Market cohesion:
why we use comps over hedonic-only.
Most valuation models smooth across a suburb. But the data is clear: properties on the same street move in the same direction the vast majority of the time, while suburb-wide averages mask outsized variance. Here's how SuburbIQ uses comp-led valuation as the primary signal, with hedonic as a sanity check.
What "cohesion" measures.
Cohesion is how often two properties at a given geographic level move in the same direction over a 12-month window. If two random houses anywhere in Australia gain or lose price in the same year, we'd expect cohesion of around 50% — a coin flip. The closer the geography, the higher the agreement. The agreement at the street level is roughly six times tighter than at the suburb level.
The implication.
If two adjacent streets in the same suburb routinely diverge, then a suburb-wide hedonic estimate has limited resolution. It's accurate on average — which means it's wrong, in opposite directions, on the streets at either end. For a buyer making one decision on one parcel, that's exactly where you don't want to be averaging.
How SuburbIQ uses this.
Tier 2 and Tier 3 reports start with three structurally-adjusted comparable sales within roughly 1 kilometre of the subject parcel. We match on land area, bedrooms, bathrooms, parking, era, condition, and frontage — 30+ features in total. The comp-led estimate is the primary fair-value read. The HTAG hedonic estimate is the sanity check: if the comps and hedonic diverge by more than 15%, we flag the report and recommend caution.
When this approach fails.
Comp-led valuation degrades in thin markets — small regional towns, prestige suburbs with few comparable sales, brand-new estates without sales history. When the comp pool drops below three and within 1 kilometre, we fall back to HTAG's parcel-level IA-CMA endpoint and clearly label the valuation as model-based rather than comp-led. We never silently substitute one for the other.
What this means for buyers.
If your suburb has thick comp data, trust the comp-led estimate. If your suburb is thin, treat the model-based estimate as a wider range — the confidence intervals are real. The report tells you which case you're in.