Mean Reversion Threshold: Technical Whitepaper
Full statistical methodology, threshold calibration, temporal consistency testing, and regional robustness results across 825,392 property sales.
1. Abstract
This paper presents a univariate threshold that measures mean reversion in Australian suburb house prices. The threshold uses a single variable: cumulative median house price growth over the preceding 10 years. Suburbs where prices grew less than 44% over the past decade are classified as "top tier" and tend to outperform in the next four years. Suburbs where prices grew more than 91.9% are classified as "bottom tier" and tend to underperform.
The top tier outperformed the national median by +1.14 percentage points per year over rolling 4-year windows. The bottom tier underperformed by -1.88 percentage points. The total spread is 3.02 percentage points per year. This was measured across 825,392 property sales from 2008 to 2021.
The signal was tested across 55 quarterly periods and 28 individual sample dates. It held at every single quarter and every single sample date. This is a 100% consistency rate. No other threshold in the Microburbs research programme achieves this level of temporal consistency. The t-statistic is 251.87 and the p-value is effectively zero.
The signal was tested across 15 GCCSA regions. It produced a positive spread in 12 of 15 regions. The three regions where it inverted are resource-driven markets (Greater Perth, Greater Darwin, and Rest of NT) where mining boom prices did not fully revert.
2. Methodology
2.1 Variable Construction
The mean reversion threshold uses a single variable: the cumulative percentage change in a suburb's median house price over the preceding 10 years.
For example, if a suburb's median house price was $300,000 ten years ago and is $420,000 today, the past growth is 40%. This suburb would fall into the top tier (below 44%). If another suburb's median was $250,000 and is now $500,000, the past growth is 100%. This suburb would fall into the bottom tier (above 91.9%).
2.2 Threshold Calibration
The thresholds were calibrated using property sales data spanning 2008 to 2021. The top tier threshold is 44% past 10-year growth. The bottom tier threshold is 91.9%. Suburbs between these two thresholds fall into the middle tier.
2.3 Performance Metric
The primary metric is the difference in median annualised 4-year growth between each tier and the national median. Statistical significance is assessed using a two-sided t-test.
t-statistic = (mean(tier) - mean(national)) / SE(tier)
p-value from two-sided t-test
2.4 Inverted Logic
This threshold is inverted relative to most other indices. The "top tier" contains suburbs with LOW past growth. These are the suburbs that outperform in the future. The "bottom tier" contains suburbs with HIGH past growth. These are the suburbs that underperform in the future.
3. Tier Performance
The model sorts suburbs into three tiers based on their past 10-year median house price growth. Each tier has a distinct forward growth profile.
| Tier | Past Growth Range | Diff vs National | p-value | N (Sales) | Significant |
|---|---|---|---|---|---|
| Top | Below 44% | +1.14% | ~ 0 | 367,662 | Yes |
| Middle | 44% to 91.9% | -0.14% | ~ 0 | 252,173 | Yes |
| Bottom | Above 91.9% | -1.88% | ~ 0 | 205,557 | Yes |
4. Temporal Analysis
We tested the mean reversion threshold across every quarter from 2008-Q1 to 2021-Q3. The top tier outperformed the bottom tier at every single quarter and every single sample date.
4.1 Date-by-Date Consistency
The result was positive at all 28 dates. Every single date showed the top tier outperforming the bottom tier.
| Sample Date | Top Tier Growth | Bottom Tier Growth | Spread | Top N | Bottom N | Significance |
|---|---|---|---|---|---|---|
| 2008-03 | +1.05% | -1.57% | +2.63% | 1,825 | 1,505 | Significant |
| 2008-09 | +1.15% | -1.38% | +2.53% | 1,727 | 1,593 | Significant |
| 2009-03 | +1.14% | -1.29% | +2.43% | 1,742 | 1,615 | Significant |
| 2009-09 | +1.04% | -0.98% | +2.02% | 1,681 | 1,676 | Significant |
| 2010-03 | +1.27% | -1.05% | +2.32% | 1,594 | 1,867 | Significant |
| 2010-09 | +1.45% | -0.96% | +2.41% | 1,513 | 2,023 | Significant |
| 2011-03 | +1.72% | -1.05% | +2.77% | 1,440 | 2,088 | Significant |
| 2011-09 | +2.18% | -1.45% | +3.63% | 1,432 | 2,003 | Significant |
| 2012-03 | +2.31% | -1.76% | +4.07% | 1,436 | 1,966 | Significant |
| 2012-09 | +2.46% | -2.16% | +4.63% | 1,439 | 1,982 | Significant |
| 2013-03 | +2.77% | -2.71% | +5.49% | 1,404 | 1,950 | Significant |
| 2013-09 | +2.86% | -3.75% | +6.61% | 1,569 | 1,574 | Significant |
| 2014-03 | +2.36% | -4.02% | +6.38% | 1,842 | 1,404 | Significant |
| 2014-09 | +1.95% | -3.71% | +5.67% | 1,876 | 1,268 | Significant |
| 2015-03 | +1.60% | -3.25% | +4.84% | 2,015 | 1,057 | Significant |
| 2015-09 | +1.31% | -2.04% | +3.35% | 2,054 | 999 | Significant |
| 2016-03 | +0.65% | -0.97% | +1.62% | 2,285 | 858 | Significant |
| 2016-09 | +0.09% | -0.47% | +0.56% | 2,629 | 802 | Significant |
| 2017-03 | +0.22% | -1.03% | +1.25% | 2,767 | 844 | Significant |
| 2017-09 | +0.16% | -1.12% | +1.29% | 3,037 | 815 | Significant |
| 2018-03 | +0.26% | -1.68% | +1.95% | 3,260 | 779 | Significant |
| 2018-09 | +0.47% | -2.13% | +2.61% | 3,344 | 779 | Significant |
| 2019-03 | +0.49% | -1.85% | +2.34% | 3,332 | 709 | Significant |
| 2019-09 | +0.49% | -1.08% | +1.57% | 3,331 | 547 | Significant |
| 2020-03 | +0.83% | -1.43% | +2.26% | 3,354 | 510 | Significant |
| 2020-09 | +1.15% | -1.78% | +2.93% | 3,416 | 397 | Significant |
| 2021-03 | +1.67% | -2.67% | +4.34% | 3,165 | 627 | Significant |
| 2021-09 | +2.52% | -3.80% | +6.32% | 2,602 | 1,167 | Significant |
5. Regional Robustness
We tested the mean reversion threshold across all 15 GCCSA regions in Australia.
| Region (GCCSA) | Top Tier Growth | Bottom Tier Growth | Spread | N (Sales) | p-value |
|---|---|---|---|---|---|
| Greater Sydney | +4.51% | -1.89% | +6.40% | 63,808 | ~ 0 |
| Greater Hobart | +5.75% | -0.57% | +6.32% | 251 | 1.3e-32 |
| Rest of Tas. | +3.11% | -0.38% | +3.49% | 364 | 1.1e-34 |
| Rest of Qld | +0.19% | -3.10% | +3.30% | 119,654 | ~ 0 |
| Rest of SA | +0.27% | -2.51% | +2.78% | 26,495 | ~ 0 |
| Rest of WA | -1.75% | -4.51% | +2.77% | 41,480 | ~ 0 |
| Greater Brisbane | +1.22% | -1.41% | +2.63% | 48,015 | ~ 0 |
| Rest of NSW | +2.00% | -0.34% | +2.34% | 86,646 | ~ 0 |
| Greater Melbourne | +2.29% | +0.14% | +2.15% | 33,702 | 4.2e-260 |
| Greater Adelaide | +0.69% | -1.14% | +1.83% | 36,174 | ~ 0 |
| Rest of Vic. | +1.71% | +0.20% | +1.51% | 56,582 | ~ 0 |
| ACT | +0.07% | +0.16% | -0.09% | 5,948 | 0.36 |
| Greater Perth | -2.13% | -1.86% | -0.27% | 48,743 | 2.8e-12 |
| Greater Darwin | -3.86% | -2.89% | -0.97% | 3,999 | 2.5e-10 |
| Rest of NT | -4.99% | -3.19% | -1.80% | 1,358 | 9.6e-21 |
6. Suburb-Level Evidence
Suburb-level comparisons for selected cities are available on the summary.
7. Defence of Method
7.1 Why Mean Reversion Works
Mean reversion in property markets is well documented in academic literature. The mechanism is straightforward. Suburbs that have grown quickly become relatively expensive. Buyers start looking at cheaper alternatives. Capital flows shift toward undervalued areas. Over time, the gap between expensive and cheap suburbs narrows.
7.2 Statistical Significance
The t-statistic is 251.87. The probability of observing a +3.02% spread across 825,392 sales by random chance is effectively zero. For context, a p-value below 0.05 is the standard threshold for statistical significance. The mean reversion threshold exceeds this by hundreds of orders of magnitude.
7.3 Consistency Over Time
The signal was positive at all 28 sample dates spanning 14 years. It was positive in 100% of 55 quarterly observations. No other threshold in the Microburbs research programme achieves this level of temporal consistency.
7.4 Geographic Breadth
The spread is positive in 12 of 15 GCCSA regions. It works in rising markets (Greater Sydney, Greater Hobart) and falling markets (Rest of WA). The three exceptions are resource-driven markets where mining booms created unusual price dynamics.
7.5 Practical Use
The mean reversion threshold provides a simple screen for suburb selection. Look for suburbs where the median house price grew less than 44% over the past decade. Avoid suburbs where it grew more than 91.9%. This is a filter, not a final answer.
8. Limitations
8.1 Resource Markets
The signal inverts in three resource-driven regions: Greater Perth, Greater Darwin, and Rest of NT. In these markets, suburbs that boomed during mining cycles did not fully revert. Investors in Western Australia and the Northern Territory should not rely on this threshold alone.
8.2 Backward-Looking Model
The model was calibrated on historical data from 2008 to 2021. Past patterns do not guarantee future results. However, mean reversion is a fundamental economic principle, not an artefact of a specific time period.
8.3 Individual Suburb Variation
Even within the top tier, individual suburb outcomes vary widely. Dysart recovered at +5.09% per year, but Nome lost -3.78% per year despite also having low past growth. The threshold provides a statistical edge across large numbers of purchases, not a guarantee for any single suburb.
8.4 Threshold Stability
The thresholds (44% and 91.9%) were calibrated on a specific data sample. Different calibration periods might produce slightly different thresholds. The underlying principle (low past growth predicts strong future growth) is robust, but the exact cutoff values may shift over time.
8.5 Confounding Variables
Suburbs with low past growth may share other characteristics that also predict future growth. The mean reversion effect may be partially explained by these confounders. This paper documents a correlation, not a causal mechanism.
Access Suburb-Level Scores
Get mean reversion data for every suburb in Australia. Identify suburbs where low past growth signals strong future performance.