What technically counts as a comp
Three filters:
- Geographic. Within ~0.5 miles in dense urban markets, ~1 mile in suburbs, ~2 miles rural. Some appraisers cross school-district or municipal lines, but only when the local market is genuinely continuous across them — pricing usually isn't.
- Temporal. Within the last 6 months, ideally 3 months. Older comps need a time-of-sale adjustment for market movement (typically 0.3–0.8% per month in a normal market). A 12-month-old comp in an appreciating market understates value if the adjustment is skipped.
- Similar. Same property type (single-family vs condo vs townhouse), same approximate size (within ~25% of square footage), same approximate age, similar lot, similar condition.
A 4-bed / 3-bath / 2,400 ft² single-family home should not be comp'd to a 2-bed / 1-bath / 1,000 ft² condo. The math doesn't recover from that mismatch no matter how many adjustments you stack on top.
What makes a "good" comp
Adjustment magnitude is the tell. Every comp has a sale price plus a set of line-item adjustments (square footage, beds, baths, lot, condition, age, garage, view, time-of-sale, special features — the full list is in comp adjustment factors explained). The total of those absolute-value adjustments is the comparability score:
| Total adjustments | Comp quality |
|---|---|
| < 5% of sale price | Excellent — direct comparable |
| 5–10% | Good — standard comparable |
| 10–15% | Fair — usable but discount weight |
| > 15% | Poor — exclude or treat as last resort |
A comp with $90k of adjustments on a $500k sale is doing 18% of the work for the model. That's a bad comp, regardless of how close it is geographically. The math is being held together by the adjustments, not by the comparability.
Sale type matters
Not every recorded sale is a market sale. The non-arm's-length sales that should be flagged or excluded:
- Foreclosure / REO sale. Distressed; below market.
- Short sale. Distressed; below market.
- Family transfer. "Sold" for $1 to a relative — not a price.
- Cash-investor flip. Often below market on the buy side, then resold above market post-renovation.
- Estate / probate sale. Sometimes below market for liquidation.
- Auction sale. Variable; depends on auction format.
USPAP requires the appraiser to investigate sale type for every comp and exclude or annotate the non-arm's-length transactions. Most modern AVMs filter automatically using sale-type flags from the MLS and the deed record, but the filter is imperfect — read the comp list yourself if the report supports it.
The 3-by-3 grid that appraisers use
Residential appraisals built off the URAR form (Uniform Residential Appraisal Report — Fannie Mae Form 1004) typically use 3 closed comps plus 3 active or pending listings. The closed sales tell you what buyers actually paid; the actives and pendings tell you what the current market looks like. The reconciled value sits inside that band.
If a report only shows you closed sales and no actives, you're seeing yesterday's market — which can be a problem in a market that's moved 5%+ in the last quarter.
Where comps come from
- MLS (Multiple Listing Service). The richest source. Has photos, full descriptions, broker remarks, days on market. Access is agent-licensed in most US markets, which is why Zillow / Redfin can show you closed comps but not always with full broker fields.
- County records / deed registry. Public; thinner data. Has the sale price and date but not photos or condition notes. Backbone of most AVMs because coverage is universal.
- CoreLogic / Black Knight / Quantarium. Aggregated commercial data sets with cleaning and standardization layered on top. Most modern AVMs license at least one of these.
What to verify before trusting a comp
A 60-second sanity check on any comp the model picked:
- Read the MLS description. Is it actually similar?
- Look at the photos. Same condition tier?
- Check the lot. Same school zone, same street type?
- Check the sale type — is it arm's-length?
- Check the date — is the time-of-sale adjustment applied?
Doing this on the top 3 comps catches 80% of the bad-comp problems that AVMs produce on atypical homes.