Executive Summary
Does buying in the ‘right suburb’ actually matter? Or is it noise? We analysed 10,874 property resales across Australia. All held for three years or less. The goal was simple. Measure how strongly properties within the same area move together.
The results are clear. At the national level, knowing that ‘the Australian property market is going up’ removes only 24% of the uncertainty about your individual property. At the suburb level, 61% of uncertainty is removed. At the street level, 89%. The direction is clear. The more granular the analysis, the more real the market.
Key Findings (Uncertainty Reduced)
- National level: 24% of uncertainty removed. Macro commentary is mostly noise for individual investors.
- State level: 36% removed. Buying ‘in NSW’ does not guarantee a uniform outcome.
- SA4 / Regional level: 38% removed. Regions start to tell a real story.
- Suburb level: 61% removed. The old saying is that suburb selection does 80% of the work. The data says 61%.
- Street level: 89% removed. Two properties on the same street, bought and sold in the same years, have a 97.3% chance of moving in the same direction.
The old saying overstates it. Suburb selection does not do 80% of the work. It does 61%. But picking the right street gets you to 89%. The more granular the analysis, the more real the market becomes.
Introduction
Everyone says suburb selection does 80% of the heavy lifting. We put that claim to the test. The answer is different from what most people expect.
Real estate podcasts, news sites, and industry reports are full of macro speculation. Interest rate movements. GDP growth. Migration data. They run this content because investors worry about it and it gets clicks. But if the ‘national market’ is barely more predictive than a coin flip, what is the point?
At the same time, suburb-level forecasts get dismissed as too granular or unreliable. And street-level analysis almost never gets discussed. In this paper, we test each level with a simple, bias-free method and real transaction data.
The core question is straightforward. If a market is truly a market, properties within it should move together. Good times, they grow together. Bad times, they fall together. If they do, you can talk about it as a real market. If they don't, you need to zoom in.
Method
Dataset
We analysed 10,874 property resales across Australia. These are properties bought and sold without renovation, held for a maximum of three years. For each resale we calculated the annualised capital growth. Buy years span 2004 to 2025. Sell years span 2006 to 2025. Restricting holding periods to three years or less means each cohort captures a single market cycle. Not a blend of multiple cycles.
Cohort Construction
Properties were grouped into cohorts by region + buy year + sell year. This means all properties in a cohort were held during the same period in the same area. The same market forces acted on all of them. Only cohorts with two or more properties were included.
Benchmark
Each property's annual capital growth was calculated. The overall median was used as the benchmark. Each property was then classified as either outperforming or underperforming it.
Metrics
1. Same-Half Probability
For each cohort, we found whether the majority of properties outperformed or underperformed the median benchmark. We then measured what percentage of individual properties matched their cohort’s majority direction. A score of 100% means perfect cohesion. Every property moves the same way.
A Note on Random Baselines and Cohort Size
The random baseline for this metric is not always 50%. It depends on cohort size. For large cohorts (hundreds of properties), the baseline sits close to 50%. A true coin flip. But for small cohorts, majority agreement happens more often by chance alone. In a cohort of just 2 properties, random draws produce same-direction agreement 75% of the time. For cohorts of 3, the baseline is also 75%. For cohorts of 10, it falls to roughly 62%.
This matters because finer geographic levels naturally produce smaller cohorts. At the national level, cohorts average 191 properties (baseline roughly 53%). At the street level, cohorts average just 2.2 properties (baseline roughly 75%). To account for this, we report Uncertainty Reduced in the comparison table. This is the percentage of the gap between the random baseline and perfect cohesion (100%) that the observed result closes. Example: if the baseline is 75% and the observed rate is 97.3%, there are 25 percentage points of possible improvement. 22.3 are realised. 22.3 / 25 = 89% of uncertainty removed.
2. Mean Absolute Deviation (MAD)
The average absolute difference between each property’s annual return and its cohort median, expressed in percentage points per year. Lower values mean tighter clustering of outcomes within an area.
Geographic Levels Tested
Five levels of geographic granularity were analysed.
- National. All of Australia as a single market.
- State. NSW, VIC, QLD, WA, SA, TAS, ACT, NT.
- SA4 (Statistical Area Level 4). Major urban regions (e.g. ‘Sydney, City and Inner South’, ‘Brisbane Inner City’).
- Suburb. Individual suburbs (SAL boundaries).
- Street. Individual street names within suburbs.
Results
Zoom in from national to street level. Markets become progressively more real.
National Level
64.3% move togetherAt the broadest level, only 64% of properties match the national trend. Cohorts average 191 properties, so the random baseline is close to 53%. That means only 24% of uncertainty is removed by knowing the national direction. Properties deviate by an average of 10.34 pp/year from their cohort median. When a commentator says ‘the market is going up’, your individual property could easily go the opposite way.
Coin flip. Not a meaningful marketState Level
71.7% move togetherBetter than national, but still weak. Roughly 3 in 10 properties go against their state’s trend. Average deviation: 9.49 pp/year. Buying ‘in NSW’ does not determine your outcome. Property selection within the state still dominates.
Weak market. Some signal, lots of noiseSA4 (Regional) Level
77.7% move togetherAt the regional level, roughly 3 in 4 properties match their area’s direction. Average deviation drops to 7.98 pp/year. The region starts to tell a real story. The Gold Coast moves differently from Newcastle. But 1 in 4 properties still goes against the grain.
Moderate. Area matters, but property selection still crucialSuburb Level
90.3% move togetherThis is where markets become real. 9 in 10 properties match their suburb’s direction. Average deviation is just 3.49 pp/year. When someone says ‘Bondi did well this cycle’ or ‘Brunswick struggled’, that is a meaningful statement. Nearly every property in that suburb matched.
Strong market. Suburb selection is a real edgeStreet Level
97.3% move togetherNearly perfect cohesion. Street-level cohorts average just 2.2 properties, so random agreement alone would produce about 75%. Yet the observed 97.3% removes 89% of the remaining uncertainty. The highest of any level. Average deviation is just 0.87 pp/year. Streets are the most cohesive market level we tested.
Very strong. Near-perfect cohesionComparison Across All Levels
| Market Level | Same-Half % | Random Baseline | Uncertainty Reduced | MAD (pp/yr) | Avg Cohort Size | Verdict |
|---|---|---|---|---|---|---|
| National | 64.3% | ≈53% | 24% | 10.34 | 191 | Coin flip |
| State | 71.7% | ≈56% | 36% | 9.49 | 43 | Weak |
| SA4 (Region) | 77.7% | ≈64% | 38% | 7.98 | 7.5 | Moderate |
| Suburb | 90.3% | ≈75% | 61% | 3.48 | 2.5 | Strong |
| Street | 97.3% | ≈75% | 89% | 0.87 | 2.2 | Very strong |
The ‘Uncertainty Reduced’ column is the most honest measure of market signal. It asks a simple question. Of the uncertainty that could be removed by geographic grouping (the gap between the random baseline and 100%), how much actually was? At the national level, only 24% of uncertainty is removed. Knowing ‘the Australian market’ barely helps. At the suburb level, 61% is removed. At the street level, 89%. And the MAD drops from 10.34 to 0.87 pp/year. Finer geographic levels are genuinely more cohesive markets. Even after accounting for the small-cohort effect.
Key Takeaways
Suburb selection does 61% of the heavy lifting. Not 80%. The old saying overstates it. Suburb selection is a real, measurable edge. But it leaves nearly 40% of the uncertainty on the table. And under matched-pair controls, the suburb signal drops to 24–27%. It may be weaker than it looks.
Street selection does 89%. Confirmed both ways. Two properties bought and sold on the same street in the same years have a 97.3% chance of moving in the same direction. It is the only level that survives both the all-properties and matched-pair tests. z = 15.3. Overwhelming.
National and state-level timing is noise. The pairwise test proves it. Approach 1 flags national, state, and SA4 as significant. That is a large-n artefact. Under matched-pair controls, state drops to 14% and SA4 to just 5%. Indistinguishable from chance.
Property selection still matters. Even on the right street. Even at the suburb level, properties deviate by 3.5 percentage points per year. Over a few years that compounds into a meaningful difference. The specific property you pick still matters.
GCCSA-Level Cohesion by Region and Holding Period
The table below shows the GCCSA-level same-half percentage for each region, broken down by maximum holding period (1, 2, and 3 years). This is cohesion measured at the GCCSA level. How often do properties within the same capital city or rest-of-state region move in the same direction? A dash means insufficient data (fewer than 20 properties in valid cohorts).
| GCCSA Region | 1 Year | 2 Years | 3 Years |
|---|---|---|---|
| Greater Sydney | 73% | 74% | 74% |
| Greater Melbourne | 88% | 78% | 77% |
| Greater Brisbane | 78% | 71% | 74% |
| Greater Perth | 74% | 69% | 75% |
| Greater Adelaide | 78% | 71% | 75% |
| Greater Darwin | – | 85% | 89% |
| Rest of NSW | 75% | 72% | 74% |
| Rest of Vic. | 80% | 75% | 78% |
| Rest of Qld | 73% | 66% | 68% |
| Rest of WA | 76% | 69% | 68% |
| Rest of SA | 86% | 73% | 67% |
| Rest of Tas. | – | 72% | 74% |
| Australian Capital Territory | 74% | 71% | 73% |
No GCCSA region consistently reaches high cohesion. Most hover between 68% and 78%. That looks decent on the surface. But random baselines for cohorts this size sit at 56 to 62%. After adjustment, GCCSA-level grouping removes only 25 to 45% of uncertainty in most regions. Knowing which capital city or rest-of-state region a property is in does not get you far. You need to go deeper.
Robustness Check: Pairwise Analysis
The results above use what we call Approach 1 (All-Properties). It maximises the sample at each geographic level. More statistical power. But different properties are analysed at each level, so cross-level comparisons are not strictly like-for-like.
To test whether the results hold under tighter controls, we ran a second analysis: Approach 2 (Matched-Pair). This fixes the dataset to properties that form valid street-level pairs. Every geographic level is tested on the exact same properties. Cross-level comparisons become genuinely like-for-like.
Approach 1: All-Properties
Maximum available sample at each level. More statistical power, especially at suburb and SA2. But different properties are analysed at each level.
Approach 2: Matched-Pair
Fixed dataset of street-level pairs. Every level tested on the same properties. Genuinely like-for-like. The more rigorous approach.
| Market Level | Approach 1 Unc. Removed | Approach 1 Sig. | Approach 2 Unc. Removed | Approach 2 Sig. | Verdict |
|---|---|---|---|---|---|
| National | 24% | large-n *** | 19% | ns | Noise |
| State | 36% | large-n *** | 14% | ns | Unreliable |
| SA4 (City region) | 39% | large-n *** | 5% | ns | Unreliable |
| Suburb / SA2† | 56–61% | Approach 1 only *** | 24–27% | ns | Contested |
| Street | 89% | *** | 91% | *** | Confirmed |
† Approach 1 has far more data at suburb/SA2 level (5,570 to 6,073 properties versus 66 to 79 in Approach 2). The wide Approach 2 confidence interval (0 to 57%) reflects small sample size, not a confirmed null result.
Why Approach 1 shows significance everywhere
With 10,000+ properties, Approach 1 has the statistical power to detect very small effects. National, state, and SA4 all appear significant (***) in Approach 1. But their uncertainty-removed values are only 24%, 36%, and 39%. Roughly two-thirds of investment uncertainty remains unexplained.
Under Approach 2's controlled test, those signals collapse. State drops to 14%. SA4 drops to just 5%. Both are statistically indistinguishable from noise. The Approach 1 significance labels at national, state, and city level are detecting negligible effects with a very large sample. This is a known statistical phenomenon. It is not evidence of a useful market signal.
Only the street survives both the All-Properties and the Matched-Pair methodology.
Street-level cohesion holds under both Approach 1 (All-Properties, 89% uncertainty removed) and Approach 2 (Matched-Pair, 91% removed). z = 15.3 in both cases. No other geographic level achieves significance under both methodologies.
Real transaction examples
The street-level pattern holds 97.6% of the time across 855 street cohorts. Below are two real pairs from the dataset. Same street. Same years. Independent buyers. Same direction.
One finding survives both methodologies. Streets: 89 to 91% uncertainty removed.
10,874 property resales. Two independent methodologies. One clear result. National, state, and regional signals collapse under matched-pair controls. Suburb shows promise but remains contested. Streets hold at 89 to 91% across both approaches. That is why Microburbs provides street-level data. Median prices, rents, renters, and turnover rates.
© 2026 Microburbs Pty Ltd. Data sourced from public property records. All analysis based on non-renovated resales held ≤ 3 years.