Finding the houses in Clapham and Wandsworth most likely to sell next
You win most of the valuations you attend. This proof of concept is about being invited to more of them for houses: using public data to show where, across your patch, house owners are most likely to sell in the coming months, so marketing and negotiator time go where they count.
About 1,000 houses change hands here every year
House sales by year and type
Standard market sales of houses in SW4, SW11, SW12 and SW18
Show as a table
HM Land Registry Price Paid Data, category A sales. The latest year is partial because sales are registered weeks or months after completion.
Median house price
Terraced houses, and semi-detached and detached houses together, by year of completion
Show as a table
HM Land Registry Price Paid Data. Medians of recorded prices, not adjusted for inflation or the mix of houses sold.
Houses in the four districts
HM Land Registry Price Paid Data. District is the first half of the postcode; the patch boundary is assumed and still to be confirmed with Noble.
Mortgage rates since 2010
Bank Rate and quoted fixed rates at 75% loan to value
Show as a table
Bank of England statistical database: IUMBEDR, IUMBV34, IUMBV42. Month-end values.
Years since the house last sold
Show as a table
HM Land Registry Price Paid Data. Earlier sales are found by exact address match within Price Paid Data back to 1995, so holds longer than about 30 years cannot be seen, and some repeat sales are missed where an address was recorded differently.
Houses here are likeliest to sell four to ten years after they were bought
The moving cycle
Share of houses that sold in the following 4 to 16 months, by years since they were last bought
Show as a table
Signals in the public record
How much likelier than an average house in the patch to sell in the following 4 to 16 months
Show as a table
HM Land Registry Price Paid Data and the EPC register, every quarter in the period, scored using only what was known at the time. A single signal on its own; the model combines them.
Where the next house sales are likely to come from
Show as a table
Model scores from the latest scoring run. Street names: OS Open Names, matched to hexagons by location only. EPC register (domestic certificates, open fields) placed with OS Open UPRN coordinates. A home counts as a house when its latest certificate describes a house or bungalow. Borough outlines: ONS. Each hexagon is about 0.1 km².
Houses put on the market and houses sold, quarter by quarter
Two counts for the patch on the same timeline. Certificates are lodged when a house is put on the market; completions follow some months later.
Show as a table
EPC register and HM Land Registry Price Paid Data. Completions stop where registration is still incomplete. The working tool will link certificates to sales home by home to measure the typical lead time; this chart does not claim one.
Score every house, mail the top of the list, then check who sold
The real test replays history. The model scores every house as it stood on a past date, using only what was known then. We then compare its list with the houses that actually sold over the following months.
Measured on your patch
Show as a table
The chances it gives are honest
Houses split into ten equal groups by predicted chance of selling in the following 4 to 16 months, against the share that actually sold
Show as a table
It holds up when we try to break it
Lift in the top tenth of houses in your patch, under each check
Show as a table
Four replay years (2016, 2019, 2022, 2024) per check. Spatial holdout: whole areas withheld from training and scored blind.
How a replay works, on invented data. The measured results above come from ranking individual houses in your patch; this simplified picture groups them into hexagons to show the idea.
Mailing houses at random reaches sellers in proportion to the share mailed: 10% of houses, about 10% of sellers. The gap between the two is the model's lift, and it is what lowers the cost of each valuation.
What lift is worth in valuations booked
Put in your own campaign figures. The lift is the number the backtest measures on houses in your patch; propensity models of this kind typically find two to three times the base rate in the top tenth of houses.
Every score comes with its reasons, in plain words
The model explains every score with the factors that push it up, worded from a short, reviewed list. Negotiators can see why a street is on the list, and the reasons suggest what to say.
Why the top-scoring houses score highly
Each house carries up to three reasons. Wording that includes a house's own figure (years owned, rooms, gain) is grouped under its rule. Only reasons shared by at least 25 houses are shown.
A fair test: target some streets, hold others back
History shows the model finds sellers. Whether contacting them wins Noble more instructions can only be shown live. The tool sets up a controlled test so the answer is not a matter of opinion.
Pick
Take the highest-scoring streets with at least 25 houses, and sort them into five bands by score.
Split by chance
In each band, a random half is held back: no targeted outreach for the length of the test. The split is recorded and cannot be changed.
Log
Every mailing goes into an outreach log, and every valuation request into a results log, with its source and whether it became an appointment and an instruction.
Compare
Valuation requests per 1,000 houses in targeted and held-back streets, quarter by quarter, with a confidence range.
An honest caveat: valuation requests are rare, so at one agency's volumes a test takes about a year to show even a large effect. We add faster measures alongside: a separate phone number or QR code for each group's mailings, and Noble's share of sales by street, which every sale in the public record eventually reveals.
From public records to a ranked list of areas
Gather
Every recorded sale since 1995, every energy certificate since 2008, local price indices and mortgage rates.
Link
Match sales and certificates to each home, so its history (how long it has been held, what has changed) is visible.
Learn
Train on houses across Greater London, where there are over a million house sales to learn from, then test on your patch.
Test
Replay past years: score houses as they stood, and check how many of the later sellers were near the top.
Plan
Roll house scores up to areas, rank them, and measure results with a controlled test once outreach begins.
What this page is, and what it is not
Status
A proof of concept prepared by The Clarion Portfolio Ltd for discussion with Noble Estates. Noble has not commissioned this work. It covers houses (terraced, semi-detached and detached); flats are left out. Market figures, the moving cycle, the signals, the map of likely sales, the backtest and the reasons are real, computed using only what was known at each date. The animated replay uses invented data and is marked as such; the calculator's cost and response inputs are assumptions.
No personal data
Nothing here identifies a home or a person. Figures are for the whole patch, for postcode districts, or for hexagons of about 0.1 km² that each contain at least 25 houses; smaller hexagons are left out. No names, addresses, postcodes or property references appear.
Limits
Past sales do not guarantee future ones. Scores are chances, not certainties: most high-scoring houses will not sell in any one year. Energy certificates cover most but not all houses, so houses that have never been certified, which are often the longest-held, are undercounted. The patch boundary is an assumption. Results from the historical test will vary year to year, and the patch is small enough that area-level figures are noisy.
This page is not legal, financial or investment advice.
Sources and attribution
Contains HM Land Registry data © Crown copyright and database right 2021. This data is licensed under the Open Government Licence v3.0.
Energy Performance of Buildings Register, Ministry of Housing, Communities and Local Government, open fields only, licensed under the Open Government Licence v3.0.
Contains OS data © Crown copyright and database right 2026 (OS Open UPRN).
Source: Office for National Statistics, licensed under the Open Government Licence v3.0 (borough boundaries). Contains OS data © Crown copyright and database right 2026.
Source: Bank of England, licensed under the Open Government Licence v3.0.