Our fair-value model's track record

Delora's suburb pages show a modelled "fair value" alongside each suburb's actual median
price — the gap between what a suburb's fundamentals say it should cost and what it actually
sells for. Most consumer property portals don't publish how well a model like this has actually
performed against the real market — independent research firms like HtAG Analytics do run and
publish their own backtests, but that kind of evidence rarely appears on a free, suburb-level
public page. We wanted ours to.

This page reruns our model exactly as it would have been at five separate points in the
past — 2013, 2015, 2017, 2019 and 2021 — using only the income, crime and socio-economic data
that was genuinely available at each point, then checks its calls against what suburbs
actually sold for through to 2024. The result is a real, sometimes-wrong track record across
more than a decade, not a highlight reel from one lucky year.

The headline numbers

Every single test, independently, points the same direction: a real, moderate edge over a coin
flip, not a guarantee for any individual suburb. In every one of the five tests, the bigger the
modelled gap, the bigger the subsequent difference in growth tended to be too — never once did
the relationship run backwards. It's also not a strong, precise predictor — the weakest test
(2019, with only 51.2% of undervalued calls beating the median) is barely better than chance,
a genuine miss worth taking seriously, not glossing over.

What "hit rate" actually means here

For each historical cutoff, we split the model's confident calls into two groups: suburbs it
said were undervalued and suburbs it said were trading at a premium. If the model's
signal is real, undervalued suburbs should tend to grow faster than the metro-wide median in
the years after, and premium suburbs should tend to grow slower — prices catching up to (or
correcting toward) fundamentals. A hit rate of 57% means just over half of those calls
turned out right — better than the 50% you'd expect from random chance, but a genuine miss on
the rest. We're publishing every test, not just the ones that look best.

2013 cutoff: the longest track record we have

Every suburb below was confidently flagged in 2013; the actual figure is each suburb's real
median-house-price growth rate through to 2024 — 11 years later.

Undervalued calls that turned out right:

Undervalued calls that missed:

Premium calls that turned out right:

Premium calls that missed:

Notably, several of the same suburbs miss in more than one test — Kensington and Point Cook
show up as undervalued misses at multiple cutoffs, and Toorak and Box Hill both outgrew the
metro median despite being confidently flagged as trading at a premium. That's not
cherry-picked; it's the same suburbs genuinely defying the model's expectation repeatedly.

2021 cutoff: the most recent, shortest test

A tougher, noisier 3-year window (2021 to 2024) — and the signal still holds up.

Undervalued calls that turned out right:

Undervalued calls that missed:

Premium calls that turned out right:

Premium calls that missed:

Toorak and Box Hill appear on opposite sides of "right" and "wrong" across the two cutoffs
shown here — flagged premium and correctly underperforming in 2021, but the same premium call
missed back in 2013. The model's signal isn't fixed to any one suburb; it moves with each
period's actual fundamentals, and neither suburb is a permanent "always right" or "always
wrong" case.

One further honest flag: a Melton rural-fringe suburb (not shown above) was confidently
flagged as trading at an extreme premium — over 300% above modelled value — in three separate
cutoffs, and then went on to decline in every one. With only a handful of house sales a year in
a small, mostly-rural suburb, a single unusual sale can distort an annual median badly. We're
naming this pattern rather than quietly filtering it out, because it's a genuine limit of the
model at very low sales volumes, not a hidden success.

How this was actually built

We reran the model as it would have looked at five separate historical points — using each
period's real income data (from the Australian Taxation Office's own postcode-level statistics,
used consistently across every test rather than mixing income measures), real recorded-crime
rate, and the relevant historical socio-economic index release — rather than today's figures.
Every other input — distance to the CBD, coast and schools, transport access, flood and
heritage overlay coverage, café density — was held at its current value, because these change
slowly at suburb level and reconstructing each one at five separate historical points was
outside the scope of this exercise. That's a real limitation, not a hidden one: it means each
historical run is a genuine test of the income/crime signal specifically, layered on a snapshot
of everything else, not a perfect time machine. As an independent sanity check, each cutoff's
own metro-wide growth figure lines up with real, separately reported Melbourne market
history — the 2017 test shows the slowest subsequent growth of the longer windows, consistent
with the well-documented Melbourne downturn that began that year, and the 2021 test shows the
slowest growth of all, consistent with the 2022-2023 rate-hike correction. Full methodology,
every data source, and the exact code: see how Delora sources and verifies its content.

What this does and doesn't tell you

A hit rate in the 50s to low 60s isn't nothing — it means the model's mispricing signal has
real, measurable value beyond chance, confirmed independently across five separate historical
periods spanning more than a decade. It also isn't a forecast for any specific suburb, a
promise that an "undervalued" tag today will pay off, or a substitute for your own research.
Close to half of undervalued calls still underperformed in most of these tests. Treat the
current fair-value figure on any suburb page the way this page treats its own past calls: as
modelled context with a stated confidence range, not a prediction.

Frequently asked questions

Does a negative relationship between the size of the gap and subsequent growth mean the
model is reliable?
It means the direction is right — more undervalued suburbs did tend to grow faster than more
overpriced ones, on average, across hundreds of suburbs and five independent historical tests.
It's a moderate signal, not a strong one, and it says nothing about any single suburb's future.

Why these five test dates specifically?
2013 is the earliest year Delora's sales-price data covers; 2021 is the most recent cutoff with
enough subsequent data (3 years) to be a meaningful test. The three years in between are evenly
spaced across that window using income data genuinely available for each point, from the
Australian Taxation Office's own historical postcode statistics.

Will this page get updated as more time passes?
Yes. As more years of sale prices accumulate, we intend to add further cutoffs and update the
results — this is a track record, and a real one keeps a record over time.

See also: why online property estimates never agree with each other
· what actually drives Melbourne house prices ·
Melbourne's most undervalued suburbs