In a recent speech, Frank Elderson, vice chair of the Supervisory Board at the ECB, noted that banks in Europe had made considerable progress in developing the capability to measure and manage climate and nature risks, enabling them “to start quantifying some capital needs.”[1] He stated:
Climate and nature risks are not just appearing at distant time horizons but are becoming an increasingly immediate concern for financial stability and economic growth. Consider that, in the next five years, extreme weather events could account for a loss of up to 5% of euro area economic output. That would be a shock similar in magnitude to the great financial crisis.
As this note demonstrates, the projection of a massive near-term loss of economic output is based on assumptions about weather disaster scenarios that appear clearly untenable. It comes from a hypothetical scenario, the Disasters and Policies Scenario, put together by the Network for the Greening of the Financial System (NGFS). The NGFS is a global consortium of central banks and other regulators, notably excluding U.S. regulators, who have left the organization.
As the NGFS states on its website:[2]
Users are reminded that neither the NGFS, nor its member institutions, nor any person acting on their behalf, is responsible or liable for any reliance on, or for any use of the NGFS scenarios and/or supplementary documentation. This also applies to the use of the data produced under the scenarios – see section 5 in https://data.ene.iiasa.ac.at/ngfs/#/license. Thus, while the NGFS climate scenarios are certainly a helpful tool, they do not alleviate the responsibility of users, including banks and other (financial) organisations, to design and implement their own risk management frameworks.
Skepticism of NGFS scenarios is warranted. A recent BPI study documented how the NGFS recently put into its long-term scenarios a model that heavily exaggerated the effect of climate change on economic growth, even though it could have or should have known the model is deeply flawed. It proceeded with those scenarios based on a journal article that experienced trouble during peer review and was debunked by three different independent analyses; the article currently carries an advisory. Hence, it is always important for users to independently validate any NGFS climate scenario.
In this note, we review the recent NGFS Disasters and Policies scenario that produces the economic collapse cited in the Elderson speech. While the scenario calibration details are non-transparent and difficult to evaluate, it is possible to estimate the likelihood of the scenario. Even under very conservative assumptions, we find that a civilization-destroying asteroid strike is generally more likely than this NGFS climate scenario. The Disasters and Policies scenario is enormously less likely than the physical risks in the Fed’s 2023 pilot climate scenario analysis, which were calibrated to a once-in-200-year weather event. An effective stress test should be severe, but plausible, but the Disasters and Policies scenario is not statistically plausible. As a result, the NGFS scenario has no risk management value.
The NGFS climate scenario at issue is one of the NGFS’s new short-term scenarios. The NGFS technical documentation explains the details.[3] The scenario assumes that in 2026 a combination of droughts, heatwaves and wildfires hits six regions around the world, followed by a combination of floods and storms in the same six regions in 2027. Extreme weather events affect the economy of each region directly when they strike. Other regions’ economies are then affected indirectly through trade and financial linkages.
Figure 1 shows the profound global economic losses produced by the Disasters and Policies scenario.[4]
Figure 1
North American and EU economies contract about 5 percent from the baseline in 2026 and are still about 3.5 percent below the baseline in 2027. Asia, Africa and South America experience even larger economic declines.
The NGFS assumes that each extreme weather event in the scenario has a 50-year return period,[5] which is the average time between occurrences of a weather event for a specified severity. For example, a 50-year return period for a drought means that a drought of a given severity would be expected to occur once every 50 years.
Since there are three extreme weather events in each region and six regions, the scenario produces 18 50-year return period climate events in 2026. An additional six 50-year return period flood/storm weather events are posited in 2027, one for each region. Thus, the Disasters and Policies Scenario includes 24 extreme — that is, once in 50-year, or with a 2% probability — weather events over a two-year period.
The problem: Too many rare events occur simultaneously
Given 24 independent climate events, each with an annual probability of 2 percent, the probability of them jointly occurring is
which implies a return period of the total scenario on the order of 1041 years. To put this into perspective, in the recent “Big Crunch” cosmological theory,[6] the universe is estimated to end in about 20 x 109 years, well before the first expected occurrence of the NGFS scenario.
We know the correlation is above zero, but we do not know much more than that. The Intergovernmental Panel On Climate Change (IPCC) report on the physical basis of climate change notes that there is strong evidence that it is now more likely that related climate events, such as droughts and heatwaves, will occur together.[7] However, we do not know how much more likely. Ridder et al. analyze the spatial distribution of “hot spots,” areas with a higher probability of joint occurrences of related climate events. [8] However, the rigorous quantification of how much joint probabilities rise is also lacking in that paper. The NGFS technical document acknowledges the issue, saying
The main challenge for the quantitative assessment of these compound events is the limited knowledge of their statistical properties. The current state of the science lacks statistical models of the joint distribution of hazards that could be used to simulate directly compound events in catastrophe risk models, even in the context of scenario-contingent projections. The estimation of tail risk (e.g. return period or quantiles) for multivariate sources of risk requires empirical samples of size that are simply not available, especially in a context of changing climate.
Given the paucity of empirical evidence, the NGFS technical document adopts a “physical climate storyline” approach that ignores the elephant-in-the-room correlation issue. The storyline approach is based on Shepherd et al.,[9] which argues that uncertainties of multiple climate events are so great that it is not possible to assign a probability to them. Shepherd et al. define a storyline as a “physically self-consistent unfolding of past events, or of plausible future events or pathways” that do not have a probabilistic interpretation.
Despite the adoption of a storyline approach, the NGFS technical document nonetheless attempted a limited analysis of the potential correlation. It pointed out that the joint return period of three 100-year return extreme weather events rises rapidly with the gaussian copula[10] correlation. A Gaussian copula is a statistical technique to determine the likelihood that a series of random events occur together. However, showing that the return period of three 100-year return climate events goes up with the Gaussian copula correlation just expresses a mathematical truism; it does not answer the crucial question of what the correlation is.
Not only do we not know what the correlation is, we also do not know whether the Gaussian copula should be used to model the correlation. The NGFS technical document did not mention a t-copula, which allows an even higher degree of correlation of rare events.
But not knowing what the correlation is or what copula should be used did not require the NGFS to throw up its collective hands and use a storyline approach. Below, we implement Gaussian and t-copulas for the 24 events to examine how they might vary with reasonable values of correlation. Doing this analysis can give some clues about the plausibility of the NGFS Disasters and Policies scenario.
What is a copula?
Before performing this analysis, it is helpful to understand intuitively what a copula is. It may be easiest to think of a copula as a statistical machine that decides the likelihood of events that are poured into it. The Gaussian[11] copula is a statistical machine with one dial, correlation. Once you have poured the twenty-four 50-year return period climate risks into the machine, you then select a correlation by turning the dial. The machine then spits out the chance of seeing the 24 climate hazards together. If, for example, the machine calculated the probability to be one in a million years, then we know that we will see the 24 events all happen on average once every million years. The return period is then 1 million years.
The t-copula[12] machine works the same way, except it has two dials: a correlation dial and a degrees-of-freedom (df) dial. The degrees-of-freedom dial affects a property of the machine called “tail dependence.” Higher tail dependence implies that rare events happen together even more often. Rare events cluster together. If you turn the correlation dial up, you get a higher probability of observing the 24 climate events and therefore a lower return period. On the other hand, if you turn the degrees of freedom dial down, you get a higher probability of seeing the events and a lower return period. Turning degrees of freedom down increases the tail dependence[13] for a given correlation.
Simulating the Copulas
We perform 10 million simulations of the Gaussian and t-copulas to derive a joint return period for the Disasters and Policies scenario under various assumptions. For each case, we show the cutoff point in which the NGFS scenario becomes less likely than a civilization-destroying asteroid. Table 1 shows the results.
Table 1
In Table 1, 10 million simulations are not enough to estimate the joint return period of the NGFS Disasters and Policies scenario for smaller correlations. For small correlations, we know the return periods are larger than 10 million years.
Table 1 shows that it is hard to avoid the conclusion that the NGFS Disasters and Policies scenario is probably less likely than a civilization-ending asteroid,[14] which has a return period of about 700,000 years. To avoid that conclusion, we would need to assume that correlation as well as tail dependence are very high, although we have no evidence to support such an assumption. Even if we did make that assumption, Table 1 shows that a t-copula with 10 degrees of freedom and a constant 50 percent correlation assumption yields a return period longer than 36,000 years, which is less absurd than an asteroid strike but still not a scenario on which one would reasonably build a capital charge or risk management changes.
Other copulas are possible too. The NGFS technical document shows, for two climate events, a chart comparing the Gaussian copula to some other copulas[15] — the Clayton copula, the Frank copula and the Gumbel copula. Each copula has a different mathematical structure and thus a different correlation effect.
The NGFS chart reveals that the Clayton and Frank copulas would make the NGFS scenario even less likely than the Gaussian copula, a point the NGFS technical document passes over without comment. The Gumbel copula, like the t-copula, makes the scenario more likely than the Gaussian copula would. The Gumbel copula, however, has only one dial — correlation. Turning up the correlation in a Gumbel copula increases the tail dependence, causing rare events to cluster together.
However, we do not have evidence to think the Gumbel copula should apply, especially to the 24 events. For example, Chen et al. found some evidence for a Gumbel copula in a joint heatwave-drought event, i.e., two events in a very localized area — the Yangtze river valley.[16] On the other hand, Tabari and Willems[17] applied the Clayton, Gumbel, and Gaussian copulas to hot-dry compound events in climate projections over various scenarios and concluded they could be characterized by the Gaussian copula. Evidence on this question is still preliminary.
Scenario Calibration
Given that the Disasters and Policies scenario assumes a set of weather events that jointly arrive, on average, over extremely long return periods, it is not surprising that the estimated economic consequences would be large. However, we have no way of checking since the NGFS is not transparent on the details of exactly how it translates the Disasters and Policies scenario into economic losses. The NGFS transforms the weather events into losses using the GEM-E3[18] world economic model by reducing capital that was damaged by the extreme weather, by making assumptions about output losses, and by also making assumptions about degradation of labor productivity in the model. But we do not know exactly how the NGFS imposed those assumptions into the model or why.
The NGFS also uses an academic model,[19] CLIMACRED-PHYS, to estimate default probabilities and changes in asset values that it adds to the economic damage. Just how they calibrated the assumptions of the CLIMACRED-PHYS model is also unclear. Thus, even though we know the NGFS fed an asteroid-risk scenario into a series of economic models, we cannot validate whether such a scenario would have led to a global depression as the NGFS claims.
Why Are the NGFS Climate Scenarios So Plainly Unrealistic?
A series of BPI reports[20] documented that the damage function adopted by the NGFS in its long-term climate scenarios is a deeply flawed exaggeration of the effects of climate change on real economic output. In addition, Bearpark et al.[21] showed that a mistake in the data in the damage function for a tiny country had inflated the estimates by a factor of about three. Schotz[22] found statistical flaws in the damage function model that were similar to the problems found in the BPI studies.
The NGFS cherry-picked a flawed model with unusually large effects. A broad review of the available evidence suggests that the economic effects of climate change are relatively small[23] — they are a risk for banks, but a relatively small risk. They are certainly not capital-level risks.[24]
The NGFS has repeated this pattern of exaggeration in its development of short-term climate scenarios. As we have seen, the NGFS took a real scientific finding, the evidence discussed by the IPCC that compound climate events have become more frequent, and then, without any empirical evidence, expanded that limited finding into a scenario that borders on the impossible. Why would the NGFS dramatically overestimate risks in both the long-term and short-term climate scenarios?
The answer cannot be that such extreme exaggeration is done for the sake of conservatism. The NGFS refuses to change the damage function in its long-term scenarios despite the evidence that it is simply wrong. Refusing to change a flawed model is not a conservative action. If the NGFS had calibrated its scenario to a once-in-200-year event, as the Fed did in its climate scenario exercise, that would be conservative. A short-term scenario that may be calibrated to a once-in-700,000-year event is not conservative.
The more likely explanation for the exaggeration is that each NGFS short-term scenario also contains an NGFS policy goal. For example, in the Disasters and Policies scenario, the world suffers climate damage not sometime in the distant future, but now, by a tumultuous set of storms so fierce that they bring down the global economy. This outcome justifies imposing immediate capital increases on banks.
In the Diverging Realities scenario, the advanced economies follow the policies the NGFS favors, but the emerging markets and lower-income countries do not. As a result, the emerging markets and lower income countries are punished by the same compound storm events featured in the Disasters and Policies scenario, but with a lower 20-year return period. The policy goal of this scenario seems to be to discourage banks from lending in countries that do not follow the NGFS’s preferred view of climate policy. The “risk” is that banks might, for example, finance real estate projects in countries that do not live up to the NGFS’s policy expectations.
Of course, climate science does not work that way. Greenhouse gas emissions affect the climate of the entire earth, not just the climates of regions. There is no scientific reason why the compound climate events should strike only those regions that do not adopt the climate policy changes that the NGFS favors.
In the Highway to Paris scenario, world governments realize that progress in green technology is not happening fast enough, and so they raise carbon taxes and use the proceeds to invest in green technology, which then improves much more quickly than anyone expected. The policy goal of this scenario is to direct lending towards green technology and away from fossil-fuel dependent companies. The policy goal is justified by the need for banks to mitigate the risk of losses if this scenario “risk” materializes.
The Sudden Wake Up Call Scenario is similarly designed to justify a climate policy the NGFS wants to see. In this scenario, a sudden event causes the world to wake up to the risks of climate change. The world lurches haphazardly towards green policies. In that case, banks that are not already lending to green technologies and prepared to lend more would be caught flatfooted. The policy goal of this scenario seems to be to direct bank lending away from countries or technologies, which, in the NGFS’s view, are ill-prepared for a more sudden transition.
Conclusion
The NGFS Disasters and Policies scenario, which predicts that a series of storms in the next few years could create a giant, global depression, is probably less likely to occur than a civilization-destroying asteroid. How this scenario was incorporated into economic models that produced the global depression is not transparent, making a fully independent audit or review futile. Even if we cannot fully validate the scenario, an asteroid-risk scenario has almost no risk management value for a bank that is serious about measuring and managing its climate risks. As a result, regulators should not require banks to run or use this scenario (or similar ones), nor use it as a justification for higher capital.
It is important for regulators to base their climate risk management supervisory expectations on established, consensus science that can be independently checked and verified, not on hypothetical scenarios that have essentially a statistically zero probability of occurrence. The NGFS scenarios are based on climate risks that are unsupported by any scientific evidence, so in attempting to direct how banks manage these (unjustified) risks, a banking regulator could end up guiding lending and capital allocation decisions surreptitiously. In developing climate risk management regulatory expectations, central banks should avoid becoming shadow climate policymakers and focus on objectively regulating banks’ management of their material risks, some of which are climate risks.
[1] Elderson, F. “From charting the course to staying the course: the path ahead for climate and nature risk supervision.” Oct 2025, available at https://www.ecb.europa.eu/press/key/date/2025/html/ecb.sp251001~ff4e3bc6dc.en.html
[2] See NGFS, 2025, at https://www.ngfs.net/en/publications-and-statistics/publications/ngfs-climate-scenarios-central-banks-and-supervisors-phase-v
[3] NGFS, “NGFS Short Term Scenarios Technical Documentation,” 2025, available at https://www.ngfs.net/system/files/2025-07/NGFS%20Short-term%20climate%20Scenarios_Technical%20Documentation.pdf
[4] Figure taken from Figure 1 in NGFS, “NGFS Short-term scenarios-Main takeaways,” 2025, available at https://www.ngfs.net/system/files/2025-05/NGFS%20Short-term%20scenarios%20-%20Main%20takeaways.pdf
[5] See pg 27 of the NGFS Short-Term Climate Scenarios Technical Documentation
[6] Cornell Chronicle, “Physicist: After 33 billion years, universe ‘will end in a big crunch’,” 2025, available at https://news.cornell.edu/stories/2025/10/physicist-after-33-billon-years-universe-will-end-big-crunch
[7] IPCC, Chapter 11, “Weather and Climate Extreme Events in a Changing Climate,” Chapter 11 in “Climate Change 2021: The Physical Science Basis, 2021, available at https://www.ipcc.ch/report/ar6/wg1/downloads/report/IPCC_AR6_WGI_Chapter11.pdf
[8] Ridder, N, Pittman, A, Westra, S, Ukkola, A, Do, H, Bador, M, Hirsch, A, Evans, J, Di Luca, A, Zscheischler, J. “Global hotspots for the occurrence of compound events,” Nature Communications, 2020, available at https://www.nature.com/articles/s41467-020-19639-3
[9] Shepherd, T, et al, “Storylines: an alternative approach to representing uncertainty in physical aspects of climate change,” Nature Climatic Change, 2018.
[10] See for example Nelson, B. “An Introduction to Copulas (Springer Series in Statistics)”, Springer, 2006
[11] The Gaussian copula assumes a multivariate normal relationship between the random events.
[12] The t-copula assumes a multivariate t-distribution relationship between the random events.
[13] If the degrees of freedom are infinitely big in a t-copula, it becomes a Gaussian copula.
[14] NASA, “Asteroid Facts” available at https://science.nasa.gov/solar-system/asteroids/facts/
[15] These copulas are the so-called “Archimedean copulas” a different mathematical family from the elliptical family that the gaussian and t-copulas belong to.
[16] Chen, D, Qiao, S, Yang, J, Tang, S, Zuo, D, and Feng, G. “Contribution of anthropogenic influence to the 2022-like Yangtze River valley compound heatwave and drought effect,” Climate and Atmospheric Science, 2024, available at https://www.nature.com/articles/s41612-024-00720-3
[17] Tabari, H and Willems P, “Global risk assessment of compound hot-dry events in the context of future climate change and socioeconomic factors,” Climate and Atmospheric Science, 2023, available at https://www.nature.com/articles/s41612-023-00401-7
[18] See https://e3modelling.com/modelling-tools/gem-e3/ for a description of the model.
[19] Mandel, A, Battison, S, and Monasterolo, I, “Mapping global financial risks under climate change,” Nature Climate Change, 2025, available at https://www.nature.com/articles/s41558-025-02244-x#Abs1
[20] See https://bpi.com/the-ngfss-new-climate-damage-function-a-flawed-analysis-with-massive-economic-consequences/, https://bpi.com/the-flawed-ngfs-damage-function-is-even-more-flawed-than-we-thought/, and https://bpi.com/troubling-timeline-of-ngfss-most-recent-climate-estimate/
[21] Bearpark, T, Hogan, D, and Hsiang, S, “Data Anomalies and the Economic Commitment of Climate Change,” Nature, (2025) available at https://www.nature.com/articles/s41586-025-09320-4
[22] Schotz, C, “Spatial correlation in economic analysis of climate change,” Nature, 2025.
[23] For a review, see Congressional Budget Office, “The Risks of Climate Change in the United States in the 21st Century,” 2024, available at https://www.cbo.gov/publication/61146. Note that the large estimates from the Kotz et al. model should be removed from the CBO estimates. The Kotz et al. model is the: flawed academic model that underlies the NGFS damage function estimate.
[24] Hopper, G., “The Fed Pilot Climate Scenario Exercise: A Review,” Bank Policy Institute, 2024, available at https://bpi.com/the-fed-pilot-climate-scenario-analysis-exercise-a-review
