Data story
How accurate are our house value estimates?
We test our own house value estimate against real sale prices every month, and publish the result, including where it is weak.
July 2026, 600 sales tested
12.2%
Typical error (median)
- Within 10%
- 41.7%
- Inside our range
- 75.8%
Our monthly backtest, HM Land Registry sales
Indicative only. This is not a valuation (and not a RICS valuation). It is built from public data and the figures you give us; we do not guarantee accuracy or completeness. Check with a qualified professional before relying on it. How accurate is this?
Key finding
The short answer
In July 2026, our estimate missed the price paid by a typical 12.2% across 600 tested sales. 41.7% landed within 10% of the price, 70.2% within 20%, and the real price fell inside our range 75.8% of the time. June 2026 was slightly better.
Data last updated 28 October 2026.
Headline results from the July 2026 test, with June 2026 underneath each figure.
Typical error
12.2%
Median, July 2026. June 2026: 11.8%
Within 10%
41.7%
June 2026: 43.3%
Within 20%
70.2%
June 2026: 73.8%
Inside our range
75.8%
June 2026: 81.2%
The figures
See it in charts and tables
July 2026
How close the estimates landed
41.7% of estimates were within 10% of the price paid and 70.2% within 20%; the price fell inside our range 75.8% of the time.
Our backtest, 600 HM Land Registry sales, July 2026
By property type
Typical error by property type
Flats had the largest typical error in July 2026 (15.7%) and semi-detached the smallest (10.6%).
Our backtest, July 2026; lower is better
June 2026 against July 2026
| Measure | June 2026 | July 2026 | Change |
|---|---|---|---|
| Sales tested | 600 | 600 | same size |
| Typical error (median) | 11.8% | 12.2% | +0.4 pts |
| Within 5% of price | 21.5% | 21.0% | -0.5 pts |
| Within 10% of price | 43.3% | 41.7% | -1.6 pts |
| Within 20% of price | 73.8% | 70.2% | -3.6 pts |
| Price inside our range | 81.2% | 75.8% | -5.4 pts |
| Median bias | -1.1% | -1.3% | -0.2 pts |
Each month is a different random sample of 600 sales, so small moves can be sampling noise.
By property type (sales tested in brackets)
| Type | Typical error June 2026 | Typical error July 2026 | Within 10% June 2026 | Within 10% July 2026 |
|---|---|---|---|---|
| Detached | 13.2% (144) | 14.1% (163) | 36.8% | 39.9% |
| Semi-detached | 12.2% (186) | 10.6% (171) | 43.5% | 45.0% |
| Terraced | 9.7% (182) | 11.5% (185) | 51.6% | 44.3% |
| Flats | 13.2% (88) | 15.7% (81) | 36.4% | 32.1% |
What it shows
What the test found
- The typical error was 12.2% in July 2026 and 11.8% in June 2026, a move of +0.4 points. Half of the estimates were closer than that to the price paid.
- About 4 in 10 estimates landed within 10% of the price paid (41.7%), and 29.8% were more than 20% out. That is a wide spread, which is why we show a range and not one number.
- The price paid fell inside our range 75.8% of the time in July 2026, down from 81.2% in June 2026. With 600 sales each month, a share like this carries a margin of roughly plus or minus 3.5 points, so part of that drop may be chance. We are watching it, not explaining it away.
- Bias was small and negative: the median estimate sat 1.3% below the price paid in July 2026 (1.1% in June 2026). On average we lean slightly low.
- Flats were the weakest type in July 2026 (15.7% typical error); semi-detached the strongest (10.6%). The order changed between months, so do not read a fixed ranking into it.
Method and sources
How we test, and how to repeat it
How it is worked out
- We draw a random sample (fixed seed, so it can be repeated) of 600 standard-price HM Land Registry sales completed in the month, leaving out the "other" property type. All 600 received an estimate; none was dropped for being hard.
- No lookahead: each home is valued as at the 1st of the month using only sales before that date and House Price Index months at least two months older. The sale being tested is never an input to its own estimate.
- The model is the postcode-and-property-type estimate, the one the free house value check uses. We compare it with the price actually paid.
- Typical error is the median absolute percentage error. "Within 10%" is the share of estimates whose gap to the price paid was 10% or less. "Inside our range" is the share of prices that fell between the low and high figures we show. Bias is the median signed error.
- To repeat it with our code, run
node --import tsx scripts/backtest-offline.mjs --month=2026-07 --sample=600 --model=current-postcodeagainst a local copy of the Price Paid file. The files behind this page are data/backtests/backtest-2026-06.json and backtest-2026-07.json.
What it does not show
- Two months, 600 sales each. Different samples, so month-to-month changes of a few points can be noise.
- England and Wales only (Land Registry covers nothing else), standard sales only, and the postcode-and-type estimate only. Estimates that use floor area or an owner's own previous sale are not tested here.
- The price paid is the price agreed, not a survey. A sale can close above or below what a surveyor would say the home is worth.
- It does not compare us with agents, surveyors or other websites. We have not tested anyone else, so we make no claim that we are better or worse.
- Condition, size, garden, noise and layout are invisible to this estimate. A past test does not guarantee the next estimate.
- Our backtest files: data/backtests/backtest-2026-06.json and backtest-2026-07.json (generated 7 October 2026 and 28 October 2026).
- Sale prices: HM Land Registry Price Paid Data, Open Government Licence v3.0.
Use this data
How to cite this page
You are welcome to quote the figures. Please link to this page and keep the source line with them.
- Attribution text
- Source: Check Labs Ltd (Your Property Check), "How accurate are our house value estimates?", 28 October 2026. Tested against HM Land Registry Price Paid Data, licensed under the Open Government Licence v3.0.
- Suggested citation
- Check Labs Ltd (2026). How accurate are our house value estimates? Your Property Check. Available at: https://yourpropertycheck.co.uk/data/house-value-accuracy (data updated 28 October 2026).
- Page address
- https://yourpropertycheck.co.uk/data/house-value-accuracy
- Licence and update
- Open Government Licence v3.0 (terms). Data last updated 28 October 2026.
Our take
What we make of it
Our estimate is a useful first look, not a price to list at.
If you are selling, use it to check that an agent's figure is in a believable range, then ask the agent for the sales they used. With 29.8% of estimates more than 20% out, a gap of a few percent between us and an agent proves nothing.
How we value a homeIf you are buying, treat the figure as a guide to what similar homes sold for, not what this home is worth.
Look at sold prices on the same street and at the condition before you offer.
Run the free check to see the range for your postcode.
Check House ValueOur opinion, drawn from the figures on this page: not advice and not a forecast. We earn from OfferCheck reports and from introducing sellers to cash buyers, so weigh it with that in mind (see our disclosure). We are not authorised or regulated by the FCA.
Questions, answered
Common questions
What does "typical error" mean here?
It is the median absolute percentage error: for each tested sale we take the gap between our estimate and the price paid, as a percentage of the price paid, then take the middle value. In July 2026 it was 12.2%. Half the estimates were closer than that and half were further away.
Is a 12.2% typical error good enough?
It is good enough to show a sensible range, not to replace a valuation. Only 41.7% of estimates landed within 10% of the price paid, so treat the figure as a starting point and compare it with recent sales on the same street. An agent or a RICS surveyor can see things a postcode-and-type estimate cannot, such as condition, size and layout.
Which homes does the estimate get wrong most often?
In July 2026 flats homes had the largest typical error (15.7%) and semi-detached the smallest (10.6%). The ranking moves from month to month because each month tests a different random sample, so read it as a pattern to watch, not a fixed rule.
Does this test cover the estimate I get for my own address?
Not exactly. The test uses the postcode and property type only, which is what the free house value check needs. An estimate that also uses details such as floor area or your own previous sale is not covered by these figures.
Can I repeat the test?
The inputs are public and the method is written out on this page, so yes: take the Price Paid file for a month, draw a random sample of standard sales, value each one using only earlier sales and the House Price Index, and compare with the price paid. We publish the summary figures, not each of the sampled sales.
Keep going
Related pages
Free tool
Check your house valueMethod
How we value a homeContains HM Land Registry Price Paid Data, licensed under the Open Government Licence v3.0.
Sale prices from HM Land Registry Price Paid Data.