Case study: Climate risk stress testing & scenario regulation — a city or utility pilot and the results so far

When the New York State Department of Financial Services (NYDFS) published its final guidance on climate risk stress testing in November 2024, it triggered one of the most ambitious regulatory pilot programs in US financial history. Fourteen domestic systemically important banks and insurers were required to conduct forward-looking climate scenario analyses using prescribed physical and transition risk pathways, submit detailed results within 18 months, and demonstrate how findings informed capital allocation, underwriting, and strategic planning. The pilot, which completed its first reporting cycle in May 2026, has produced granular data on how climate hazards interact with financial portfolios and revealed both the promise and the limitations of scenario-based regulation in practice.

Why It Matters

Climate risk stress testing has moved from a voluntary exercise for a handful of progressive institutions to a regulatory mandate shaping capital flows across the US financial system. The Federal Reserve's pilot climate scenario analysis in 2023 covered six of the largest US bank holding companies and produced directional findings, but regulators acknowledged that methodology gaps prevented prescriptive supervisory conclusions. NYDFS advanced the approach by requiring standardized scenarios, consistent time horizons, and comparable output metrics that allow cross-institutional comparison for the first time in a US regulatory context.

The financial exposure at stake is substantial. According to the Federal Reserve's Financial Stability Report (2025), US bank holding companies hold approximately $2.4 trillion in commercial real estate loans, $1.8 trillion in residential mortgages, and $340 billion in energy sector exposures, all categories with significant climate sensitivity. Insurers supervised by NYDFS underwrite roughly $600 billion in annual premiums across property, casualty, and life lines where physical climate risks directly affect loss ratios. Without rigorous stress testing, these exposures remain opaque to both supervisors and investors, creating systemic vulnerabilities that traditional risk management frameworks were not designed to capture.

The NYDFS pilot also matters because it established a template that other state regulators are actively studying. California's Department of Insurance launched its own climate risk assessment mandate in January 2026, drawing directly on NYDFS methodology. The National Association of Insurance Commissioners (NAIC) Climate Risk Disclosure Survey, updated in 2025, incorporated stress testing elements informed by the New York experience. For investors evaluating financial sector exposures, understanding how these pilots work and what they reveal has become essential for portfolio risk management.

Key Concepts

Physical Risk Scenarios model the financial impact of acute and chronic climate hazards on asset portfolios and insurance liabilities. The NYDFS pilot prescribed three physical risk pathways aligned with IPCC Shared Socioeconomic Pathways: SSP1-2.6 (orderly transition, approximately 1.8C warming by 2100), SSP2-4.5 (middle of the road, approximately 2.7C), and SSP5-8.5 (high emissions, approximately 4.4C). Each scenario specifies hazard intensities for hurricanes, flooding, wildfire, extreme heat, and sea level rise across defined geographic zones, enabling institutions to estimate direct asset losses, insurance claims, and collateral impairment under each pathway.

Transition Risk Scenarios assess how policy, technology, and market shifts associated with decarbonization affect loan portfolios, investment holdings, and underwriting profitability. The pilot adopted the Network for Greening the Financial System (NGFS) scenarios, requiring analysis under "orderly transition" (gradual carbon pricing reaching $150 per ton by 2050), "disorderly transition" (delayed action followed by sharp policy intervention after 2030), and "hot house world" (minimal additional policy). Institutions modeled impacts on carbon-intensive sectors including fossil fuel extraction, utilities, transportation, and real estate through revenue impairment, stranded asset write-downs, and shifting demand patterns.

Climate Value-at-Risk (CVaR) quantifies the potential portfolio loss attributable to climate factors over specified time horizons. Unlike traditional VaR calculated over days or weeks, CVaR operates over multi-decade horizons that match climate projection timescales. The NYDFS pilot required CVaR calculations at 2030, 2040, and 2050 time horizons for both physical and transition risks, providing supervisors with a standardized metric for cross-institutional comparison. This approach acknowledges that climate risks are fundamentally different from market risks: they are non-stationary, fat-tailed, and spatially correlated in ways that historical data alone cannot capture.

Sectoral Sensitivity Analysis breaks portfolio exposures into climate-sensitive categories and models how each responds to scenario conditions. The pilot defined 22 sectors using NAICS codes, with prescribed sensitivity coefficients for physical and transition risks. High-sensitivity sectors (fossil fuels, agriculture, coastal real estate) received detailed bottom-up analysis, while moderate-sensitivity sectors (manufacturing, transportation, hospitality) used standardized top-down approaches. This tiered methodology balanced analytical rigor with practical feasibility for institutions managing diverse portfolios.

Pilot Design and Implementation

The NYDFS pilot launched formally in March 2024 with the release of Technical Guidance Bulletin No. 2024-3, a 147-page document specifying scenarios, methodologies, output templates, and submission requirements. Fourteen participating institutions included four bank holding companies with combined assets exceeding $3.2 trillion and ten insurance groups writing approximately $280 billion in annual premiums. Each institution appointed a dedicated climate risk officer reporting directly to the chief risk officer, with cross-functional teams averaging 12 to 18 professionals drawn from risk management, actuarial, investment, and sustainability functions.

The implementation timeline spanned 18 months, divided into three phases. Phase 1 (March to September 2024) focused on data infrastructure, requiring institutions to geocode physical asset exposures, map counterparty revenues to climate-sensitive sectors, and establish connections to hazard databases including First Street Foundation's flood and wildfire models, Moody's RMS hurricane models, and NASA's downscaled climate projections. Phase 2 (October 2024 to March 2025) involved running prescribed scenarios through internal models and producing preliminary results for supervisory review. Phase 3 (April to August 2025) required institutions to refine models based on supervisory feedback, conduct sensitivity analyses, and submit final reports with board-approved management responses.

Data requirements proved to be the most resource-intensive element. Banks reported spending an average of 4,200 staff hours on data remediation alone, primarily geocoding commercial real estate collateral that legacy systems tracked only by zip code rather than precise latitude and longitude. Insurers faced comparable challenges mapping policy locations to hazard zones at sufficient resolution. Several institutions contracted with specialized climate analytics firms including Jupiter Intelligence, Moody's Analytics, and Rhodium Group to fill internal capability gaps.

Measured Outcomes

The pilot's first reporting cycle produced several findings with direct implications for investors and portfolio managers.

Physical risk losses under the SSP5-8.5 scenario averaged 6.2% of total bank commercial real estate portfolios by 2050, with concentration in coastal Florida (12.8% loss rate), the Gulf Coast (9.4%), and wildfire-prone California (7.1%). These figures exceeded internal risk estimates that most institutions had previously generated by a factor of 2.5 to 3.8, primarily because the standardized scenarios required modeling compound events (simultaneous hurricane and flooding) and chronic stressors (sea level rise plus subsidence) that prior analyses had treated independently.

Transition risk impacts were most pronounced in energy sector exposures. Under the disorderly transition scenario, bank lending portfolios to fossil fuel companies experienced modeled credit losses of 14 to 22% by 2040, driven by revenue declines, asset impairment, and refinancing risk as capital costs increased. Utility sector exposures showed bifurcated outcomes: utilities with diversified generation portfolios and credible transition plans experienced modest losses of 3 to 5%, while coal-dependent utilities faced losses exceeding 30%.

Insurance underwriting profitability showed significant geographic redistribution under physical risk scenarios. Property insurers modeled premium adequacy gaps of 15 to 40% in high-hazard zones under SSP2-4.5 by 2035, meaning current pricing does not reflect expected losses even under moderate warming pathways. This finding reinforced trends already visible in markets like Florida and Louisiana, where private insurers have been withdrawing coverage and state-backed insurers of last resort have absorbed growing portfolios.

Cross-institutional comparison revealed substantial methodology dispersion. For identical asset exposures, modeled losses varied by a factor of 2 to 4 across institutions, driven by differences in hazard model selection, damage function calibration, and assumptions about adaptation and insurance recovery. NYDFS supervisors flagged this dispersion as a priority for harmonization, noting that inconsistent methodologies undermine the comparability that standardized scenarios were designed to achieve.

Challenges and Lessons Learned

Model uncertainty remains substantial. The pilot exposed fundamental limitations in translating climate science into financial risk metrics. Damage functions relating hazard intensity (wind speed, flood depth, temperature extremes) to asset losses are calibrated on historical events that may not represent future conditions. Correlation structures between hazards, sectors, and geographies are poorly understood. Participants reported that model uncertainty dominated results for time horizons beyond 2040, limiting the precision of long-term capital planning.

Data granularity gaps persist. Despite significant investment, most institutions could not achieve property-level physical risk assessment across their full portfolios. Approximately 35% of commercial real estate collateral and 25% of insurance policy locations could only be mapped to census tract or zip code level, introducing significant spatial averaging that obscures localized risk concentrations. Residential mortgage portfolios performed better due to existing property appraisal data, but even these lacked elevation data critical for flood risk assessment.

Board engagement varied widely. Institutions where boards actively engaged with scenario results and challenged assumptions produced more credible analyses and more substantive management responses. Institutions where climate risk remained delegated to compliance functions produced technically adequate but strategically disconnected submissions. NYDFS supervisors noted that board-level climate literacy correlated strongly with the quality and actionability of stress test outcomes.

Integration with business decisions lagged. Only four of fourteen participants demonstrated concrete examples of stress test results influencing actual lending, underwriting, or investment decisions during the pilot period. The majority treated the exercise as a regulatory compliance obligation rather than a strategic planning input. This gap between analysis and action represents the central challenge for the next phase of climate risk regulation.

Cost was significant but manageable. Participating banks reported total pilot costs of $8 to $15 million each, including personnel, external data, consulting support, and technology infrastructure. Insurers reported $5 to $10 million. While substantial, these costs represent less than 0.1% of annual revenues for the participating institutions and are expected to decline by 30 to 50% for subsequent reporting cycles as infrastructure and processes mature.

Key Players

NYDFS Sustainable Finance Division designed and administered the pilot, drawing on expertise from seconded staff from the Bank of England's Prudential Regulation Authority, which conducted pioneering climate stress tests in 2021. Superintendent Adrienne Harris positioned the initiative as a model for federal adoption.

Jupiter Intelligence provided hyperlocal physical risk analytics to eight of fourteen pilot participants, offering forward-looking hazard projections at property-level resolution for flood, wind, heat, and wildfire under multiple climate scenarios.

Moody's Analytics supplied transition risk modeling tools integrating NGFS scenarios with credit risk frameworks, enabling institutions to translate macroeconomic pathway assumptions into counterparty-level probability of default and loss-given-default estimates.

Rhodium Group contributed economic impact modeling that translated physical hazard projections into GDP, employment, and property value impacts at county level, providing context for portfolio-level loss estimation.

Action Checklist

• Review NYDFS Technical Guidance Bulletin No. 2024-3 for methodology details applicable to your institution or portfolio
• Assess data infrastructure readiness, particularly geocoding precision for physical asset and collateral locations
• Evaluate whether internal risk models incorporate compound climate events and chronic stressors, not just acute hazards
• Benchmark physical risk exposure concentrations against pilot-reported loss rates for comparable geographies and asset classes
• Engage boards and investment committees on climate scenario results with focus on strategic implications, not just compliance
• Map fossil fuel and carbon-intensive sector exposures against disorderly transition scenario assumptions for 2030 to 2040 horizons
• Establish relationships with climate analytics providers to fill internal capability gaps before regulatory deadlines
• Monitor California Department of Insurance and NAIC developments for expanding state-level stress testing requirements

FAQ

Q: How does the NYDFS pilot differ from the Federal Reserve's 2023 climate scenario analysis? A: The Fed's 2023 exercise was exploratory, covering six banks with internally developed methodologies and no prescribed scenarios, producing qualitative findings without supervisory consequences. The NYDFS pilot is prescriptive, specifying standardized scenarios, output metrics, and submission templates for fourteen institutions, with results informing supervisory assessments and potential enforcement actions. The NYDFS approach enables cross-institutional comparison that the Fed's pilot could not achieve.

Q: What is the estimated cost for a mid-size bank or insurer to prepare for climate stress testing requirements? A: Based on pilot participant reporting, expect $5 to $15 million for the first cycle, with 40 to 50% allocated to data infrastructure, 25 to 30% to personnel and consulting, and 20 to 25% to technology and modeling tools. Subsequent cycles should cost 30 to 50% less as infrastructure matures. Institutions with existing ESG data programs and geocoded asset databases will see lower costs.

Q: Are climate stress test results publicly disclosed? A: NYDFS published aggregated, anonymized results in its December 2025 supervisory summary but did not disclose institution-level results. However, several participating institutions voluntarily published summaries in their annual reports and TCFD disclosures. Investor pressure for greater transparency is increasing, and future regulatory cycles may require public disclosure of institution-level results, mirroring the approach used for traditional bank stress tests under CCAR.

Q: How should investors use climate stress test disclosures in portfolio construction? A: Focus on four elements: geographic concentration of physical risk exposures, particularly in high-hazard coastal and wildfire zones; sector concentration in carbon-intensive industries under transition scenarios; the magnitude of gap between current internal risk estimates and standardized scenario results; and the quality of management response, specifically whether stress test findings are influencing actual capital allocation, underwriting, and lending decisions.

Q: Will federal regulators adopt the NYDFS approach? A: The Federal Reserve has signaled interest in expanding its climate scenario analysis program but has not committed to prescriptive stress testing requirements. The SEC's climate disclosure rules, effective for large accelerated filers in 2026, require scenario analysis disclosures but do not prescribe specific methodologies. The most likely near-term trajectory is continued state-level expansion (California, Connecticut, and Washington have active initiatives) with federal harmonization following once methodology consensus solidifies.

Sources

• New York State Department of Financial Services. (2024). Technical Guidance Bulletin No. 2024-3: Climate Risk Stress Testing for Regulated Financial Institutions. Albany, NY: NYDFS.
• Federal Reserve Board. (2025). Financial Stability Report, November 2025. Washington, DC: Board of Governors of the Federal Reserve System.
• Network for Greening the Financial System. (2024). NGFS Climate Scenarios for Central Banks and Supervisors, Phase IV. Paris: NGFS Secretariat.
• Jupiter Intelligence. (2025). ClimateScore Global: Methodology and Validation Report. San Mateo, CA: Jupiter Intelligence Inc.
• Moody's Analytics. (2025). Climate Risk Scenario Analysis: Integrating NGFS Pathways with Credit Risk Models. New York: Moody's Corporation.
• National Association of Insurance Commissioners. (2025). Climate Risk Disclosure Survey: 2025 Update and Analysis. Kansas City, MO: NAIC.
• Bank of England Prudential Regulation Authority. (2022). Results of the 2021 Climate Biennial Exploratory Scenario. London: Bank of England.