UK home insurance risk map

Perils — real open data. The subsidence score is BGS 625k bedrock geology classified for shrink–swell susceptibility (London Clay/Thames Group, Gault, Oxford & Kimmeridge Clays, Mercia Mudstone…; pre-Mesozoic mudstones down-weighted), read at each unit postcode and averaged over the district. The weather score blends Met Office grids interpolated to each district: winter mean wind speed (5 km), the annual wind-driven rain index for SW-facing walls (5 km), annual count of ≥10 mm rain days and annual precipitation (HadUK-Grid observations 1991–2020).

Marginals. Annual aggregate loss per policy per peril is compound Bernoulli × LogNormal, with claim frequency and severity driven by the scores.

Flood. Official flood maps sampled at every unit postcode: rivers & sea from the EA's Risk of Flooding from Rivers and Sea (England, 13 m, risk bands decoded), NRW FRAW (Wales, its polygons) and SEPA maps (Scotland, polygons at 5 m); surface water from EA RoFSW (13 m, category colours decoded), NRW FRAW surface water and SEPA surface water maps. The shares of postcodes in each band drive the flood claim frequency directly; surface-water claims are modelled shallower and cheaper than river/sea.

Groundwater. EA groundwater flood alert target areas via the postcode search tool data: per district, the share of unit postcodes flagged ‘Possible’. England only (UK groundwater flooding concentrates on the English chalk); a nominal background elsewhere.

Coastal erosion. EA NCERM National 2024 frontages: the strip of each district projected to be lost to the sea by 2105, under the adopted Shoreline Management Plan and under no-further-intervention. Not part of the premium — standard household policies exclude gradual erosion — so it is shown as blight and valuation exposure only. England only.

Theft. Three years of police.uk street-level burglary, located point by point into each district. The police category includes commercial premises, so the rate divides by households plus VOA non-domestic premises — each district carries only the residential share of its break-ins, instead of shop-dense cores wearing retail burglaries as household risk — with a 99.9th-percentile cap kept as a tiny-denominator backstop. Scotland is not on police.uk and takes recorded housebreaking from Recorded Crime in Scotland instead — published for all 32 council areas and apportioned to districts by household share, averaged over 2023–24 to 2025–26. The rate is scaled to the ABI's national claim count, so the burglary→claim propensity cancels out and never has to be assumed.

Escape of water. A flat base — plumbing and appliance failure has no open spatial predictor — plus a freeze-attributable slice (~15% of claims) that scales with each district's annual air-frost days (Met Office HadUK-Grid 1991–2020). The level is anchored to the ABI's £1.8m-a-day escape-of-water figure (~£657m/yr, ~£4,000 average claim). The spread is deliberately modest — the Highlands carry about 1.5× the frequency of the mildest coasts — because that is all the open data supports.

Vine copula. The four weather-driven perils and erosion are joined by a C-vine with weather at the root. Tree 1: Gumbel(θ_WF up to 3.0) weather–flood, Gumbel(θ_WG up to 2.4) weather–groundwater (aquifer recharge follows prolonged rain), Gumbel(θ_WS up to 2.5) weather–subsidence, Gumbel(θ_WE up to 2.6) weather–erosion (coasts retreat in storms) — all upper-tail dependent, so extreme years hit perils together. Tree 2 (given weather): Gaussian ρ=0.35 flood–erosion (the same surges drive both, making this the strongest of the three), ρ=0.25 flood–groundwater, ρ=0.15 flood–subsidence; tree 3 independent. 20,000 simulated years per district (Marshall–Olkin for the root pair, conditional h-function inversion for the other margins). A tau-matched 5-dim Gaussian copula and an independence run are computed for comparison. Theft, escape of water, fire and accidental damage sit deliberately outside the vine: none of burglary, burst pipes, dwelling fires or dropped televisions co-move with the storm track (the vine's root is storm/rain-driven; cold is the one weather axis it does not carry), so all four are simulated as independent legs. Theft adds essentially no capital; escape of water adds a small charge, because a hard national freeze (winter 2010 roughly doubled the year's burst-pipe bill) is genuinely systemic; fire adds none at all — forty-five years of Home Office counts show no systemic fire years — and neither does accidental damage, whose sternest test on record (the 2020–21 lockdowns, the whole country at home) moved claims by only ~6%. There is deliberately no per-district copula uplift layer: at calibrated frequencies that difference is inside Monte Carlo noise, so it would be drawing noise across the country. The dependence effect is real, but it lives in the portfolio.

Calibration. Each peril's claim frequency is scaled to the ABI's published domestic payouts — storm £244m, flood £312m, subsidence £307m (2025), theft £450m (2018, the last year the ABI published it), escape of water ~£657m (from the ABI's £1.8m-a-day figure), fire ~£434m (a documented triangle: Home Office incident counts × the ABI's average fire payout), accidental damage ~£227m (a triangle of its own: the ABI's claim count × the market's at-home AD share × Aviva's average payout) over ~15.5m policies — and severities take the published average claim as their mean. Together the eight insured perils come to ~£164/policy/yr, around 75% of what all home claims cost. Coastal erosion is not in that figure and is not calibrated to ABI at all — it has no published payout, because it is not covered. These are eight-peril premiums, not home insurance premiums: a real policy also carries expenses, commission and profit. What is not fitted is the geography and the dependence.

Rating groups. Technical premium = expected loss + 6% cost-of-capital on the district's Euler-allocated share of portfolio tail risk — its expected loss given the whole book is in its worst 1% of years — so cross-district diversification is credited rather than charging every home for its own private worst year. TVaR (expected shortfall) rather than VaR: with rare compound losses, tail dependence concentrates claims into fewer, worse years, so VaR can move the wrong way while TVaR stays monotone in dependence. Rating groups are deciles of THIS map's own units, so a group 7 sector and a group 7 district are not the same population — compare the premiums, not the group numbers.

Reading this map. This is the same technical premium as the model map, re-expressed: each district is shown as a multiple of the UK median premium (£162/yr, household-weighted), so 2.00× means homes there carry twice the median modelled risk. Nothing new is modelled on this page — it exists because “2.4× the UK median” answers “how exposed is my area?” more directly than a £ figure does.

Hazard data: BGS Geology 625k, Met Office HadUK-Grid / Climate Data Portal (incl. air-frost days), EA / NRW / SEPA flood maps, EA flood-risk postcode data, police.uk street-level burglary (all OGL). Loss frequencies/severities and copula parameters are illustrative assumptions, not calibrated to claims. Boundaries: uk-postcode-polygons. See DATA_SOURCES.md in the repo for every endpoint, licence and access date.

The data behind this map: every district and column as a CSV download, or look up a single district with the postcode search.