RESEARCH DOSSIER
Actuarial Climate Scenario Analysis
Abstract. Actuarial climate scenario analysis translates physically and economically coherent climate pathways into conditional insurance outcomes. Its central problem is the conversion of changes in hazards, exposures, vulnerabilities, claims costs, policyholder behavior, investments, and macroeconomic conditions into defensible assumptions about insurance liabilities and financial resilience. This monograph provides an integrated examination of scenario selection, physical and transition-risk transmission, catastrophe modeling, claims trend decomposition, pricing, reserving, reinsurance, capital, asset-liability management, life and health insurance, and governance. It distinguishes scenario-conditioned loss distributions from unconditional forecasts and explains why climate model outputs alone cannot yield reliable insurance loss projections. A formal multistage representation connects hazard event sets, insured exposure, vulnerability, contract terms, insurance cash flows, and balance-sheet effects. Two worked numerical exercises illustrate nonstationary frequency-severity pricing and the sensitivity of capital proxies to distributional assumptions. The study critically evaluates international supervisory and reporting materials and emphasizes the December 2025 retraction of a macroeconomic climate-damages study incorporated into selected NGFS Phase V outputs. The original contribution is a comparative analytical synthesis of assumptions, temporal horizons, source dependencies, contractual mechanisms, and validation tests. No new proprietary empirical dataset, causal estimate, actuarial theory, or calibrated proprietary catastrophe model is claimed.
Keywords: Actuarial science; climate scenario analysis; climate change; physical risk; transition risk; catastrophe modeling; insurance pricing; claims inflation; technical provisions; nonstationarity; asset-liability management; solvency; reinsurance; ORSA; model risk; insurance protection gap.
Introduction
Climate conditions affect insurance through insured events, changes in insured values, policyholder and insurer responses, and the financial environment in which liabilities are valued and paid. The actuarial question is consequently one of conditional projection: how would the distribution of insured cash flows change if a specified physical and socioeconomic pathway were realized, holding explicitly identified commercial and contractual conditions fixed or allowing those conditions to evolve? An answer requires connecting multiple scientific and institutional domains while preserving the distinct object of each.
The physical foundations are substantive. The Intergovernmental Panel on Climate Change (IPCC, 2021) assesses an observed increase in hot extremes across most land regions and in heavy precipitation over most land areas with sufficient observational records; its assessment of climate extremes uses peril-specific confidence language and does not imply that every weather peril has changed in the same direction in every territory. IPCC (2023) also assesses growing losses and damages under climate change, subject to substantial regional and socioeconomic differences in its synthesis of climate impacts. This evidence motivates forward-looking insurance analysis; it is not a ready-made insurer claims triangle.
Insurers face a distinctive conversion problem. A projected change in precipitation intensity might affect a flood hazard map; the hazard change influences insured buildings according to elevation and flood defenses; the insurer then applies deductibles, exclusions, limits, and reinsurance; incurred claims are developed and settled under changing repair costs. A change in macroeconomic climate damages does not flow mechanically into a particular insurer's loss ratio. Similarly, an emissions-policy shock may affect the market value of corporate bonds held by an insurer while also altering transportation, construction, and liability exposures within the underwriting portfolio.
The International Association of Insurance Supervisors (IAIS, 2025) treats scenario analysis as a means of examining long-run risk, strategy, and capital vulnerabilities in its climate supervision application paper. The document states that its examples and recommendations support application of existing Insurance Core Principles rather than creating new binding requirements. European Insurance and Occupational Pensions Authority (EIOPA, 2022) guidance demonstrates the translation of climate scenarios into an insurer's own risk and solvency assessment (ORSA). These approaches share an emphasis on materiality, governance, and the practical limits of model outputs.
This dossier addresses four questions. First, which climate and socioeconomic scenario variables can legitimately modify actuarial assumptions? Second, how should an insurer reconcile climate timescales, policy durations, renewal discretion, accounting rules, and capital horizons? Third, which statistical and modeling tests separate scenario sensitivity from unsupported precision? Fourth, what can an insurer infer for underwriting, reserves, solvency, and strategy when the main uncertainties are structural and correlated?
Scientific Object, Scope, and Method
Actuarial climate scenario analysis is the conditional mapping from a coherent climate-economic scenario to quantities relevant to insurance contracts and balance sheets. Its objects include claim frequency and severity, loss-development patterns, recoveries, premiums and expenses, policy persistency, investment returns, discount curves, capital requirements, liquidity, and insurance availability. The model must specify whose exposure is measured, over which period, under which contract terms, in which currency and price level, and relative to which baseline.
A useful notation is S for a scenario narrative and associated trajectories, H for hazard, E for exposed assets or persons, V for vulnerability, C for contract structure, M for insurer management response, and Y for the resultant cash flows or losses. The central conditional quantity is P(Y | S, E, V, C, M, D), with D identifying the data and model assumptions. This expression describes a conditional distribution, not the probability that S itself occurs. The distinction matters because a policymaker's pathway, a stress test, and a probabilistic weather generator have different epistemic status.
The method combines critical examination of primary climate assessments, official supervisory and accounting standards, and peer-reviewed research on disaster-loss normalization; formal mathematical analysis; and transparent synthetic stress calculations. It does not estimate new parameters from claims data, perform proprietary downscaling, or replicate unpublished insurer portfolios. Every numerical illustration below is hypothetical and is used to reveal an inferential mechanism, not to state a market-wide prediction.
For each important inference the manuscript asks whether the source actually establishes the proposed claim, whether the measurement scale is appropriate, whether the causal variables are identified, and whether the uncertainty can be carried forward into an actuarial decision. The Actuarial Standards Board's (ASB, 2020) ASOP No. 56 on modeling provides a useful professional reference for model purpose, input selection, validation, checking, governance, and communication in its applicable U.S. context. International use requires attention to local professional obligations and law.
A scenario exercise has multiple distinct outputs. A sensitivity calculation asks how results change when selected assumptions change; a stress test examines adverse but specified combinations; a probabilistic model integrates over an explicitly defined probability distribution; and a strategic scenario compares robust choices under several plausible futures. These outputs may coexist but should not be reported as if each measured an unconditional probability of the future.
Physical Climate Evidence and Scenario Families
The climate system provides an essential boundary for actuarial narratives. IPCC (2021) finds high confidence in widespread increases in the frequency and intensity of hot extremes, and high confidence that heavy precipitation has intensified over much of the land area where observations permit analysis. The same physical science assessment expresses lower confidence for some aspects of tropical-cyclone frequency and regional drought trends. Actuarial use must preserve this differentiation by peril, geography, season, and time horizon. A uniform percentage uplift to every hazard is an expedient sensitivity, not a climate-science conclusion.
The Copernicus Climate Change Service reported on November 11, 2026 that 2026 was on track to become the second calendar year exceeding 1.5 degrees Celsius above the 1850-1900 reference period. Its November 2026 temperature assessment concerns observational and statistical global temperature measures. A single-year threshold exceedance must be distinguished from a sustained multidecadal Paris Agreement temperature benchmark and from a local insurance peril. Temperature observations are an input to, rather than a replacement for, event and vulnerability modeling.
The IPCC scenario families describe alternative emissions and socioeconomic pathways. The Network for Greening the Financial System (NGFS) connects physical and transition risks with macro-financial trajectories. Its long-term scenario portal organizes illustrative pathways into orderly, disorderly, and hot-house-world categories and documents Version 5 of the long-term scenarios published in November 2024. An orderly trajectory typically assumes earlier, more coordinated transition action; a disorderly trajectory exposes economies to delayed or uneven adjustment; and a hot-house-world narrative entails stronger eventual physical risks. None should be read as a guaranteed future.
NGFS (2025) updated its guide to climate scenario analysis, reflecting experience with both short- and long-term frameworks. Scenario pathways should be chosen for decision relevance: a near-term underwriting plan may need climate-conditioned hazard shifts and claims inflation during the next several renewals; a long-duration life portfolio may need heat, mortality, investments, and policyholder behavior over decades; strategic capital planning may need disorderly transition and severe physical-loss cases. A single named global scenario cannot supply all of these local quantities without an explicit translation layer.
The translation layer must account for downscaling. Global mean temperature, country GDP, and economy-wide carbon prices describe different quantities from a parcel-specific flood depth, local construction-cost inflation, or the credit spread of a particular issuer. Downscaling introduces spatial assumptions; peril-model transformation introduces physical and engineering assumptions; and insurance contract mapping introduces exposure and financial assumptions. Each step has its own error and should be documented independently.
Scenario provenance is a material scientific issue. NGFS (2026) provides explanatory notes on Phase V, describing the assumptions and uses of its scenarios. Its dedicated Phase V data notice identifies particular integrated physical-damages variables affected by the subsequent retraction of an underlying empirical climate-damages paper. This status must accompany any reuse of those variables.
Scenario Probability and Decision Horizons
A scenario can be coherent and relevant without possessing a defensible likelihood. Emissions, government policy, innovation, migration, adaptation, and underwriting choices are partly endogenous to decisions. Assigning probabilities to a handful of narratives merely because a financial model requires numerical weights creates an unsupported bridge from possibility to expectation. A modeler may instead report conditional results separately, use explicitly expert-elicited weights with sensitivity ranges, or estimate a probabilistic distribution only where an inferential basis exists.
In catastrophe practice, an annual exceedance probability expresses the frequency with which an annual insured loss crosses a specified threshold under a defined hazard-exposure-vulnerability distribution. A 1-in-200 annual return period means a modeled annual exceedance probability of 0.005 under that distribution. It does not mean the loss occurs once every exactly 200 years. Under nonstationarity, the annual exceedance function changes with time and scenario: EP_t(x | S) = P(L_t > x | S). A current 200-year return level can have a different future annual exceedance probability, and a future "200-year loss" is meaningful only after the time-dependent distribution is specified.
Horizon mismatch is a decisive risk in interpreting results. General insurance contracts may reprice annually, while assets, reinsurance programs, catastrophe exposure, and franchise decisions evolve over other horizons. Liability claims may be reported and settled long after the policy period. Technical provisions recognize defined contract cash flows under applicable accounting and prudential rules, while ORSA may examine longer-term business viability. The same projected 2050 hazard should not be treated automatically as a 2026 claim reserve adjustment.
A consistent architecture has several clocks: climate and policy transition time; hazard-event time; policy inception and coverage time; claim occurrence, reporting, and settlement time; investment maturity; and risk-capital measurement time. A scenario-conditioned premium indication for 2035 may use exposure expected in 2035, while a present reserve for already incurred claims requires estimating future payments arising from past coverage. Keeping these clocks explicit prevents double counting and false inference.
IPCC (2022) discusses insurance as one component of wider risk-financing and adaptation strategies in its assessment of risk-management options. Insurance purchase, risk reduction, public support, and risk retention are jointly shaped by incentives and affordability. An insurer's pathway therefore incorporates choices that may change the risk distribution: adaptation investment, exclusions, price adjustments, nonrenewal, geographic withdrawal, and public-private intervention. A fixed-portfolio stress and a dynamic-business strategy answer different questions and should be shown side by side.
Hazard, Exposure, Vulnerability, and Insurance Contracts
An insured catastrophe loss is produced by the interaction of event characteristics with the location and susceptibility of covered assets and persons, followed by contractual financial transformation. A conventional event-loss representation is L_e = sum_i g(H_e,i, E_i, V_i, C_i), where g combines intensity, asset value, vulnerability, and policy terms for insured item i under event e. The function may include damage thresholds, repair-cost ratios, deductibles, limits, coinsurance, exclusions, sublimits, and reinstatements. Gross physical damage, policyholder economic loss, gross insured loss, and net insurer loss are distinct measurements.
Hazard is multidimensional. Flood risk depends on precipitation, runoff, river flow, coastal level, drainage, and local defenses. Wildfire loss depends on ignition, weather, fuel, topography, and the built environment. Wind damage depends on event tracks, local gusts, building standards, and occupancy. Heat-related claims depend on duration, nighttime temperature, air conditioning, demographics, and medical access. The selected scenario variables must connect physically to the insured peril; an aggregate global temperature adjustment is usually too remote to modify individual insured claim rates directly.
Exposure is not constant merely because a modeled portfolio is frozen. Replacement costs may rise, insured values may diverge from market values, properties may be constructed in expanding hazard zones, and migration may alter the distribution of lives and assets. A fixed-exposure experiment may be scientifically useful for isolating the effect of changing hazard, but an underwriting business plan requires exposure dynamics as a separate component. Insurer data need geocoding, construction attributes, occupancy, deductibles, insured values, policy terms, and records of coverage changes.
Vulnerability captures damage conditional on hazard intensity. Adaptation modifies this relationship. A strengthened roof, defensible space, elevated electrical equipment, or a flood barrier may materially change severity conditional on an event. Structural vulnerability differs from realized loss after claims adjustment and contractor inflation. A climate scenario that assumes improved defenses should disclose their construction dates, coverage of the insured portfolio, maintenance, potential failure modes, and possible adverse effects under more extreme events.
Insurance contracts create nonlinearities. Consider a covered loss of $150,000 with a $25,000 deductible and a $100,000 limit: a simplified indemnity is min(max(150000 - 25000, 0), 100000) = $100,000. If physical damage rises 10%, the insurer still pays $100,000 under the same terms. Where losses sit near a deductible, small hazard changes may instead cause a large proportional change in claims. Consequently, applying a uniform percentage to historical net paid claims can misstate scenario consequences even when gross damage changes are known.
An insurer should also examine coverage definitions and legal constraints. Business interruption may require insured physical damage or have special contingent triggers; parametric contracts pay according to an index; liability insurance may react to changing litigation and regulatory environments. These distinctions are operationally material. Contract inventories and coverage wording are necessary inputs to catastrophe scenarios, not post-processing details.
Catastrophe Loss Modeling and Tail Dependence
Catastrophe models usually combine an event set, intensity footprints, exposure inventory, vulnerability functions, and insurance-financial terms. In a simple independent-event annual compound model, N_t | S is the event count and X_j,t | S represents individual insured event loss, giving aggregate S_t = sum_{j=1}^{N_t} X_j,t. If N_t is Poisson with rate lambda_t and independent severities with mean mu_t and finite second moment, E[S_t] = lambda_t mu_t and Var(S_t) = lambda_t E[X_t^2]. These identities are conditional on the model and do not establish that catastrophe event counts are Poisson in nature.
For rare perils, severity may be strongly heavy-tailed, event rates may vary with shared climate drivers, and losses may exhibit spatial clustering. A storm or flood event damages many policyholders together; common construction-cost inflation raises severity across events; distinct perils can respond to the same atmospheric conditions. Appropriate simulations may require common factors, event clustering, non-Poisson frequency processes, extreme-value treatment, dependence structures, and portfolio-specific vulnerability. Parameter uncertainty can become dominant in the far tail.
Tail metrics serve different decisions. Expected annual loss represents an average conditional cost. Value at Risk at a stated percentile represents a threshold of a modeled loss distribution. Expected Shortfall captures average loss beyond the specified quantile under an appropriate convention. The annual occurrence exceedance curve for a single event and the annual aggregate exceedance curve for the sum of multiple events are also different objects. Catastrophe capital is sensitive to the choice, and reports must name the metric precisely.
A climate scenario affects not only marginal event intensities but possibly dependence. When drought increases wildfire risk, agriculture exposure and property exposure may deteriorate simultaneously; when coastal storm surge coincides with precipitation and high river flow, flood losses may compound. Dependence must be tied to physical and geographical mechanisms before an assumption of correlation is propagated into capital. An arbitrary adjustment to a copula parameter can illustrate sensitivity, but it should not be represented as a measured change.
Model error has several layers: event-generation uncertainty, peril-intensity uncertainty, vulnerability uncertainty, exposure-data error, financial-module approximation, and uncertainty about future adaptation. Increasing precision in one layer does not eliminate uncertainty in the others. For long-term climate risk, structural comparison across several credible models may be more decision-useful than a single highly detailed exceedance curve with narrow confidence bands.
Claims Trend Recalibration and Empirical Identification
Historical claims experience combines hazard, exposure, vulnerability, contract changes, inflation, and claims operations. A schematic multiplicative decomposition is Loss_t = Hazard_t x Exposure_t x Vulnerability_t x Coverage_t x CostLevel_t x ClaimsProcess_t, with each component appropriately normalized and interactions recognized. This factorization is analytical; it is not a claim that the components are statistically independent. Changes in average insured loss can also reflect claim frequency, reporting thresholds, fraud controls, legal costs, litigation patterns, and reserve strengthening.
Normalization is indispensable. Bouwer (2011), in a review of disaster-loss trends, found that rising reported disaster losses could not be attributed simply to anthropogenic climate change using the studies reviewed at that time, and highlighted the role of growing exposure. Pielke (2021) revisits the empirical normalization literature in an assessment of economic disaster losses. These studies address historical loss attribution under their data and methods; their findings cannot negate independently assessed physical changes in specific hazards. The analytically defensible position is to separate physical hazard evidence from attribution of realized economic and insured losses.
A credible claims trend exercise constructs consistent exposure bases, deflates relevant costs, adjusts policy limits and deductibles, tracks coverage and claims-handling changes, and compares results by peril and territory. Weather observations and peril intensity may enter generalized linear or hierarchical models, while exposure and repair-cost variables control financial trends. For example, log E[ClaimCount_it] may be modeled using an offset log Exposure_it, peril indicators, geographically appropriate climate covariates, building attributes, calendar effects, and an interaction between hazard and adaptation. The chosen link, distribution, overdispersion treatment, and spatial dependence require testing.
Claims inflation must be separated into general and social or sector-specific components where data permit. After a major catastrophe, labor demand, material shortages, and surge pricing can make repair costs rise faster than broad consumer inflation. A scenario may shift the distribution of claims settlement time as well as indemnity amount. For long-tailed casualty claims, evolving legal and social conditions may dominate direct physical hazard effects. A precise climate signal can be overwhelmed by uncertainty in policyholder behavior or legal interpretation.
Causal identification is difficult. Hazard trends vary geographically and may covary with settlement patterns, infrastructure investments, economic activity, and reporting technologies. Comparing claims from climate-exposed and less-exposed regions requires attention to selection, coverage changes, different asset mixes, and changing penetration. Without matched exposure and hazard data, a regression slope on calendar year should not be labeled a climate effect.
For validation, use time-based and geographically blocked out-of-sample comparisons rather than only random policy-level splits, which may leak spatial and event information across training and test sets. Compare a stationary benchmark, an exposure/inflation-adjusted benchmark, and a climate-conditioned specification; inspect not only average error but tail calibration and decision loss. Structural breaks after changes to coverage or claim handling should be diagnosed before treating observed trend as persistent.
Pricing and Underwriting Under Climate Nonstationarity
An actuarial rate indication reflects anticipated claims and expenses over the coverage period, together with reinsurance, capital, and profit provisions under a chosen pricing framework and applicable law. A climate scenario can shift prospective claim frequency and severity, but its influence on price depends on policy renewal flexibility, competition, regulation, risk classification, and affordability. A purely retrospective average may be inadequate where hazards or insured exposure have changed materially.
For a stylized homogeneous portfolio, expected annual losses are E[L_t | S] = Npol_t q_t(S) mu_t(S), where Npol_t is the number of insured units, q_t the annual insured-claim frequency per unit, and mu_t mean insured severity. More generally, sum over policies, perils, and geographical units, with joint event effects modeled explicitly. This baseline supports a transparent decomposition of premium changes into hazard, exposure, vulnerability, claims inflation, contractual retention, and risk-load contributions.
The elasticity of expected losses to physical intensity is often nonlinear because vulnerability functions have thresholds and insurance terms alter the paid proportion. Under an annual renewable contract, the insurer may progressively reprice, raise deductibles, reduce limits, alter underwriting criteria, or withdraw from a market. A scenario that assumes automatic premium increases should identify the statutory and commercial feasibility of doing so. Insurance demand and anti-selection may change in response to those adjustments, feeding back into the next portfolio.
The most relevant economic comparison is frequently between an unchanged portfolio and an explicitly adaptive portfolio. In a fixed-book scenario, worsening hazards raise expected losses while policy terms and exposures are held constant. In a management-action scenario, repair standards, reinsurance, underwriting, deductibles, and premium levels change subject to operational feasibility. The difference measures the modeled effects of responses conditional on assumptions; it is not automatically the causal benefit of adaptation.
Rate adequacy is not identical to market accessibility. Even technically sufficient premiums may be unaffordable for households and businesses. The IPCC's insurance and adaptation discussion emphasizes the interaction of financial protection, physical risk reduction, and public policy. A scenario assessment should quantify protection-gap implications and distinguish a reduction in insured claims caused by risk mitigation from one caused by coverage withdrawal. The latter can improve an insurer's loss ratio while increasing uninsured social losses.
Reserving and Technical Provisions
Reserving begins with identifying which obligations belong to which contractual and accounting boundary. Incurred claims reserves relate to insured events that have already occurred, including reported and unreported claims, while liabilities for remaining coverage concern future insured service within defined contract groups. A 2040 physical scenario may be highly relevant to franchise strategy and long-term risk appetite but have limited direct bearing on 2026 incurred claims unless it changes projected settlement costs or claim development for existing events.
Climate transmission into incurred claims can occur through secondary damage, delayed repairs, re-opening of claims, inflation after successive catastrophes, or future claims settlement conditions. For a long-tailed line, changes in medical, legal, economic, and occupational conditions may affect future payments from past insured events. Scenario selection must reflect which future variables remain relevant to current contractual liabilities.
Under IFRS 17, insurance contract measurement uses current estimates of fulfillment cash flows, discounting, and a risk adjustment for nonfinancial risk, together with the contractual service margin where applicable. The IFRS 17 standard overview specifies that the measurement and recognition of obligations depend on groups of contracts and their coverage period. Therefore a strategic climate-stress loss over several decades does not map one-to-one into an IFRS 17 liability. The insurer must establish which scenario inputs influence estimates at the reporting date and how uncertainty is reflected under the standard.
Reserving analysis should retain a baseline model and calculate clearly defined sensitivities: expected settlement inflation, claim closure delays, large-loss incidence, reinsurance recovery collectability, and exposure to peril-specific late reporting. Where development triangles include extreme events, changes in mix and inflation can invalidate stable development assumptions. The actuary should evaluate calendar-year patterns, event-level information, and the maturity of relevant cohorts rather than inserting a universal climate factor into all reserve cells.
A practical approach is to compute conditional cash flows at each future payment date, apply relevant contract boundaries, account for uncertainty according to the measurement regime, and document which climate signals are sufficiently evidenced to influence current estimates. Several regimes can yield different recognized liabilities from the same economic scenario because accounting and prudential purposes differ. That distinction is a substantive part of the actuarial analysis, not an inconsistency to be suppressed.
Transition Risk, Investments, and Asset-Liability Management
Transition pathways affect insurers through carbon policy, technology, energy prices, sector competition, regulation, labor markets, and economic growth. An insurer's corporate-bond portfolio may lose value if transition exposure weakens issuers' cash flows or credit quality; other issuers may benefit from adaptation and new infrastructure demand. Real estate collateral and equity portfolios can face both physical damage and transition-related repricing. The timing of market recognition can precede physical losses by years.
An asset-liability management scenario should calculate both sides of the balance sheet under a consistent economic state. A simplified economic surplus measure is A_t(S) - L_t(S), where A_t denotes scenario-conditioned asset value and L_t denotes liability value. If asset cash flows are long-duration and liabilities short-duration, an interest-rate shock has a different effect than for a long-duration annuity portfolio. Credit-spread and discount-rate changes should be modeled according to the relevant valuation conventions; the change in market asset prices cannot be automatically offset by an identical change in insurance liabilities.
Liquidity is a separate channel. A severe wildfire or flood season can accelerate claim payments while market dislocation reduces the liquidity of some investments. Reinsurance recoveries may arrive after gross claim payments, requiring bridge funding. In an adverse transition scenario, collateral or margin requirements and cash outflows may increase before long-term asset values adjust. Balance-sheet solvency at a reporting date does not establish liquidity resilience throughout the scenario.
The IAIS (2025) supervision guidance on concentrations highlights the importance of concentrations in underwriting, investments, and reinsurance, including potential common shocks. A complete analysis examines whether the same geography or industrial sector appears in both invested assets and insured liabilities. For example, a regional insurer may insure coastal property while holding municipal obligations or real estate investments whose values also deteriorate after a coastal disaster.
The NGFS long-term categories can supply coherent macroeconomic background, but translating them into yield curves, default probabilities, sector spreads, and policyholder lapse rates requires additional assumptions. The NGFS guide on scenario design supports the distinction between macroeconomic pathways and institution-specific risk measurement. An insurer should not call a resulting asset value an observed or consensus market forecast merely because it was generated from an established scenario family.
Reinsurance, Aggregation, and Capacity
Reinsurance changes the distribution of net losses, the timing of recoveries, and counterparty risk. A catastrophe excess-of-loss layer responds to the distribution of losses crossing a retention, the limit, reinstatement terms, and other contract conditions. A climate-driven increase in gross expected loss need not produce a proportional increase in ceded or retained losses because layers attach nonlinearly. Erosion of a layer by multiple events may be more important than one isolated extreme loss.
Consider a treaty attaching at $20 million and exhausting at $60 million per occurrence. A $15 million event produces no recovery; a $40 million event produces $20 million before contractual adjustments; a $100 million event yields at most $40 million under a single unreinstated layer. Future hazard change can push more events across the attachment, altering reinsurance demand and cost disproportionately. Aggregate annual covers may respond differently to multiple moderate events than to one large occurrence.
Diversification assumptions also require scrutiny. Geographical diversification may weaken if the portfolio expands into correlated hazard regions, or if shared inflation and supply-chain shortages amplify several events. Global reinsurers face simultaneous exposure to physical perils, asset repricing, and cedent credit quality. Correlation among gross losses, reinsurance prices, and available capital may create a second-order effect that is missed by an isolated gross-to-net calculation.
Contract continuity over long horizons is uncertain. Reinsurance renews; limits, prices, exclusions, collateral conditions, and counterparty participation change. The IAIS (2025) explicitly discusses climate scenarios in relation to future reinsurance capacity and product viability. Modeling unchanged reinsurance through 2050 is therefore a controlled assumption, not a guaranteed mechanism of future risk transfer.
Net-loss scenario analysis should report the gross event-loss distribution, ceded recoveries under actual treaty terms, net retained loss, counterparty concentration, reinstatement costs, and sensitivity to unavailable renewal capacity. For long-term strategic scenarios, a range of plausible reinsurance market states is more informative than simply applying the current ceded percentage to every future year.
Life, Health, Agriculture, and Liability Portfolios
Physical climate effects extend beyond property catastrophe claims. Heat and air pollution episodes can influence morbidity and mortality, but exposures depend on age, occupation, housing, health care, adaptation, and local conditions. A life insurer's projection requires linking geographically appropriate health evidence with insured population characteristics, underwriting selection, benefit definitions, competing risks, and long-term medical change. Broad climate indicators cannot be substituted for life-table parameters without explicit estimation.
Long-term life and retirement products add asset risk and liability-duration effects. Mortality and longevity impacts may move in different directions across groups and periods, and the effect on an insurer's net financial position depends on the product mix. A term-life benefit and a life-contingent annuity can respond in opposite financial directions to a mortality shock; disability insurance may be driven by morbidity, economic dislocation, and claims-duration changes rather than mortality alone. Even when the physical hazard change is well established, the portfolio outcome may remain ambiguous.
Agricultural insurance exposes the sensitivity of actuarial assumptions to crop phenology, rainfall timing, drought, heat, soil, pests, irrigation, farm adaptation, and government support. An insurer must identify crop and location, growing season, coverage trigger, yield or revenue base, premium subsidy, and basis risk. A seasonal drought forecast does not directly estimate the cost of a multi-year farm insurance book. Parametric indices may be easier to simulate under some climate inputs, but the payout/actual loss relationship can change under nonstationarity.
Liability portfolios may be affected by climate litigation, transition disclosure claims, infrastructure damage, or changing duties of care. These mechanisms depend on jurisdiction, policy wording, claims-made or occurrence basis, retroactive dates, aggregation clauses, and evolving legal precedent. The existence of climate litigation does not establish a numerical future insured loss. Scenario exercises should therefore identify legal pathways, policy triggers, plausible aggregation and defense costs, and genuine coverage uncertainty.
Multiline groups need an integrated view. A flood can create property claims, business-interruption payments, temporary displacement impacts, and economic disruption to invested assets; a heat event can affect agriculture, health, workers' compensation, and power infrastructure. Interdependence changes the solvency assessment. At the same time, some risks remain diversifying, and the model must avoid assuming perfect dependence simply because all exposures share a climate label.
Worked Portfolio Scenarios and Calculation Checks
The following stylized examples are transparent mathematical diagnostics. Every assumption is invented for illustration. No value is fitted to an actual insurer, hazard projection, or market series.
Consider 10,000 insured property units. Assume an annual insured claim probability q0 = 0.01 per policy, mean insured severity mu0 = $40,000, and a total annual premium of $5 million. Under a simplified independent frequency-severity representation, lambda0 = 100 expected claims, E(L0) = 100 x $40,000 = $4 million, and the expected loss ratio is 4 / 5 = 80%, excluding expenses and reinsurance. The baseline has no modeled climate trend or growth.
Now specify a future conditional scenario with claim probability increased by 25%, mean severity increased by 15%, and insured units increased by 10%, holding policy terms constant. The multiplicative expected-loss factor is 1.25 x 1.15 x 1.10 = 1.58125. Expected claims are 10,000 x 1.10 x 0.0125 = 137.5 and mean severity is $46,000. The future expected annual loss is therefore $6.325 million; the increase relative to $4 million is 58.125%. With nominal premiums frozen at $5 million the expected loss ratio is 126.5%. To maintain an 80% claims-only loss-ratio target, required premium is $6.325 million / 0.8 = $7.90625 million, before considering expenses, investment income, reinsurance and regulatory constraints.
Suppose now that verified adaptation reduces modeled claim severity by 20% from the stressed level while the modeled frequency and exposed-unit count remain unchanged. Expected insured losses become $6.325 million x 0.80 = $5.06 million. The reduction is $1.265 million relative to the unadapted scenario and the claims-only premium required at an 80% target becomes $6.325 million. This calculation illustrates why adaptation and underwriting withdrawal must be distinguished: an identical reduction in retained losses could arise from mitigation or from covering fewer policyholders, but the social outcomes differ.
For a distributional sensitivity, assume claim count is Poisson and individual insured severity has coefficient of variation CV = 1.5, independent across claims. Then E[X^2] = mu^2(1 + CV^2) = 3.25 mu^2. At baseline, Var(L0) = 100 x 3.25 x $40,000^2 = $520 billion-squared, giving standard deviation about $0.7211 million. Under the stress, Var(L1) = 137.5 x 3.25 x $46,000^2, with standard deviation about $0.9724 million. The exact variance algebra follows the assumed compound-Poisson structure; neither severity CV nor Poisson event count is empirically supported in this example.
A normal approximation gives a deliberately limited 99.5% quantile proxy of mean plus 2.576 standard deviations. The baseline proxy is about $5.858 million and the stressed proxy about $8.830 million. These values are not regulatory solvency capital, a calibrated 1-in-200 catastrophe metric, or a credible heavy-tail quantile. A genuine property-catastrophe model would evaluate event clustering, spatial accumulation, severe event tails, contract terms, reinsurance, and parametric uncertainty. The toy approximation demonstrates only that a higher mean and a changing dispersion can raise an extreme-loss proxy by more than the mean shift.
A second diagnostic concerns reinsurance attachment. Let gross annual occurrence losses in a stylized scenario be $12 million, $35 million, and $85 million. Under an unreinstated $20 million excess $40 million per occurrence layer, simplified single-event recoveries would be $0, $15 million, and $40 million. If the annual contract contains a single aggregate exhaustion limit of $40 million instead of independently replenished occurrence limits, actual aggregate recovery can be markedly lower when several events occur. Therefore every net-loss scenario must carry the precise treaty structure. Illustrations with aggregate exhaustion cannot be treated as if they were three independent fully reinstated layers.
Three limits govern interpretation. First, the frequency, severity, and exposure multipliers are assumed, not inferred from IPCC or NGFS numerical outputs. Second, the expected-loss and variance identities rely on a selected toy distribution. Third, the calculation establishes conditional financial arithmetic, not the probability of the future climate pathway. These restrictions prevent the example from becoming spurious empirical evidence.
Capital, Liquidity, and Solvency Decision-Making
Climate scenario analysis can inform capital adequacy, but conditional results must be reconciled with the legal capital framework, own risk assessment, liquidity planning, and business strategy. The IAIS Insurance Core Principles and ComFrame establish an internationally oriented framework including enterprise risk management, capital adequacy, and valuation; their adoption and binding effect are jurisdiction-specific. ICP 16 integrates risk identification, quantitative techniques, capital, asset-liability management, ORSA, and recovery planning.
An ORSA climate exercise should begin with materiality. Where a line of business has little material climate transmission on the relevant horizon, effort may be proportionately limited, though the rationale should be documented. Where risks are material, analysis should include gross and net underwriting impacts, investments, asset-liability matching, liquidity, strategic response, risk appetite, and possible effects on future insurability. EIOPA (2022) provides detailed examples for mock life and non-life insurers in its climate scenario ORSA guidance.
Scenarios for regulatory stress testing and insurer internal risk management have different objectives. EIOPA's 2024 insurance stress test is principally a geopolitical/macroeconomic adverse-stress exercise rather than a physical-climate study, but it demonstrates a useful method distinction between fixed-balance-sheet and constrained-balance-sheet management responses, and between capital and liquidity components. Those features can inform climate stress design without mischaracterizing the exercise's actual topic.
Capital sufficiency at the start of a scenario can be expressed as available financial resources relative to a specified risk requirement; survival through the scenario also depends on how losses, investment marks, reinsurance recoveries, and management actions unfold. Management responses must be feasible at the time of stress. Selling illiquid assets instantaneously, renewing reinsurance at unchanged terms after widespread losses, or repricing legally constrained policies without delay can artificially improve modeled capital.
Climate risk may also affect policyholder security indirectly through a contraction in available insurance. The Organisation for Economic Co-operation and Development (OECD, 2026) discusses protection gaps, insured catastrophe losses, and public-private arrangements in its Global Insurance Market Trends assessment. Reporting a stronger capital ratio following withdrawal from exposed communities is incomplete without accompanying measures of coverage availability, affordability, and retained societal risk.
International Standards and Institutional Applications
International actuarial scenario practice draws on several types of normative source. A scientific assessment summarizes evidence and uncertainty; a supervisory standard frames prudential expectations; a jurisdictional regulation imposes legally binding duties when adopted; an accounting standard governs recognition and measurement for entities applying it; and a professional actuarial standard governs work within its specified professional scope. These sources should be mapped to their actual authority rather than placed in one undifferentiated compliance list.
The IAIS April 2025 climate risk application paper provides practical examples of supervisory application of existing principles and explicitly does not itself impose a new global legal duty. EIOPA's 2022 materiality and ORSA guidance serves a European supervisory-convergence purpose in the Solvency II setting. It should not be imported as a universally binding requirement on insurers elsewhere.
For financial reporting, IFRS 17 addresses measurement and presentation of insurance contract liabilities and profits under that accounting framework. IFRS S2 addresses disclosure of climate-related risks and opportunities and requires scenario analysis of climate resilience in an approach proportionate to the entity's circumstances. The IFRS S2 disclosure standard became effective for annual reporting periods beginning on or after January 1, 2024, subject to the extent of jurisdictional adoption and entity applicability. The IFRS Foundation's March 2026 scenario analysis factsheet clarifies that scenario analysis is used to assess climate resilience. Disclosure requirements are distinct from directly booking all hypothetical adverse-scenario losses as liabilities.
Professional modeling guidance matters for communication. ASOP No. 56 (ASB, 2020), in its model validation and disclosure provisions, addresses purpose, model risks, assumptions, validation and the communication of limitations for work in scope. Good international research practice similarly identifies when a result depends on a model's intended use, input credibility, and known failure modes. A regulatory name is not a substitute for demonstrating whether the selected scenario is fit for the relevant actuarial decision.
The practical synthesis is a mapping of each decision to both a scientific basis and an applicable normative basis. Scenario choice requires defensible climate and economic narratives; conversion to claims requires relevant peril and insurance evidence; valuation requires the applicable measurement standard; capital planning requires applicable prudential obligations; public disclosure requires the relevant reporting rules; actuarial communication requires professional transparency. The distinctions are especially important in cross-border groups.
Model Validation, Governance, and Evidence Quality
Validation should proceed along the chain rather than only at the final loss number. At the physical stage, test hazard-model outputs against observed events and independent scientific estimates, including geospatial resolution and regional confidence. At the exposure stage, reconcile sums insured, insured locations, policy counts and building characteristics to the underwriting administration system. At the vulnerability stage, assess claims and engineering evidence after controlling for event intensity and coverage differences. At the financial stage, reproduce deductibles, limits, reinsurance recoveries, and settlement timing for known event examples.
At the statistical stage, compare alternative distributions, test sensitivity to event frequency and tail shape, examine blocked temporal out-of-sample performance, and assess parameter uncertainty. At the economic stage, reconcile the scenario's price levels, real or nominal variables, exchange rates, discount assumptions and sector classifications. At the governance stage, identify owners for source versions, model changes, overrides, expert judgment, materiality, and signoff. The ASB's model risk mitigation guidance recognizes that validation depth should be consistent with complexity and intended purpose.
Data quality is a first-order issue, particularly for catastrophe losses. Location coordinates may be imprecise, insured values stale, cover terms incomplete, historical loss records inconsistent, and low-frequency severe events sparsely observed. Geographical and social differences in insurance penetration mean the insured-loss series is a selected subset of economic damage. Comparing historical insured loss totals without coverage and inflation normalization can produce false climate attribution and misleading model calibration.
An important real-world evidence integrity problem emerged in 2025. Kotz, Levermann, and Wenz (2024), whose empirical macroeconomic damages approach informed selected physical-risk measures in NGFS Phase V, retracted their paper on December 3, 2025. The Nature retraction notice attributes the retraction to sensitivity to the removal of Uzbekistan because of inaccuracies in underlying 1995-1999 economic data. Bearpark, Hogan, and Hsiang (2025) had identified data anomalies and uncertainty issues in their published methodological critique. This example demonstrates how high-profile estimates and apparent statistical precision can change after source-level audit.
The NGFS itself published a statement on affected Phase V estimates. It distinguishes affected long-term integrated physical-damages variables from unaffected components and short-term scenario outputs. Its official affected-variable notice identifies affected integrated physical-damage REMIND-MAgPIE outputs, certain NiGEM physical and combined outputs, and selected post-processed GDP-damage fields. Accordingly, a present-day insurer should not treat these particular affected values as independently validated loss drivers; it should either replace them with supported alternatives or mark the affected exercise as sensitivity-only, with explicit evidence limitations. It would also be incorrect to characterize all NGFS variables as invalid.
The editorial rule from this episode is substantive. Trace each numeric assumption to the exact variable, source version, and model dependency; check for published corrections and retractions; evaluate whether the disputed component drives the financial result; and preserve the original and alternative outcomes in the audit trail. Source citations must support the specific proposition asserted. A scenario branded by an authoritative institution cannot substitute for the integrity of its underlying inputs.
Uncertainty Disclosure and Interpretation
Four uncertainties deserve separate disclosure. Physical uncertainty concerns future emissions, climate response, internal variability, and local hazard structure. Socioeconomic uncertainty concerns population, asset values, technology, regulation, and adaptation. Insurance uncertainty concerns policy terms, demand, underwriting and claims operations, reinsurance, and litigation. Model uncertainty concerns statistical specification, downscaling, data limitations, dependence, and validation. A single confidence interval on an expected loss cannot summarize all four without an explicit aggregation method.
Conditional scenario ranges should be displayed alongside assumption tables. The reader should see the base period, projection horizon, currency and price basis, hazard and exposure definitions, contractual retention, management actions, and source versions. If the pathway lacks a defensible probability, the result should be described as a scenario-conditioned outcome, not an expectation weighted by an invented likelihood. A probability derived from weather-generator events within a fixed pathway should likewise not be confused with the likelihood of the pathway.
A robust disclosure distinguishes climate-conditioned physical damage, economic loss, insured loss, insurer gross loss, reinsurer ceded recovery, and insurer net retained loss. Different numbers can move differently with adaptation, coverage restriction, inflation, or contract thresholds. Ambiguous use of the word "loss" is a common source of mistaken comparison. The same problem arises in confusing expected annual loss with an extreme annual exceedance threshold or with present-valued technical provisions.
Materiality and uncertainty are jointly relevant. A source with broad confidence may support a directional vulnerability assessment even if it cannot support a precise reserving factor; a well-measured hazard change may have an uncertain contractual impact. Disclosure should explain which part of the decision is supported by strong evidence, which is a sensitivity, and which remains a strategic assumption.
Integrated Analytical Findings
The first finding is that actuarial climate scenario analysis is a translation problem with multiple scientific boundaries. Climate science supplies conditional physical pathways; economic scenarios supply transitions and socioeconomic context; actuaries translate them through exposure, vulnerability, contracts, claim development, finance, and legal obligations. Each transformation must preserve provenance, time horizon, and uncertainty. The quality of the final result is constrained by the weakest consequential step in the chain.
Second, nonstationarity is peril-, geography-, portfolio-, and time-specific. The evidence does not justify one global climate factor applied across every insured claim type. Where physical evidence is strong but insurance mapping weak, stress sensitivities may be informative without supporting a precise price or reserve estimate. Where claims data are strong but physical trend attribution weak, a well-calibrated near-term predictive model can be useful without making a causal statement about climate change.
Third, horizon reconciliation is integral to insurance economics. The same 2050 climate event set can illuminate long-term franchise strategy, future premium adequacy, and capital planning, while producing a different present impact on incurred-claim reserves or current IFRS 17 measurements. These are consequences of different decision and liability boundaries, not contradictions. An actuarial report should always label the date, contract period, event period, and settlement horizon.
Fourth, climate risk affects both the numerator and the denominator of apparent insurer performance. Underwriting actions can reduce retained claims by changing insured exposure; that may improve measured loss ratios while increasing protection gaps. Better modeling therefore compares outcomes for insurers, policyholders, and the distribution of residual loss. The IPCC's wider risk-financing perspective and the OECD's recent protection-gap analysis support treating the accessibility of insurance as an explicit strategic outcome.
Fifth, reinsurance and investment portfolios can transmit correlated climate losses through channels that are absent from standalone underwriting simulations. Both asset quality and insured losses may deteriorate in the same region or sector. Reinsurance prices and recoverability may worsen when insurers most need protection. An internally consistent solvency scenario must consider those interactions while avoiding unjustified assumptions of universal correlation.
Sixth, evidence maintenance is part of scientific quality. The retraction affecting selected NGFS Phase V physical-damage outputs shows why version control, correction surveillance, and variable-specific dependency analysis are necessary. Scenario analysis is valuable precisely when its results can be interrogated and revised as knowledge changes.
Seventh, useful actuarial decisions can be made without claiming false precision. A clearly identified sensitivity range, robust underwriting rule, adaptation comparison, or capital contingency can be more defensible than a narrow actuarial loss estimate derived from unvalidated macroeconomic proxies. Results become stronger through reproducible transformations and explicit evidence, not through the addition of unsupported decimal places.
Conclusion
Actuarial Climate Scenario Analysis establishes an integrated method of examining how hypothetical climate and transition pathways affect insurance assumptions and financial resilience. A scientifically defensible exercise identifies the scenario, separates its conditional assumptions from event probabilities, maps hazard to exposure and vulnerability, applies insurance and reinsurance terms, projects cash flows on appropriate timelines, and evaluates accounting, capital, liquidity, and strategic outcomes under the relevant institutional rules.
The synthesis demonstrates that insurance losses cannot be inferred directly from global warming levels or macroeconomic damage trajectories. Local physical mechanisms, insured values, adaptation, contract thresholds, settlement costs, portfolio selection, and insurer behavior determine the transmission. Models must represent interactions across underwriting, investments, and reinsurance while disclosing which quantities are observations, which are estimates, and which are scenarios.
The worked illustrations verify elementary frequency-severity arithmetic and show how changes in frequency, severity, exposure, and adaptation can alter pricing and a limited distributional proxy. They are explicitly synthetic, and their capital approximations are unsuitable as calibrated catastrophe-risk estimates. The evidence review also establishes the need to screen scenario inputs for retractions and corrections, including the documented limits of affected NGFS Phase V physical-damage variables.
The resulting practical conclusion is an evidence-governed approach to actuarial climate reasoning: use scenario analysis to test decisions, expose vulnerabilities and response options, and document consequential uncertainties; retain independent validation for hazard, claims and financial modules; and apply every result within its proper legal, accounting and supervisory scope. This is an original analytical synthesis of existing scientific and actuarial knowledge, with no claim to a new theoretical system or original causal estimate.
References
Actuarial Standards Board. 2020. Actuarial Standard of Practice No. 56, Modeling. Effective for work commenced on or after October 1, 2020. https://www.actuarialstandardsboard.org/asops/modeling/
Bearpark, Tom, Dylan Hogan, and Solomon Hsiang. 2025. "Data Anomalies and the Economic Commitment of Climate Change." Nature 644: E7-E11. https://www.nature.com/articles/s41586-025-09320-4
Bouwer, Laurens M. 2011. "Have Disaster Losses Increased Due to Anthropogenic Climate Change?" Bulletin of the American Meteorological Society 92 (1): 39-46. https://doi.org/10.1175/2010BAMS3092.1
Copernicus Climate Change Service. 2026. "2026 and the 4-Year Global Temperature Average on Track to Exceed 1.5 Degrees Celsius Above Pre-industrial Levels." November 11. https://climate.copernicus.eu/2026-and-4-year-global-temperature-average-track-exceed-15degc-above-pre-industrial-levels
European Insurance and Occupational Pensions Authority. 2022. Application Guidance on Running Climate Change Materiality Assessment and Using Climate Change Scenarios in the ORSA. August 2. https://www.eiopa.europa.eu/publications/application-guidance-climate-change-materiality-assessments-and-climate-change-scenarios-orsa_en
European Insurance and Occupational Pensions Authority. 2024. Insurance Stress Test 2024. Technical background, scenarios, and methodology. https://www.eiopa.europa.eu/insurance-stress-test-2024_en
IFRS Foundation. 2023. IFRS 17 Insurance Contracts, current issued-standard summary. https://www.ifrs.org/issued-standards/list-of-standards/ifrs-17-insurance-contracts/
IFRS Foundation. 2023. IFRS S2 Climate-related Disclosures, issued standard and amendment history. https://www.ifrs.org/issued-standards/ifrs-sustainability-standards-navigator/ifrs-s2-climate-related-disclosures/
IFRS Foundation. 2026. Climate Resilience and Climate-related Scenario Analysis Requirements in IFRS S2. Factsheet Series, March. https://www.ifrs.org/content/dam/ifrs/supporting-implementation/ifrs-s2/climate-related-scenario-factsheet.pdf
Intergovernmental Panel on Climate Change. 2021. "Summary for Policymakers." In Climate Change 2021: The Physical Science Basis, Working Group I Contribution to the Sixth Assessment Report. https://www.ipcc.ch/report/ar6/wg1/chapter/summary-for-policymakers/
Intergovernmental Panel on Climate Change. 2022. "Chapter 17: Decision-Making Options for Managing Risk." In Climate Change 2022: Impacts, Adaptation and Vulnerability, Working Group II Contribution to the Sixth Assessment Report. https://www.ipcc.ch/report/ar6/wg2/chapter/chapter-17/
Intergovernmental Panel on Climate Change. 2023. "Summary for Policymakers." Climate Change 2023: Synthesis Report. https://www.ipcc.ch/report/ar6/syr/summary-for-policymakers/
International Association of Insurance Supervisors. 2024. Insurance Core Principles and Common Framework for the Supervision of Internationally Active Insurance Groups, updated December 2024. https://www.iais.org/activities-topics/standard-setting/icps-and-comframe/
International Association of Insurance Supervisors. 2025. Application Paper on the Supervision of Climate-related Risks in the Insurance Sector. April. https://www.iais.org/uploads/2025/04/Application-Paper-on-the-supervision-of-climate-related-risks-in-the-insurance-sector.pdf
Kotz, Maximilian, Anders Levermann, and Leonie Wenz. 2025. "Retraction Note: The Economic Commitment of Climate Change." Nature 648: 764. December 3. https://www.nature.com/articles/s41586-025-09726-0
Network for Greening the Financial System. 2024. NGFS Long-term Climate Scenarios, Phase V. Version 5 published November 2024. https://www.ngfs.net/ngfs-scenarios-portal/
Network for Greening the Financial System. 2025. Guide to Climate Scenario Analysis for Central Banks and Supervisors, updated edition. November 13. https://www.ngfs.net/en/publications-and-statistics/publications/2025-guide-climate-scenario-analysis-central-banks-and-supervisors
Network for Greening the Financial System. 2025. Statement Regarding Physical Risk Estimates in Phase V of NGFS Long-term Scenarios. December 3. https://www.ngfs.net/en/press-release/statement-regarding-physical-risk-estimates-phase-v-ngfs-long-term-scenarios
Network for Greening the Financial System. 2026. NGFS Scenarios Portal: Data and Resources, Phase V Notice to Users. Accessed November 14. https://www.ngfs.net/ngfs-scenarios-portal/data-resources/
Network for Greening the Financial System. 2026. Explanatory Notes on NGFS Long-term Climate Scenarios. Updated September 24. https://www.ngfs.net/en/publications-and-statistics/publications/explanatory-notes-ngfs-long-term-climate-scenarios
Organisation for Economic Co-operation and Development. 2026. Global Insurance Market Trends 2026, "Risks and Vulnerabilities Facing the Insurance Sector." https://www.oecd.org/en/publications/global-insurance-market-trends-2026_c8be6585-en/full-report/component-8.html
Pielke, Roger Jr. 2021. "Economic 'Normalisation' of Disaster Losses 1998-2020: A Literature Review and Assessment." Environmental Hazards 20 (2): 93-111. Published online August 5, 2020. https://doi.org/10.1080/17477891.2020.1800440