Paradigm shift to more reliable climate risk assessment
Abrams et al. (2025) have reported that economic models used by governments, central banks, and investors are increasingly understating physical climate risk because they rely on assumptions that break down as the world moves toward higher levels of warming. Using expert judgment from 68 climate scientists, they find that many economic models are failing to capture the extreme events, compounding shocks, and rising uncertainty that are likely to dominate impacts in a hotter world. They also show how closer alignment between scientific estimates and economic modeling of physical risks may be possible as the world moves towards a temperature increase of 2°C.
The report was led by the University of Exeter’s Green Futures Solutions team, in partnership with Carbon Tracker Initiative and funding from the Aurora Trust. It synthesises expert judgement from 68 climate scientists, who were consulted through a combination of survey and workshop approaches. Together, they represent universities, research institutions, and government agencies from 12 counties (USA, UK, Germany, Australia, France, China, Netherlands, Spain, Norway, Canada, Austria and Sweden). The results were collected anonymously in conformance with standard practice for expert elicitation exercises. This briefing summarises the findings that are described in Adams et al. (2025) and Carbon Tracker (2025).
The authors conclude that their expert elicitation reveals a fundamental disconnect between current economic models and climate scientists’ understanding that beyond 2°C they are not dealing with manageable economic adjustments that assume that the future will behave like the past. Consequently, current economic models systematically underestimate climate damage because they cannot capture what matters most: the cascading failures, threshold effects, and compounding shocks that define climate risk in a warmer world and which could undermine the foundations of economic growth.
Given the presence of this disconnect, it seems that the term “recalibration” in the title of the report is better described as a “paradigm shift,” which is indeed identified but not until page 53 of the report. When the authors argue against the core assumption in many economic models that economic growth under climate change continues indefinitely, merely at reduced rates, they imply that the current approach is based on recalibration, undermining the call for recalibration in the title of the report.
The authors conclude that flaws in economic modelling have drawn attention in recent years, with influential models criticised by UK actuaries and scientists for understating climate impacts that many scientists now anticipate. The report aims to reduce that shortcoming by establishing early consensus on how to best improve those estimates and calling for closer collaboration between climate scientists and economists. The result is an assessment that examines in detail how uncertainty is treated, the extent to which GDP-based estimates remain meaningful, and what this means for financial regulators and investors as the world moves toward 2°C.
The survey results demonstrate a fundamental disconnect between what climate scientists understand about climate impacts and how these impacts are represented in economic models. The respondents expressed deep concerns about oversimplified temporal dynamics, limited spatial resolution, inadequate treatment of tipping points, and the failure to capture extreme events and cascading risks. Addressing these limitations requires research investments spanning years, yet the window for preventing catastrophic warming is narrower, indicating that policy action cannot await perfected models but must proceed on the basis of precautionary risk management, physical climate science, and observed impacts.
The main conclusions of the report are summarised under the following eight topics, interspersed with notes on our experience at Risk Frontiers. This is followed by commentary on how our practice in natural hazard catastrophe loss modelling has addressed many of the concerns epressed by Abrams et al. (2025).
Climate damages are structural and compounding, not marginal
At higher levels of warming, climate impacts are increasingly likely to disrupt multiple sectors at once, as physical risks cascade across trade, finance, and migration. These non-linear, structural impacts are expected to increasingly reshape entire economies and undermine the conditions necessary for economic growth.
This undercuts a core assumption in many economic models, which assume that economic growth continues indefinitely, merely at reduced rates.
Extremes, not averages, define the future of climate damages
While economic modelling has traditionally linked damages to changes in global mean temperature, societies and markets experience climate change through local and regional extremes, such as heatwaves, floods and droughts, which drive the bulk of economic and financial disruption while often barely registering in global averages. The reliance on smooth polynomial relationships between global mean temperature and aggregate GDP obscures the mechanisms through which climate impacts are actually manifested: destroyed capital stock, degraded labour productivity, disrupted supply chains, and cascading system failures.
Gross domestic product (GDP) underestimates the full extent of harms
GDP can mask the full extent of harms by failing to account for impacts on mortality and morbidity, inequality, ecosystem loss and social disruption: all factors that undermine societal, human and economic health. As these risks rise, continuing to rely on GDP-based assessments can give policymakers and financial institutions a false sense of resilience even as underlying vulnerability increases. Several respondents urged replacing GDP-based “black box” functions with non-GDP-centred frameworks that integrate multiple dimensions of welfare, inequality, and human security. The GDP measurement problem is particularly acute: natural disasters often increase measured GDP through reconstruction spending while actual prosperity declines. This systematic bias means econometric calibration to historical data will perpetuate underestimation.
In our experience, official recovery narratives don’t match lived experience. Recovery is often described in economic terms, but Risk Frontiers’ surveys show a different reality on the ground: long-term health impacts, loss of connection to place, permanent out-migration, and environmental, business and farming damage that affects future livelihoods. When these impacts are overlooked, communities feel invisible in decision-making, leading to frustration, mistrust, and a gradual erosion of social cohesion.
Scale and heterogeneity matter fundamentally
Respondents consistently emphasised that oversimplified temporal dynamics and limited spatial resolution are critical limitations. Survey responses called for bottom-up, process-based models at appropriate spatial scales – grid-cell or regional rather than global averages – and with temporal resolution distinguishing seasonal impacts, gradual versus abrupt changes, and slow-onset versus rapid-onset hazards. The distributional consequences are particularly concerning. Even if global GDP losses appear moderate, catastrophic damage concentrated on vulnerable populations create humanitarian crises, migration pressures, and security threats that aggregate functions miss. This reveals a critical feedback loop: first-round effects on GDP may seem manageable in aggregate, but by failing to capture capital destruction, labour productivity losses, ecosystem degradation, and institutional breakdown, standard models miss the devastating second-round effects. These cascading impacts – supply chain failures, credit market stress, insurance withdrawal, and governance collapse – amplify initial shocks far beyond what aggregate GDP metrics suggest, transforming seemingly moderate first-round impacts into compounding economic crises.
In our experience, repeated disasters are wearing communities down. Many NSW East Coast communities are no longer recovering from one disaster before the next occurs. In numerous communities, there are examples of disasters where a combination of events -floods, bushfires, and storms have stacked up, creating disaster fatigue. Risk Frontiers repeatedly hears about people being displaced multiple times, leading to a heavy reliance on volunteers who are themselves burning out, and rising anxiety and mental health stress, all of which weaken community resilience over time.
Expanding scope beyond aggregate damage functions
Some impacts cannot be adequately captured within aggregate damage functions and require complementary approaches. Priority actions identified by the respondents include establishing a working group of climate scientists, social scientists, and economists to build consensus on synthesizing extreme events, human mortality, and tipping point risks; gathering empirical evidence through calibration to observed impacts and validation against historical events; and determining how to present these analyses to decision-makers.
In our experience, risk is shifting from systems to households and businesses. As insurance becomes unaffordable or unavailable, risk is increasingly being borne by households and businesses. Property values fall, rebuilding slows, and people feel unfairly penalised for living in places they may not be able to leave. This shift undermines confidence in insurance, government, and recovery systems, and fuels a growing sense of injustice. Places like Taree, where businesses have flooded three times in two years, illustrate how unpredictability and lack of insurance are leading some businesses to close permanently, with longer-term impacts on local economies.
Key questions not resolved by the respondents include whether process-based approaches can be incorporated into damage functions or must operate alongside them; how stochastic extreme event projections should be visualized; and what communication methods best convey cascading risks to financial institutions. They outlined a medium-term timeline, up to five years for consensus-building, and empirical validation, though progress on individual components can proceed in parallel.
Uncertainty rises sharply with warming
With temperatures trending towards a 2°C future, impacts become increasingly unpredictable as tipping points and tail risks increase. Even as models continue to produce seemingly precise point estimates, climate impacts will probably undermine the assumptions of continuous growth fundamental to many economic models. Policymakers should be wary of climate scenarios extending beyond certain temperature levels and take account of tail risks.
In our experience, insurance is adding to uncertainty, not reducing it. Insurance is intended to provide certainty after disasters, but for many flood-affected communities it is now doing the opposite. As risks become harder to predict, insurers are pulling back. Risk Frontiers’ community surveys consistently show premiums doubling or tripling, flood cover being excluded, and widespread underinsurance, leaving households uncertain about whether they will be protected in future events. Under-insurance is also well documented from our post-disaster bushfire work where risk mitigation policies, such as Building Attack Level (BAL) ratings, impact the ability to rebuild.
Recommendations for financial regulators and supervisors
Climate risk as a financial stability issue
Abrams et al. (2025) see strong evidence that climate change amplifies the traditional drivers of financial instability, including macroeconomic downturns, geopolitical tensions, supply-chain disruptions, and destruction of human and physical capital. These interactions occur even when global averages might appear moderate. Regional extremes, temporal clustering of shocks, and compounding effects can generate system-wide stress disproportionate to headline GDP impacts. This supports the view that climate risk is not merely a micro-prudential concern, but a core threat to long-term financial stability.
Implications for stress testing and supervision
Abrams et al. (2025) identify several limitations in current supervisory climate stress tests:
- Over-reliance on mean temperature pathways.
- Use of smooth damage functions that suppress tail risk.
- Point estimates that mask deep structural uncertainty.
To address these weaknesses, Abrams et al. (2025) suggest reporting ranges rather than single outcomes in climate stress tests, testing resilience across multiple plausible damage trajectories, and explicitly incorporating non-linearities, thresholds, and compounding mechanisms where possible. Where modelling cannot reliably quantify outcomes, particularly at higher warming levels, supervisors are encouraged to acknowledge limits explicitly rather than allowing false precision to shape risk perception.
Tail risk and prudential risk management
The respondents indicated that low-probability, high-impact outcomes dominate climate risk. From a prudential perspective, this strongly suggests that median outcomes are insufficient guides to stability alone, as even small probabilities of catastrophic loss warrant attention. Accordingly, climate supervision should align with actuarial approaches to ruin risk, taking the precautionary principle. The report reinforces the rationale for supervisory approaches that treat climate change analogously to other sources of systemic tail risk, where the objective is not to price risk accurately, but to prevent destabilising outcomes.
Recommended research to improve damage modeling
Respondents called for greater integration of political, geopolitical, and societal dynamics into climate damage modelling, deeper connections between empirical evidence and complex systems analysis, and a new generation of interdisciplinary frameworks linking Earth-system and socioeconomic models. They also urged that modelling communities amplify heterodox and cross-disciplinary perspectives to better capture social, ecological, and economic heterogeneities.
Respondents emphasised the need for better integration of alternative climate metrics beyond temperature, explicit modelling of tipping points and cascading effects, and improved process-based impact channels. Cross-disciplinary collaboration and enhanced representation of extreme events also ranked highly, as shown in Figure 1.
How our practice in natural hazard catastrophe loss modelling has addressed many of these concerns
Risk Frontiers’ work over more than 30 years in natural hazard catastrophe loss modelling speaks directly to many of the gaps identified in Abrams et al. (2025). Our modelling has always been grounded in the physical characteristics of extreme events — floods, bushfires, tropical cyclones, severe convective storms and heatwaves — rather than relying solely on smooth relationships between global mean temperature and economic output. By combining event-based catastrophe models with high-resolution hazard data, detailed exposure information and vulnerability functions, we focus on the extremes that drive loss, rather than averages that can obscure it.
Our expertise in weather and climate datasets — including regional climate model simulations and stochastic event sets — allows us to explore how changing hazard frequency and intensity alter risk distributions over time. In this sense, our approach is inherently process-based: capital destruction, supply-chain interruption, insurance withdrawal and changing rebuilding conditions are treated as mechanisms, not abstract GDP shocks.
We also recognise the importance of scale, heterogeneity and uncertainty highlighted in the report. Risk Frontiers’ work is typically undertaken at regional and asset-level resolution, enabling us to capture spatial variation in hazard, exposure and vulnerability, and to identify distributional effects that aggregate national statistics may conceal. Our engagement with the research community, application of advanced statistical and machine learning techniques, and development of hazard- and climate-extreme–specific metrics are part of an ongoing effort to better characterise tail risk and compounding events.
Through collaborative, co-designed research projects with clients, we test assumptions transparently and communicate uncertainty explicitly, rather than relying on single-point forecasts. At the same time, we acknowledge that important challenges remain — particularly in modelling deep uncertainty, tipping points, and cascading cross-sector failures at global scale. These are active areas of research, and while no modelling framework can yet fully resolve them, grounding economic and financial analysis in physically realistic hazard processes is a necessary step toward narrowing the gap identified in the report.
REFERENCES
[1] Abrams, J.F., S. Hu and B. D. Bampton (2025).
Recalibrating Climate Risk. Aligning Damage Functions with Scientific Understanding.
[2] Carbon Tracker (2025).
Governments must fix ‘faulty radar’ in economic climate models as storm approaches, scientists warn. Press release, 5 February 2026.
