Board of Governors of the Federal Reserve System

Pilot CSA Exercise Insights

This section summarizes the key insights from the pilot CSA exercise. Observations and conclusions apply to only the six participants and are not necessarily indicative of practices at other banking organizations.

Participants use climate scenario analysis to consider the resiliency of their business models to a range of climate scenarios and to explore potential vulnerabilities across short- and longer-term time horizons. Many participants had conducted climate scenario analysis exercises prior to the pilot CSA exercise to identify risks and vulnerabilities, facilitate internal dialogue, inform strategic planning, or meet supervisory expectations in foreign jurisdictions. At a high-level, participants followed the stylized approach to model climate-related risks shown in figure 2.

Participants' approaches to the pilot CSA exercise varied significantly. Participants used different approaches to develop the physical and transition risk scenarios and to translate these scenarios into climate-adjusted credit risk parameters. Differences in approach were driven largely by participants' business models, views on risk, access to data, and prior participation in climate scenario analysis exercises in foreign jurisdictions. In constructing the common physical hazard shock, for example, some participants used external vendors to run simulations of thousands of potential hurricane events consistent with the severity parameters provided. Others studied historical hurricane events and created bespoke storm paths tailored to hit areas with material exposures.

Participants generally used existing credit models to estimate the impact of climate-related risks on credit risk parameters. In both the physical and transition risk modules, most participants adjusted inputs to their existing credit risk model frameworks to better capture climate-related risks, rather than adjusting the models themselves. They then used the climate-adjusted inputs in their existing credit risk models to generate climate-adjusted credit risk parameters. Some participants acknowledged that reliance on existing models assumes that historical relationships between model inputs and outputs continue to hold in future states even as the climate and the structure of the economy evolve. Some suggested that models could be enhanced to better capture climate transmission channels and associated impacts going forward.

Participants faced data challenges as they conducted the exercise. As summarized in table 3, participants noted a range of data gaps, including gaps related to real estate exposures, insurance, obligors' transition risk management, and infrastructure. Participants filled these gaps by sourcing data from third-party vendors or public sources or by using proxies to provide an estimate. Going forward, participants reported plans to capture additional data from clients, to source data from vendors, and to use proxies where necessary.

Table 3. Examples of participant-identified data gaps

Most participants worked with third-party vendors to conduct the pilot CSA exercise. For the physical risk module, some participants used catastrophe models provided by third-party vendors to simulate a large number of physical risk events and related damages to individual properties under the pilot CSA exercise scenarios, while others used vendors to estimate the property damage caused by their internally-generated, bespoke scenarios.³ For the transition risk module, most participants used external databases or vendors to help expand and customize the scenario variables, to access greenhouse gas (GHG) emissions data, and/or to capture the impact of transition risk on corporate financial statements or ratings. Participants noted that the lack of historical data and the proprietary nature of vendor models inhibited their ability to independently assess model performance.⁴ Some indicated a desire to further develop in-house modeling capabilities in order to reduce reliance on third-party vendors, while others plan to continue to explore vendor solutions.

Most participants considered indirect impacts and/or chronic risks in the physical risk module but faced modeling challenges. Most participants attempted to estimate at least one indirect impact, although this was not required as part of the exercise. Some participants made adjustments to macroeconomic variables (MEVs), such as county- or state-level gross domestic product (GDP), unemployment, or real estate prices. Some considered the effects of higher insurance premiums or elevated labor and raw material costs associated with rebuilding efforts.

Participants noted the importance of understanding insurance market dynamics when modeling the impact of physical risk hazards on credit exposures. The pilot exercise considered the credit impact of physical hazards assuming current insurance coverage and assuming no insurance coverage. While exploring these cases can be helpful, some participants noted the importance of developing a more nuanced understanding of insurance markets, including understanding the evolution of insurance pricing and its impact on property prices and obligors' cashflows, in order to manage climate-related risks.

Some participants conducted deep dive analysis to understand how obligors expect to manage transition risks over time, although this was not required as part of the exercise. Participants noted that information obtained through public disclosures, engagement with obligors, and/or third-party vendors led to a more nuanced understanding of potential transition risk effects on obligors' business strategies, profitability, and capital needs.

Participants intend to incorporate climate scenario analysis into their risk-management processes over time. Participants plan to continue to invest in data, models, and expertise to better identify, estimate, and monitor climate-related financial risks through the use of scenario analysis exercises and other tools. Participants' specific plans for future investments include acquiring more granular climate and exposure data, enhancing modeling capabilities, designing more customized scenarios that are better suited to test participants' unique business models and vulnerabilities, and shifting from vendor models to in-house solutions. Participants identified the high degree of uncertainty inherent to climate risk modeling, as well as the challenges created by such uncertainty in reliably and consistently quantifying the impact of climate-related risks, as factors impacting how the results of climate scenario analysis exercises could be used going forward.

Designing climate scenario analysis exercises requires consideration of tradeoffs. Standardized scenarios built on a consistent set of assumptions and variable pathways—for example, common storm paths, wind speeds, precipitation levels, or sea level rise—may result in greater consistency and comparability of specific estimates across participants. However, participants are uniquely exposed to different types of physical and transition risks, and standardization limits their ability to tailor scenarios to risks most material to their portfolios. Similarly, greater prescription in methods and data foster comparability of estimates but could obscure idiosyncratic risks and stifle innovation as practices continue to evolve.

Participants pointed to key design choices that impacted their approaches and estimates. These include

• Scope of the shock. The pilot CSA exercise physical risk module focused primarily on estimating the direct impacts of a single, acute hazard on RRE and CRE exposures over a one-year time horizon, limiting consideration of the longer-term impacts of the hazard.⁵ The exercise did not require participants to capture indirect impacts, such as damage to critical infrastructure (e.g., bridges and power stations) or supply chain disruptions, that could lead to prolonged disruption to local economies. In addition, the exercise did not require participants to capture the effects of chronic risks like sea level rise or higher average global temperatures. Participants noted that consideration of different types of physical risk shocks, the cumulative effects of multiple hazards over time, or a more comprehensive incorporation of indirect impacts and chronic effects, could meaningfully affect the nature of the exercise and the channels through which physical risks could impact their portfolios.
• Scenario severity. For the physical risk module, participants described the scenarios (e.g., a 200-year return period loss consistent with SSP 8.5/RCP 8.5) as relatively severe acute physical hazard events, particularly when assuming no insurance coverage. By contrast, most participants viewed the two scenarios used in the transition risk module (i.e., Current Policies and Net Zero 2050) as orderly, rather than stress, scenarios, and they noted the limited differentiation between the scenarios over the 10-year horizon. As credit risk parameters are sensitive to the macroeconomic environment, the benign MEVs included in the NGFS scenarios resulted in limited aggregate impact across the transition risk scenarios despite the significant increase in the carbon price in the Net Zero 2050 scenario.
• Starting point. The pilot CSA exercise focused on the impact of the scenarios on participants' 2022:Q4 exposures. Climate-estimated impacts were applied at a point of the credit cycle when credit quality was strong and loan-to-value (LTV) ratios were low across the portfolios, acting as a credit risk mitigant. As with other external shocks, the effects of adverse climate shocks could be different if the shocks were to occur during an economic downturn.
• Insurance assumptions. Financial protections, such as effectively functioning insurance markets, can limit the credit risk that large banking organizations face from climate-related physical risks in their real estate portfolios. To consider the sensitivity of the physical risk impact to insurance assumptions, the exercise asked participants to estimate the credit impact of physical hazards assuming no public or private insurance coverage, including no coverage from the National Flood Insurance Program (NFIP). An assumption of no insurance coverage may help to understand tail risk scenarios, but it represents an extreme outcome.
• Balance sheet assumptions. The pilot CSA exercise prescribed a static balance sheet approach. Static balance sheet assumptions hold the size and risk characteristics of the balance sheet constant over the projection horizon. This approach can build capacity around the measurement of potential risks by isolating the scenario impacts on measurements of PD and loss given default (LGD) for current exposures, but it does not account for management actions that could reduce the impact of climate risks. Participants noted their ability to rebalance their portfolios over the forecast horizon could significantly mitigate risk.

Additional investment and analysis could improve participants' risk-management capabilities. For example, participants faced challenges in modeling indirect impacts, chronic risks, and insurance dynamics related to physical risks. Participants also reported challenges in modeling the broader macroeconomic and sectoral implications of various transition pathways. Participants' modeling approaches varied significantly across these areas, and their estimates suggested that some of these factors could meaningfully magnify or mitigate credit risk impacts. Further research in these and other areas would help participants better understand their potential exposure to climate-related financial risks.

Highly uncertain risks are challenging to measure, and thus hard to incorporate into risk-management frameworks. The degree of uncertainty around the timing and magnitude of climate-related risks is high, making it difficult for participants to determine how best to account for and manage these risks on a business-as-usual basis. Those uncertainties can generate considerable variation in estimates of expected impacts, which complicates use of some common risk-management tools, such as quantitative risk limits, and strategic decisions. The high degree of uncertainty is a significant factor in considering how participants could use the insights gained from climate scenario analysis going forward.

 

References

3. A catastrophe model is a computerized process that simulates a large number of potential catastrophic events in order to assess losses due to the events. Return to text

4. The Participant Instructions stated that the pilot CSA exercise was designed to build capacity. Thus, participants were not prohibited from using models that had not been fully integrated into their model risk-management framework, including those that had not yet been subject to comprehensive model validation, unless participants also relied on a model used in this exercise for business-as-usual decisionmaking or to estimate risks on a regular basis. Return to text

5. The Participant Instructions stated that the review would focus primarily on the direct impacts of physical risks. Participants were encouraged, but not required, to incorporate indirect impacts of the event where possible. Return to text