How accurate is ARVCast?
Most home-value tools ask you to trust a number. We’d rather show our work: ARVCast is backtested against real, closed sale prices in every county we cover, and the results are published here — including the counties where we’re weakest.
Latest backtest results by county
| County (TN) | Sales scored | Median error | Within ±10% | Within ±20% | As of |
|---|---|---|---|---|---|
| Robertson | 377 | 5.2% | 72.4% | 88.9% | 2026-07-23 |
| Cheatham | 366 | 6.7% | 61.7% | 81.7% | 2026-07-23 |
| Dickson | 382 | 9.1% | 53.4% | 76.4% | 2026-07-23 |
| Fayette | 385 | 9.1% | 53.8% | 75.3% | 2026-07-14 |
| Tipton | 379 | 9.6% | 52.2% | 76.3% | 2026-07-14 |
| Shelby | 391 | 15.8% | 37.9% | 55.2% | 2026-07-08 |
| Haywood | 106 | 19.4% | 31.1% | 51.9% | 2026-07-14 |
- Robertson: Best accuracy in the fleet — dense, homogeneous commuter subdivisions.
- Cheatham: Nashville-commuter towns along the Cumberland — compact, accurate pool.
- Dickson: County-seat market on the I-40 corridor west of Nashville.
- Fayette: Towns spanning distinct price bands; more accurate than the Shelby baseline.
- Tipton: Dense suburban pool — the best mean-error profile of any covered county.
- Shelby: Sampled from the most recent arms-length sales of a ~119k-sale market.
- Haywood: Thin rural market: many addresses return an honest "insufficient comps" instead of a low-confidence number.
Methodology
For each county we hold out recent arms-length sales — genuine market transactions between unrelated parties, priced above $10,000 — that the model never saw. ARVCast then estimates each property blind, and we compare the estimate to the price the home actually sold for.
The headline metric is the median absolute percent error (MdAPE): half of all estimates land closer than that figure, half land further. We use the median because it’s honest — a handful of unusual properties can’t drag it around the way they can an average. We also report the share of estimates landing within ±10% and ±20% of the true sale price.
Two honesty rules. First, thin markets: where a neighborhood lacks enough comparable sales, ARVCast returns “insufficient comps” rather than a low-confidence guess — you’ll see that reflected in the notes above. Second, what we estimate: ARVCast’s number is the after-repair value (the home in updated condition), so backtests naturally score best on homes sold in market-ready shape, and estimates for heavily distressed sales will read high — by design.
For context, Zillow publishes a nationwide median error of about 7% for off-market homes. In several of our counties ARVCast’s published error is at or below that; in others (notably Shelby’s very diverse housing stock) it’s higher. The difference: we show you the comparable sales behind every single estimate, so you never have to take the number on faith.
Check it yourself with a free lookup on the homepage, or see everywhere we cover.