Insurers refused to renew 1.9 million US homeowners policies in six years. California's last-resort plan now carries $768bn of risk. Who holds it now.
The Retreat, Measured
The shortest contract in housing was the first to reprice the climate
Between 2018 and 2023, American insurers declined to renew the policies of more than 1.9 million homeowners, a count assembled from 23 companies covering roughly two-thirds of the market and 249 million policy records✓ Established [4]. By 2024 the West was running 25.1 non-renewals per 1,000 policies and the Southeast 22.0, against 11.7 in the Northeast✓ Established [2]. This is not a forecast about the 2050s. It is an accounting record of the last eight years.
The homeowners policy is the shortest-dated contract attached to the longest-dated asset most households own. It is repriced or cancelled every twelve months, while the building behind it has a service life measured in decades and a mortgage measured in thirty years. That duration mismatch is why insurance became the first financial market to mark physical climate risk to market. The U.S. Treasury's Federal Insurance Office assembled ZIP-code data from more than 330 insurers covering over 246 million policies between 2018 and 2022, and found non-renewal rates in the highest-risk ZIP codes running about 80% above those in the lowest-risk ones✓ Established [1].
The same dataset priced the gap. Households in the fifth of ZIP codes with the highest expected annual losses from climate-related perils paid an average of $2,321 a year, 82% more than households in the lowest-risk fifth✓ Established [1]. No national ZIP-level collection of this kind existed before January 2025, which is why the debate ran for a decade on anecdote✓ Established [1]. The first thing the data established was that the withdrawal has a geography, and that the geography matches the hazard maps◈ Strong Evidence [56].
The National Association of Insurance Commissioners' regional analysis, released on 31 July 2026, tracked what carriers did rather than what they announced. In the Northeast, non-renewals rose from 6.2 per 1,000 policies in 2022 to 11.7 in 2024, a 147% increase against the 2018 baseline✓ Established [2]. Of 263 insurers writing continuously in that region between 2018 and 2024, 141 — 54% of them — cut their policy counts, by an average of 38.7%✓ Established [2]. Fifty-six carriers entered the market over the period and sixty-one left✓ Established [2].
California is where the arithmetic is most visible. The FAIR Plan — the state's insurer of last resort, created to cover what no admitted carrier will write — held 696,562 policies in June 2026, up 8% in nine months and 157% since September 2022✓ Established [8]. Its total exposure reached $768 billion, a 250% increase over the same four years✓ Established [8]. A pool designed as a residual has become one of the largest single aggregations of property risk in the United States, and it is not underwritten by anyone who chose it.
The same instrument is failing under different names elsewhere. Natural catastrophes caused $107 billion of insured losses across 190 events in 2025✓ Established [18]. Wildfire, storm and flood accounted for 92% of that total, a record share for those three perils✓ Established [20]. Uninsured losses rose more than 7% to $424 billion, and in North America alone the gap between economic and insured loss widened 6% to $140 billion, against an 11% widening to $90 billion across Europe, the Middle East and Africa✓ Established [21] [19].
Treasury's ZIP-code analysis found non-renewal rates roughly 80% higher in the highest-risk fifth of ZIP codes than in the lowest, with premiums 82% higher in the same places✓ Established [1]. The NAIC's regional data reproduces the gradient at coarser resolution: 25.1 non-renewals per 1,000 policies in the West and 22.0 in the Southeast against 11.7 in the Northeast in 2024✓ Established [2]. Withdrawal is tracking modelled hazard, not general cost inflation◈ Strong Evidence [30].
This report asks what happens to a property market when the annual contract underneath it is repriced or withdrawn. The answer runs through four channels that can now be measured separately: the premium, the residual market, the mortgage, and the valuation. Each has produced published evidence since 2024, and the evidence is more specific than either side of the public argument allows. What none of it has produced is a policy response that addresses the exposure rather than the price signal that revealed it◈ Strong Evidence [53].
What the Actuaries Repriced
The models stopped assuming the past was a guide
The catastrophe models in commercial use through the 2010s were built between 2005 and 2015 on hazard histories running from roughly 1960 to 2005, and they assumed event frequency and severity were stationary over time◈ Strong Evidence [41]. That assumption has been abandoned. What followed was not a panic about the future but a revision of the expected loss for next year, and premiums are a twelve-month instrument✓ Established [42].
A premium is a forecast with money behind it. It resolves into three components: the expected loss produced by a catastrophe model, the insurer's expenses, and the cost of the capital held against the tail. Regulators treat the first as a technical input and the third as a market price, and both moved sharply between 2022 and 2026✓ Established [42]. Understanding the retreat requires separating them, because they move on different clocks and reverse for different reasons.
The peril mix changed first. Since 2020, severe convective storm losses — hail, straight-line wind and tornado — have outpaced hurricane losses in the United States, and a single hailstorm can now produce insured damage rivalling the fallout from a Category 4 landfall◈ Strong Evidence [41]. Swiss Re put 2025 severe convective storm losses at $51 billion✓ Established [18]. Wildfire is the fastest-growing peril of any, with insured losses rising an estimated 12% a year◈ Strong Evidence [19].
Both were historically classified as secondary perils and given thin model treatment, because the models were calibrated on the events that dominated the twentieth-century record◈ Strong Evidence [41]. Regulators have since built model-governance frameworks of their own, and residual-market boards now vote explicitly on which vendor models to weight✓ Established [42]. The Texas Windstorm Insurance Association approved revised catastrophe model weights and set a 1-in-50 probable maximum loss of $4.3 billion for the 2026 storm season on that basis✓ Established [45].
Catastrophe models were calibrated on 1960-2005 hazard records and treated event frequency as stationary, while the perils now driving losses — wildfire and severe convective storm — were classed as secondary and modelled thinly◈ Strong Evidence [41]. Regulators and residual-market boards now set model weights by formal vote, as the Texas Windstorm Insurance Association did in fixing a $4.3 billion probable maximum loss for 2026✓ Established [45]. When the model changes, the premium changes the following renewal, whether or not that year's weather cooperates◈ Strong Evidence [42].
The second input is the cost of capital, and it runs on a market cycle rather than a climate one. After the hard market of 2023, the January 2026 renewals delivered what Guy Carpenter called accelerated softening: its global property catastrophe rate-on-line index fell 12%, with the United States and Asia-Pacific also down 12% and Europe down 15%✓ Established [16]. Howden Re measured property catastrophe reinsurance down 14.7% and retrocession down 16.5% at the same renewal✓ Established [17].
This is the part most commentary reads backwards. Softening reinsurance lowers the price of transferring risk; it does not lower the risk. The declines were attributed to excess capital and investor appetite for insurance-linked securities, not to any reassessment of hazard✓ Established [17]. In the same period, Swiss Re's measured protection gap widened rather than narrowed✓ Established [19] [21]. Cheaper reinsurance and a growing uninsured tail are not contradictory findings; they are the same market at two different layers.
A house that cannot be insured cannot be mortgaged. No bank will provide a mortgage for uninsurable property. Credit markets freeze.
— Günther Thallinger, Member of the Board of Management, Allianz SE, March 2025Thallinger's argument is structural rather than rhetorical, and it specifies a sequence: insurance withdraws, mortgage credit follows, and the valuation adjusts last◈ Strong Evidence [43]. It is also testable, which is unusual for warnings of this kind. Each step has since been examined separately in the microdata, using administrative records rather than projections. The results are more qualified than the warning and more alarming than the industry's rebuttal.
The remainder of this report follows that sequence in order — what the evidence shows about transmission, what households are actually paying, what the state absorbed when private capital left, and what governments did in response. The distinction that matters throughout is between measures that change the price of risk and measures that change the risk◈ Strong Evidence [51].
The Evidence in the Microdata
Fragile insurers, foreclosures, and an 11% repricing
The strongest work on insurance retreat does not use projections. It uses regulatory filings, loan-level records and ZIP-code administrative data to trace what happened after carriers left◈ Strong Evidence [24]. The finding is that the transmission from insurance to mortgage to house price is real, measurable, and narrower than the rhetoric on either side◈ Strong Evidence [26].
Parinitha Sastry, Ishita Sen and Ana-Maria Tenekedjieva built the reference study, circulated in December 2023 and awarded the 2025 Marshall Blume Prize. It links the cost of extreme weather to insurer flight, insurer insolvency and mortgage delinquency in a single chain✓ Established [24]. Their central claim is that mispricing of climate risk in both property insurance and mortgages generates large taxpayer exposures and pushes excess credit into high-risk areas◈ Strong Evidence [24].
The mechanism they identify is institutional rather than meteorological. Financially weak, state-regulated insurers took disproportionate share in high-risk states, and severe or persistent mispricing pushed them toward losses and insolvency, prompting exit or retrenchment◈ Strong Evidence [25]. Lenders responded to the resulting coverage gaps by selling the affected mortgages onward to Fannie Mae and Freddie Mac, which transfers the residual risk to the federal balance sheet◈ Strong Evidence [49]. The retreat does not end at the insurer.
Work from New York University's Stern School and the University of British Columbia takes the next link. Insurer-initiated non-renewals are associated with higher foreclosure rates, falling home values, weaker retail spending and declining homeownership in the affected areas◈ Strong Evidence [28]. The effect is not a housing crash; it is a persistent local drag that compounds. What makes it consequential is that the trigger is administrative — a letter from a carrier — rather than a physical event◈ Strong Evidence [28].
Non-renewals initiated by insurers are associated with higher foreclosure rates, lower home values, weaker retail spending and falling homeownership rates in the affected areas◈ Strong Evidence [28]. The originating study links climate losses to insurer fragility and then to mortgage delinquency, and finds that lenders offload the resulting exposure to the government-sponsored enterprises◈ Strong Evidence [24] [49]. The channel from a cancelled policy to a defaulted loan is short and now documented in loan-level data◈ Strong Evidence [25].
The valuation channel has the cleanest number. Research summarised by the National Bureau of Economic Research finds a relative home price decline of about 11% in ZIP codes that are both highly exposed to rising risk and in the top decile of catastrophe exposure◈ Strong Evidence [26]. That is a repricing of the asset to reflect the running cost of insuring it, and it happens without any disaster occurring◈ Strong Evidence [27]. Insurance is the transmission belt; the loss is booked in the deed.
The forward estimate is larger and more contested. First Street's twelfth national risk assessment projects $1.47 trillion of net property value loss by 2055 across 70,026 census tracts — 84% of the country — driven by insurance costs and shifting demand rather than by physical destruction◈ Strong Evidence [22]. It projects average premiums rising 29.4% in real terms over the same period and more than 55 million Americans relocating domestically toward lower-risk areas◈ Strong Evidence [23].
The most consequential figure in the First Street work is a decomposition, not a total. Of the projected 29.4% real premium increase by 2055, 18.4 percentage points correct present underpricing and only 11 points reflect growing hazard◈ Strong Evidence [22]. On that accounting, most of the increase now reaching households is not the future arriving early — it is the past being settled◈ Strong Evidence [23]. The finding is unwelcome to regulators and consumer advocates alike, because it implies the price is not the thing that went wrong.
The household-level evidence is consistent with all of this and undramatic. ICE's mortgage data show average annual property insurance payments rising 6.6%, or $149, in 2025 to an all-time high — though at the slowest pace since 2020✓ Established [48]. More usefully, it measures the transmission directly: for every percentage point of housing expense shifted into insurance, the non-current loan rate rose by roughly 0.14 percentage points across credit-score quintiles◈ Strong Evidence [48].
Taken together, the microdata supports a narrower claim than the headlines and a firmer one than the rebuttals. Insurance retreat does not collapse housing markets. It applies a persistent, geographically concentrated drag on values, credit performance and local spending, and it does so through a contract that can be withdrawn in a single renewal cycle◈ Strong Evidence [26] [28].
The Household Ledger
Who pays, who drops out, and who was never covered
In Australia, the number of households in home-insurance affordability stress — those paying more than four weeks of gross income for cover — rose 30% in a year to 1.61 million, or 15% of all households✓ Established [31]. Average premiums rose 28% to A$1,894, and by about 50% for the highest-risk properties✓ Established [32]. The country has no wildfire-driven residual market and no Proposition 103. It has the same arithmetic.
The Australian data is the most granular affordability series any country publishes, which is why it is worth reading closely. Stressed households spend an average of 9.6 weeks of gross income on home insurance — seven times what unstressed households spend✓ Established [31]. Riverine flood is the single largest driver: for 171,000 households, flood risk alone accounts for more than half of the home insurance premium✓ Established [32]. Affordability stress is not spread thinly; it is concentrated on a hazard.
The American series is coarser but points the same way. Insurify recorded average annual home insurance at $2,948 at the end of 2025 and projects $3,057 by December 2026, a fifth consecutive year of increase◈ Strong Evidence [47]. The state-level moves in 2025 were the striking part: Minnesota up 34% to $3,530, Colorado up 33% to $3,996, Iowa up 28% to $2,802✓ Established [47]. None of those states is coastal, and none of them is where the public conversation has been.
The distributional pattern is counter-intuitive and matters for policy. Brookings' analysis of the Treasury data found the largest dollar increases between 2018 and 2022 in adaptable, higher-wealth ZIP codes — up $204, or 10.66% — against $54 in both vulnerable and low-risk ZIP codes, on a national average of $104, or 6.75%✓ Established [30]. The absolute burden fell on wealthier places; the relative burden fell on poorer ones, where a $54 increase consumes a far larger share of income◈ Strong Evidence [30].
The second household response to price is exit, and the United States has run a national experiment on it. FEMA's Risk Rating 2.0, introduced in 2021, moved the National Flood Insurance Program to property-level risk pricing; 77% of policyholders saw an increase in the first year✓ Established [37]. New policy take-up subsequently fell by between 11% and 39% and renewals by 5% to 13%, depending on the size of the premium increase, with the largest coverage losses in lower-income communities◈ Strong Evidence [38].
Hurricane Helene showed what that looks like when the water arrives. In Buncombe County, North Carolina, where more than fifty people died in the flooding, fewer than 1% of homes carried federal flood insurance✓ Established [39]. CoreLogic estimated $20 billion to $30 billion of uninsured flood losses against $10.5 billion to $17.5 billion of insured wind and flood losses including the NFIP✓ Established [40]. The uninsured number was roughly double the insured one, in a state nobody had classed as high-risk.
Insurance market instability is unlikely to affect all communities equally.
— Manann Donoghoe, Brookings Institution, April 2026That inequality has a demographic shape in the American South. Black residents make up 28.8% of the population of vulnerable ZIP codes across states including Louisiana, Florida, South Carolina and Mississippi, against about 10% in adaptable ones✓ Established [30]. The relationship between premium burden and neighbourhood composition intensifies above a 20% threshold◈ Strong Evidence [30]. Where adaptive capacity is lowest, the same dollar increase does the most damage, and the coverage lost is hardest to replace◈ Strong Evidence [56].
Across Europe, only about a quarter of losses from extreme events were insured between 1980 and 2024, and EIOPA's 2025 Eurobarometer found just 17% of respondents holding cover for natural-catastrophe property damage✓ Established [35]. In emerging economies, 80% to 90% of catastrophe losses are typically uninsured✓ Established [19]. The insurance retreat is a visible crisis in the small part of the world that had cover in the first place; everywhere else the same losses have always been absorbed by households and public budgets◈ Strong Evidence [36].
The household ledger therefore has two columns that behave differently. Where cover exists and is mandatory — attached to a mortgage — the adjustment shows up as price. Where cover is voluntary, it shows up as coverage loss, and the loss concentrates among the households least able to self-insure◈ Strong Evidence [38] [58]. The first column is politically visible. The second is where most of the money actually is✓ Established [21].
The State Becomes the Insurer
Residual markets, assessments, and the socialisation of the tail
When private carriers withdraw, the risk does not disappear; it migrates to a state entity with no ability to decline it. California's FAIR Plan now carries $768 billion of exposure✓ Established [8]. In February 2025 it levied a $1 billion assessment on its member insurers, half of which the Insurance Commissioner permitted carriers to recoup from policyholders statewide✓ Established [9].
The residual market is the pressure valve that makes the retreat politically survivable, and it works by converting a priced risk into an unpriced obligation. The January 2025 Los Angeles firestorm tested it. UCLA Anderson put total insured losses at as much as $44.5 billion, of which the FAIR Plan alone accounted for $4.8 billion✓ Established [10]. Economic losses were estimated at $76 billion to $131 billion — a gap of roughly two to one against what was insured◈ Strong Evidence [10].
The pricing response was immediate and constrained. State Farm was granted an emergency interim increase of 17% on California homeowners policies in May 2025, having initially requested 22%, with 15% on renters and condominium cover and 38% on rental-owner policies✓ Established [11]. The company committed to a $400 million capital infusion from its parent and to pausing some non-renewals in exchange✓ Established [11]. In March 2026 a three-party settlement left the 17% in place but cut the condominium increase to 5.8%, with refunds backdated to June 2025✓ Established [12].
Texas shows the same migration without the wildfire narrative. The Texas Windstorm Insurance Association held 286,251 policies and $127.1 billion of exposure at the first quarter of 2026, having added more than 50,000 policies in two years, with exposure up roughly 19% in 2024 and 11% in 2025✓ Established [44]. It now writes 53% of residential windstorm and hail cover in its designated coverage area, against 47% for the entire private market✓ Established [44].
Florida is the counter-case, and it is the strongest evidence against a purely climatic reading of the retreat. Citizens Property Insurance held 1.42 million policies at its October 2023 peak and 936,182 at the start of 2025; by the end of January 2026 it was below 392,000, a 73% fall and the lowest level in its history✓ Established [14]. The depopulation programme moved more than 546,000 policies to private carriers during 2025 alone✓ Established [14].
The state pool is no longer Florida's largest property insurer, and in December 2025 Citizens recommended rate cuts for most policyholders✓ Established [13]. It expects lower reinsurance costs in 2026 on the back of reduced exposure and a softening market✓ Established [15]. Louisiana has moved the same way at smaller scale: residual enrolment is down nearly 20% from its 2022 peak to about 114,000 policies, premium increases are roughly flat in 2026 after 4.6% in 2025, and three new carriers were licensed in the first four months of the year✓ Established [46].
Florida Citizens fell from 1.42 million policies in October 2023 to under 392,000 by the end of January 2026, a 73% reduction achieved after litigation reform and rate adequacy allowed private carriers to re-enter✓ Established [14]. Louisiana's residual pool is down nearly 20% from its 2022 peak, with fourteen insurers licensed across 2024 and 2025 and three more in early 2026✓ Established [46]. Neither state's hurricane exposure fell. What changed was the legal and pricing environment in which private capital was asked to carry it◈ Strong Evidence [15].
France represents the fully socialised alternative and shows its own cost. On 1 January 2025 the mandatory CatNat surcharge attached to every property policy rose from 12% to 20%, with the motor equivalent going from 6% to 9% — the largest change to the regime in more than twenty-five years✓ Established [33]. The system had accumulated a deficit of €1.9 billion since 2015 and faces a projected financing shortfall of at least €420 million a year by 2050✓ Established [33] [34].
The French design buys universality and pays for it by deleting the signal. Because the surcharge is flat and statutory, a household in a flood plain and a household on a hill pay the same rate, which is equitable in the year of the flood and inflationary in every year before it◈ Strong Evidence [34]. The regime cannot decline a risk, cannot price one, and has now had to raise the levy on everyone by two-thirds to stay solvent✓ Established [33].
What Governments Did About It
Five responses, ranked by what they actually change
Policy responses to the retreat fall into five families: suppress the price, socialise the tail, subsidise the demand, harden the asset, or move the asset◈ Strong Evidence [53]. Only the last two change expected loss. They are also the two that have received the least money in every jurisdiction examined here◈ Strong Evidence [51] [52].
Price suppression is the oldest instrument and the best documented failure. California's Proposition 103 prior-approval regime historically barred insurers from using forward-looking catastrophe models or the net cost of reinsurance in setting rates, which held approved prices below modelled expected loss◈ Strong Evidence [55]. The consequence was not cheaper insurance but less of it, and analysis after the 2025 fires warned that insolvencies and further withdrawals were the plausible next step◈ Strong Evidence [55].
The Sustainable Insurance Strategy adopted in response is an explicit trade. Carriers gain the right to use catastrophe models and to pass through net reinsurance costs; in exchange they commit to writing in distressed and high-risk areas✓ Established [11]. The State Farm settlement is that trade in miniature — a 17% increase and a pause on some non-renewals, financed in part by a $400 million infusion from the parent company✓ Established [11] [12]. Whether the commitments outlast the next fire season is the open question.
Socialising the tail is the second family and the most widely used. It works until the assessment arrives: the FAIR Plan's $1 billion levy in 2025 was partly recouped from policyholders across California, including those who had never been near a fire✓ Established [9]. Europe is designing a larger version of the same idea, with EIOPA and the European Central Bank proposing a public-private reinsurance scheme to narrow a gap in which only about a quarter of extreme-event losses have been insured since 1980✓ Established [36] [35].
| Policy Approach | Structural Risk | Assessment |
|---|---|---|
| Rate suppression in prior-approval states | Holding approved rates below modelled expected loss does not reduce expected loss; it changes who is willing to write the policy◈ Strong Evidence [55]. California's exclusion of catastrophe models and net reinsurance cost from ratemaking coincided with the largest carrier withdrawal in the state's history◈ Strong Evidence [55]. | |
| Residual-market growth without exposure limits | The FAIR Plan's exposure rose 250% in under four years to $768 billion, and its first billion-dollar assessment was partly recouped from policyholders statewide✓ Established [8] [9]. The pool cannot decline a risk or reprice one◈ Strong Evidence [52]. | |
| Reliance on soft reinsurance pricing | January 2026 renewals fell 12-15% on excess capital and appetite for insurance-linked securities, not on any reassessment of hazard✓ Established [16] [17]. Capital cycles reverse considerably faster than exposure does◈ Strong Evidence [19]. | |
| Subsidised demand without mitigation conditions | Premium support keeps cover affordable but preserves occupancy in the hazard, and the evidence is that insurance pricing alone does not steer development away from wildfire-prone land◈ Strong Evidence [50]. | |
| Mitigation standards and structural hardening | Building to the IBHS Wildfire Prepared standard measurably restores availability: a rebuilt property in Paradise, California can be insured for about $2,000 against roughly $8,000 through the FAIR Plan◈ Strong Evidence [54] [50]. The constraint is coverage and funding, not efficacy. |
The fourth family is the only one with a demonstrated effect on the loss itself. The Insurance Institute for Business and Home Safety's Wildfire Prepared Home standard targets three failure points — the roof, specific building features, and the zero-to-five-foot ignition zone around the structure✓ Established [54]. Where communities have rebuilt to it, private capacity has returned: a Paradise, California property built to the standard can obtain cover for roughly $2,000 where the FAIR Plan would charge about $8,000◈ Strong Evidence [50].
The fifth family is relocation, and it is the least funded relative to the exposure it addresses. FEMA's voluntary buyout programmes pay 75% of acquisition costs with 25% from local government, cannot use eminent domain, and are chronically slow✓ Established [52]. Because the local match is a precondition, buyouts concentrate in municipalities that can afford them, which are not the places where the need is greatest◈ Strong Evidence [53]. Legal scholarship now describes public insurance itself as the most plausible available lever for semi-managed retreat◈ Strong Evidence [52].
In every jurisdiction examined here, the dominant response has been to act on the price rather than on the exposure◈ Strong Evidence [53]. France raised a mandatory surcharge by two-thirds; California suppressed rates for three decades and then traded model access for writing commitments; Florida rewrote its litigation rules; the European Union is designing a public reinsurance backstop✓ Established [33] [11] [14] [36]. Only mitigation standards and relocation change expected loss, and they remain the two smallest line items◈ Strong Evidence [51] [52].
The National Flood Insurance Program is the natural experiment on what happens when a government does the opposite and lets the price through. Risk Rating 2.0 improved the accuracy of the signal and reduced the coverage: new policies fell 11% to 39% and renewals 5% to 13%, with the sharpest losses among lower-income households◈ Strong Evidence [38] [37]. Accurate pricing and universal coverage are not simultaneously achievable without a subsidy that is explicitly named as one◈ Strong Evidence [51].
That is the trade every regulator in this report is making, usually without stating it. A price that reflects the risk will shrink the pool. A price that does not will shrink the number of carriers. The only exit from that trade is to change the risk, which requires capital spending on buildings and land use rather than on premiums◈ Strong Evidence [51] [54].
The Argument That Will Not Resolve
Two readings of the same administrative record
The Government Accountability Office reported in 2026 that national average homeowners premiums rose about 3% in real terms between 2019 and 2024 — while rising 25% or more in southern coastal areas⚖ Contested [29]. Both halves of that sentence are true, and each supports a different account of what is happening⚖ Contested [6].
The GAO's findings are the strongest evidence against a generalised crisis narrative. Its analysis found premiums broadly tracking inflation nationally, with the divergence concentrated by hazard: homes in high wind-risk areas paid about 58% more than comparable homes at medium wind risk, while moving from medium to high wildfire risk was associated with an 8% premium increase⚖ Contested [29]. Premiums as a share of median household income were highest in Florida, Louisiana and Oklahoma✓ Established [29].
The industry's position is that climate is one input among several and not the largest. Trade bodies attribute the deterioration to a combination of exposure growth in hazard zones, rebuilding-cost inflation, litigation and legal system abuse, and argue that reinsurance is a transmission mechanism rather than a cause — if expected losses rise, premiums must rise whether the risk is retained or ceded⚖ Contested [7] [57]. When the Senate Budget Committee published its non-renewal data, the industry's response was that the hearing had ignored that mix⚖ Contested [6].
The counter-position rests on the gradient rather than the level. If cost inflation were the driver, withdrawal would be broadly distributed; instead non-renewal rates run about 80% higher in the highest-risk ZIP codes and roughly twice as high in the West as in the Northeast✓ Established [1] [2]. The Senate dataset covering 249 million policies from 23 insurers found the rise concentrated in climate-exposed counties across all fifty states◈ Strong Evidence [4] [5].
The Repricing Reading
The annual policy is the only instrument in the housing stack that reprices every year, so it registered the change in expected loss first◈ Strong Evidence [26].
Non-renewal rates run about 80% higher in the highest-risk ZIP codes, and more than twice as high in the West as in the Northeast✓ Established [1] [2].
Of a projected 29.4-point real premium rise to 2055, 18.4 points correct present underpricing rather than future hazard◈ Strong Evidence [22].
The Cost-Inflation Reading
Nationally, average homeowners premiums rose about 3% in real terms between 2019 and 2024, with divergence concentrated regionally⚖ Contested [29].
More property of higher value has been built in hazard zones; part of the loss growth is a denominator problem rather than a climate one⚖ Contested [7].
Industry bodies attribute much of the deterioration to legal system abuse and construction cost inflation rather than to physical risk⚖ Contested [6] [57].
Florida cut its state pool by 73% in twenty-seven months with no change whatsoever in its hurricane exposure✓ Established [14].
The Florida result is the single most inconvenient fact for the repricing reading, and it deserves to be stated without hedging. A 73% reduction in the state residual pool in twenty-seven months, achieved while hurricane exposure was unchanged, demonstrates that a large part of what looked like climatic uninsurability was legal and actuarial✓ Established [14] [13]. Any account that treats withdrawal as a direct function of hazard has to explain that number⚖ Contested [15].
The repricing reading's answer is that Florida changed the two variables it could change — litigation exposure and rate adequacy — and thereby made the underlying hazard bearable at a price. That is a genuine achievement and it is not a refutation, because the hazard is unchanged and the price is now being paid◈ Strong Evidence [15]. What Florida shows is the size of the wedge between physical risk and institutional risk, not the absence of the former⚖ Contested [46].
What is not contested is narrower than the volume of argument suggests. Nobody disputes that non-renewals rose sharply between 2018 and 2024, that they rose fastest in high-hazard geographies, that residual markets absorbed the difference, or that the uninsured share of global catastrophe loss is growing✓ Established [2] [8] [21]. The dispute is over the weight assigned to physical hazard within that, and it will be settled by the NAIC's ZIP-code dataset rather than by argument✓ Established [3].
What the Evidence Tells Us
The signal was accurate; the response was to the signal
Four findings survive contact with the data from both sides of the argument. Withdrawal tracks modelled hazard✓ Established [1]. Transmission into credit and value is real but bounded◈ Strong Evidence [26]. Residual markets grow where private capital is barred from pricing and shrink where it is not✓ Established [14]. And the uninsured share of loss is growing even as reinsurance gets cheaper✓ Established [21].
The first finding is the most robust and the least politically useful. Non-renewal rates are about 80% higher in the highest-risk fifth of ZIP codes than in the lowest, premiums 82% higher, and the regional gradient reproduces at every level of aggregation regulators have tried✓ Established [1] [2]. Whatever share of the cost increase is attributable to litigation or construction inflation, the pattern of who gets dropped is not distributed that way◈ Strong Evidence [30].
The second is real and smaller than advertised. Insurer exit is associated with higher foreclosure rates, lower home values, weaker local spending and falling homeownership, and roughly an 11% relative price decline in the most exposed ZIP codes◈ Strong Evidence [28] [26]. That is a serious, compounding local drag. It is not the systemic event the more dramatic warnings describe, and the loan-level evidence supports the smaller claim rather than the larger one◈ Strong Evidence [25] [48].
The third finding is the one governments could act on tomorrow. Florida cut its residual pool 73% in twenty-seven months and Louisiana by nearly 20%, in both cases by changing the legal and pricing terms rather than the weather✓ Established [14] [46]. California's residual pool grew 250% in exposure over four years under a ratemaking regime that excluded catastrophe models and reinsurance costs◈ Strong Evidence [8] [55]. The residual market is a policy output, not a climate reading◈ Strong Evidence [52].
The fourth finding is the one that should worry anyone reading the 2026 renewal season as a recovery. Property catastrophe reinsurance fell 12% globally and 15% in Europe in January 2026, and Florida Citizens recommended rate cuts✓ Established [16] [13]. In the same twelve months, global uninsured catastrophe losses rose more than 7% to $424 billion and wildfire insured losses continued growing at an estimated 12% a year✓ Established [21] [19]. Capital returned; exposure did not fall.
The structural problem is a duration mismatch that no party to the argument disputes. A twelve-month contract is being used to underwrite a thirty-year mortgage on a hundred-year building in a hazard environment that is changing on a decadal scale◈ Strong Evidence [43]. Every institutional response so far — surcharges, assessments, residual pools, prior approval, backstops — operates on the twelve-month contract. None of them operates on the building or the location◈ Strong Evidence [53] [52].
Property catastrophe reinsurance rates fell 12% globally and 15% in Europe at the January 2026 renewals, driven by excess capital and appetite for insurance-linked securities✓ Established [16] [17]. Over the same period the value of uninsured natural-catastrophe losses rose more than 7% to $424 billion, with the North American gap widening 6% to $140 billion✓ Established [21] [19]. Cheap reinsurance is a statement about the supply of capital, not about the quantity of risk◈ Strong Evidence [18].
There is one intervention with demonstrated effect on the loss rather than the price, and it is the smallest programme in this report. Building to the IBHS Wildfire Prepared standard restores private capacity at roughly a quarter of the residual-market price in the communities that have adopted it◈ Strong Evidence [54] [50]. Nothing in the evidence suggests it is not scalable. Everything in the funding record suggests nobody has tried to scale it◈ Strong Evidence [51].
The insurance market did not fail. It did the one thing it exists to do — reprice an asset when the expected loss changes — and it did so faster than any other institution attached to housing◈ Strong Evidence [26]. Every major response since 2024 has addressed the price rather than the exposure: a two-thirds increase in the French surcharge, suppressed rates then negotiated model access in California, litigation reform in Florida, a proposed public reinsurance backstop in Europe✓ Established [33] [11] [14] [36]. The two instruments that change expected loss — hardening buildings and moving people out of the hazard — remain the least funded of the five◈ Strong Evidence [51] [52].
The measurable questions for the next five years are already specified. Whether the NAIC's ZIP-code dataset reproduces the Treasury gradient on 2018-2025 data✓ Established [3]. Whether Florida's residual pool stays below 400,000 policies through a landfall season✓ Established [14]. Whether California's FAIR Plan exposure passes $1 trillion before the Sustainable Insurance Strategy returns private capacity◈ Strong Evidence [8]. And whether any jurisdiction moves money from premium subsidy to structural mitigation at a scale that shows up in the loss data◈ Strong Evidence [51] [54].