Which backtest is this?
This page documents the selection-efficacy backtest: does the Dex ranking sort stronger markets from weaker ones before the fact? It is a different exercise from the forecast-accuracy backtest, which measures how close the published numbers land to what happened. Both are summarised side by side in how HtAG validates its property data and models.
Quick Summary
HtAG back-tested its Dex suburb-selection strategy against every house suburb in Australia across 14 years (2012–2025). The top-decile picks beat the market at every holding period, and outperformed in all 14 individual years. On identical cash at 80% LVR, the growth difference compounded into 38.7% more equity after fifteen years in the affordable band. This article sets out exactly how the test was run, what the edge is worth in dollars, and what it does not establish.
Every property data company claims its model finds the next growth suburb. Almost none show the receipt. So HtAG Analytics put its own engine — the Dex strategy — on trial against the hardest benchmark available: every house suburb in Australia, across 14 years of booms, flat years and the 2022 correction. (New to the concept? Start with what backtesting in property forecasting means — and how to spot a fake one.)
The one-line version: across 14 years and every Australian house suburb, Dex top-decile picks grew 17.6% over the following year against 9.1% for the market, beat their own price band in 87.1% of cases, and did it in all 14 years tested. The edge is real, it is largest at one year and in the affordable band, and it narrows as the horizon lengthens — which is exactly why the dollar value depends on how long you hold and how the purchase is funded.
Table of Contents
- What the 14-Year Backtest Actually Tested
- The Results: Dex Picks vs the Market
- Three Findings, and What Each One Does Not Cover
- In Dollar Terms: What the Edge Is Actually Worth
- Why a Validated Ranking Beats a Single Metric
- How the Backtest Was Built (and Its Limits)
- The Statistical Basis, Stated Plainly
- What This Backtest Does Not Establish
- Surface This Data Inside Your AI Agent
- From Data Signal to Portfolio Decision
- Key Takeaways
- Frequently Asked Questions
What the 14-Year Backtest Actually Tested
The backtest asked one question at every point in history: which suburbs would the Dex strategy have flagged as top picks using only the data available on that day — and what did those suburbs actually do next? The word that matters is “only.” There is no hindsight in the selection.
At each historical date, HtAG rebuilt the Dex selection signal from the data that existed then — price momentum, gross yield, the growth cycle, affordability (years-to-own), demand and turnover, and IRSAD socio-economic quality — selected the top 10% of suburbs, then measured their real forward price growth over 1, 3 and 5 years. The strategy was matched to the holding period: the short-term “hot spots” play for the 1-year test, the mid-term strategy for 3 years, and the long-term lower-risk strategy for 5 years.
Crucially, each pick was judged against other suburbs in the same price band — a $450,000 pick measured against other $450,000 suburbs, not against blue-chip markets. According to HtAG Analytics, the test pooled roughly 150,000 suburb observations across the 2012–2025 window.
What This Means in Plain English
“Top decile” just means the top 10% of suburbs the strategy liked best at the time. We then watched what really happened to those suburbs afterwards — using only information that was available on the day they were picked, so there’s no cheating with hindsight.
The Results: Dex Picks vs the Market
HtAG Dex top-decile picks beat the whole-market benchmark at every horizon tested. The table below shows annualised growth (CAGR), the percentage-point edge, how often picks beat their price-band benchmark, and how rarely they lost money.
The table below is the primary result. Read it as three separate strategies benchmarked independently, not as one strategy held for three different lengths of time.
| Holding period | Dex picks (CAGR) | Whole market | Edge | Beat-market rate | Holds that lost money |
|---|---|---|---|---|---|
| 1 year (short-term) | 17.6% | 9.1% | +8.9 pp | 87% | 3.1% |
| 3 years (mid-term) | 9.4% | 7.0% | +2.5 pp/yr | 67% | 6.0% |
| 5 years (long-term) | 8.1% | 6.8% | +1.2 pp/yr | 64% | 3.9% |
Source: HtAG Analytics backtest, houses, suburb level, 2012–2025. CAGR = compound annual growth rate. Each horizon is benchmarked against the same universe measured over the same forward window, and the edge column is computed against a price-band-matched benchmark, so a $450k pick is judged against other $450k markets. Each horizon uses a different Dex archetype: short-term, mid-term and long-term respectively.
Over a one-year hold, HtAG Dex top-decile suburb picks grew 17.6% versus 9.1% for the same universe over the same window — and beat their price-band benchmark 87.1% of the time.
HtAG Analytics, 14-Year Dex Backtest (2026)
Three Findings, and What Each One Does Not Cover
The headline outperformance matters, but three deeper findings are what separate a genuine edge from a lucky window.
1. It won in every one of 14 years
The short-term Dex picks beat the universe in all 14 anchor years from 2012 to 2025, with the annual edge ranging from +4.2 to +16.4 percentage points. The result is not driven by a single boom — it holds through the flat market of 2018 and the 2022 correction alike.
2. The edge is largest and tightest at the short horizon
As the horizon lengthens the raw edge narrows, from +8.9 percentage points at one year to +2.5 at three and +1.2 at five. Downside is not monotonic and should not be presented as such: the worst 5% of picks returned +2.0% per year at the one-year horizon, -0.5% at three years and +0.9% at five, and the share of holds that lost money was 3.1%, 6.0% and 3.9% respectively. The three-year archetype carries the widest downside of the three. Each horizon is a different strategy rather than the same picks held longer, so these rows compare three methods, not one method over time.
3. The edge is biggest in affordable suburbs
In the sub-$550,000 band, Dex picks beat their price-band benchmark by +12.1 percentage points over a single year — the strongest edge of any band. For investors and buyers’ agents working in accessible price points, that is where the data advantage concentrates.
What This Means in Plain English
A short-term pick can swing hard in a single year. But hold the strategy’s picks for five years and the outcomes become remarkably dependable — nearly all of them made money, and even the unluckiest ones still grew. Time turns the edge from “bigger” into “safer.”
In Dollar Terms: What the Edge Is Actually Worth
A percentage-point edge is not a number anyone can act on. What it converts to depends on two things the backtest does not decide for you: how long the asset is held, and how much of the purchase price was your own cash. Both are set out below, conservative case first.
Every figure in this section is measured against a price-band-matched benchmark — a $450,000 pick is judged against other $450,000 markets, not against the national average. That is the harder test and it produces smaller headline numbers than a whole-of-market comparison would.
Stage one: one property, bought and held
Taking each archetype at its own horizon and comparing it with its own price band, this is the difference in the value of the asset itself, growth only:
| Purchase price | Held 1 year | Held 3 years | Held 5 years |
|---|---|---|---|
| $450,000 (sub-$550k band) | +$52,000 | +$63,000 | +$52,000 |
| $650,000 ($550–850k band) | +$48,000 | +$28,000 | +$23,000 |
| $850,000 ($850k+ band) | +$55,000 | +$36,000 | +$39,000 |
Source: HtAG Analytics. Each horizon uses a different Dex archetype benchmarked against its own price band, so these are three strategies rather than one strategy held longer. Both legs compounded at their measured CAGR. Rounded to the nearest $1,000. Growth only; acquisition costs, holding costs and taxes excluded.
Two things stand out, and neither is what a marketing table would show. The edge is largest in the affordable band, not the expensive one. And it is largest at one year, then narrows — the sub-$550k band aside, the five-year archetype in the mid band adds only +0.54 percentage points a year.
That is where most published backtests stop, and it is the point at which the numbers look least impressive. It is also the wrong place to stop, because almost nobody buys property with cash.
Applying the same band-matched growth rates to a leveraged purchase — 80% LVR, interest-only, 5.5% acquisition costs — changes the size of the result without changing a single input. The gap in the asset is unchanged; what changes is the amount of the investor’s own money that gap is a return on.
| $450,000 in the sub-$550k band | Held 5 yrs | Held 10 yrs | Held 15 yrs | Held 20 yrs |
|---|---|---|---|---|
| Equity, Dex pick | $319,144 | $664,969 | $1,186,892 | $1,974,583 |
| Equity, same-band market buy | $266,737 | $512,887 | $855,711 | $1,333,180 |
| Equity gap | +$52,407 | +$152,083 | +$331,181 | +$641,403 |
| More equity than the market buy | +19.6% | +29.7% | +38.7% | +48.1% |
Source: HtAG Analytics. Identical $114,750 cash in on both legs (20% deposit plus 5.5% acquisition costs). Interest-only, so the loan balance is constant and every dollar of the difference is equity. Growth only; holding costs, rates, maintenance and taxes excluded.
The compounding is the point. A growth difference of 1.73 percentage points a year is worth 19.6% more equity at five years and 48.1% more at twenty — on identical cash. The percentage gap does not stay flat as the hold lengthens; it widens every year, because the difference compounds on a base that is itself compounding.
The same arithmetic in the other two bands is smaller but moves the same way: the $550–850k band reaches +11.0% more equity at fifteen years, and the $850k+ band +15.5%.
For a lender, the more relevant number is the loan-to-value trajectory. On that $450,000 purchase, after fifteen years the Dex-picked property sits at an LVR of 23.3% against 29.6% for the same-band market buy, on the same original loan. Faster deleveraging on identical debt is a lower-risk book, not just a better return.
Stage two: running the strategy, rather than holding one house
The figures above hold one property and let it compound. A different question is what the selection method itself returned over the full test window — chaining each anchor year’s measured forward one-year growth from 2012 to 2025.
| Chained 2012–2025 | Dex picks, re-selected annually | All Australian house markets |
|---|---|---|
| Cumulative multiple of capital | 9.61x | 2.53x |
| Compound annual growth rate | 17.54% | 6.84% |
| $650,000 grown over the window | $6,245,682 | $1,642,087 |
Source: HtAG Analytics, chained from the fourteen anchor years of the backtest.
This figure must be read for what it is. It is the return to running the strategy — re-selecting each year — and not the return to buying one property and holding it for fourteen years. It assumes the median top-decile pick is achieved every year and it excludes transaction costs, which are substantial if the strategy is executed by selling and rebuying rather than by holding and borrowing against equity. Quoted as a property return it would be wrong; quoted as a strategy return it is what the fourteen anchor years actually produced.
Note also that the universe compounds to 6.84% a year on this equal-weighted basis, against the 9.1% quoted earlier in this article. Both are correct. The 9.1% is weighted by the number of picks, so the strong 2021–22 market carries more of it; the chain weights each of the fourteen years equally. They answer different questions and should not be mixed in the same sentence.
What these figures exclude
Growth only. No rental income, no holding costs, no rates, no maintenance, no tax treatment and no selling costs. They also assume the median outcome of a top-decile pick, not a specific property — suburb-level medians do not transfer automatically to any individual purchase. And the leveraged figures assume the borrowing is available: on a single median income, servicing constrains a second purchase long before equity does, so multi-property projections built on these growth rates should be treated with considerable caution.
On identical cash, a Dex-picked property in the sub-$550,000 band left an investor with 38.7% more equity after fifteen years than a same-band market buy — $331,181 on a $450,000 purchase — from a growth difference of 1.73 percentage points a year.
HtAG Analytics, Dex Backtest Equity Model (2026)
Why a Validated Ranking Beats a Single Metric
Most “suburb picks” rest on a single number — last year’s growth, a yield figure, or an auction clearance rate. The problem is that any one metric is easy to game and easy to misread. The Dex strategy is different: it is a multi-signal, horizon-matched, price-band-aware ranking that weighs momentum, yield, the growth cycle, affordability, demand and socio-economic quality together.
This backtest is the evidence that the combined ranking carries genuine forward predictive power — not just a flattering description of the past. It outperformed consistently, across cycles, with downside protection built in. That is the difference between a score that looks intelligent and a process that actually compounds capital faster than the market. To understand the cycle signal that feeds it, see the HtAG Growth Rate Cycle explainer, and to see selection in action explore the GeoDex heatmap.
How the Backtest Was Built (and Its Limits)
The test was run as a point-in-time signal reconstruction with strict no-look-ahead controls: every signal at a given date used only data available on or before that date, and forward returns used only actual, never forecast, prices. Forward growth was measured directly from HtAG’s monthly price series, which runs back to 2007. One qualification matters and is stated rather than buried: the price series itself is a current-vintage index. HtAG does not hold stored historical vintages of it, so the 2012 price path reflects the index as estimated today rather than as it stood in 2012. This applies identically to the picks and to the benchmark, so it does not create a directional advantage, but it does mean the exercise is a point-in-time reconstruction rather than a true archival replay.
Two honest caveats matter. First, this is a faithful reconstruction of the Dex selection signal from HtAG’s deep data history — not a literal replay of the live endpoint, because point-in-time snapshots of the full 40-metric Dex set only extend back six months. Second, when extended to a full 10-year hold, the long-term edge largely converges with the market as mean-reversion sets in. The reliable outperformance window is therefore 1 to 5 years. HtAG Analytics reports the result with these limits stated rather than hidden.
What This Means in Plain English
We didn’t let the test peek at the future. We rebuilt what the strategy would have chosen on each day using only what was known then, then checked what actually happened. And we’re upfront that the edge is strongest over one to five years, not forever.
The Statistical Basis, Stated Plainly
Three properties of this backtest matter more to a professional reviewer than the headline growth number: the edge is not a disguised market bet, the benchmark can be rebuilt from source data, and the result survives the strictest available treatment of sample size. Each is stated below with the figure and its basis.
The edge is not a market-timing tilt
Regressing annual pick growth on whole-universe growth across the 14 anchor years returns a beta of 1.01 and an intercept of +10.65 percentage points. The correlation between the annual excess and market conditions is 0.016, with a p-value of 0.96. In plain terms: Dex picks carry the same market exposure as the universe and add a return on top of it, and that added return does not depend on whether the market was strong or weak in a given year. The 2022 anchor year, the weakest of the fourteen, still produced a +4.2 percentage point edge.
The benchmark reproduces from source data
A backtest is only as trustworthy as the benchmark it beats. The universe series used here was independently rebuilt from HtAG’s locality-level monthly price panel, covering 6,236 Australian house markets, and compared against the published series year by year. The two agree at a correlation of 0.992, with a mean absolute error of 0.38 percentage points and a mean bias of zero. The residuals change sign across years rather than running in one direction, which is what an honest reconstruction looks like and what a systematically understated benchmark does not.
The result survives a strict sample-size treatment
The one-year test contains 3,769 picks, but those picks overlap in time and cluster geographically, so treating them as 3,769 independent observations would overstate the evidence. The defensible unit is the anchor year, of which there are 14. On that basis the mean annual excess is +10.73 percentage points unweighted and +9.00 percentage points when weighted by the number of picks in each year, with a t-statistic of 12.50 and a p-value of 1.3 x 10-8. A bootstrap over the 14 annual excesses puts the 95% confidence interval at +9.10 to +12.35 percentage points. Nothing is lost by being conservative here: the result is decisive either way.
The unweighted and pick-weighted means differ by 1.73 percentage points because pick counts vary sharply across years, from 15 in 2012 to 845 in 2022. Both figures are published rather than only the more flattering one.
What This Backtest Does Not Establish
A backtest that reports only its strengths is a marketing document. These are the limits of what the evidence above supports, stated so that a reader can price them rather than discover them.
- No stored point-in-time price vintage. Signals are reconstructed point-in-time, but the underlying price index is the current vintage. HtAG does not hold archived historical vintages of it, so the exercise is a reconstruction rather than an archival replay. This applies identically to picks and benchmark, so it does not create a directional advantage.
- Costs are excluded. All figures are growth only. Acquisition costs, holding costs, financing and taxes are not modelled, and they fall on both legs of the comparison.
- Year-by-year detail exists only for the short-term strategy. The 14-of-14 consistency record is a property of the one-year archetype. Annual tables for the three-year and five-year archetypes are not yet published, so the consistency claim should not be generalised to them.
- Thin anchor years carry thin evidence. The 2012 result rests on 15 picks. It is included for completeness, not because a 15-market sample is decisive on its own.
- Selection, not valuation. This validates a suburb-level ranking’s ability to sort markets. It is not a property-level valuation model and carries no per-property error attribution.
- Liquidity is not modelled. The test assumes a pick can be acted on. In thin regional markets that assumption is weaker than in metropolitan ones.
Full technical paper
The Dex Selection Backtest: Methodology, Results and Limitations
The complete methodology note written for analysts and risk reviewers: construction of the selection signal, benchmark definition and reproduction, full results by horizon and price band, statistical treatment of the effective sample, a stated limitations register, and the boundary of what the result can and cannot support in a credit or risk context.
Read the technical paperSurface This Data Inside Your AI Agent
The exact Strategy endpoint that powers this backtest is now live in the HtAG Developer Portal — and it is queryable directly by AI agents through MCP (Model Context Protocol). Investors and buyers’ agents using Claude, Perplexity, Manus AI, or any MCP-compatible AI tool can run the Dex strategy in plain English and get a ranked suburb shortlist back, without leaving the AI tool they already use.
HtAG’s MCP-enabled Developer Portal puts the same Strategy endpoint validated in this backtest inside your AI agent. Ask “run the short-term Dex hot-spots strategy on suburbs under $550k in South-East Queensland” and get the ranking back instantly.
HtAG Analytics Developer Portal (2026)
Browse the endpoint catalogue at developer.htagai.com and submit the HtAG Developer Portal application — approved members receive an API key and an MCP setup guide for their preferred AI tool.
From Data Signal to Portfolio Decision
The Dex strategy, the Growth Rate Cycle, and the supply and demand signals described in this article are live inside the HtAG Analytics platform — updated each quarter as new valuation data flows in. Professional buyers’ agents use these signals to time entries, validate briefs, and build conviction before making offers; you can see real outcomes in the HtAG Evidence Portal.
If you’re building a portfolio and want the exact data powering articles like this one, the HtAG Starter Plan gives you suburb-level analytics across every Australian market — no lock-in, cancel any time.
Start your HtAG Analytics membership → · Apply for Developer Portal access →
Key Takeaways
- HtAG Dex top-decile picks beat the market at every horizon, each measured against the same universe over the same forward window: 17.6% vs 9.1% over 1 year, 9.4% vs 7.0% per year over 3 years, and 8.1% vs 6.8% per year over 5 years.
- The short-term picks beat the market in all 14 years from 2012 to 2025 — no cherry-picked window.
- Downside is contained but not monotonic in horizon: 3.1% of one-year holds, 6.0% of three-year holds and 3.9% of five-year holds lost money, with the three-year archetype carrying the widest downside.
- The edge is largest in affordable suburbs (sub-$550,000), at +12.1 percentage points over a year.
- In dollars, before costs and against a price-band-matched benchmark: a $450,000 purchase in the sub-$550k band was worth about $52,000 more after five years. Financed at 80% LVR and held, the same growth difference compounds to +$331,181 of equity at fifteen years — 38.7% more equity on identical cash — and +48.1% at twenty.
- The edge is not a market-timing tilt: beta to the universe is 1.01 with an intercept of +10.65 percentage points, and the excess is uncorrelated with market conditions (r = 0.016, p = 0.96).
- The benchmark reproduces independently from HtAG’s locality price panel at a correlation of 0.992, with a mean absolute error of 0.38 percentage points and zero bias.
- Treating the 14 anchor years rather than the 3,769 picks as the unit of observation, the mean annual excess is +10.73 percentage points (t = 12.50, p = 1.3 x 10-8).
- The Strategy endpoint is now live in the HtAG Developer Portal and queryable by AI agents like Claude, Perplexity and Manus AI.
Frequently Asked Questions
Does the HtAG Dex strategy actually beat the market?
Yes. In a 14-year backtest across every Australian house suburb, HtAG Dex top-decile picks grew 17.6% over the following year versus 9.1% for the same universe over the same window, and beat their price-band benchmark 87.1% of the time. The outperformance held across 1, 3 and 5-year holds and in all 14 individual years from 2012 to 2025.
How much more money would Dex picks have made?
Against a price-band-matched benchmark, a $450,000 purchase in the sub-$550k band was worth about $52,000 more in the asset after five years. Because almost nobody buys with cash, the more useful figure is equity: at 80% LVR on identical cash of $114,750, the Dex pick produced $331,181 more equity after fifteen years, or 38.7% more, rising to 48.1% at twenty years. Separately, chaining the strategy across all fourteen anchor years returned 9.61x against 2.53x for the market — that is the return to re-selecting annually, not to holding one property. All figures are growth only and exclude acquisition costs, holding costs and taxes.
Is the backtest free of hindsight bias?
Yes. Every selection signal used only data available on or before the pick date, and forward returns used only actual (not forecast) prices. It is a faithful reconstruction of the Dex selection signal rather than a literal replay of the live endpoint, and the reliable outperformance window is 1 to 5 years.
How do I access HtAG Dex strategy data inside Claude or Perplexity?
The HtAG Strategy endpoint is live in the HtAG Developer Portal and works with any MCP-compatible AI agent, including Claude, Perplexity and Manus AI. Browse the endpoint catalogue at https://developer.htagai.com/ and apply for access at https://links.htag.com.au/widget/form/GFVegAaXzeTUH7QzRl1T to receive an API key and an MCP setup guide.
What is the Dex strategy?
Dex is HtAG Analytics’ suburb-selection engine. It is a multi-signal, horizon-matched and price-band-aware ranking that combines price momentum, yield, the Growth Rate Cycle, affordability, demand and socio-economic quality into a single score, rather than relying on any one metric.
Disclaimer: This article is general information only and does not constitute financial or investment advice. Past performance and backtested results are not a guarantee of future returns. Property investment carries risk. Consider your own circumstances and seek independent professional advice before making any investment decision.
