How we work — sources, method, and what the figures mean

At a glance

Source and how the retrofit gap is computed

The index is built entirely on the EPC Open Data register — the official GOV.UK open data record of Energy Performance Certificates for England and Wales. It is the origin of every certificate figure shown on this site.

The Retrofit Gap for an area is the percentage of its domestic dwellings whose current EPC band is below band C — that is, band D, E, F, or G. It is the count of those below-C dwellings divided by the count of dwellings with a valid band (A to G), expressed as a percentage.

Data vintage and methodology version

Data vintage: 2026-Q2

Methodology version: 0.3.0

How the cost estimate is derived

The cost estimate uses a documented per-band-gap cost-to-band-C model. For each below-C dwelling we take the published cost of raising it from its current band up to band C only, chosen by the size of the band gap (D→C, E→C, F→C, or G→C), and we sum those costs across an area. It is a modelled estimate of the investment the band C standard requires, not a quotation for any individual home.

The cost estimate reaches band C only — it does not aim for a dwelling's full potential rating. The EPC recommendations dataset and the potential-rating improvement measures it lists are out of scope for the MVP: those measures target the potential rating and would overstate the cost to reach band C, so they are not used in the headline figure.

The per-band cost-to-band-C figures are taken from the English Housing Survey: Energy Report 2019–20 (Ministry of Housing, Communities & Local Government, Table 3.2), which models the average cost to bring a dwelling up to at least band C by its current band: D→C £6,472, E→C £13,285, and F or G→C £18,858. These are the survey's cost-to-band-C estimates — deliberately not its higher “all recommended measures” figures, which target a dwelling's full potential rating.

Two things to note about these figures. The survey reports bands F and G combined, so both use the same £18,858 estimate. The figures are 2019–20 prices and are derived from English stock; we apply them to Welsh dwellings as an approximation, as there is no equally current band-by-band Welsh equivalent.

These band-average figures are now the fallback. Where a home has the data for it, we replace this single per-band constant with a per-home cost-to-band-C estimate built from that home's own EPC recommendations — described next. The English Housing Survey band averages are still used wherever a home lacks the recommendation data the per-home model needs.

The per-home cost to reach band C

Most homes that fall short of band C carry their own list of EPC recommendations — the specific measures a surveyor suggested, such as loft insulation, a new boiler, or solid-wall insulation. Each recommendation comes with a cost, but the certificate does not say how many SAP points (the 1–100 efficiency score behind the band) each individual measure would add. So we cannot read off, measure by measure, what it takes to cross into band C. Our per-home model fills exactly that gap, and it gives a cost that varies from home to home rather than a single band-wide average.

The method has two steps. First, we learn a typical point value for each measure type across the whole EPC register: given which measures each home was recommended and how far its score actually rose from its current to its potential rating, we work out the average SAP contribution that best explains those gains for each measure type (loft insulation, glazing, heating, and so on). We constrain every contribution to be zero or positive — a measure can help or do nothing, but never count against the score. Second, for each individual home we take those average contributions and reconcile them to that home's own current-to-potential gap, so our numbers never contradict what its certificate already states. We then search its recommended measures for the cheapest combination that lifts it to band C (SAP 69) and report that combination's total cost as the home's cost to reach band C.

Crucially, this is honest about homes it cannot help. Some homes' recommended measures do not add up to band C even when all of them are applied — typically hard-to-treat fabric where the listed improvements simply do not close the gap. We flag those homes as not reaching band C on their recommendations rather than inventing a cheaper-looking number or pretending the target is within reach.

The assumptions we are making, stated plainly. SAP contributions are modelled as additive and linear — we add up each measure's typical points — whereas the real RdSAP calculation has interactions (measures can reinforce or partly overlap one another). The per-measure point values are population averages, then reconciled to each home; they are not a bespoke re-assessment of that specific property. And we treat each recommended measure as independently installable, ignoring sequencing or physical dependencies between works. Because of this, a per-home figure is a modelled estimate, not a quotation. Where a home lacks the recommendation data this model needs, we fall back to the published English Housing Survey band-average cost-to-band-C described above.

Sale-price signals (HM Land Registry)

Where available, we fuse each dwelling's EPC floor area with the most recent HM Land Registry Price Paid sale for the same property to derive four area-level sale-price signals: the median price per square metre, a modelled matched-sale exposure estimate for below-band-C homes, the observed C-or-above price per square metre spread, and the transaction-rate ratio for below-C homes versus the rest.

A dwelling is matched to its most recent non-stale HM Land Registry Price Paid sale within the same postcode using a building-number token. On the Land Registry side we use SAON where present, otherwise PAON; on the EPC side we derive the comparable token from the certificate address. The match must also be corroborated by a shared building or street-name token. If a postcode has more than one candidate for the same number, the dwelling is left unmatched rather than guessed.

Sale prices are uprated to the data vintage using a national annual index derived from Price Paid transactions. Price per square metre is the uprated sale price divided by the EPC total floor area. The match is partial by design: numbered houses match best; flats, named buildings, missing floor area, stale sales, and ambiguous postcode-number combinations reduce coverage. Every area reports match coverage and areas below the coverage floor omit these signals.

Current fusion metadata: Price Paid vintage 2026-04; national match coverage 44.1%; per-area coverage floor 5.0%; stale-sale window 10 years.

When citing the index, separate EPC-derived claims from matched-sale claims. EPC gap and cost are based on retained EPC dwellings. Matched-sale exposure is based only on the matched subset and should be cited with coverage.

The Home Index, 2026-Q2. National EPC gap: cite the share of rated dwellings below C. Matched-sale exposure estimate: cite the value alongside national or local match coverage.

The current artifact records the matcher as: EPC dwelling matched to its most recent non-stale HM Land Registry Price Paid sale within the postcode on building number (Land Registry SAON, else PAON; the EPC address otherwise), corroborated by a shared building/street name token; sale prices are uprated to the data vintage by a national annual index derived from Price Paid, and £/sqm = uprated price ÷ EPC total floor area. Postcodes with more than one candidate number are left unmatched. Coverage is partial and reported per Area.

Contains HM Land Registry Price Paid Data © Crown copyright and database right, licensed under the Open Government Licence v3.0.

The green premium — what it is, and why the simple version misleads

The green premium is how much more an energy-efficient home (EPC band C or better) sells for than a comparable inefficient one. The naive way to measure it — just comparing the average price per m² of C-or-above homes against below-C homes — is confounded: efficient homes also tend to be newer, larger, and in pricier areas, so that raw gap mixes the energy effect with everything else. To isolate the part that tracks the rating, we fit a hedonic model that holds size, property type and location constant. The result is smaller and more honest: nationally, homes at band C or above carry roughly a +7% sale-price premium over below-C homes once those factors are controlled.

The model regresses log sale price on the EPC band (relative to a reference of band D) together with floor area, property type and area fixed effects, using matched EPC × HM Land Registry sales. The band coefficients are monotonic: each step up the scale is associated with a higher price — nationally about −9.5% (G), −3.1% (F), −2.6% (E), 0% (D, the reference), +2.7% (C), +11.7% (B), +13.7% (A) — and each estimate carries a 95% confidence interval. We publish the modelled premium per local authority wherever an area has enough matched sales to estimate it reliably, and fall back to the national figure otherwise.

This modelled premium is also what informs value at risk. The discount we apply to a below-C home defaults to the national modelled green premium (about 7%), not the per-local-authority figure: in cheaper, lower- variation markets the local hedonic cannot fully separate the band effect from the “newer-estate” effect, so a few areas carry implausibly large local premiums. We anchor the headline £-at-risk to the national estimate — the one we identify most reliably — while the per-area premium is still published as a labelled local association. The £ figure still varies by property, because it is the discount applied to each home's own modelled value.

The honest caveats. This is an association, not a clean causal effect: even controlled, the band correlates with unobserved quality (better maintenance, glazing, layout), so both this cross-sectional estimate and the alternative repeat-sales approach are best read as upper bounds on the pure energy effect — improving a specific home's rating will not necessarily move its price by the headline figure. The naive per-area spread shown elsewhere on the site is locally unreliable — it swings on a handful of sales and on the local mix of housing — which is exactly why the modelled premium is the figure to cite for the green premium.

The Cold Homes Index (hard to heat × hard to fix)

For scored local authorities we combine four measures. Two come from the EPC certificates themselves: the off-gas-grid share (certificates flagged as not on mains gas, among those with a known flag) and the solid-wall share (wall descriptions containing “solid” but not “cavity”, the hard-to-treat fabric). Two are official local-authority datasets: the DESNZ sub-regional fuel poverty statistics (Low Income Low Energy Efficiency) and the English Indices of Deprivation 2025 (the share of a local authority's LSOAs in the most-deprived national decile). All are joined by ONS local-authority code.

The Cold Homes Index is a transparent, equal-weighted composite of four percentiles, grouped into two axes: hard to heat (off-gas-grid + fuel poverty) and hard to fix (solid-wall + deprivation). The index is the mean of the two axes — equally, the mean of all four percentiles (0–100, higher = worse). Fuel poverty and deprivation are ranked separately inside each nation before their standardised scores can be combined; raw English LILEE and Welsh measures are never compared. It highlights where homes are both expensive to heat and expensive to fix; it is a cross-reference, not a causal claim.

England uses DESNZ's 2026 sub-regional release (2024 LILEE data) and IoD 2025. Wales has WIMD 2025 LAD profiles, but its latest fuel-poverty release provides no LAD estimate and uses a different definition. All 22 Welsh authorities are therefore explicit “missing fuel poverty” rows and are excluded from the composite; the obsolete 2018 LAD model is not substituted. Barnsley and Sheffield use a documented one-to-one code alias for their April 2025 code-only changes. Fabric shares use the latest valid certificate per UPRN, with certificates lacking a UPRN treated as separate dwellings.

Contains public sector information licensed under the Open Government Licence v3.0: DESNZ sub-regional fuel poverty statistics and the Ministry of Housing, Communities & Local Government English Indices of Deprivation 2025 and Welsh Government WIMD 2025 LAD profiles.

Energy bills now and to 2030 (a scenario calculator, not a forecast)

For each home we estimate a modelled annual energy bill — what it costs to run today — and how that bill tracks to 2030 under a small set of clearly-labelled price scenarios. We then roll those per-home figures up to a per-area exposure cut: the median bill now and at 2030 for each local authority and postcode district. This is a scenario calculator, not a price forecast — we do not predict energy prices; we apply illustrative cap scenarios to each home's modelled energy use.

The consumption model. The EPC records each home's regulated energy use in kWh/m²/yr; we multiply that by the certificate's total floor area to get the home's annual delivered energy. “Regulated” is the key scope word: the EPC figure covers heating, hot water and lighting — the loads the assessment governs — and not cooking, appliances or other plug loads. So our bill is deliberately a regulated-load bill, lower than a full household bill, and we say so wherever it appears.

The fuel-split assumption (disclosed). The certificate gives total kWh, not a clean breakdown by fuel, so we model the split and disclose it. We assign a standard electricity baseline for lighting (calibrated by floor area to a typical home's electricity use), then put the remaining energy on the home's main fuel — mains gas, or electricity for an all-electric home — and price each at the relevant Ofgem price-cap unit rate and standing charge. Homes on heating oil or LPG are priced on published unit rates for those fuels. Homes on solid fuel, biomass or community heating are approximated with an electricity-and-gas-equivalent estimate and flagged as lower-confidence, rather than left out or given false precision.

The scenarios. We define three named, illustrative paths from 2026 to 2030, expressed as multipliers on today's cap (standing charges projected on the same paths): Flat — the cap holds at today's level (a real-terms anchor); Central — a modest illustrative rise; and High — an upside-stress path. These are configuration, not predictions: they are transparent multipliers chosen to bracket a plausible range, so a reader can see how exposed a home or area is to bills going up — not a claim about what prices will do.

The honest caveats, in one place. The bill is a modelled estimate over regulated loads only (heating, hot water and lighting), not a quotation and not a total household bill. The fuel split is modelled and disclosed, not read off a meter, and the non-gas/electric fuels carry an explicit lower-confidence flag. And the 2030 figures are illustrative scenarios, not a forecast of energy prices.

Time-to-Target

Time-to-Target answers a simple question: at each area's current rate of improvement, the year its retrofit gap falls to 10% — the point at which roughly nine in ten homes are at EPC band C or above, a practical “effectively retrofitted” end-state. We fit a robust (Theil–Sen) trend — the median of all pairwise year-on-year slopes — over the area's most recent eight years of gap figures, then project that line forward to the crossing year.

It is an at-current-rate projection, not a forecast: it assumes the recent pace simply continues. Areas that are flat or worsening, or whose projection lands more than fifty years out, are reported as “not on track” rather than given a misleading far-future year. The threshold (10%), the look-back window (eight years) and the horizon (fifty years) are fixed and applied identically to every area.

Limitations

• The published statistics cover England and Wales domestic dwellings only. Scotland, Northern Ireland, and non-domestic buildings are not included.
• The cost estimate is a modelled estimate, not a quotation. It is derived from the per-band-gap cost model described above and should not be read as a price for any specific property or works.
• Public statistics are aggregated to Postcode District level or above. No individual certificate or dwelling-level record is published.
• The sale-price signals (price per m², matched-sale exposure, C-or-above price/m² spread, transaction-rate ratio) are based on partial sale matching and are modelled estimates, not valuations or causal claims. Each area reports its match coverage; areas below a minimum coverage omit these figures rather than publish an unreliable one.
• The fuel-poverty and deprivation figures and the cold-homes index are England-only and at local-authority level. They are a cross-reference, not a causal claim, and are absent for Welsh authorities and any authority missing from the source.
• The energy-bill figures (the bill now and the 2030 scenarios) cover regulated loads only — heating, hot water and lighting — so they are lower than a total household bill and exclude cooking, appliances and other plug loads. They are a modelled estimate over a disclosed fuel split, not a quotation, and the 2030 figures are illustrative scenarios, not a price forecast.
• The private-rented-sector (PRS) cut is computed only over dwellings with a recognised tenure value, and is not a census of the private rented sector. Each area also reports its tenure coverage — the share of its dwellings with a recognised tenure — so the PRS cut is not mistaken for a complete count.

Constituency boundary vintage

Constituency figures use the 2024 Westminster; comparisons are only valid within one boundary vintage.

Source attribution

Energy Performance of Buildings Data: England and Wales, get-energy-performance-data.communities.gov.uk, GOV.UK

The EPC Open Data is published by the current EPC service at get-energy-performance-data.communities.gov.uk.