ML Suburb Forecasts:
what we'll publish, what we won't.
Other platforms publish "12-month predicted growth" numbers for thousands of suburbs. We don't, and we're not going to. Here's why directional confidence is honest, why point forecasts aren't, and how the SuburbIQ Cycle Score (RCS) gives you the same signal without the false precision.
The problem with point forecasts.
A "predicted growth: 7.2% over 12 months" number reads as a single confident prediction. The reality behind that number is a model with a typical mean absolute error of 4-6 percentage points. The prediction's confidence interval — if it were published, which it usually isn't — runs from roughly 1% to 13%. Saying "7.2%" out loud lets users anchor on it. Showing the interval would tell users what they actually need to know: this could be a strong year, or a flat one.
What we do instead.
The RCS (Recovery Cycle Score) gives you a 0-100 read on whether the suburb is in early-cycle recovery, mid-cycle expansion, late-cycle peak, or correction. The DDI (Demand Depth Index) tells you whether the affordability is structurally supported or hollow. The STGS (Short-Term Growth Signal) ranks 1-year forward signals against research-validated thresholds. The BBT (Bull-Bear Tide) detects directional regime — vendor patience vs. seller flood.
Together, these tell you whether a suburb is set up for growth. They don't tell you exactly how much. We think that's the right level of resolution.
Where we'd publish a forecast.
One: directional only. Will the suburb median be higher or lower in 12 months? At what confidence level? Two: ranged. We can predict that 75% of suburbs in the BUY bucket will outperform the national median over 12 months. That's a defensible statement. We're working toward both of these for 2026.
Where we won't.
Suburb-level point forecasts. Property-level point forecasts. Any forecast horizon beyond 24 months. Any forecast where we can't show the confidence interval. If you see a competitor publishing precise 5-year predictions with 0.1% resolution, ask to see their out-of-sample accuracy.