A UK home-insurance risk model that scores every postcode district for subsidence, weather damage, flood, groundwater flooding, theft, escape of water, fire and accidental damage — the weather-driven four joined by a vine copula, because the expensive years are the ones where several perils strike at once; the other four priced independently, because burglary does not arrive on the storm track, burst pipes answer to cold snaps, and kitchen fires and dropped televisions pay no attention to the sky at all. Coastal erosion is modelled alongside them but kept out of the premium, because no standard policy covers it.
Separately, 352 districts sit on a coast the sea is projected to take land from by 2105 — England's, and three Welsh border districts its estuary frontages reach. For 80 of them the adopted defence plans hold that at zero — the clearest thing in the model that sea walls do. Worst case is HU19, losing 28% of its area even with its plan in place. Scotland adds 179 districts and 80 km² by 2100, on a different and clearly-marked basis; the Welsh coast itself has no published projection at all. None of this is in the premium: no standard policy covers coastal erosion.
And re-running the whole model on the Environment Agency's climate-change flood extents — calibration held fixed, so only the hazard map moves — shifts the technical premium by +4% across the 2,087 districts it covers, worst DN32 at +82%. A scenario, not a forecast, and England only: what it does and does not mean.
Type the outward part of a postcode — the bit before the space — to see how the model rates it.
What actually happened, what the model makes of it, the same model relative to the median, what a warming climate does to it — and the methodology page as the full audit trail behind all four.
What UK home insurance actually paid out, year by year, from the ABI’s own releases — revisions included — beside the model’s account of what separates a quiet year from a catastrophic one, and the 35 years of real weather that test it.
See the record →Eleven switchable layers — rating group, technical premium, each peril score, 1-in-200 loss, surface-water depth, coastal erosion and the climate repricing. Click any district for its full risk breakdown, right down to the bedrock formation beneath it.
Open the map →The same technical premium, re-expressed: every district as a multiple of the UK median, so “2.4× the national median” answers “how exposed is my area?” without needing a £ figure interpreted first.
Compare your area →Frost days are down a fifth since the 1961–1990 normal and summer soil deficits are up. What 66 years of measured weather did to the two perils temperature actually reaches — one of them re-mapped on the evidence, the other deliberately left alone, with the reasoning for both.
See what changed →Every peril score, the vine copula structure, why the risk measure is TVaR and not VaR, all twenty-eight open datasets with their endpoints and quirks — and the commands to rebuild the whole thing.
Read the methodology →Findings that fall out of the simulation.
Re-running the whole model at postcode-sector resolution — 10,398 units instead of 2,736 — reproduces the district answer when aggregated back, but shows what the district average was hiding: in 52 districts (2%) the cheapest and dearest sectors differ by more than 2×, and in DN14 by 4.0× (£162–£657). The splits are physical: a river through half a district, a clay boundary under the other. See the sector map →
Storms, floods and shrink–swell droughts share a climate. Modelling that properly makes it 99 times more likely that the same home is hit by two or more perils in one year — 0.12% of years against 0.0012% under independence. But both are rare, so a single home's own 1-in-100 loss barely moves: measured at ±0 within simulation noise. The dependence surfaces where it should, in the aggregate — catastrophic years cost 4.6% more than an independence assumption implies.
Weather is the workhorse loss in every year, but it is the secondary perils that turn a typical year (£70 per policy) into a bad one (£137). Going from typical to bad, flood cost multiplies 2.4× and subsidence 2.0×, while weather multiplies just 1.3×.
With rare compound losses, tail dependence concentrates claims into fewer, worse years — pushing losses beyond the 99.5% quantile. Value-at-Risk can therefore fall as dependence rises. Premiums use expected shortfall (TVaR 99%), which is monotone in dependence and doesn't hide the clustering.
Perturbing the copula — dependence strength ±25%, conditional correlations zeroed or doubled — moves at most 0.7% of districts into a different rating group. Perturbing the flood assumptions moves up to 24%. Flood's depth-damage curve alone, which has no published anchor, moves 12% of districts when depth is ignored altogether. The vine is what makes the tail honest, but if you want the ranking right, spend your effort on claim frequencies and severities. Every scenario is tabulated on the year analysis page.
Storm, flood and subsidence are calibrated separately to the ABI's 2025 payouts — £244m, £312m and £307m across ~15.5m policies — theft to the floor of its documented vintage envelope (£342m), escape of water to the ABI's £1.8m-a-day figure (~£657m a year), fire to a documented triangle of Home Office incident counts and the ABI's average fire payout (~£434m a year), and accidental damage to a triangle of its own — the ABI's claim count, the market's at-home AD share and Aviva's average payout (~£227m a year) — rather than to a single blended number. Together the eight insured perils come to £164 per policy per year, 75% of what all home claims cost. The rest of a real premium is expenses — which this does not model — plus commission and profit. How the calibration works →
BGS geology, Met Office climatologies, Environment Agency, Natural Resources Wales and SEPA flood maps, EA groundwater alert areas — all Open Government Licence, all fetched by scripts in the repo. Reconstructing 35 years of those drivers picks out 1990, 2015 and 2020 as the worst storm years and 1995, 2018 and 2022 as the driest summers — the Burns' Day storm, Desmond, Ciara and Dennis, and the known subsidence-surge years. See the backtest →
Top of the premium ranking, and what is actually driving each one. No two are alike.
| District | Where | Premium | Subs. | Weather | Flood | Gr.water | What drives it |
|---|---|---|---|---|---|---|---|
| DN7 | postcode area DN | £452 | 0.17 | 0.21 | 1.00 | 0.00 | Flood-dominated (94% of E[loss]) |
| DN14 | postcode area DN | £411 | 0.35 | 0.18 | 1.00 | 0.00 | Flood-dominated (91% of E[loss]) |
| WC2N | postcode area WC | £394 | 0.56 | 0.15 | 0.63 | 0.00 | Subsidence + flood (53% / 30%) |
| W1K | postcode area W | £392 | 0.53 | 0.14 | 0.78 | 0.00 | Subsidence + flood (48% / 36%) |
| W1J | postcode area W | £388 | 0.53 | 0.15 | 0.61 | 0.00 | Subsidence + flood (54% / 27%) |
| W1C | postcode area W | £386 | 0.55 | 0.15 | 0.28 | 0.00 | Subsidence + weather (61% / 19%) |
Technical premium per policy per year = expected loss + 6% cost of capital on the district's allocated share of portfolio tail risk. Scores are 0–1. Download the full table (CSV) · how these numbers are calibrated
Twenty-eight datasets from a dozen public bodies, fetched by the scripts in the repository and documented endpoint-by-endpoint — including the quirks that cost hours to discover, and the dead ends so nobody repeats them.