We show our working.
Read the research.
Six papers explaining how SuburbIQ's numbers are built — what the model looks at, what we threw away, what we'd change. We'd rather you understand our methodology and disagree than trust us blindly.
GPT vs HTAG hedonic: who picks suburbs better?
We ran 500 suburb picks through GPT-4 and compared each against actual 12-month forward growth. LLM-only suburb selection underperforms by ~1pp per year — consistent with prior studies.
Market cohesion: why we use comps over hedonic-only.
Properties on the same street move in the same direction the majority of the time. Suburb-average models smooth over the most important signal. Here's how we built around it.
ML Suburb Forecasts: what we'll publish, what we won't.
Why SuburbIQ doesn't ship a "predicted growth" number — and what we do instead. The short version: directional confidence is honest. Point forecasts are not.
The 7-rate APRA stress test, explained.
APRA requires lenders to assess serviceability at customer rate + 1.5pp. We run seven scenarios. Here's exactly what each one tells you and why APRA's floor isn't enough.
Growth signal research: 20 thresholds, 25 years.
Every metric in HTAG has been tested against historical growth outcomes. Here's how we pick the thresholds — what counts as a green light, what counts as red — and how that maps to our BUY / REVIEW / WALK verdicts.
CMA: matching comparable sales on 30+ features.
A comparable sale isn't a property nearby. It's a property nearby with similar land area, bedrooms, bathrooms, parking, frontage, era and condition. Here's the matching engine.
Disagree with the methodology? We want to hear it.
If you've read a paper and think we got something wrong — or if there's a metric you want tested against historical outcomes — email us. Methodology improvements ship as version notes attached to the relevant pages.
Send feedback →