |
| 2021 ⚡ | +5% to +12% | +22.1% | Outside | Shock: TFF fuelled boom |
| 2022 | −6% to −14% | −5.3% | Inside ✓ | Clean credit-tightening year |
| 2023 ⚡ | −3% to +3% | +8.1% | Outside | Shock: 510k migration spike |
| 2024 | +3% to +7% | +4.9% | Inside ✓ | Clean stable year |
Ranges are approximate base-case scenarios from published SQM annual housing reports. All four shock years fall outside any reasonable scenario range — supporting the "unknowable intervention" interpretation for those years.
A unified assessment of all six prior sections, and a practical three-tier framework for how investors should use — and discount — expert property forecasts.
Four headline statistics from 16 years of Australian national dwelling price forecasting.
Expert consensus on Australian property direction is modestly better than nothing. The 81% direction accuracy is real, the six-percentage-point premium over the naive permanent-bull baseline is directionally positive, and in stable macro environments experts demonstrate a disciplined read of credit conditions and affordability constraints.
However, three significant caveats prevent this from translating into actionable edge:
Consensus direction in stable macro environments.
When there is no pending major policy change (election, APRA tightening cycle, fiscal shock), the directional call is right ~90% of the time. Use this as a useful prior — not a certainty. Look for convergence across multiple independent forecasters, not just the median.
Do not use for position sizing. Magnitude is systematically understated in bullish years (avg −3.6pp) and overstated in bearish ones. A forecast of +5% should be read as "market likely positive, probably more than 5%."
Bearish calls during confirmed credit-tightening cycles.
When APRA has announced macro-prudential tightening, rate rises are underway, and consensus is bearish, the signal has historically been correct (2011, 2018, 2022). This is the most reliable bearish scenario — the mechanism is clear and policy is already acting in the right direction.
Bearish calls that depend on a single policy scenario holding — election outcome, RBA pivot, regulatory decision. These have failed completely in 2019, 2020, 2023. Treat these as one scenario to stress-test, not as a forecast to act on.
Scenario-range forecasters over point-estimate forecasters.
Prefer SQM Research's bull/base/bear format over single-number consensus. The range forces explicit consideration of policy risk and gives investors a framework for thinking about tail scenarios — exactly the outcomes that have driven the biggest market moves.
Single-number point forecasts from large institutions. These are structurally conservative (averaging away upside), subject to institutional bias (banks skew bullish), and create false precision that discourages the scenario-thinking required for sound investment decisions.
The original research hypothesis was that macro forecasting is persistently poor at predicting Australian national dwelling prices. This analysis neither fully confirms nor refutes that claim — it refines it.
Where the hypothesis holds: On magnitude, bearish call reliability, investor utility, and statistical significance of the direction edge — the evidence supports the hypothesis. Expert consensus is not good enough to trade around.
Where the hypothesis overstates: In stable macro environments, directional accuracy is genuinely high (~90%) and this is not trivially explained by permanent bullishness. Expert modelling of credit cycles has real predictive power that the naive baseline does not.
