What happened

UK home insurance, year by year — first what was actually paid out, from the ABI’s own published figures with their revisions shown, then what the model makes of good and bad years, and whether the drivers it watches match the years Britain remembers. No per-district claims data is published anywhere, by any insurer, at any price — so the observed record here is national, and the maps on the other tabs are model output calibrated to it.

What the ABI actually paid

Weather-related payouts to UK homes, as the Association of British Insurers announced them — the observed numbers this model is calibrated against. Hollow markers are figures as first published where the ABI later revised them: claims keep settling after a year closes, so the first announcement and the eventual cost differ — for 2023 by a quarter.

Weather claims paid to homes (£m) All home claims paid (£m, context) as first published, since revised

How many homes claimed

Homes making claims per year by weather peril, where the ABI has published a count. Burst pipes are the freeze component of escape of water — announced in the winter releases rather than the annual ones, which is why their years differ.

Storm Flood Subsidence Burst pipes

The subsidence record, quarter by quarter

The ABI’s subsidence announcements, assembled from its press releases — one recovered from the Internet Archive after the original page 404ed. Gaps are gaps in what the ABI chose to announce, not quiet quarters. The 2018 surge — from 2,500 notified claims to more than 10,000 in a single quarter — is the sharpest jump since the series began, and 2022’s record-heat summer is the modern echo.

The model’s view: good years vs bad years

Everything below is simulated — the model’s answer to what separates a quiet year from an expensive one. UK home portfolio · subsidence × weather × flood (incl. surface water) × groundwater · C-vine copula · simulated years · per-policy averages, these four weather-driven perils only — coastal erosion is in the model but not here, being chronic rather than episodic, so it has no good or bad years; theft, escape of water, fire and accidental damage are priced but sit outside the vine — theft's ~0.1% systemic loading means its bad years barely exist (a 1-in-100 theft year is ~1.3× average), fire's is smaller still (no systemic fire years in forty-five years of Home Office counts), accidental damage's is the smallest of all (the 2020–21 lockdowns, the biggest stay-at-home shock on record, moved AD claims by only ~6%), and while escape of water's freeze loading is real (winter 2010 doubled the year's burst-pipe bill), its bad years are cold snaps, not storm seasons, so all four would only blur the weather story told here

How much worse do bad years get?

Average annual cost per policy by return period. The gap between the two lines is the portfolio cost of tail dependence. It is much smaller than the same gap for a single home (see the map popups): across 2,736 districts, diversification between places largely washes out dependence between perils. Hover for values.

C-vine (modelled dependence) Perils independent

What drives the cost, year by year

Years are ranked by portfolio cost and bucketed: good (best 50%), typical (50–90th percentile), bad (90–99th), catastrophic (worst 1%). Bars show the average per-policy cost split by peril.

Weather Subsidence Flood Groundwater

How widespread are claims in each kind of year?

Share of districts with at least one claim per peril. The spread between good and catastrophic years is deliberately modest: the spatial loading is calibrated so a 1-in-100 year runs at about twice the claim frequency of an average year, which is already generous against the record. UK weather is strongly correlated across the country, but UK claims are not — a storm crosses every county while only a small, weakly correlated fraction of homes end up claiming.

Weather Subsidence Flood Groundwater

The numbers

How many claims, and how big?

The same simulated buckets, split the other way: how many claims each kind of year produces, and what each claim costs. Counts are claims, not claimants — a home claiming on two perils counts twice, which is the basis the ABI’s own published figure uses.

Method: 20,000 simulated portfolio years; perils joined per district by a C-vine copula (weather-root; Gumbel weather–flood, weather–groundwater and weather–subsidence; Gaussian flood–groundwater and flood–subsidence given weather). Each district's year combines the national systemic factor per peril — which carries the vine dependence — with district-level noise (spatial loadings: weather 50%, flood 40%, subsidence 60%, groundwater 70%; storm, drought and aquifer-recharge years are large-scale in the UK, flood more localised), so good years still see scattered claims while bad years cluster nationally. Hazard data: BGS, Met Office, EA/NRW/SEPA (OGL) — see DATA_SOURCES.md. Loss frequencies/severities are illustrative assumptions.