Boulder County Home Prices: The Numbers We Won't Print
Your neighbour says the house behind you got $1.6M. A portal says yours is worth less than you think. The paper says the market is down. Three numbers, three different kinds of wrong, and four questions that tell you which is which.
Every one of those numbers can be accurate and still be the wrong thing to act on.
Over the last year we have declined to publish a handful of figures about this market, each in the piece where it came up. Side by side they are not one objection repeated. They are four different failures, and each has a different fix.
1. It is about one house, and a model made it
The most common bad number anyone brings us, by a wide margin, is an automated estimate of their own home.
An automated valuation is built from public records and nearby sales. So it inherits the county's square footage even when the county is stale, counts a basement as finished whether or not it was ever permitted, and cannot see condition at all. It does not know about the 2018 kitchen, the roof that is one hailstorm from a claim, or the wildland edge behind the fence that decides what an insurer will do. It has never been inside.
That is not a sourcing problem, so a better source does not fix it. The remedy is a comp read with the adjustments shown — how a defensible list price actually gets built — which is why our home value page asks for an address and gives you a person rather than an instant number. The neighbour's sale is the same failure at a smaller sample: one house is an anecdote until someone adjusts it for what differs about yours.
2. It does not exist at the scale you asked about
Sometimes there is no figure, and what you found is a derivation wearing a statistic's clothes.
Lyons is the clean case. The Colorado Association of Realtors publishes no Lyons market report, only Boulder County, because a handful of homes close there in a month. The public figures in circulation have spanned more than $300,000 in a single year, which is not a market moving. It is four methods disagreeing.
So the Lyons guide publishes a method rather than a median: anchor to the county qualitatively, then build value up from the specific parcel. The same absence sits under the US-36 corridor towns, which is why that guide refuses per-node medians instead of inventing them.
The remedy: find out who publishes the underlying data, and at what geography. If nobody publishes it at your scale, every number you find is somebody's derivation, and derivations vary.
3. It exists, it is accurate, and it answers a different question
This one survives every sourcing check and still misleads, because the figure is true about something you did not ask. It shows up three ways.
Wrong product. A single-family average set beside a blended median is two different things measured two different ways, which is why we declined to print a Boulder–Longmont price gap. A master-planned community's "median" can blend a low-$300,000s townhome into the number pricing a house three times that, which is why the Centerra guide prices the in-band slice separately.
Wrong quantity. A median measures the mix of what sold, not price change. Where roughly half the housing stock is new construction, as in Berthoud, the median moves when a builder changes what it releases. For appreciation the instrument is a repeat-sales index, which follows the same homes.
Wrong population. A town median is the middle of everything. In Boulder our $800,000–$1.5M band straddles it; in the corridor towns the whole band sits above it. Two neighbouring towns can post medians moving on entirely different stock, describing two middles that neither buyer is shopping; on Lake Loveland a single number averages away the one variable a buyer is paying for, which is lake position.
The remedy: name your question first, then go find the instrument that answers it. Level, direction, rate-of-change and "what does my price range do" are four questions with four different tools.
4. The sources disagree, and the spread is the finding
When published figures for one place span six figures, the useful information is not their average. A wide spread tells you the market is thin, the definitions are inconsistent, or both, and that is genuinely useful. The average of four disagreeing figures belongs to nobody and describes no house.
The remedy: read the spread, do not resolve it. A number from a market like that has to be built from the parcel up.
How do I tell whether a number I found somewhere else is trustworthy?
Five questions, in order. In practice the first two do most of the work.
- Is it about one house, or about a market?
- On what basis — single-family or blended, median or average, sold or asking?
- Over what window, and how many sales are in it?
- From what source can you open and read it?
- What is not in the number — seller credits, a rate buydown, the condition of the house?
That last one matters more than it sounds in this market. A recorded sale at $1.25M with a $30,000 seller credit and a rate buydown is not a $1.25M sale, and none of that appears in the price.
A figure that survives all five is usually worth acting on. A figure missing any one of them is a rumour with a dollar sign in front of it.
What we do publish
Town-level single-family figures from the Colorado Association of Realtors, sourced and dated, refreshed on the release cycle, plus a quarterly cross-town read that applies these tests to a specific month's numbers. This piece is the method behind that read; the read is the method applied.
The declined list is short on purpose. It is not a position against market data. It is a position about which figures can carry the weight people put on them.
Frequently asked
Why do Zillow and Redfin show different values for the same house?+
Because each is a model, and the models disagree. An automated valuation is built from public records and recent nearby sales, so it inherits whatever the county has on file for your square footage, bed and bath count, and it cannot see condition. It does not know your basement was finished without a permit, that the kitchen was redone in 2018, or that the lot backs a wildland edge that changes what an insurer will do. Two models with different training data and different comp selection produce different answers about the same house, and neither has been inside it.
Do you refuse to publish market data generally?+
No, the opposite. We publish town-level single-family figures from the Colorado Association of Realtors every month, sourced and dated, and those reports are the backbone of the site. What we decline is a short list of specific figures that cannot be made accurate at the scale people ask for them. The distinction is the point: a number you can stand behind and a number you cannot are different objects, and most sites publish both without marking which is which.
Why won't you give a year-over-year appreciation figure for Berthoud?+
Because in Berthoud that figure moves on construction mix rather than value. Roughly half the housing stock there is new build, so the median moves when a builder changes what it releases, not when existing homes become worth more. A median measures the mix of what sold; it is not a price-change index. For appreciation the right instrument is a repeat-sales index such as the FHFA House Price Index, which follows the same homes over time.
The town median is published. Why isn't it the number for my price range?+
Because a median is the middle of everything that sold, and it describes a house at that middle. In Boulder our $800,000 to $1.5 million band straddles the median, so an in-band buyer is shopping at or below the middle of that market. In the corridor towns the whole band sits above the median. The town figure is real and useful for level and direction. It is the wrong instrument for a band, and a band figure is only honest attached to a specific situation, which is why we run it for a price range rather than publish it as a headline.
How do I tell whether a number I found somewhere else is trustworthy?+
Five questions, in order. Is it about one house or about a market? On what basis was it measured, meaning single-family or blended, median or average, sold or asking? Over what window, and how many sales are in it? From what source can you open and read it? And what is not in the number, such as seller credits, a rate buydown, or the condition of the house? A figure that survives all five is usually worth acting on. A figure missing any one of them is a rumour with a dollar sign in front of it.
If you are weighing a move in Boulder, Longmont, Lyons, Loveland or Berthoud and you want to know what a number you have found actually says, send it to us and we will tell you what it measures. The Boulder market report carries the figures we can stand behind, with their basis and their date attached.
Daniel Hsieh is a licensed Colorado real estate broker with True North Boulder, brokered by eXp Realty.
Nothing here is an appraisal or an opinion of value on any specific property.
Sources & data notes
- Town-level single-family data throughout this site: Colorado Association of Realtors Local Market Update (IRES data), published monthly via the Longmont-Boulder Association of Realtors. Sold prices, single-family, with the month stated on each figure. Lyons is not among the towns published.
- Appreciation: a median measures the mix of what sold and is not a price-change index. The repeat-sales instrument referenced is the FHFA House Price Index.
- Each refusal described above was made first in the piece it came from, and each is linked at the point it is described. The individual Lyons figures are deliberately not reproduced: the spread is the datum, and one value in wide circulation is an asking price that has repeatedly been requoted as a sale.
- Our in-band ($800,000–$1.5M) cut is run per situation rather than published as a headline; a band figure is only meaningful attached to a specific price range and property type.
