This map shows the average price per m² of houses in each postcode district in England & Wales.
It extends the pioneering work of Anna Powell-Smith from 2017. I've re-run the analysis with data to Jan 1995–Jan 2026, matching records between the Price Paid dataset and floor area from the Energy Performance Certificate dataset. The result is a national dataset of price per m² for 2,276 postcode districts.
| District | £/m² |
|---|
| District | £/m² |
|---|
The underlying datasets are available as CSV files, licensed under the
Open Government Licence
v3.0.
A machine-readable index with row counts and file sizes is available as
data.json.
One row per postcode district × property type, plus an ALL rollup row per
district. 2,276 districts.
| Column | Type | Description |
|---|---|---|
postcode_district |
string | UK postcode district (e.g. SW1A) |
property_type |
string | Property type: ALL (rollup), D detached,
F flat/maisonette, S semi-detached,
T terraced, O other |
num_sales |
integer | Number of residential sales matched to an EPC record |
total_floor_area |
float | Total floor area of matched properties (m²) |
total_price |
float | Total transaction value of matched sales (£) |
price_per_sqm |
integer | Nominal price per m² = total_price / total_floor_area (£) |
adj_price_per_sqm |
integer | CPI-adjusted price per m², base January 2026 (£), using ONS CPI series D7BT |
Sale prices are taken from HM Land Registry's Price Paid dataset (standard residential sales to individuals, Jan 1995–Jan 1995–Jan 2026). Floor area per property is taken from Energy Performance Certificates.
Each sale is matched to its EPC in two stages. First, an exact match on the property's Unique Property Reference Number (UPRN) using the UBDC Price Paid–UPRN lookup table, which covers transactions up to January 2022. For each matched sale, the certificate closest in time is used — preferring the most recent one lodged before the sale date, with a fallback to the earliest post-sale certificate if none exists within a ten-year window. For sales after January 2022 — where no UPRN link is available — a second stage matches on postcode plus a normalised address string (combining sub-address, building name or number, and street name), and the most recent certificate for that address is used.
The price per m² for each postcode district is the total transaction value of all matched sales divided by their total floor area — not a mean of per-property ratios. Districts with fewer than 10 matched sales are excluded. Some districts have residential sales but fall below this threshold — for example parts of the City of London (EC2R, EC3V, W1C) and TR23 (Bryher, Isles of Scilly), each with only 2–5 recorded transactions. Other gaps are districts with no matching residential transactions at all, such as industrial estates.
The year-range filter switches the map from its default all-time view to a chosen year window, showing CPI-adjusted prices for matched sales within that period. Each district appears in a given year only where at least 10 matched sales were recorded; sparsely-traded years or early years before widespread EPC adoption (pre-2008) may be absent for some districts. Prices are adjusted to January 2026 pounds for direct comparison across years.
The map uses a hybrid colour scale. The first seven colour bands use quantile breaks, dividing the 2,276 districts into roughly equal groups — this maximises visual contrast across the bulk of the distribution. The top two bands use fixed thresholds: £5,000/m² and £10,000/m². Districts above £5,000/m² are predominantly inner London; above £10,000/m² is almost exclusively central London (Mayfair, Covent Garden, and the City).
Hyphens in PPD and EPC address fields are now replaced with a space before matching, rather than silently removed. Previously, a hyphenated address such as CROSS-O-THE-HANDS was collapsed to CROSSOTHEHANDS, which no longer matched the same address written with spaces in the other source.
The fix applies to both the Python normaliser and the DuckDB macro used in tier-2 and tier-3 joins. Non-hyphen punctuation (apostrophes, periods) is still stripped without adding a space.
Measured gain on the March 2026 dataset: +2,428 matched records across all property types.
The word "THE" is now stripped from both PPD and EPC addresses before the normalised match key is computed. "THE OLD RECTORY" and "OLD RECTORY" — the same property in different sources — previously produced different keys and went unmatched.
The removal uses a whole-word match (\bTHE\b) so property names that contain "THE" as a substring (e.g. THETFORD, THEYDON) are unaffected. Both sides receive the same transformation, so false positives require two distinct properties to have the same name minus the article at the same postcode — structurally very unlikely.
Measured gain on the March 2026 dataset: +4,995 matched records across all property types.
EPC assessors inconsistently split or join compound words in property names. The following pairs are now collapsed to a single canonical one-word form before matching:
| Two-word form | One-word canonical |
|---|---|
| FARM HOUSE | FARMHOUSE |
| GATE HOUSE | GATEHOUSE |
| SCHOOL HOUSE | SCHOOLHOUSE |
| MILL HOUSE | MILLHOUSE |
| ALMS HOUSE | ALMSHOUSE |
Unmatched detached counts in the March 2026 data show significant splits: SCHOOL HOUSE vs SCHOOLHOUSE (3,947 vs 130), MILL HOUSE vs MILLHOUSE (2,218 vs 54), FARM HOUSE vs FARMHOUSE (3,364 vs 4,048).
Measured gain on the March 2026 dataset: +342 matched records across all property types.
A new fourth matching tier recovers rural named properties that tier 2/3 cannot reach due to a structural mismatch between PPD and EPC address fields.
The gap: where PPD has no street field, the tier-2/3 normalised key equals the property name alone ("ROSE COTTAGE"). The EPC key is ADDRESS1 + ADDRESS2, and ADDRESS2 often contains a village name ("Rose Cottage Testvillage"). These keys never match even when ADDRESS1 exactly equals the PPD property name.
Sampling (March 2026): **35.7% of unmatched named-paon detached records** (210,285 of 589,685) have a null street in PPD. 21.8% of unmatched ROSE COTTAGE PPD transactions have an EPC at the same postcode with ADDRESS1 = "ROSE COTTAGE" exactly — a real match that was unreachable.
Tier 4 approach:
paon contains no digits are considered.(postcode, normalise_addr(paon)).(postcode, normalise_addr(ADDRESS1)) — ADDRESS2 is ignored.positives at postcode granularity.
The tier benefits automatically from the normaliser improvements in #113, #114, and #115 (THE-stripping, hyphen→space, compound words).
Yield to be measured on next full pipeline run.
E20 (Queen Elizabeth Olympic Park / East Village, Stratford) was created by Royal Mail circa 2012 — after our Geolytix boundary snapshot — by carving it out of E15. Because no polygon exists for E20 in our boundary file, any PPD or EPC record with an E20 postcode was previously silently dropped at aggregation.
As an interim fix, all E20 records are now folded into E15 before aggregation. E20 and E15 are geographically contiguous and were administered as a single district before 2012, so this is a reasonable approximation. The workaround will be removed once a proper E20 polygon is available (issue #81).
Affected output: price_per_sqm_postcode_district.csv — E15 figures now include 884 previously lost E20 sales.
| Source | v0.1.0 | v0.2.0 |
|---|---|---|
| HM Land Registry PPD | to Jan 2026 (~22.5M rows) | to March 2026 (~29.3M rows) |
| EPC bulk export | per-LA ZIPs, epc.opendatacommunities.org | single monolithic CSV, get-energy-performance-data.communities.gov.uk (March 2026) |
| UBDC PPD→UPRN lookup | unchanged — covers to Jan 2022 | unchanged |
| ONS CPI | D7BT Jan 1988–Jan 2026 | unchanged |
The EPC source changed to the new MHCLG GOV.UK One Login API which delivers a single 5.7 GB CSV assembled from all local authority feeds. Four CSV parsing issues were encountered and fixed; see research/epc-csv-data-quality.md.
All sale prices are now inflation-adjusted to January 2026 pounds before aggregation, using the ONS CPI All Items monthly series (series ID: D7BT, Jan 1988–Jan 2026, OGL v3.0). The adjustment converts each transaction price to real terms: adjusted_price = price × CPI[Jan-2026] / CPI[sale_month].
This avoids mixing 1995 pounds with 2024 pounds in the same aggregate. Postcode districts where most sales occurred before 2010 were systematically understated in nominal terms; the real-terms figure makes districts comparable regardless of when their housing stock turned over.
The headline column in both output CSVs is now adj_price_per_sqm (real Jan-2026 £/m²). The nominal price_per_sqm is retained as a reference column.
research/cpi-deflator-choice.md| Column | Type | Description |
|---|---|---|
adj_price_per_sqm |
int | Headline. Real Jan-2026 £/m² (CPI-adjusted) |
price_per_sqm |
int | Nominal £/m² at time of sale — retained for reference |
| Tier | Count | Share |
|---|---|---|
| Tier 1 — UPRN exact match | 9,255,768 | 31.6% |
| Tier 2 — address normalisation | 7,321,554 | 25.0% |
| Unmatched | 12,705,953 | 43.4% |
| Total PPD sales | 29,283,275 |
The match rate dropped from 76.9% (v0.1.0) to 56.6%. This is structural: the PPD gained ~6.8M rows (all post-Jan 2022 transactions) that fall outside the UBDC UPRN lookup's coverage window. The absolute number of matched records is broadly unchanged (~16.6M vs ~17.3M). See research/uprn-coverage-in-epc-data.md for a detailed breakdown.
| Metric | Value |
|---|---|
| Districts included | 2,277 |
| Total matched sales | 16,577,322 |
| Median real adj price | £3,058/m² |
| Most expensive district | W1S — £35,462/m² (real Jan-2026) |
| Least expensive district | TS2 — £733/m² (real Jan-2026) |
| Top 5 district rankings | W1S, WC2A, WC2R, W1B, W1K — stable vs v0.1.0 |
| Bottom 5 district rankings | TS2, TS1, BD3, CF43, DN31 — stable vs v0.1.0 |
| Commit / PR | Description |
|---|---|
PR #72 ebe8619 |
CPI inflation adjustment |
| PR #78 | EPC download: migrate to new MHCLG API |
4e09bf1 |
fix: atomic download — no partial file on interrupt |
bc2e1a5 |
fix: strict_mode=false — handle backslash-escaped JSON in EPC CSV |
943efbe |
fix: pin quote/escape — prevent single-quote column misdetection |
6630a73 |
fix: null_padding=true — handle short/stub rows |
fed70dd |
fix: parallel=false — allow null_padding with quoted newlines |
f6ff4ab |
perf: two-pass EPC dedup to stay within 2 GB memory limit |
---
First versioned release of the UK house price per m² analysis.
| Source | Coverage | Licence |
|---|---|---|
| HM Land Registry Price Paid Data | Standard residential sales to individuals, Aug 2007–Jan 2026 | OGL v3.0 |
| DLUHC Energy Performance Certificates | All domestic lodgements, England & Wales | OGL v3.0 |
| UBDC PPD–UPRN lookup | Transactions up to Jan 2022 | Open data, University of Glasgow |
| GeoLytix postcode district boundaries | — | OGL v3.0 |
Transactions are filtered to standard residential sales to private individuals (transaction_category = 'A'). New-build and established property are both included. Leasehold and freehold are both included.
Tier 1 — UPRN exact match Each sale is joined to the UBDC Price Paid–UPRN lookup table (doi:10.20394/agu7hprj), which provides a Unique Property Reference Number for transactions up to January 2022. That UPRN is then joined to the EPC register to retrieve the most recent certificate lodged _before or on the sale date_. This is the highest-confidence tier.
Tier 2 — Address normalisation match For sales after January 2022 (where no UPRN link exists) and for any earlier sales that failed tier 1, the sale address is normalised: sub-address, building name/number, and street name are concatenated and lowercased, punctuation stripped. The normalised string is matched against the same normalised form in the EPC register, restricted to the same postcode. The most recent EPC before the sale date is selected.
Tier 3 — Postcode fallback Remaining unmatched sales are assigned the median floor area of all EPCs within the same postcode. This tier is lower confidence and flagged in the data.
_Temporal selection note:_ in all tiers, the EPC selected is the most recent certificate lodged on or before the sale date. This avoids using post-sale improvements to infer the floor area at the time of sale.
Price per m² for each postcode district is computed as:
price_per_sqm = Σ(sale_price) / Σ(floor_area_m²)This is a value-weighted aggregate — not a mean of per-property ratios — so that large properties contribute proportionally to the result.
Districts with fewer than 10 matched sales are excluded from the output.
price_per_sqm_postcode_district.csv — primary output: one row per postcode districtprice_per_sqm_lsoa.csv — LSOA-level aggregation (experimental)postcode_districts.geojson — boundary-joined GeoJSON for the web mapTo reproduce this analysis, check out commit da82e13 (the HEAD at time of release) and follow the pipeline instructions in PLAN.md and CLAUDE.md.
| Commit | Description |
|---|---|
91920ab |
Initial pipeline setup |
31aeb42 |
First full pipeline run — outputs committed |
cee3740 |
Add tier-3 normalisation (address-string matching) |
8929d3b |
Regenerate CSVs incorporating tier-3 matches |
30ef34a |
Confirm UBDC lookup coverage ends at Jan 2022 |
049aa57 |
Add output page with choropleth map |
da82e13 |
Temporal EPC selection in tier-1 join (HEAD at v0.1.0) |
| Metric | Value |
|---|---|
| Total sales | 22,503,694 |
| Matched with floor area | ~77% |
| Districts included | 2,279 |
| Median price per m² | £1,986 |
| Data range | Aug 2007–Jan 2026 |
| Most expensive district | W1S — £24,184/m² |
| Least expensive district | TS2 — £504/m² |
Price Paid data produced by HM Land Registry © Crown copyright 2026, licensed under the Open Government Licence v3.0. EPC data © Crown copyright DLUHC, licensed under the Open Government Licence v3.0. UBDC PPD–UPRN lookup published as open data by the Urban Big Data Centre, University of Glasgow. Contains Ordnance Survey data © Crown copyright and database right 2026. Contains Royal Mail data © Royal Mail copyright and database right 2026. Contains National Statistics data © Crown copyright and database right 2026. Postcode district derived boundaries reproduced under the Open Government Licence v3.0. Postal boundaries © GeoLytix copyright and database right 2012. Map tiles © CARTO, data © OpenStreetMap contributors.