Forecast Track Record
We hold our forecasts to a hard standard: a model only ships if it beats a momentum-persistence baseline out-of-sample. This page shows the measured accuracy — nothing is hidden. Lead with the track record, not the claim.
12-month price forecast
| Level | Samples | Avg error (MAE) | Direction right | 80% interval covers | vs baseline |
|---|---|---|---|---|---|
| County | 25552 | 17.1 pts | 72% | 74% | Beats ✓ |
| Metro | 18265 | 11.3 pts | 74% | 71% | Beats ✓ |
| ZIP | 21999 | 16.7 pts | 72% | 71% | Beats ✓ |
36-month price forecast
| Level | Samples | Avg error (MAE) | Direction right | 80% interval covers | vs baseline |
|---|---|---|---|---|---|
| County | 26061 | 30.4 pts | 85% | 59% | Beats ✓ |
| Metro | 18277 | 23.3 pts | 91% | 51% | Beats ✓ |
| ZIP | 22583 | 29.9 pts | 86% | 59% | Beats ✓ |
How to read this
- Avg error (MAE) — the typical gap between forecast and realized price change, in percentage points. Lower is better.
- Direction right — how often the forecast got the up/down direction correct. A coin flip is 50%.
- 80% interval covers — how often the realized value landed inside the stated 80% range. Closer to 80% means better-calibrated ranges; we report the real number rather than assume it.
- vs baseline — whether the model beat momentum-persistence (“the market keeps doing what it’s been doing”), the bar every forecast must clear to ship.
These figures are a development backtest on the latest data vintage (some hindsight remains because sources revise history). The live track record — each forecast recorded today and scored when it comes due — is the true, leakage-free measure and builds over time. Read the methodology →