Unpacking the Federal Reserve’s Climate Scenario Analysis

• The Federal Reserve recently conducted its inaugural climate scenario analysis, revealing significant vulnerabilities to physical risks, though the study lacked comprehensive depth and variable range.
• Challenges with data quality and modeling assumptions hindered banks' estimations of climate-related shocks.
• Moving forward, banks aim to enhance their data collection and modeling methods to improve understanding and management of climate risks.

On May 9, the Federal Reserve published the findings of its first climate scenario analysis (CSA) of Wall Street banks, joining the ranks of other supervisors including the European Central Bank and Monetary Authority of Singapore in showcasing how the firms under its watch could be impacted by physical and transition risks.

The results are a touch paradoxical.

On the one hand, they show that US banks are highly exposed to physical risks. Asked to imagine the impact of a 1-in-200 year hurricane making landfall in the US north east, in aggregate the banks estimated such a storm would impact 20% of their commercial real estate portfolios and 50% of their residential real estate loans in the region.

On the other hand, the Fed’s report does not go deep enough on the findings, nor did the CSA explore a wide enough range of variables, to produce truly granular data on banks’ exposures to physical climate shocks.

In addition, the Fed reported that participants experienced “significant data and modelling challenges” putting together their results. This underscores the ongoing difficulty even the largest banks have calculating their climate risk exposures, and the usefulness of advanced modelling techniques, which can surface the kind of asset-level insights that firms require to build operational resilience.

The CSA was announced by the Fed back in 2022.

Its purpose was to enhance the supervisor’s understanding of the state of climate risk management among the US’ largest banks – Bank of America, Citigroup, Goldman Sachs, JPMorgan Chase, Morgan Stanley, and Wells Fargo.

The exercise included a physical and transition risk module. Each tasked the banks to evaluate the robustness of specific portfolios against a range of forward-looking scenarios, reflecting “plausible future outcomes.”

Zeroing in on the physical risk module, this challenged banks to estimate how an array of “common” and “idiosyncratic” climate shocks would upset their residential and commercial real estate portfolios over a one-year horizon.

As these labels suggest, the “common” shocks were assigned by the Fed, while the “idiosyncratic” ones were chosen by each bank. The Fed-selected shocks consisted of three hurricane scenarios of escalating severity. Under the most severe iteration, banks had to assume no help from insurance coverage.

Participants took different approaches to modelling their exposure to these hypothetical storms. Some used catastrophe model projections acquired from third-party vendors. Others used data from past hurricanes and amplified their historic impacts using climate risk tools.

When it came to the idiosyncratic shocks, again a range of choices was on display. While some banks stressed their portfolios against a whole slate of climate risks – including wildfires, floods, severe storms, and hurricanes – others took a narrower approach.

Methods of estimating the financial implications of these shocks differed too.

Some banks used land value estimates as a starting point, based on expert judgment or country records, while others used property values, and still others external estimates at the zip code level.

Across the board, the Fed found that participants used their existing credit risk modelling frameworks to guess what the projected climate damage to properties would do to the default probabilities and loss-given default values of their loans.

With all these caveats and differences in approach, the aggregate findings disclosed by the Fed can only provide limited insights into the participants’ true exposures.

Under the most severe common shock, the results show that average probabilities of default jumped 40 basis points (bps) for commercial real estate loans, and 10bps for residential home loans. The idiosyncratic shock numbers were higher – with 260bps and 110bps of impact reported on average. This makes sense. After all, the idiosyncratic shocks were supposed to be uniquely damaging for each participant.

Unsurprisingly, the common and idiosyncratic scenarios where no insurance coverage was assumed produced higher impact estimates.

The banks acknowledged shortcomings in how they approached estimating their physical risk exposures, and raised familiar concerns with the quality and accessibility of climate data needed for this kind of assessment.

On data, the banks flagged issues with real estate exposures, insurance, and data infrastructure. Plugging data gaps and ensuring data consistency and reliability were oft-cited challenges, too.

On the modelling front, the banks said that by using existing credit models, the results assumed that historical relationships between model inputs and outputs would stay constant even as the type, severity, scale, and frequency of climate risks evolves, something a vocal group of academics and scientists deem unlikely.

Furthermore, banks took a scattershot approach to modelling the indirect and knock-on effects of physical shocks, like economic disruptions and a change in insurance availability. Some participants left indirect impacts out of their analysis completely.

The CSA results mark the end of the beginning of US banks’ climate risk management journeys – and the Fed’s report shows there’s a great deal they still have to do to become climate competent.

The participants told the Fed they plan to obtain more granular climate and exposure data and to enhance their modelling capabilities going forward, as part of their efforts to overcome some of the hurdles discovered in the course of the exercise.

Indeed, for climate risk analysis to be truly decision-useful, banks need access to asset-level data and the ability to calculate potential financial losses from a wide range of climate hazards at varying levels of intensity.