RESEARCH DOSSIER
Insurance Market Climate Risk Disclosure Reporting Comparability
Abstract. Climate risk disclosures have become an important component of insurance market information, but their economic interpretation depends on the population, contract perimeter, legal mandate, measurement unit, and conditional scenario represented in the report. This monograph examines the comparability of insurer disclosures across international standards and selected supervisory and reporting systems in the United States, Canada, the European Union, Australia, and the United Kingdom, distinguishing public financial disclosure from regulatory returns and exploratory exercises. It combines documentary interpretation, an institutional crosswalk, critique of prior empirical research, and reproducible hypothetical calculations. Common disclosure headings support comparison of governance and risk-management processes; quantitative comparison requires substantially stronger alignment of exposure, risk transfer, valuation and scenario assumptions. Insurance-associated emissions, financed emissions, annual expected claims, tail losses, and prudential capital sensitivities represent distinct economic measures. Differences in business mix, reinsurance, coverage and reporting eligibility can reverse comparative rankings. The analysis identifies the conditions under which harmonized disclosures support cross-border supervision and market discipline, and the limits of aggregation when evidence is selective, modeled, or missing. No original market-level causal estimate is claimed.
Keywords: Climate risk disclosure; insurance markets; reporting comparability; insurance supervision; scenario analysis; physical risk; transition risk; insurance-associated emissions; financial materiality; cross-border insurance.
Introduction
An insurance market discloses its climate exposure through institutions whose contractual obligations, investment portfolios and geographical operations connect present financial decisions to future physical and economic conditions. A report makes these connections observable only insofar as its underlying definitions and measurements permit a reader to reconstruct what is being described. Consequently, reporting comparability is an evidentiary property of particular observations. It exists between sufficiently aligned observations of the same economic phenomenon and supports qualified comparison when differences are transparent and correctable.
The relevant population includes insurers and reinsurers, insurance groups and regulated legal entities, and the policies, assets, liabilities and risks attributable to those units. Climate-related market statistics concern a different population again: the households, businesses and territories that require protection, including those without insurance. Confusing the population of insurers with the population of insurable exposures creates systematic errors. An insurer can reduce its recorded catastrophe exposure by ceasing business in highly exposed locations while the corresponding community becomes less financially resilient.
The research addresses three questions. Which institutional documents define the observations available to analysts? How do scope, template and scenario choices alter their interpretation? Under which conditions can a cross-border comparison support an inference about climate exposure, risk management or insurance market resilience? The investigation uses documentary evidence available on November 19, 2026, selective source-level examination and transparent analytical examples. Jurisdictions are selected for contrasting reporting functions, not statistical representation of the world. The analysis examines instruments rather than giving jurisdiction-specific legal advice or a compliance determination about any company. Its quantitative analysis distinguishes documentary extracts of issuer-reported figures from constructed illustrations; neither is presented as a representative market estimate.
Conceptual and Historical Foundations
Insurance as an intermediary of climate-related financial consequences
Physical climate risk concerns losses associated with changes in hazards, including abrupt extreme events and slower changes in environmental conditions. Transition risk concerns the effects of changing technologies, regulations, economic activity and behavior during adjustment to a lower-emissions economy. For insurance institutions, each may affect the asset and liability sides of the balance sheet. Property claims can rise through altered hazard frequency and severity; asset values can change as borrowers or investees face new costs; liability claims may emerge through legal disputes; mortality, morbidity and operational burdens may change across different horizons.
These mechanisms are represented by several layers of observation. A hazard measure describes an environmental process. An exposure measure describes the assets, persons or obligations within reach of that process. Vulnerability expresses the susceptibility of those exposures. Contract terms determine the covered portion of economic damage. Financial risk metrics reflect the resulting claims, financing costs and capital consequences. Moving from a hazard indicator to a financial loss requires assumptions at each intervening layer. Comparability of the first layer cannot establish comparability of the final layer.
The international baseline developed from the climate-related financial disclosure agenda. The ISSB (2023) explains that IFRS S2 climate disclosures serve users of general-purpose financial reports and integrate earlier TCFD recommendations. Their subject is the entity's prospects over different horizons. Insurance-specific analysis must extend this framework into its actual financial transmission channels: contract design, insured exposures, retention, reinsurance, asset-liability matching and solvency.
Different audiences and different decisions
Public financial disclosures inform providers of capital about material effects on an entity's prospects. Policyholders seek evidence of contract reliability, continuing protection, claims-paying capacity and availability of coverage. Supervisors may require granular data to assess individual undertakings and risks across the system. A report suited to one decision can be incomplete for another. An investment analyst's central concern may be retained profitability; a supervisor may also care about the social consequences of withdrawal from particular markets.
The IAIS (2025) climate supervision application paper treats public disclosure and supervisory reporting jointly while identifying their distinct users and purposes. Its status is implementation guidance for Insurance Core Principles, rather than a new universally binding domestic statute. The distinction explains why a multinational group can produce one consolidated sustainability narrative and submit several locally defined regulatory returns. Those returns cannot be presumed to represent either duplicated public information or complete data accessible to researchers.
Market discipline requires the reader to connect statements to economic consequences. A governance declaration can demonstrate that a committee exists, while a measured claim-cost sensitivity provides information about a specified exposure. A target is an intention under assumptions. A retrospective observation is a result over a defined period. A scenario presents a conditional outcome. These observations can coexist in a single climate report and should remain distinguishable in analysis.
Historical continuity and discontinuity
The California Department of Insurance (accessed 2026) describes the evolution of the NAIC insurer climate survey from the earlier questionnaire to a TCFD-aligned approach adopted in 2022. Changes in questions can increase recorded awareness and disclosure without an equivalent change in firms' underlying exposures or controls. A time series crossing a redesign requires either a documented mapping or parallel reporting under both versions. Continuity of the institutional name alone is inadequate.
Disclosure standardization also changes the incentives of reporters. A board may allocate resources to matters for which disclosure is required. Common definitions reduce repeated requests and improve internal information flows. At the same time, a firm may optimize presentation, select favorable denominators or emphasize targets relative to results. A research design that equates formal alignment with actual risk reduction misses these institutional incentives.
Primary Institutional Documents and Comparative Research Design
Documentary case selection and the unit of analysis
The international baseline, a public insurer survey, mandatory sustainability reporting, prudential returns and exploratory supervisory scenarios are complementary documentary cases. The case selection compares reporting mechanisms and legal reach. It does not estimate an average national quality of disclosure. The evidence hierarchy places official instruments and reporting specifications first for requirements, issuer publications first for reported company characteristics, and peer-reviewed research first for its own reported methods and findings. Regulatory descriptions and industry guidance establish expectations, not independent proof of implementation.
Each observation should retain at least the original document version, its effective period, reporting date, issuer and legal entity, consolidation boundary, geographic scope, insurance activity, public/private accessibility, measurement unit, numerator, denominator, model assumptions, and identification of estimates or omissions. A data extraction should preserve the original value and transformation separately. This separation permits a later analyst to reverse a conversion, recover an original classification, or inspect an uncertainty that was suppressed during a data merge.
Institutional crosswalk
| Framework or exercise | Reporting function | Main observation unit | Comparison limitation |
|---|---|---|---|
| ISSB IFRS S2 | Investor-useful climate financial information | Reporting entity | Local adoption, materiality, relief and versions |
| NAIC climate survey | Public insurance-sector reporting | Participating eligible insurers, with attention to group reports | State participation and premium eligibility |
| OSFI Guideline B-15 | Climate risk management and disclosures | Federally regulated financial institution | Entity classes and implementation dates |
| OSFI climate returns | Structured prudential data | Reporting institution and applicable subtable | Return access, field applicability and scope |
| EU ESRS | Sustainability reporting | In-scope undertaking or group | Double materiality, scope and transitional versions |
| EIOPA ORSA guidance | Prudential risk assessment | Insurance or reinsurance undertaking | Materiality and applicable supervisory context |
| Australia AASB S2 | Climate-related financial reporting | Reporting entity within relevant mandate | Phased population and Australian text |
| UK CBES 2021 | Exploratory scenario analysis | Participating financial institutions | Exercise assumptions and participant selection |
The table is an analytical crosswalk of the documents discussed below. Its categories describe the dominant function rather than a claim that legal or reporting obligations are exhausted by these instruments. The appropriate comparator is the observation produced, not the mere name of the jurisdiction.
North American reporting populations
The California survey results portal reports a reporting-year 2022 response count of 1,698, compared with 1,541 in 2021, and describes a threshold of more than USD 100 million in nationwide direct written premiums for insurers licensed in participating jurisdictions. The increased count can be reproduced by dividing 157 additional submissions by 1,541, yielding approximately 10.2%. The portal separately states 9.2%; that figure differs from conventional growth calculated using the stated starting count. Dividing by the ending count yields approximately 9.25%. The discrepancy is documented rather than silently adopted. The counts describe submissions and are insufficient to measure unique consolidated reporting groups or the change in disclosure quality.
OSFI (2025) establishes distinct expectations through Guideline B-15. The Canadian regulatory population is defined by federal supervision rather than by the United States survey's participating-state mechanism. OSFI's insurer climate risk returns supply an example of machine-readable structural discipline. The specifications describe exposure and emissions information relevant to prudential supervision and identify field applicability by subtable. A blank for an inapplicable field is meaningful metadata. Replacing it with zero would insert an observed absence of exposure where the specification intended a structural absence of a measurement.
European and Australian version control
The European Commission (2023) established the original sustainability reporting standards through Delegated Regulation 2023/2772. Their materiality architecture includes the undertaking's impacts as well as financial consequences. This can produce a broader reporting object than an investor-oriented framework even where the same climate metric is used. The European Parliament and Council (2026) subsequently amended the population through Directive (EU) 2026/470. Its revised provisions include consolidated thresholds for parent undertakings of more than EUR 450 million net turnover and more than 1,000 employees. A researcher must apply relevant transitional and national implementation provisions rather than treating the new threshold as a universal filter for all historical insurance reports.
A further 2026 ESRS amendment provides for a revised standard set and a transitional choice for undertakings subject to the earlier standards in 2026 financial years. The available indexed legal text supports this version-level observation; it does not support reconstruction of every revised datapoint. Preserve version and effective period as independent variables. Comparing 2025 and 2026 filings solely by an indicator's label risks confounding economic change with reporting reform.
EIOPA (2022) issued climate scenario guidance for ORSA following its 2021 supervisory opinion. The application guidance supports consistent and proportionate climate risk assessment. It should be understood within prudential own-risk and solvency assessment rather than as a globally standardized public spreadsheet. In Australia, the AASB (2024) AASB S2 standard contains an entity-prospects objective and disclosure architecture similar to IFRS S2. The original version has a specified operative period. An inference about an insurer's legal obligation requires the Australian statutory population rules and specific reliefs, not a presumption based on its commercial size alone.
Climate Reporting Templates
Semantic correspondence in climate templates
Common themes in climate disclosure include oversight, strategy, risk-management processes and metrics. Agreement at this level makes a search for relevant information easier. Actual comparability requires interpretation of the content beneath these themes. A board-level risk committee, a documented escalation process, an independently validated model and a portfolio remediation program represent different interventions. Coding each as a single affirmative governance statement discards meaningful distinctions between their operational effects.
A sound qualitative inventory distinguishes the existence of a stated policy, the scope of its application, its operating evidence, and realized outcomes. The existence of a target should be recorded separately from its base year, target year, coverage, interim trajectory and progress. Documents may contain conditional language, exceptions or qualifications. Text extraction should retain the sentence or paragraph supplying those qualifications rather than isolating a reassuring phrase. A model cannot infer that every occurrence of risk terminology expresses a policy or an outcome.
Multiple meanings of quantitative disclosure
Three families of emissions observations require careful separation. An insurer's operational emissions concern its own activities and applicable value-chain categories. Financed emissions attribute emissions to financing and investment relationships. Insurance-associated emissions attribute emissions in connection with underwriting activity. PCAF (2025) maintains separate materials for these latter functions in its financial industry emissions standard. Even when expressed in tonnes of carbon dioxide equivalent, the resulting values arise from different economic relationships, populations and methodologies.
The ISSB (2025) greenhouse gas disclosure amendments permit limiting Category 15 measurement and disclosure to financed emissions as defined in IFRS S2, and allow alternatives to the formerly specified industry classification system. Their effective date is the beginning of 2027 reporting periods, with earlier application permitted. A dataset must therefore distinguish original requirements, early application and later mandatory effective periods; a changed emissions boundary is capable of producing a break even if the underlying business remains constant.
A source that distinguishes financed and insurance-associated emissions is especially important where a jurisdiction expects information on both. OSFI's guideline specifically addresses relevant portions of Scope 3 and identifies PCAF or a comparable industry-accepted approach. Comparable methodology is a subject of evidence, not an automatic consequence of using a recognized standard. Attribution choices, covered assets and contracts, estimation coverage, and emission-factor vintage should remain available to downstream users.
Emissions measures are also distinct from financial loss measures. Absolute emissions may help describe transition-related exposure, yet a high-emitting insured company could possess strong physical resilience or a credible transition investment program. A low-emitting portfolio could be concentrated in locations susceptible to flood. The underwriting relationship may transfer only narrow perils, leaving many transition costs outside insurance coverage. It is therefore analytically unsound to interpret an emissions inventory as an expected claim-cost inventory.
Coverage, denominator and data quality
Consider a constructed insurer A reporting 80,000 tonnes on USD 400 million of covered annual premium, and insurer B reporting 70,000 tonnes on USD 700 million. Their illustrative reported intensities are 200 and 100 tonnes per million dollars, respectively. The comparison is meaningful only if premium is the same valuation basis, the emissions boundary matches the premium population and the relevant accounting approach is sufficiently consistent.
If insurer A's covered premium represents half its eligible portfolio and B's represents 90%, it would be misleading to treat those intensities as comprehensive portfolio measures. Multiplying missing portions by reported intensities silently assumes identical emissions characteristics for reported and unreported risks. That assumption can fail where data availability depends on sector, client size or geography. Disclosing the percentage of portfolio covered by a calculation makes the comparison more informative without solving the selection problem.
Data quality should be assessed independently of numerical magnitude. Direct measurement, third-party estimates, sector averages and unavailable data imply different uncertainty. A measurement with more primary data may be less uncertain yet differ from a prior proxy-based estimate in either direction. A legitimate improvement in estimation may result in a reported increase even when physical emissions remain constant. Historical restatement or a reconciliation would distinguish the measurement effect from changes in exposure.
Public reports and supervisory returns
The Swiss Re reporting website (Swiss Re reporting scope, 2026) describes its 2025 sustainability report as group-wide and organizes climate disclosure around TCFD recommendations. The Travelers reporting archive (Travelers historical report archive, accessed 2026) lists separate 2024 TCFD and SASB reports. These documentary examples establish reporting organization and the availability of different frameworks. They do not establish that the companies' emissions, catastrophe losses or scenario results are directly comparable; the underlying issuer PDF metrics were not independently audited for this study.
The distinction has methodological consequences. A researcher gathering documents through issuer sites may identify several publications for one entity-year. The extraction process must determine which is the authoritative version for each metric and detect repeated content across reports. One sustainability report linked by multiple regulated subsidiaries should not be counted as multiple independent risk assessments. A documented common group process can coexist with separate local exposure observations.
The NAIC survey eligibility description (NAIC, n.d.) states that the threshold is at least USD 100 million in nationwide direct written premium in the preceding year, with jurisdictional participation discretionary. The California portal's wording should not be used to resolve the exact boundary case in every participating jurisdiction; the applicable local instructions control. This distinction also prevents survey respondent counts from being interpreted as the total regulated population.
Documentary Extraction from Two Insurer Reports
The two issuer cases below are purposively selected demonstrations of measurement boundaries, not a representative sample or a comparative performance ranking. The reporting periods differ. Extraction identifies the published indicator and its denominator, rather than constructing a common numerical climate score.
The Travelers 2024 TCFD report (The Travelers Companies, Inc., 2025, p. 14) describes a U.S. hurricane analysis using RCP 4.5 and RCP 6.0, with average annual loss and a 100-year return-period measure at 2030, 2050 and 2100 for an exposure portfolio dating to the second quarter of 2020. The company reports greater confidence in the direction of projected change than in its size. A reader can compare the horizon, pathway and risk statistic with another insurer's disclosure; a numerical comparison additionally requires the baseline loss distribution, portfolio date, contract terms and model configuration. None of the graph values is transcribed here because the page image could not be retrieved.
Reported pretax catastrophe losses were USD 3,335 million in 2024 and USD 2,991 million in 2023 (Travelers catastrophe metrics, p. 28). The change is USD 344 million, or approximately 11.50%. This is a change in a defined accounting and event-classification outcome. The report's catastrophe designation and loss threshold restrict its scope. A single annual change cannot identify the causal contribution of climate change: exposure growth, inflation, event realization, coverage and other factors also matter. Ceded premium and reinsurance recoverable are separately reported; subtracting a balance-sheet recoverable from an income-period total would mix measurement periods and stock-flow concepts.
The Travelers operational emissions table (p. 30) reports 19,069 metric tons of CO2 equivalent in Scope 1 and 18,938 in Scope 2 for 2024, totaling 38,007 for owned operations. The Scope 2 figure is location-based. The 2023 total was 38,635; the calculated decline is 1.63%. The separately identified business-travel Scope 3 amount in 2024 is 20,210 metric tons. It is an identified subcategory, not a measurement of all value-chain emissions. The report also restates the 2023 business-travel amount. Comparisons extending across publication vintages therefore require both category boundaries and revisions.
The Swiss Re 2025 sustainability report (Swiss Re, 2026, p. 52) identifies single-risk business as 26% of total property and casualty gross written premium, with selected interim targets covering 24% of the single-risk premium. Multiplication gives approximately 6.24% of total P&C premium, consistent with the issuer's rounded 6%. Thus a percentage describing target coverage within an eligible subportfolio cannot be read as whole-portfolio coverage. Treaty business has different information constraints and may call for a different disclosure unit.
The Swiss Re climate alignment metric (p. 61) for a specified listed-company, nonfossil-fuel single-risk portfolio reports 54% of eligible premium associated with third-party-validated science-based targets in 2025, against a revised 52% in 2024. The company attributes a nine-percentage-point upward restatement of the 2024 figure to enhanced ownership information and matching. The originally implied 43% baseline would yield an apparent eleven-point improvement; the revised series yields two points. The discrepancy quantifies the effect of data-vintage choice, not an independently demonstrated decrease in emissions.
| Extracted observation | Measurement basis | Permissible inference |
|---|---|---|
| Travelers: 2024 catastrophe loss, USD 3,335 million | Realized annual pretax loss under issuer definition | Accounting outcome, subject to event and reporting scope |
| Travelers: 2024 Scope 1 and 2, 38,007 tCO2e | Owned operations; location-based Scope 2 | Operational inventory with specified perimeter |
| Swiss Re: approximately 6% target coverage | Total P&C GWP | Breadth of specified interim underwriting targets |
| Swiss Re: 54% alignment metric | Eligible selected single-risk premium | Client target alignment within a defined subset |
These observations concern four different quantities. A cross-company ranking derived by treating them as interchangeable would have no stable economic interpretation. Their comparability can instead be assessed in terms of whether the same field is disclosed elsewhere with a matching definition and whether the residual differences can be measured. The documentary extracts do not establish the relative climate resilience of either firm. They demonstrate why reporting-vintage control, denominator reconciliation and explicit distinction between exposures, risk outcomes and transition proxies are necessary even when the issuer reports are detailed.
Scope and Scenario Definitions
Contractual geography and consolidation boundaries
A climate-risk observation may refer to the domicile of an insurer, the location of a policyholder, the geographical location of insured property, the country of an underlying investee or the region affected by a reinsurer's cedants. Each geographical definition may answer a legitimate question. They produce different maps. Aggregating by headquarters alone assigns an international group's distributed underwriting exposure to a single country and can obscure concentrated losses elsewhere. For physical risk, hazard-relevant location is generally more informative than legal domicile; for legal jurisdiction, the domicile or governing law may be central.
Group boundaries pose a similar problem. Legal-entity reporting allows reconciliation to local capital and supervisory duties. Group-level reporting can identify net economic exposure and common governance. Both are important. Intragroup reinsurance is a transfer for the local company but is typically internal to the consolidated group. An analyst adding the local company's net exposure to the group's consolidated exposure would count overlapping business. External reinsurance, in contrast, may reduce the cedant's retained loss while adding risk to another group's portfolio.
Conditional scenario outcomes
The Bank of England's (2022) Climate Biennial Exploratory Scenario examines divergent climate pathways and differentiated consequences for financial institutions. An exploratory exercise creates common shocks or pathways to illuminate sensitivities; it does not deliver an unconditional forecast. The insurer's starting balance sheet, model assumptions, contract renewal and management-action rules shape the result. An asset valuation shock incurred today is not interchangeable with a sum of underwriting losses incurred over thirty years.
A scenario comparison must preserve scenario provider, version, model, geographic and sectoral resolution, horizon, baseline and adaptation assumptions. It should identify whether the balance sheet is static, evolves without management action, or evolves through specified responses. For nonlife insurance, a static pricing assumption may reveal vulnerability of the existing portfolio, while dynamic repricing shows how losses are redistributed over time through premiums or coverage changes. Customers' capacity to afford repricing is a separate question. Life insurers' long-duration obligations can retain exposure long after a nonlife contract would normally renew.
Consider a hypothetical property portfolio with annual covered loss of USD 50 million under an explicit baseline. Under a given scenario the same contracts and sums insured produce expected annual covered loss of USD 75 million. The change is 50%. If management subsequently reduces written exposure so that expected covered loss becomes USD 54 million, the latter is only 8% above the original baseline. The change from 75 to 54 million is attributed to the illustrative scenario's altered portfolio rather than a demonstrated reduction in hazards. A reader needs both numbers to understand business vulnerability and policyholder consequences.
Expected values, extremes and financial position
For a fully specified annual loss random variable L, expected loss E(L) concerns the mean of its distribution. A 99th percentile q0.99(L) describes a severity threshold exceeded in approximately one percent of modeled years under regularity assumptions. The average loss conditional on exceeding a high threshold concerns tail severity; it cannot be reconstructed from the percentile alone. Annual aggregate and single-occurrence distributions can also differ when several events occur during a year.
Gross claims, ceded recoveries and retained losses require a contractual bridge. In a simplified illustration with one USD 100 million covered loss, an applicable USD 60 million excess-of-loss layer attaching at USD 20 million would pay USD 60 million and leave USD 40 million retained, before other adjustments. If a USD 100 million annual aggregate results from five USD 20 million events with an occurrence attachment at USD 20 million, the same layer could pay nothing. The illustration ignores reinstatement, recovery risk, limits, exclusions and expenses. It establishes why an identical annual claims total cannot determine the retained loss without the event structure and treaty terms.
A climate-related capital sensitivity can be affected by asset changes, reserve remeasurement, diversification, counterparty credit risk and the capital measure's definition. Neither a loss amount nor an emissions percentage directly yields an insurer's solvency ratio. A ratio comparison requires the same numerator and denominator or a defensible reconciliation of different capital regimes.
Model revisions and research integrity
Scenario assumptions can change because scientific evidence changes. The NGFS (2025) issued a Phase V physical-risk statement following retraction of a paper underpinning selected long-term physical-damage estimates. Its statement distinguishes outputs depending on that damage function from unaffected outputs, including its short-term scenarios. A version label and model provenance are therefore material components of a report. This monograph does not use the retracted damage estimates as positive evidence.
If a reported series incorporates an affected output, an analyst should determine which variable is derived from the disputed function and when the firm used it. Recalculating that variable with an alternative function changes the observation basis and must be marked as a sensitivity or restatement. It does not automatically invalidate every physical-risk estimate in the report. Local catastrophe models and other scenario components can depend on separate evidence. Conversely, keeping a legacy number without a warning can give a false impression of precision. The scientifically defensible approach retains the old value, identifies its dependency and records the result of any new calculation.
Cross-Border Comparison
Building a comparable population
A comparative dataset should begin with a census of legally eligible entities or a well-defined sampled population at each observation date. Identify firms through stable identifiers, then map them to ownership groups and subsidiaries over time. The sampling frame must distinguish reinsurance, primary insurance, life, nonlife, composite and specialized firms; the balance-sheet and underwriting exposures central to their risk differ.
A public survey with an eligibility threshold disproportionately represents large reporters. A stock-exchange reporting sample represents publicly listed businesses. A prudential return may include institutions whose observations are unavailable outside a regulatory system. A pooled set of these observations describes the intersection or union of chosen samples, not a probability sample of world insurance. Market-wide extrapolation requires separately established coverage and weights.
Define a disclosure participation rate as N_reported/N_eligible for unique entities in an explicit period. Define premium-weighted coverage as the sum of consistently defined premium for reporters divided by the corresponding premium of eligible entities. The first measures institutional participation; the second measures economic coverage within the selected premium population. Neither indicates how much of a reporter's underwriting emissions or hazard exposure is measured. A third coverage variable is needed for the metric's own portfolio perimeter.
Illustrative aggregation and reversals
Suppose two markets are each represented by two reporting insurers. The capital and modeled losses in the following table are invented and measured in millions of a common currency under a common scenario.
| Market and firm | Capital | Retained loss | Loss/capital |
|---|---|---|---|
| X, small | 100 | 30 | 30% |
| X, large | 900 | 45 | 5% |
| Y, small | 100 | 10 | 10% |
| Y, large | 900 | 90 | 10% |
The unweighted mean sensitivity is 17.5% for X and 10% for Y. The capital-weighted aggregate is 75/1,000 = 7.5% for X and 100/1,000 = 10% for Y. Both estimates are accurate descriptions of their respective constructed statistics. The apparent reversal is a consequence of weights, not contradictory evidence. The unweighted average describes the typical sampled institution with equal entity weight; the ratio of totals describes aggregate retained loss relative to aggregate capital, subject to common capital measurement and nonoverlapping exposures.
A hypothetical missing exposure introduces a second selection problem. If only X's large insurer reports its scenario loss, the observed capital-weighted ratio is 5% but covers 90% of total capital. The omitted 10% has a disproportionately high sensitivity in the constructed case. The observed average cannot be called the market estimate. Report coverage and sensitivity bounds where credible assumptions support them. If no credible bounds exist, preserve the comparison as incomplete.
Temporal normalization and uncertainty
Cross-border financial data should retain original currency, valuation date and exchange-rate convention. Converting a stock at period end and a flow at an annual average rate may be appropriate for different questions; combining those conversions without disclosure complicates comparisons. Inflation adjustment requires a stated base year and index, and care when losses reflect construction costs rather than broad consumer prices. Premium and claim trends require consistency in gross/net and written/earned or paid/incurred bases.
A technique useful in longitudinal analysis is matched-entity comparison. Hold the population constant across dates as a diagnostic, then compare it with the full eligible population for each date. The difference indicates sensitivity to entry and exit but does not by itself identify causal determinants. Where the reporting instrument changes, a separate version effect must be considered. If firms enter a sample following a new mandate, improvements in report coverage can reflect eligibility expansion rather than changed management practices among incumbents.
Uncertainty has several sources: missing observations, measurement errors, sampling error, scenario-model uncertainty and genuine ambiguity concerning future management decisions. A conventional statistical interval derived from a random sample addresses only some of these. A model-based sensitivity range from alternative damage functions addresses a different problem. Combining them would require an explicit joint probabilistic assumption that is rarely warranted in public disclosures. A useful report identifies each source of uncertainty and makes the basis of any range transparent.
Lessons from empirical literature
Gatzert and Reichel (2022), in their insurer climate awareness study, use a panel of fifty firms with five hundred firm-year observations and account for repeated firm observations with clustered errors in their logistic analysis. The study illustrates the importance of firm characteristics and longitudinal design. Its choice of awareness indicator and historical population cannot be transplanted directly into a 2026 standardized survey. Observed associations are not identified causal effects of implementing a reporting mandate. An empirical replication would require the underlying data and original coding decisions, which this documentary study does not possess.
Kraus (2024), in a climate disclosure text-mining study, reports increased climate-related mentions over the period examined and a relationship between company size and the frequency of relevant terms. The university abstract also reports relatively early development of stranded-asset discussion. These findings motivate controls for report length and firm size. Because the publisher full text was unavailable in this session, this use is explicitly limited to the indexed abstract and bibliographic record; no numerical coefficient or detailed model specification is adopted.
Text length can be an important confounder. In an illustrative corpus, a 100,000-token report with 200 relevant terms yields two mentions per thousand tokens; a 20,000-token report with 80 terms yields four mentions per thousand. Absolute frequency ranks the former higher; relative frequency ranks the latter higher. Neither determines the economic relevance or precision of each disclosure. The appropriate text analysis combines normalized frequency with manual assessment of passages, distinction between risk types and evidence of the specific reporting function.
For a causal evaluation of a mandate, a credible design would need a clearly dated change applied to firms differently, comparison units with plausible pre-intervention trends, and evidence about anticipation and other concurrent reforms. Firms can change ownership and exposure in response to regulation, making selection endogenous. Staggered adoption complicates simple pooled before-and-after inference. The documents examined here support questions and data design for such an evaluation, but they do not supply the microdata necessary to estimate one.
Integrated Scientific Synthesis and Future Research
Interpreting equivalence
The documentary comparison supports three kinds of judgment in ordinary comparative practice. Some records can be compared directly when they measure the same phenomenon under sufficiently aligned definitions. Some can be reconciled where the differences and necessary transformations are observable. Others support only qualitative or contextual comparison because a necessary component of the economic measurement is absent. These are practical distinctions for this study, not a newly proposed theory or universal scoring system.
For example, the existence of a documented board oversight process can be examined across different disclosure frameworks, provided that criteria for identifying the process are defined. A monetary loss metric can be reconciled between currencies if the underlying time, valuation and exposure bases agree. But an insurance-associated emissions figure cannot be converted into retained catastrophe loss using currency normalization. Nor can the absence of a confidential return in a public data repository be converted into evidence that the insurer did not assess the associated risk.
The international framework is most valuable where it reduces semantic ambiguity and duplicated extraction while preserving local details needed for financial interpretation. Harmonization should make distinctions visible rather than suppress them. Insurance exposures differ substantially in duration, spatial distribution, coverage and risk transfer. Consequently, a common core of observations can support a comparison only when specialized extensions retain characteristics material to the result.
Implications for supervision and market competition
Climate reporting creates costs for collection, consistency review and modeling. Larger groups may spread those costs across several subsidiaries, while small insurers can be constrained by external model subscriptions and limited internal data teams. Proportionate requirements can reduce onerous work. However, poorly defined exceptions can produce a dataset in which missingness correlates with organizational resources. A useful supervisory design distinguishes reduced scope or frequency from an unobserved or low level of exposure.
Competition affects the meaning of a firm's response to climate risk. Premium increases, different deductibles and coverage restrictions may protect a company from retained risk, while shifting costs toward customers or other parts of the financial system. A strong disclosure of operational preparedness can coexist with a growing protection gap. Evaluating the market as an institution requires observation of what happens to risks outside an individual insurer's book. The protection gap itself has several denominators: total economic losses, eligible assets, potential households or businesses, or estimated exposures for a specified peril. These cannot be substituted for one another.
Supervisory data sharing offers a way to reduce duplication for internationally active groups. The IAIS application paper discusses group cooperation and integrated reporting. A minimum common description of the original legal entity and metric boundary would facilitate coordination without requiring identical prudential regimes. Confidential information can be analyzed by authorized supervisors while public aggregates are produced under safeguards. Public limitations should be disclosed to independent researchers; aggregate disclosure that leaves the population or method unidentified may be less useful than a carefully defined narrower statistic.
Practical implications for reporting systems
A durable reporting system should keep source data and transformation lineage. A metric definition must remain linked to its jurisdiction, year, version, contract and asset classification, value and unit, population, missingness code, and assumptions. Changes in a material measure should be explained by a reconciliation identifying additions and removals from the perimeter, changes in estimations and economic changes. Where the components cannot be calculated reliably, the limitation should be stated.
A machine-readable template is beneficial when it expresses these distinctions consistently. It is not a remedy for undefined concepts. A perfectly valid machine-readable file can contain an incomparable figure. The researcher should first establish the meaning of a field, then evaluate its representation in the data format. Automated validity checks ought to detect duplicate entity-period keys, impossible coverage rates, inconsistent numerators and denominators, unexpected unit changes and unmapped template versions. The audit trail should remain reproducible if original definitions change.
Assurance supports confidence within the engagement's defined scope. A limited-assurance conclusion about particular sustainability information does not constitute validation of every scenario model or proof of future financial resilience. Users should identify the assurance provider, reporting period, level and specific information covered. Model validation, source-data verification and external sustainability assurance are different exercises. Their presence can strengthen interpretation when their precise objects are visible.
Future empirical investigations
A first feasible research project would construct a matched longitudinal dataset of insurer groups and legal entities from a fixed set of public reporting populations. Establish a version-aware taxonomy of disclosure observations. Measure prevalence by legal entity, group, premium and metric coverage, and test robustness to alternative weights. The dataset should preserve source document locations and publicly reproducible transformations.
A second project would investigate the relationship between comparable exposure measures and observed claims or coverage outcomes. It requires policy or exposure records with geographic and peril-specific detail, contract conditions, event data and financial observations. A multilevel design could separate local hazard differences, insurer portfolio choices and reporting-regime characteristics if sufficient observations exist. Data confidentiality, nonrandom insurance take-up and changes in cover are substantial research constraints.
A third project would examine scenario sensitivity across insurers using matched model releases and explicit fixed-portfolio and management-response assumptions. It would record scientific corrections and their effect on reported variables. Reporting a range of results under clearly stated assumptions would be more defensible than a single universal risk score. Comparative researchers can then examine whether differences are attributable to exposure, business response or methodological choices.
Conclusion
Climate risk disclosure comparability emerges from alignment of economic meaning, population, boundary, measurement and time. International institutions have advanced the availability of climate-related reporting by promoting common themes, sector-specific information and supervisory learning. The documentary cases show the value of shared language and the continuing importance of legal scope and measurement design. A common heading is a starting point; common measurement depends on the actual underlying observation.
Insurance is a distinctive subject because climate-related financial effects travel through contractual terms, assets, underwriting decisions, reinsurance and market participation. Emissions accounting, physical exposure, expected losses, extreme losses and solvency effects answer different questions. A reduction of a firm's reported risk can occur through genuine risk mitigation, transfer to other institutions, portfolio withdrawal or methodological change. The difference is central to interpreting market resilience.
Cross-border evidence is strongest where the original population and measurements can be reconstructed, appropriate denominators selected, and scenario assumptions and version changes explicitly documented. Where observations cannot be aligned, the analysis retains contextual comparison and uncertainty rather than inventing harmonized figures. The original contribution of this dossier is the integration of institutional reporting analysis with contractual and quantitative interpretation of insurance exposure. It offers a reproducible rationale for determining which inferences are supported by disclosed information. Its documentary evidence and illustrative calculations do not constitute a new theory, a causal assessment of disclosure regulation, a global insurer ranking, or an independent actuarial model.
References
International Sustainability Standards Board. (2023, June). IFRS S2 Climate-related Disclosures. IFRS Foundation. Standard description and history. Source record.
International Sustainability Standards Board. (2025, December 11). ISSB issues targeted amendments to IFRS S2 to support implementation. IFRS Foundation. Source record.
International Association of Insurance Supervisors. (2025, April). Application Paper on the Supervision of Climate-Related Risks in the Insurance Sector. Basel: IAIS. 126 pp. Source record.
California Department of Insurance. (n.d., accessed November 19, 2026). NAIC Climate Risk Disclosure Survey Results. Public survey portal. Source record.
Office of the Superintendent of Financial Institutions. (2025). Climate Risk Management: Guideline B-15. Government of Canada. Annex 2. Source record.
Office of the Superintendent of Financial Institutions. (2025). Business Specifications for the Climate-Related Risk Returns for Insurers. Government of Canada. Source record.
European Commission. (2023, July 31). Commission Delegated Regulation (EU) 2023/2772 supplementing Directive 2013/34/EU as regards sustainability reporting standards. Official Journal of the European Union. Source record.
European Parliament and Council of the European Union. (2026). Directive (EU) 2026/470 amending Directives 2006/43/EC, 2013/34/EU, (EU) 2022/2464 and (EU) 2024/1760 as regards certain corporate sustainability reporting and due diligence requirements. Official Journal of the European Union. Source record.
European Commission. (2026, July 3). Commission Delegated Regulation (EU) 2026/1563 amending Delegated Regulation (EU) 2023/2772 as regards sustainability reporting standards. Official Journal of the European Union. Source record.
European Insurance and Occupational Pensions Authority. (2022, August). Application Guidance on Climate Change Materiality Assessments and Climate Change Scenarios in ORSA. EIOPA. Source record.
Australian Accounting Standards Board. (2024, September). AASB S2 Climate-related Disclosures. Original version, reporting periods beginning on or after January 1, 2025, and before January 1, 2027. Source record.
Bank of England. (2022, May 24). Results of the 2021 Climate Biennial Exploratory Scenario (CBES). London: Bank of England. Source record.
Network for Greening the Financial System. (2025, December). Statement Regarding Physical Risk Estimates in Phase V of NGFS Long-Term Scenarios. Source record.
Partnership for Carbon Accounting Financials. (2025). The Global GHG Accounting and Reporting Standard for the Financial Industry. Parts A and C, third and second editions respectively. Official standards index. Source record.
Gatzert, N., & Reichel, P. (2022). Awareness of climate risks and opportunities: Empirical evidence on determinants and value from the U.S. and European insurance industry. The Geneva Papers on Risk and Insurance—Issues and Practice, 47(1), 5–26. doi:10.1057/s41288-021-00227-5. Source record.
Kraus, A. (2024). A text mining analysis of European banks' and insurers' disclosures on climate-related risks. Risk Management and Insurance Review, 27(3), 257–286. doi:10.1111/rmir.12268. Source record.
Swiss Re. (2026). Sustainability Reporting: Reporting Scope and Frameworks for the Sustainability Report 2025. Corporate reporting website. Source record.
The Travelers Companies, Inc. (n.d., accessed November 19, 2026). Past Reports. Sustainability reporting archive. Source record.
The Travelers Companies, Inc. (2025). Travelers Task Force on Climate-related Financial Disclosures Report 2024. pp. 14, 28, 30. 2024 TCFD Report.
Swiss Re. (2026). Annual Report 2025: Sustainability Report. Climate Transition Plan and Climate Targets and Metrics, pp. 52, 61. 2025 Sustainability Report.
National Association of Insurance Commissioners. (n.d.). Climate Risk and Resiliency. Center for Insurance Policy and Research. Accessed November 19, 2026. Climate Risk and Resiliency.