The Fed Pilot Climate Scenario Analysis Exercise: A Review

In September 2022, the Federal Reserve announced it was requiring six large banks[1] to conduct a pilot climate scenario exercise in order for the Fed to evaluate the capabilities of the banks to measure climate risks. The Fed followed up in January 2023 with a definition of various physical risk and transition risk scenarios for each bank to run. The physical risk scenarios required banks to measure the effects of extreme weather events on the credit risk of their residential and commercial lending exposures. The transition risk scenarios required banks to measure the effects on the credit risk of their corporate and commercial real estate lending exposures resulting from a sudden, global climate policy change consistent with the fulfillment of the Paris Agreement.[2] The Fed recently released the results, which include its assessment of banks’ climate risk practices and capabilities along with a summary of the impacts to probabilities of default measured by the banks under the climate scenarios. 

The purpose of this blog post is to review in detail the Fed climate scenario exercise, focusing on what the results mean both for bank risk management as well as the safety and soundness of the financial system. The climate scenarios specified by the Fed can best be understood as a means of getting an order-of-magnitude estimate on the effects of climate change on the credit risk of a bank’s most potentially climate-sensitive exposures under very extreme conditions: the physical risk scenarios used in the exercise are literally impossible while the transition risk scenarios are exceedingly unlikely. In this review, we first describe the scenarios, then assess the quantitative results, putting them into context by comparing them to CCAR results used by the Fed to measure capital adequacy of banks.

From the perspective of bank risk management, the exercise showed that the banks’ climate risk practices and methodologies are varied and that banks face serious modeling and data challenges when measuring climate risks.[3] In terms of safety and soundness, the results add to the growing evidence that climate risks do not rise to the level of capital level risks for banks and do not pose systemic risks to the financial system.  Rather, climate risks are more akin to the typical risks faced by banks. They should be managed and monitored by banks like all other risks, but there is no justification for elevating them to be more important than the other risks that banks must control, such as interest rate, FX or credit risk.

The Physical Risk Scenarios

The physical risk exercise can be thought of as a set of Rip Van Winkle scenarios in which the global economy falls asleep in 2022, only to awaken in 2050 with everything the same except for the climate, which is changed to be consistent with 2050 conditions. In these scenarios, the economy is exactly as it was in 2022, buildings are constructed to 2022 or earlier building standards, bank exposures are exactly the same as in 2022 and insurance markets reflect 2022 climate conditions. However, the banks face a 2050 climate. Of course, that is an impossible scenario.

The Fed specified two physical risk scenarios: a common scenario run by all banks in which banks must simulate the effects of a severe hurricane in the Northeast; and an idiosyncratic scenario that each participant bank must select based on the materiality of its exposures. The idiosyncratic scenario might be another hurricane, but it could also be chosen to be some other physical risk hazard, such as a wildfire or severe flood. The idiosyncratic scenario must be run in a region of the U.S. other than the Northeast.

In the Northeast, there were three hurricane scenarios:

  • A hurricane with a 100-year return period loss under climate scenario SSP2-4.5 or RCP 4.5 with insurance coverage, assuming 2050 climate conditions for the hurricane
  • A hurricane with a 200-year return period loss under climate scenario SSP5-8.5 or RCP 8.5 with insurance coverage, assuming 2050 climate conditions for the hurricane
  • A hurricane with a 200-year return period loss under climate scenario SSP5-8.5 or RCP 8.5 with no insurance coverage, assuming 2050 climate conditions for the hurricane

The return period loss can be thought of as analogous to value at risk metrics in market risk. A 100-year return period loss would be an annual loss in dollars at the 99 percent confidence level (although it is not clear from the results whether all participants interpreted it that way). The 200-year return period loss would be an annual loss in dollars at the 99.5 percent confidence level. 

The SSP2-4.5 or RCP 4.5 scenario is a standardized projection of GHG emissions and temperature increase projections that is a little better than the path the world is currently on. The SSP5-8.5 or RCP 8.5 scenario is a standardized projection of GHG emissions and temperature increase projections that is a good deal worse than the path the world is currently on. However, the differences between these scenarios tend to manifest over long horizons. In practice, there is not too much difference between them by 2050.

The second set of scenarios in the physical risk exercise is exactly like the three hurricane scenarios above, except that the scenarios must be run in a different region of the U.S. and the hazard may be a hurricane or some other physical risk hazard. The Fed review indicates that banks chose a variety of hazards for the idiosyncratic portion of the exercise.

Overall, banks were obligated to run six physical risk scenarios. Under each scenario, banks were required to compute the effect on one-year probability of default and loss given default for their residential real estate and commercial real estate exposures. In addition, banks were required to calculate one-year rating transitions for CRE exposures.

Results of the Physical Risk Scenarios

The Fed results reported averaged PD estimates across the six banks under the various physical risk scenarios. Table 1 below shows the PD increases for the 100-year and 200-year return periods compared to the implied PD derived from the loss rates predicted by the 2023 Fed CCAR CRE and first-lien loss rates methodology.[4] Of course, these comparisons are subject to the caveat that the PDs of the banks, which are averages, vary across banks, even for the same exposures. Moreover, the hypothetical portfolio in the CCAR methodology is not the same as the portfolios in the climate scenario test. Nonetheless, the comparisons should give a rough idea of the relative magnitudes.

Table 1[5]

table 1-the fed pilot climate scenario

For the common hurricane scenario, we can see that the PD increases are a good deal smaller than the implied CCAR PD changes. The estimates for the idiosyncratic scenario are also significantly lower than CCAR PD changes, but not by as much as the common scenario. What accounts for the difference? The idiosyncratic scenario is inherently more severe since it requires banks to choose the physical risk hazard by type and location based on the risk profile of its exposures. Thus, each bank is picking the worst hazard and location for its risk profile and then the PDs from different hazards and locations are averaged together. Average PD changes will naturally be larger when each bank is picking the worst-case hazard and location compared to a common scenario in which all banks have the same hazard and location.

The requirement to pick different hazards and locations in the idiosyncratic scenario introduces additional unreality to a physical risk scenario that is already logically impossible. Although it is not impossible in principle for a 200-year return period hurricane to strike for one bank in a single year while a 200-year return period wildfire strikes in a different part of the country for a different bank, it is certainly exceedingly unlikely.

Similarly, the common and idiosyncratic scenario potentially share an additional subtle significant exaggeration of the results. Turning to the common scenario first, because the return period of the hurricane is defined in loss terms rather than in terms of a physical characteristic of the hurricane such as windspeed, different banks that used catastrophe models could have selected the hurricane characteristics, such as the geographical area it struck, based on the concentration of loan exposures in different areas, since loss is the metric being used. Thus, two banks using catastrophe models would choose where and how the hurricane strikes and moves across land in a different manner, based on where its exposures and thus losses are located. If there is one 200-year return period hurricane during the year, that is impossible. The hurricane does not know where banks’ assets are located. It just strikes where it strikes, in the same places for all banks regardless of where the exposures are.

The alternative interpretation is that there are as many as six separate 200-year return hurricanes striking in a single year, each one tailored to the exposures of each bank. That interpretation is so unlikely that it is effectively impossible. Similar observations apply to the idiosyncratic case for any banks that choose the same idiosyncratic hazard and who have applied the return period definition literally.

Why is Physical Climate Risk Not a Capital-Level Risk?

Given all the assumptions built into the Fed’s physical risk scenarios—the Rip Van Winkle premise, the very rare and extreme hazards, the tailoring of the hazard to the bank’s risk profile, and the return period metric that implies that hazards implicitly know where the assets are located—it may be very surprising that even under those circumstances the results do not get close to the PD changes used in CCAR for capital purposes. Clarifying the differences between a capital scenario and a climate physical risk scenario shows that this result is not surprising at all.

The CCAR PD changes, the product of a financial crisis scenario, reflect losses that result from a very deep recession that lasts over a longer period, nine quarters. In contrast, a physical risk hazard is not associated with an accompanying recession nor do its acute effects last for a long time. A financial crisis scenario like CCAR will affect all mortgages and CRE loans, albeit not equally. A physical risk hazard is localized, on the other hand, affecting only exposures in the geographical regions in which it strikes, leaving other exposures unaffected. Large banks in particular tend to have geographically diversified real estate exposures, limiting the breadth of a physical risk scenario. Given these differences, we should not expect physical risk loss rates to rise to the level of loss rates in a financial crisis scenario.

More importantly, capital-level risks that are produced by financial crisis scenarios or similar profound market disruptions are universal, i.e., the risk affects all asset classes simultaneously. Physical risk, in contrast, depends on geographic location. Mortgages securing buildings will be affected but other asset classes or risk types should see limited or no effect. For example, operational risk losses play an important role in capital level risks, but the operational risk impact from the Fed climate scenario exercise is minimal.[6] Similarly, there is no effect analogous to the Global Market Shock from a physical risk.

It might be objected that the PD change estimates were not as large as they could have been since some effects were excluded. For example, participants were encouraged but not required to include indirect effects such as disruptions of supply chains and infrastructure and damage to the local economy. However, as the Fed results document noted, many (but not all participants) included indirect effects in their analysis.

When considering whether all effects were included, it is also important to note that scenario by construction omits very important mitigating factors. As an example, the Rip Van Winkle assumption necessarily requires that older buildings in Florida still exist in 2050 with the same distribution as in 2022. The Florida building code has continued to evolve, though, with new construction being more and more impervious to severe storms the more recently it is built. Newer residential housing, for example, is built using Concrete Block System rather than the more vulnerable wood frame, and includes hurricane impact windows, doors and roofs. Research confirms that more recent homes built under the more stringent Florida building codes suffer significantly less damage.[7] In general, there would be many mitigating adjustments to climate hazards by 2050. If the physical risk scenario really did occur in 2050, the damage and therefore PD change estimates would likely be significantly lower than reported in the Fed results.

The Transition Risk Scenario

Although the physical risk scenario was defined by the Fed, the transition risk scenario is one of six scenarios available at the Network for the Greening of the Financial System website.[8] The Net Zero by 2050 scenario meets the Paris Agreement ambition by limiting the global temperature rise to 1.5°C. To accomplish this objective, it assumes that there are international policy changes as well as technological advances that drive CO2 global emissions to zero by 2050. Besides the Net Zero scenario, the Fed also required banks to run the Current Policies scenario that extrapolates current climate policies into the future. Figure 1 shows the assumed emission path for both scenarios. Figure 2 shows carbon prices in each scenario.

Figure 1

fig 1-GHG Emissions

Figure 2

fig 2-price of carbon emissions

Participants were required to find the PD, LGD and risk rating impacts of both scenarios on their CRE and corporate credit exposures over a 10-year horizon. For the net zero scenario, participants were required to assume that the scenario represents a sudden change in global climate policy that is credibly expected to be enforced over a long horizon, i.e., the global price of a ton of CO2 emissions would rise to over $400 in 2010 dollars and CO2 emissions globally would decline to essentially zero by 2050. The PD, LGD, and ratings were to be calculated for each year over the next 10 years, starting with 2022 exposures. Participants were also required to assume that their CRE and corporate exposures over each of the next 10 years in the scenario remained essentially static at 2022 exposures, so that no portfolio risk management was allowed in the exercise.

Results of the Transition Risk Exercise

The Fed defined the transition risk impact as the largest annual change in PD in some particular one-year period over the 10-year period. The change in PD was calculated as the PD estimated under the net zero scenario minus the PD estimated under the current policies scenario. The Fed did not explain why it defined transition risk as the largest annual PD change chosen from the ten estimated PD changes. Although likely an overestimate, the choice of the largest one-year PD change does have the benefit that is adjusts for the unrealistic assumption that participants were required to hold their portfolios constant at 2022 risk levels. Reporting the full 10-year change or some combination of annual changes during the 10 years would have been an extreme exaggeration rather than merely an overestimate, because portfolios are assumed to be held fixed over 10 years in the scenario. In reality, of course, banks would have anticipated the policy change and aggressively risk-managed their exposures over the decade.

As in the Physical Risk scenarios, the Fed reported the largest change in PD averaged across the participants. Table 2 shows the results, compared to the typical PD changes the Fed assumes in its CCAR calculations.[9] 

Table 2

table 2-the fed pilot climate scenario

Table 2 indicates that transition risk is not a capital-level risk for banks. Transition risk is concentrated in a few areas such as CRE, since buildings need to incur retrofitting costs to reduce their emissions, and in corporate lending exposures, since company production processes or supply chains may depend on GHG emissions to varying degrees. In both cases, the effect of transition risk on the exposures is much lower than the risks the Fed assumes in CCAR calculations. In contrast, a capital-level risk scenario affects all risks profoundly. Thus, as in the case of physical risk, a transition risk would not produce losses even close to a capital or systemic risk scenario, such as a financial crisis scenario.

It should be noted that the Net Zero by 2050 scenario, while not impossible, is extremely unlikely. The scenario presumes a sudden, global change in emissions policy that is completely unexpected, given that portfolios are assumed to be equivalent to 2022 exposures. How could that happen? It is widely accepted that current and projected world climate policy will miss the Paris 1.5°C and indeed the 2°C target. GHG emissions should fall by 45 percent by 2030 to meet Paris goals, but instead they are expected to rise about 9 percent by 2030.[10] The scenario would require a sudden, dramatic change in all countries’ climate policies and, in particular, the policies of the largest emitters.

In Figure 3, the percent of annual GHG emitted by high income countries is depicted in blue and the percent of annual GHG emissions of the BRIC countries is depicted in red. 

Figure 3

fig 3- percent of world co2 equivalent

Figure 3 shows that the BRIC countries emit about 43 percent of global annual GHG emissions while the high-income nations emit about 25 percent of the total. Together, the group emits 68 percent of the total GHG emissions, with a large number of nations with small annual GHG emissions making up the other 32 percent.  Thus, the Net Zero by 2050 scenario requires sudden, unanticipated and unprecedented agreement on climate policy between the BRIC countries and the high-income countries, even though they are at odds on a substantial number of other issues. Moreover, the large numbers of smaller countries must also suddenly come into agreement, an exceedingly unlikely scenario.

What Are the Lessons For Bank Risk Management?

The results of the exercise are instructive, since they summarize the estimates of six large banks. A comparison of Table 1 and Table 2 shows that CRE is the largest risk identified for both physical and transition risk, with about the same magnitude of loss, a PD change in the neighborhood of 100 bps. RRE has much smaller physical risks and corporate exposures have much smaller transition risks. It should also be remembered that these estimates are the result of a physical risk scenario that is impossible and a transition risk scenario that is exceedingly unlikely. As a consequence, RRE under physical risk and corporate exposures under transition risk are likely de minimis in more realistic scenarios. CRE exposures are also likely much smaller in realistic scenarios.

However, it is important for banks to measure and manage all material risks, even if relatively small. For bank risk management purposes, climate scenarios should be used, but it is important for the scenarios to be more realistic. Although a global policy change affecting all commercial real estate simultaneously is very unlikely, a policy change in a particular city in which a bank’s commercial real estate exposure is concentrated may well happen. Climate change policies directed at commercial real estate have already been put into place in many cities. For example, local law 97, passed by the City Council in New York City in 2019, requires most buildings over 25,000 square feet to reduce emissions gradually to achieve net zero emissions by 2050. Bank risk managers can look to this example and others as a guide in formulating targeted transition risk scenarios for their commercial real estate exposures. The transition risk scenarios could be incorporated into risk management processes. Similarly, physical risk scenarios, such as hurricanes in South Florida, could be developed to test the resiliency of CRE exposures. These scenarios should simulate current rather than future hurricane impacts, since real estate exposures tend to be more short-term. In general, climate scenario analysis should evolve from the large-scale macro scenarios of the NGFS to more targeted, realistic micro climate scenarios aimed at specific risk vulnerabilities.

In developing these targeted, micro scenarios, it is important that climate risks be analyzed in the context of the other risks a bank faces so that risk resources are allocated efficiently. Climate risks should be treated the same as other risks. They should be assessed for materiality, and the more material risks should be identified and managed according to their magnitudes and likelihoods, with larger, more likely risks getting more attention and resources and smaller, less likely risks getting less focus.

Implications For Regulatory Policy

The results of the exercise show that climate risks are not of sufficient magnitude to threaten the capital adequacy of banks or the safety and soundness of the financial system. As a consequence, there is no reason for the regulatory authorities to elevate these risks into a special category, such as in the recent Pillar 3 disclosure requirements proposed by the Basel Committee. Enhanced disclosure along with the ongoing requirements for climate scenario exercises in various jurisdictions are predicated on the belief that climate is a much larger risk for banks than it really is. These enhanced requirements are not costless: they consume risk resources that might better be allocated to larger, more probable risks.

The climate scenario results show that physical and transition risk are relatively small for the most important asset classes affected. However, the resources needed to quantify these small risks under impossible or highly unrealistic scenarios are very significant, distracting attention and resources from much more likely and larger risks. Considering CRE, for example, one potentially significant non-climate risk is the move to remote work. Gupta, Mittal, and Nieuwerburgh estimate that the effect of remote work in New York City has reduced the value of buildings by 49 percent.[11] Another potential risk in CRE is that many loans must soon be refinanced at higher interest rates, which can also affect the credit risk of loans. Both risks to CRE are probably more significant than climate risk and are also more likely. By emphasizing climate risks of exposures in scenario exercises and in other proposals, the regulatory authorities are signaling to bank risk managers that they consider climate risks to be the key risks, and that banks should allocate resources and attention in that direction, potentially distracting bank risk management from larger and more likely risks.

Governor Bowman recently expressed the concern that focusing on climate risk can distract supervisors from emphasizing traditional risks that may be accumulating:

One example of a supervisory and regulatory distraction is from the Fed’s recent focus on climate risk. Recently, the Federal Reserve conducted a pilot climate scenario exercise with a subset of the largest U.S. global systemically important banks (G-SIBs) and issued guidance to large institutions on managing climate-related financial risk. This regulatory attention and focus on one specific, non-core risk could reasonably call into question whether regulatory priorities are focused sufficiently on key risks. Firms are already required to manage all material risks, and prioritizing climate risk in this way could lead to the misallocation of risk-management resources.[12]

Conclusions

Despite this admonition, it may be tempting for the regulatory community to continue to require banks to run large-scale climate scenarios, given an underlying belief that with the proper assumptions and the inclusion of all possible risks, climate change may yet rise to the level of a capital level or systemic risk. The results of the scenario exercise indicate that this belief is misplaced. Unlike scenarios that create systemic risks—e.g., financial crisis scenarios, bank runs, severe pandemics, or other severe, systematic market stress—climate risk scenarios cannot create risks of sufficient magnitude that affect all asset classes simultaneously. RRE and CRE assets are too diversified geographically to be simultaneously affected by a physical risk hazard. Risk exposure profiles are also too short for physical risks to be meaningfully amplified by climate change. To achieve simultaneous application of transition risks, it is necessary to make exceedingly unlikely assumptions about potential global policy changes. Even then, the risks are not very large and would not affect most asset classes. Without producing simultaneous, very large effects across all asset classes, a climate risk scenario cannot build up into a capital level, systemic risk for banks. Many of the risks that contribute profoundly to a capital-level loss scenario have no analogue in climate risk. How would the Global Market Shock in CCAR be produced by climate change? How could climate change produce credit losses at the level of a severe, global recession in CCAR? How could climate change produce large credit valuation adjustment losses, since there would not be any large market risk effects? 

The regulatory authorities should shift their focus from climate risks as potential systemic risks and reorient climate risk to be handled within the portfolio of risks that banks are already obligated to manage. The climate scenario results indicate that banks are still facing data and modeling difficulties and have not fully integrated climate risks into risk management processes. The regulatory community should not require banks to run any additional large-scale climate scenarios but, as part of day-to-day risk management, it should encourage banks to develop stressful but realistic, targeted climate scenarios in cases in which the risk is sufficiently large and likely.


[1] The banks were Bank of America, Citigroup, Goldman Sachs, JPMorgan Chase, Morgan Stanley and Wells Fargo.

[2] The Paris Agreement is a global pact adopted by 196 nations in 2015 to limit the increase in global temperatures to well below 2°C above pre-industrial levels with the ambition to limit the increase to no more than 1.5°C.

[3] This is not a surprising result. For a review of data and modeling challenges for physical risk, see, for example, Hopper, G, “A Primer on Climate Physical Risk,” Bank Policy Institute (2024), available at https://bpi.com/a-primer-on-climate-physical-risk/.

[4] See Table 28, Typical loss rate for the hypothetical portfolio and Table 33, Typical loss rate for the hypothetical portfolio available at https://www.federalreserve.gov/publications/2023-june-supervisory-stress-test-methodology-modeled-loss-rates.htm. Loss rates were converted to implied PDs by assuming a recovery rate of 50%.

[5] Note these PD changes are estimated from the charts in the Fed results by visual inspection and so may be a little off from the precise values.

[6] See Hopper, G, “Are Banks’ Operational Risks Significantly Affected By Climate Change?”, (2024), available at https://bpi.com/are-banks-operational-risks-significantly-affected-by-climate-change/

[7] See Done, J, Simmons, K, and Czajkowski, K, “Relationship between Residential Losses and Hurricane Winds: Role of the Florida Building Code,” Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil Engineering, (2018)

[8] See https://www.ngfs.net/ngfs-scenarios-portal/explore/

[9] For typical losses for corporate exposures, see Table 23 in https://www.federalreserve.gov/publications/2023-june-supervisory-stress-test-methodology-modeled-loss-rates.htm. Loss rates were converted to PD changes assuming a recovery of 50%.

[10] See UN press release, “World ‘Massively Off Track to Limiting Global Warming to 1.5C’, Secretary-General Stresses, at Launch of United Nations Convention on Climate Change Report,” November 2023, available at https://press.un.org/en/2023/sgsm22031.doc.htm#:~:text=Under%20current%20national%20plans%2C%20global,rise%20to%201.5%C2%B0C.

[11] Gupta, A, Mittal, V, and Nieuwerburgh, S, “Work From Home and the Office Real Estate Apocalypse,” revised from NBER Working Paper 30526, (2023), available at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4124698

[12] Governor Bowman, Speech to Texas Bankers Association 2024 Annual Meeting, May 10, 2024, available at https://www.federalreserve.gov/newsevents/speech/bowman20240510a.htm