Reading a CMA without fooling yourself
A comparative market analysis is a small argument built from a handful of sales. Here is how ours is built, which numbers to trust, and the three places a CMA quietly goes wrong.
Every comparative market analysis is the same argument: homes like this one, near this one, sold recently for about this much, so this one is worth about that. The argument is only as good as the word "like", the word "near", and the word "recently". This note is about how Estaio's CMA handles each, so you can read one with the right amount of trust.
Which comps get in
The engine pulls a pool of candidates from the listings index, sold and active, in the subject's ZIP first and then the city, and asks for the subject's property type. A condo is compared with condos; a house with houses. If there are too few of the same type to work with, other types are admitted and the report says so.
Every candidate is scored from 0 to 100 for similarity. Location carries the most weight and decays with distance, then size, then bedrooms, bathrooms, age and type. A closed sale gets a bonus over an active listing. Recency counts too, and it is measured properly: for a sold home it is months since the closing, not how many days the listing took to sell.
The six strongest make the report, with closed sales pulled in ahead of higher-scoring actives when they exist.
Which comps carry the number
Not every comp in the report counts toward the value. Three kinds are shown for context and excluded from the statistics, and each is labelled:
- Sales older than a year. The market a year ago is a different market.
- Outliers. A comp whose price per square foot sits far from the others, measured with a method that a single extreme value cannot drag.
- Reference-only actives. When three or more recent closings exist, the valuation rests on closings alone, and active listings are shown only for context.
When closings are scarce, active list prices are blended in, discounted by a few percent and given less weight, and the report says the valuation is blended. With no closings at all it says so too. This is why confidence tops out at "medium" for a blended valuation and "low" for an active-only one. High confidence needs closed sales under the number.
How each comp is adjusted
Each comp is first normalized to the subject's size: its price per square foot times the subject's square footage. Then three small deltas are applied against that figure: four percent per bedroom of difference, two and a half percent per bathroom, and for houses a lot-size adjustment that grows with the difference and is capped. The report prints the deltas next to each comp so nothing is hidden.
What is not adjusted: condition, view, finishes, renovations. Those need an inspection, and an estimate that pretends otherwise is guessing. The caveats say this every time.
The number and the range
The midpoint is a weighted median of the adjusted values, where stronger, more recent, closed comps weigh more. A median, not an average, because one unusual sale should not move the answer. The range around it comes from how spread out the comps are: tight comps give a narrow range, never narrower than five percent; scattered comps give a wide one, never wider than fifteen; and fewer than three comps always widen it to at least ten.
The prose in the report is written by a model from those computed facts, and every figure in the prose is checked against them before you see it. Prose that cites a number the engine did not compute is discarded and replaced with plain sentences built from the facts alone.
Three places a CMA goes wrong
Anchoring on the list price. The subject's asking price is the thing under examination; it must not leak into the answer. Our adjustments are sized against each comp's own value, never the subject's list price, and the model is told not to reason from it.
Counting yesterday's market. Stale sales flatter a falling market and undersell a rising one. Excluding them is the single biggest improvement a CMA can make.
Mistaking a list price for a sale. An active listing is a hope. Treating it as a data point at full weight lifts every valuation in a slow market. Discount it, weigh it less, and say you did.
Read the confidence factors before the number. They tell you which of these the report had to compromise on, and how much to lean on it.
Field Notes by email
The next note, when it is written.
Something in this note worth arguing with? Write to us