How Accurate Is HouseCanary’s AVM? Testing Methodology and Results

Overview

Every automated valuation model (AVM) vendor says its values are accurate. Few say how they check. This page is how HouseCanary checks: what we test the AVM against, how often, who tests it besides us, what the latest results show, and where the model is weakest. If you've ever wondered why your home's HouseCanary value differs from another website's estimate or from the list price, the answer is further down too.

Key Takeaways

• As of August 18, 2026, HouseCanary's AVM had a 2.5% median absolute error on the purchase benchmark and 6.8% on the harder prelist benchmark, across six months of sales.
• HouseCanary tests its AVM weekly against real arm's-length sales the model couldn't see, using a trailing six-month nationwide sample, and joins quarterly third-party tests run by AVMetrics and Mercury Network.
• In the same August 18 prelist test, covering 1.2 million home sales from March through mid-August 2026, HouseCanary's values showed almost no lean high or low, and 63.5% came within 10% of the sale price.
• Every HouseCanary value comes with its own confidence measure. Homes the model is most confident about came in with a median absolute error of about 3%, and the least confident ones above 15%.

How accurate is HouseCanary's AVM?

As of August 18, 2026, HouseCanary's AVM had a median absolute error of 2.5% on the purchase benchmark and 6.8% on the prelist benchmark, across the previous six months of arm's-length sales. Independent third-party tests generally agree with these internal results, and often come in slightly better.

Median absolute percentage error (MdAPE) is the standard way to score an AVM. Take every home in the test, measure how far the AVM's value was from the price the home actually sold for, as a percentage, and find the middle of that list. A 2.5% median absolute error means half of the HouseCanary values in the test landed within 2.5% of the sale price.

The two numbers measure two different tests, and the gap between them is the point. The next sections explain both. If you're new to AVMs, start with what an automated valuation model is; this page covers how the HouseCanary AVM is tested.

How does HouseCanary measure AVM accuracy?

AVM testing measures how close the AVM's value came to an actual arm's-length sale price that was unknown to the model when the value was generated. An arm's-length sale is an ordinary market sale between unrelated parties, as opposed to a sale between family members or a foreclosure transfer.

Three rules make the test honest:

• The model can't see the answer. Each sale is matched to the HouseCanary value that existed before that sale was known.
• It runs constantly. HouseCanary generates fresh test results weekly.
• The sample is big and national. Each test uses a trailing six months of arm's-length sales nationwide. A test published June 1, for example, covers every sale from December 1 through May 31. Six months gives enough sales to measure areas where few homes sell and areas where sale prices are reported late.

Purchase vs. prelist: what's the difference?

HouseCanary tracks two internal benchmarks. They differ in one thing: when the value was produced relative to the listing.

The purchase benchmark compares the AVM that existed just before an arm's-length sale became known to the eventual sale price. Most of those homes were already on the market at that point, so the model usually knew the list price.

Figure 1. The purchase benchmark. HouseCanary scores the value it had after the home was listed and before the sale was known.

The prelist benchmark compares the AVM that was available before the home was listed to the eventual sale price from that listing. The model has never seen a list price, by construction.

Figure 2. The prelist benchmark. HouseCanary scores the value it had before the home was listed, so the model never saw a list price.

That's why the prelist number is higher. Prelist is the harder test: the model has to value the home before anyone has named a price. A list price is a strong signal, and most purchase-benchmark observations have it. If you're judging an AVM for a decision made before a home is listed, such as a home-equity loan, a portfolio review or a listing appointment, prelist is the number that matches your situation. We covered why in why prelist benchmarks matter more for AVM accuracy.

Who tests HouseCanary's AVM besides HouseCanary?

HouseCanary's AVM is also tested every quarter by two independent firms:

• AVMetrics. Its PTM test is comparable to HouseCanary's prelist benchmark.
• Mercury Network. It runs two tests. The contract test is comparable to HouseCanary's purchase benchmark. The appraisal test scores the AVM against appraised values on refinances.

Their results generally agree with HouseCanary's internal numbers, and often come in slightly better, which validates HouseCanary's internal testing process. Third-party samples differ from HouseCanary's own, so the two sets of results never match exactly.

Third-party tests don't publicly publish rankings for vendors. For specific results, including how HouseCanary compares with other AVMs in the same tests, contact AVMetrics or Mercury Network directly. Fitch Ratings also validates HouseCanary's AVMs, most recently in 2023 (Fitch's November 2023 update). For its earlier review, see nearly half of HouseCanary's AVMs get top marks from Fitch Ratings.

What did HouseCanary's latest accuracy test show?

HouseCanary's headline figures come from its August 18, 2026 test, covering arm's-length sales from March 1 to August 16, 2026: a 2.5% median absolute error on the purchase benchmark across 1,898,015 sales, and 6.8% on prelist across 1,137,075 sales. Weekly runs move a few tenths of a point from there. Here is the full breakdown of that test.

HouseCanary AVM benchmarks: sales March 1 – August 16, 2026 (test of Aug. 18, 2026)

A signed median error near zero (0.1% on purchase, −0.2% on prelist) means the HouseCanary AVM isn't tilted: the value was about as likely to come in above the sale price as below it.

The breakdowns below use the prelist benchmark from the same test, because it's the harder one and the one that matches decisions made before a home is listed.

Results by property type

Homes that look like their neighbors are easier to value. Townhouses and condos share floor plans and sell often, so the HouseCanary AVM is tightest there. Small multifamily and manufactured homes sell less often and vary more, so errors are wider.

Results by price

The HouseCanary AVM is most accurate in the middle of the market, between $300,000 and $800,000, where most homes trade. It is less accurate at the high end, where homes are more unique, and below $200,000.

Results in non-disclosure states

In twelve non-disclosure states, sale prices don't appear in public records, which leads some people to assume no AVM can be accurate there (how non-disclosure states work). The prices still exist: when a listed home closes, its sale price is reported through the MLS. HouseCanary uses those MLS-reported prices to measure its AVM's accuracy in these states, and the prelist results hold up:

The cost shows up in coverage rather than accuracy: HouseCanary had a pre-listing value for 84% of the Texas homes that sold, against 91% nationally. (Montana, Alaska and North Dakota had 2,000–3,000 sales each, at 7.0%, 6.0% and 6.5%. New Mexico and Wyoming had too few sales to report, and Missouri is non-disclosure only in some counties.)

Why is HouseCanary's value different from another website's estimate?

Because two AVMs are two different models. Each one starts from different data, weighs it differently, and updates on a different schedule. Two AVMs can disagree by several percent on the same house and both still hit their accuracy benchmarks, because a benchmark measures the aggregate across millions of homes, not any one home.

Two things are worth checking when they disagree:

• When was each value produced? A value from before a recent sale or price change on the street will lag one produced after.
• How confident is each model? HouseCanary reports its confidence on every value (see below). When it's low, a large gap between two estimates is expected.

Why might an AVM differ from the list price or the last sale?

List price is an ask, not a value. Sellers and agents price above, at or below market for strategy reasons. Because the prelist benchmark is blind to list price, a large gap between the HouseCanary value and a new listing's price is information in itself, not an error.

The last sale is history. A sale from four years ago reflects that market. The AVM moves with the market every time HouseCanary updates it.

An AVM isn't an appraisal. An AVM is a statistical estimate of market value built from data. An appraisal is a licensed appraiser's opinion of value, usually after inspecting the home. Lenders use AVMs where the rules allow it and appraisals where an inspected opinion is required. That's also why Mercury Network scores AVMs against refi-appraisal values: it shows how closely the model tracks what an appraiser concludes.

How do you read the confidence score on a single property?

The median absolute error describes aggregate measures across millions of homes. For the one in front of you, use the confidence measure that comes with every HouseCanary value: the forecast standard deviation (FSD). It is the model's own estimate of how far off it's likely to be on that home. A low FSD means a tight range and a value you can lean on. A high FSD means a wide range and a home to check by hand.

The August 18 prelist test shows the FSD does its job:

Each row is homes whose FSD rounds to that value. As the FSD rises, the error rises with it, in step. More than half the homes tested (57%) had an FSD of 0.10 or lower. What widens it: unique homes, few recent nearby sales, and thin data. In Property Explorer, the range sits next to the value. For a worked example on three homes, see Start a CMA with One Question.

What these numbers don't tell you

• A national median hides local variation. Some counties run well above 6.8% and some well below. Your market's error isn't the national one.
• The purchase figure is list-price informed. That's why it's lower. Use the prelist figure when there's no list price yet.
• Low-priced homes are the weak spot. Prelist median absolute error is 37% for sales between $50,000 and $100,000, and above 100% for sales below $50,000. At these prices, a small difference in dollars is a large difference in percent.
• Coverage isn't 100%. In the August 18 prelist test, HouseCanary had a pre-listing value for 91% of sold homes. The rest had too little data.
• Results move. Weekly runs drift above and below the published figures. That's normal, and it's why each figure here carries its test window.
• An AVM is a starting valuation. It isn't an appraisal, and it doesn't see the new kitchen or the foundation crack.

How we update this page

We review this page every quarter against new third-party results, and every figure on this page carries the date or sales window it was measured on.

FAQ

See the confidence range on a home you know

Pull up a property you know well in Property Explorer and look at the value next to its confidence range. That's the number this page is about. Agent plans start at $190 a year (see agent pricing). Lenders and investors: explore HouseCanary's AVM or talk to our team.