Insurance Dependency · Infrastructure Dependency Atlas

The backstop grew, then shrank. Nobody voted on either.

Florida's insurer of last resort went from 1.42 million policies to 395,144 and is now cutting rates. Louisiana's increases flattened to 0.1%. Texas held rates unchanged. California went the other way — 696,562 policies and $768bn of exposure, with a 29.1% rise landing this autumn. Both directions were set by carriers and residual-market boards, not by anyone the affected households elect. Retreat and recovery are the same dependency seen twice.

states assessed

backstops receding as carriers return

backstops still absorbing risk

record no policy count at all

EVIDENCE WINDOW OPENS 2025-01-01 CITATIONS 14 STATES 6

● observed · ◌ derived · ▌ interpreted · ⊘ refused — marks appear beside claims; defined under Method.

Primary evidence · observed

One field, six states, six different measurements — each now saying so.

Each state's rate_change_pct with the period and basis it actually measures, both recorded in the corpus as of the 2026-08 refresh. Colorado's is a six-year cumulative rise; Florida's is a single filing, and negative; Louisiana's is a year-to-date average; Texas's is a decision to hold rates. They are still not drawn on a shared axis, because they remain different quantities — but the corpus now says which.

Colorado

57.9%

2018-2023 — Statewide average homeowners premium increase over six years. Cumulative, not annual.
Colorado premiums rose 57.9% statewide from 2018 to 2023, and the state's FAIR Plan only opened to residential applicants in 2025 — capped at $750,000, on actual cash value.

California

29.1%

2026, effective autumn — Approved FAIR Plan homeowners rate increase.
It is raising rates 29.1% this autumn.

Louisiana

0.1%

2026 year to date — Statewide average annual homeowners premium change. Prior years: 4.6% (2025), 6.6% (2024), double digits in 2022-23.
Four years on, premium increases have flattened to 0.1%, nine carriers have filed decreases, and Louisiana Citizens is down to about 114,000 policies.

Texas

+0%

2026 — TWIA rates held unchanged for 2026. Not a measure of market-wide premiums.
Rates were held unchanged for 2026. TWIA reported a $41.6M surplus in Q1 2026, reversing a year-end deficit, with about $4.3B of total funding available for the 2026 hurricane season.

Florida

-2.6%

2026 filing — Recommended statewide average change to Citizens personal-lines rates. Negative: a decrease.
Citizens went from a 1.42M-policy peak in October 2023 to 395,144 at the start of 2026 — a 72% fall — and is now recommending a rate cut.

North Carolina

none

period not recorded — No verifiable statewide figure located in the 2026-08 refresh. The previous entry carried 15% with no period, basis or source; it has been withdrawn rather than republished unverified.
This is the thinnest record in the atlas: no published policy count and no verifiable rate figure were found.

n 6 states · 5 with a recorded figure WINDOW 2025-01-01 → 2026-07-11
DERIVED the anchoring flag: a figure counts as anchored when its stored value appears in one of its own state's rate sentences. Colorado's stored 77 is the prose's 76.6% rounded, and its six-year window is the only period the corpus states anywhere. No shared axis and no severity sort — the values are not one quantity.

Texas's stored figure is 3. The only rate sentence in its record says residential rates were “deemed inadequate by 38%” — a regulator's judgment about a shortfall, not an increase paid by anyone. Whatever the 3 is, the corpus does not say, and the published map paints it in the same scale as Colorado's 77.

▌Interpreted

What the six records describe

The backstop is becoming the market

Underneath the schema argument there is a story the corpus tells plainly: what insurers did, what the state did back, and what it costs the people who live there. Verbatim, in the corpus's own words.

Florida

Residual market receding

Insurer behaviourA dozen-plus carriers went insolvent after Hurricanes Laura, Delta, Ida and Zeta (2020-2022), and Citizens became the largest insurer in the state at its October 2023 peak of 1.42 million policies. Since then private capacity has returned: the depopulation programme moved more than 546,000 policies to private insurers during 2025 alone. State responseDepopulation continued through 2025, cutting Citizens by 73% from its peak to roughly 385,000 by year-end and 395,144 at the start of 2026 — its lowest since the corporation was created in 2002. In December 2025 Citizens recommended a statewide average rate decrease of 2.6%, with three in five policyholders seeing an average reduction of 11.5% (about $359). What it costs householdsCoastal property values remain under pressure and insurance is still a material factor in affordability, but the direction has reversed: premiums at the residual insurer are falling rather than rising, and Citizens is no longer the state's largest insurer.

California

Accelerating insurer withdrawal

Insurer behaviourState Farm stopped writing new policies in 2023 and non-renewed 72,000. Allstate and Farmers retreated. The FAIR Plan covered about 1.5% of California single-family homes in December 2020 and roughly 5% by March 2026. In the highest wildfire-risk ZIP codes it now covers about 41% of residential structures, against 4% in lower-risk areas — a more than nineteen-fold rise in those ZIP codes between 2009 and 2024. State response25 moratorium declarations. SB 824 (2018) prohibits fire-risk non-renewals after a declared disaster. Nine consumer-protection laws took effect in 2026. The 'sustainable insurance strategy' trades rate increases for writing commitments. The FAIR Plan itself was approved for a 29.1% homeowners increase effective autumn 2026. What it costs householdsTotal FAIR Plan exposure reached $768B by June 2026, an 11% rise since September 2025 and 250% since September 2022. Growth is finally slowing — roughly 16,000 residential policies added in Q1 2026, against 35,000-50,000 per quarter over the preceding two years — but the book is still expanding, unlike Florida's or Louisiana's.

Texas

Growing residual market exposure

Insurer behaviourPrivate carriers continue to limit coastal wind exposure, leaving TWIA as the windstorm insurer of last resort along the Texas coast. Policies in force fell 1.9% in Q1 2026 to 286,251 from 291,756 at year-end 2025, while total insured exposure rose to $127.1B from just over $117B. State responseRates were held unchanged for 2026. TWIA reported a $41.6M surplus in Q1 2026, reversing a year-end deficit, with about $4.3B of total funding available for the 2026 hurricane season. A statutory change reduced the funding target from the 1-in-100-year probable maximum loss to 1-in-50 — a lower solvency bar, reached partly by lowering the bar. What it costs householdsCoastal exposure keeps growing even as policy count falls, because insured values rise faster than TWIA sheds policies. Galveston County alone accounts for roughly a third of the $127.1B.

Louisiana

Residual market receding

Insurer behaviourA dozen-plus carriers went insolvent after Hurricanes Laura, Delta, Ida and Zeta, and many others stopped writing in high-risk parishes. Capacity has since returned: around 20 companies have completed full licensing, three more were licensed in the first four months of 2026, and enrolment at Louisiana Citizens has fallen nearly 20% from its 2022 peak to roughly 114,000 policies as of June 2026. State responseThe three-year protection rule was repealed in January 2025, allowing up to 5% annual cancellations per insurer — a deliberate shift toward market access over consumer protection. Rate filings have since reversed direction: four carriers filed increases in the first part of 2026 against 27 in 2025 and 50 in 2024, while nine filed decreases covering more than 100,000 policyholders and about $25M of premium. What it costs householdsLouisiana premiums remain among the highest in the country, but the trajectory has changed. Greater New Orleans home sales rose 10% in the first half of 2026, which local brokers attribute partly to easing insurance costs.

Colorado

Post-disaster underinsurance crisis

Insurer behaviourNot primarily an exit story — Colorado is the underinsurance story. The Marshall Fire destroyed more than 1,000 homes in a single day in December 2021 with about $2B in insured losses. Afterwards 74% of affected homeowners were found underinsured and 36% severely underinsured, carrying less than 75% of replacement cost. In parts of Boulder and El Paso counties and along the foothills, premiums have since climbed 150-300%. State responseA Colorado FAIR Plan was signed into law in 2023 but did not open to residential applications until spring 2025. It caps dwelling coverage at $750,000 and pays on an actual cash value basis rather than replacement cost — so the state's own backstop reproduces the underinsurance problem it was created to answer. A 2026 legislative proposal would raise the cap to $1.5M and add a state reinsurance programme. What it costs householdsColorado has the largest increase in home insurance premiums in the country since 2020. Underinsured households are markedly less likely to rebuild, and the FAIR Plan's actual-cash-value basis carries that risk forward for anyone who ends up on it.

North Carolina

Growing residual market exposure

Insurer behaviourPrivate carriers continue reducing coastal exposure across the nine eligible beach counties — Brunswick, Carteret, Currituck, Dare, Hyde, New Hanover, Onslow, Pamlico and Pender — leaving the NCIUA Beach Plan as the windstorm and hail insurer of last resort. The rate environment hardened through 2024-2026 on storm losses, reinsurance costs and construction inflation. No current NCIUA policy count was located in published sources. State responseRate regulation remains commonplace and Beach Plan claims fall under North Carolina's unfair-claims-settlement rules like any carrier's. Coastal building codes were updated after recent hurricanes. Development pressure on the barrier islands continues largely unchecked. What it costs householdsOuter Banks and coastal households face rising premiums and shrinking options, with complete hurricane protection typically running $2,000-$5,000 a year across the three separate policies required. Flood cover through the NFIP is priced separately and also rising. Compound exposure: wind, flood and sea level.

OBSERVED reproduced verbatim from the corpus, per-state receipts

Two of these accounts carry figures the record cannot check. California's headline — FAIR Plan up 139% in 4 years — is a claim about a trend, and the corpus stores one snapshot with no baseline. Florida's 1.41M policies to 395K is the largest movement on the page, and both of its citations are home pages. Neither is likely to be wrong. Neither can be verified from here.

▌Interpreted

Secondary evidence · interpreted

The classification, mapped. The shading is an authored class, not a measurement.

Fill encodes the corpus's stress_band — an assessment class with no derivation rule. The colours are categorical, not a scale. Hatching marks the states with no FAIR-plan policy counts recorded at all.

n 6 states SUBSTRATE boundaries are reference geography, not evidence (source in Method)
INTERPRETED every fill. Neither shade nor land area encodes a measured quantity. OBSERVED only the hatch — whether FAIR-plan counts exist.

Policy counts

Two numbers, no dates

FAIR plans are the insurer of last resort — the residual market that absorbs what private carriers drop. Their policy counts are the nearest thing here to a measurement. They are not a series: no date is attached to either number, so the interval between them is unknown, and half the states record neither.

State	Plan	Policies	Recorded peak	Reading
Florida	Citizens Property Insurance	395,1442026-01-01	1,420,0002023-10-01	72% below the recorded peak
California	California FAIR Plan	696,5622026-06-30	696,5622026-06-30	at its recorded peak — still rising
Texas	Texas Windstorm Insurance Association (TWIA)	286,2512026-03-31	291,7562025-12-31	2% below the recorded peak
Louisiana	Louisiana Citizens Property Insurance	114,0002026-06-30	not recorded	current count only; the corpus holds no peak to compare against
Colorado	Colorado FAIR Plan (residential applications opened spring 2025)	none recorded	none recorded	no count published
North Carolina	North Carolina Insurance Underwriting Association (Beach Plan)	none recorded	none recorded	no count published

OBSERVED both counts, where recorded DERIVED the “% below peak” reading
DENOM policies in force. No date is attached to either count anywhere in the corpus, so “peak” is undated and the interval between the two numbers is unknown.

State evidence

The six

Each claim as the corpus states it. Every class is an authored judgment.

State	Assessment class	Conf.	Rate	FAIR policies
Florida	Residual market receding	high	-2.6%	395,144
Citizens went from a 1.42M-policy peak in October 2023 to 395,144 at the start of 2026 — a 72% fall — and is now recommending a rate cut. The state transferred risk back to a market that had fled, and so far the market has taken it. Citizens policy counts and rate filings are published primary documents; the October 2023 peak and the 2026 recommended decrease are both from Citizens itself.
Louisiana	Residual market receding	high	0.1%	114,000
A dozen-plus insurers went insolvent and the state repealed consumer protections to attract replacements. Four years on, premium increases have flattened to 0.1%, nine carriers have filed decreases, and Louisiana Citizens is down to about 114,000 policies. Rate-filing counts and the Citizens enrolment figure come from Louisiana Department of Insurance reporting relayed by state press; the 2022 peak value itself is not published in the sources located, so no peak is recorded here.
California	Accelerating insurer withdrawal	high	29.1%	696,562
The FAIR Plan reached 696,562 policies and $768B of exposure by June 2026 — up 250% in under four years — and now covers about one in twenty California single-family homes. It is raising rates 29.1% this autumn. The insurer of last resort is becoming an ordinary one. FAIR Plan policy counts, exposure and the approved rate increase are published; the 1.5%-to-5% household share comes from separate analyses with stated dates.
Colorado	Post-disaster underinsurance crisis	high	57.9%	none
Three in four Marshall Fire households were underinsured and more than a third were severely so. Colorado premiums rose 57.9% statewide from 2018 to 2023, and the state's FAIR Plan only opened to residential applicants in 2025 — capped at $750,000, on actual cash value. The 57.9% (2018-2023) figure and the 74%/36% underinsurance findings are from Colorado State University REDI and state reporting. The previous entry's figures — 87% holding extended coverage, 75% still short, and 76.6% over 2019-2024 — could not be reproduced from any source located in the 2026-08 refresh and have been replaced rather than carried forward unverified.
Texas	Growing residual market exposure	high	0.0%	286,251
TWIA carries $127.1B of coastal exposure on 286,251 policies, a third of it in Galveston County. It swung to a $41.6M surplus before the 2026 season, and the legislature cut the funding target from a 1-in-100-year loss to 1-in-50. TWIA publishes quarterly fact books and board materials; policy counts, exposure, surplus and funding figures are all from those primary documents.
North Carolina	Growing residual market exposure	medium	None%	none
Barrier island development continues while private carriers retreat from the coast, pushing wind risk onto the Beach Plan. This is the thinnest record in the atlas: no published policy count and no verifiable rate figure were found. Weakest evidence in this atlas. NCIUA does not publish policy counts as accessibly as Florida, Texas or California; no statewide rate-change figure could be verified. Recorded as absent rather than estimated.

AuditEverything below is the page checking itself: limits, method, and the source record.

Audit · what the marks mean

● Observed

Read off a cited source. Has a unit and a receipt.

◌ Derived

Computed here. Shows its rule; carries no receipt.

▌ Interpreted

Authored judgment. Publishable — and not evidence.

⊘ Refused

Declined or unavailable. Shown, not omitted.

What this page does not claim. That one state is worse off than another, that the assessment classes rank anything, or that recovery is permanent — Florida's residual market has swung by a million policies in three years and could swing back. It reports what the corpus can and cannot substantiate.

Audit · boundary conditions

What this page does not show

The atlas is six individually researched essays wearing one schema. The schema is the part that does not fit.

Unverifiable is not wrong. Of the 62 figures asserted in prose, 8 appear in a structured field. The rest may be perfectly accurate; the record simply cannot confirm them, and neither can this page.
rate_change_pct has no period and no base. One state's prose gives a window (Colorado, six years). The other five give none, so the field cannot be compared across rows — and this page does not compare it.
Half the states record no FAIR-plan count — Colorado, North Carolina — and two of those carry confidence: high anyway.
Colorado disclaims the atlas's own subject. Its record opens “Not primarily an exit story — Colorado is the underinsurance story,” inside an atlas about insurer exit. It is kept because the corpus keeps it, marked refused rather than dropped.
None of 14 citations point at a home page, not a document — including both of Florida's, the state with the largest figures on the page.

Audit · method

What is computed here, and what is not

Nothing on this page ranks the six states, because the corpus provides no basis on which they could be ranked.

stress_band is interpreted. No derivation rule was ever specified for it — the repository history records the class assignments and no threshold, weight, or ordering. The classes are also not commensurable — “systemic insurer retreat” is a market state, “accelerating withdrawal” a rate-of-change claim, “post-disaster underinsurance crisis” a different mechanism entirely. So the map's colours are categorical.

Two derivations, both stated where they appear. The rate-anchoring flag (rule under the first figure) and the prose-token backing flag (rule under the second). The evidence window (2025-01-01 → 2026-07-11) is min/max of all receipt dates. Everything else is presentational.

Geography is substrate. The corpus's own state outlines are coarse hand sketches; drawing them implied a precision it does not have. Boundaries here are vendored — Natural Earth 1:50m Admin 1, v5.1.2, public domain (vendor/boundaries/PROVENANCE.md) — and carry no claim. The corpus polygons are untouched in the data.

Confidence is the corpus's own field, unchanged. It rates a state's inputs, never the class assignment — which is why Colorado can be confidence: high with no FAIR-plan count recorded.

Audit · sources

Source material

All 14 citations — exactly two per state, for every state, which is itself a sign of a quota rather than a search.

Seven of these twelve point at a home page rather than a document. A citation to propublica.org/ or ncdoi.gov/ cannot be checked and will never break, which makes it invisible to link-checking. No citation in this corpus has a retrieval date or an archive snapshot either.

Supports	Publisher	Published
California	KQED	2026-06-01
California	Stanford Woods Institute for the Environment	2026-06-01
California	Insurance for Good	2026-05-05
Colorado	Colorado State University REDI	2026-01-01
Colorado	Colorado Public Radio	2026-01-15
Colorado	The Colorado Sun	2026-07-11
Florida	Citizens Property Insurance Corporation	2025-12-10
Florida	Florida Realtors	2026-01-15
Louisiana	NOLA.com / The Times-Picayune	2026-07-01
Louisiana	KPLC / Louisiana Department of Insurance	2026-03-24
North Carolina	North Carolina Insurance Underwriting Association	2025-01-01
North Carolina	R Street Institute	2025-06-01
Texas	Texas Windstorm Insurance Association	2026-04-01
Texas	Insurance Business / Reinsurance	2026-05-01

Built from docs/insurance/data/states.geojson by tools/build_atlas_insurance.py. Case prose is reproduced verbatim from the corpus; every derived value states its rule on the page.
Hormuz · Cloud & CDN · All atlases