Board of Governors of the Federal Reserve System

Physical Risk Module

The physical risk module required participants to estimate the credit risk impact of different types of acute physical hazards of varying degrees of severity on their RRE and CRE portfolios in certain National Climate Assessment (NCA) regions. The Participant Instructions prescribed a common shock for all participants as a hurricane in the Northeast region. Participants selected an idiosyncratic shock based on the materiality to their business model and exposures. Table 4 provides a summary of the six iterations of the physical risk shocks considered in the exercise. See the Participant Instructions for details on the severity parameters for each iteration.

Table 4. Summary of physical risk shocks

Translating physical risk shocks into credit risk parameters for real estate exposures is a complex, multi-step process. As illustrated in figure 3, participants projected future physical shocks under changing climate conditions, estimated the vulnerability of their in-scope exposures to these physical shocks, and estimated the credit risk impact, including the impact on PD, LGD, and internal risk rating grade (where applicable). The following sections discuss how participants approached each of these steps.

Define Physical Risk Shock

The first step of the exercise involved identifying the physical hazard and its detailed characteristics (left side of figure 3). Given the complexity of modeling physical climate systems and related financial impacts, most participants relied on external vendor models to define physical shocks consistent with the severity parameters outlined in the Participant Instructions and to forecast property-level damages caused by the physical shocks. While there was wide variation in approach among the participants, several participants used external catastrophe models for this step. Catastrophe models simulate a large number of extreme events to quantify the financial impact of a range of potential disasters.

Common Shock Selection

The Participant Instructions defined the common shock to be a hurricane event (or series of events) in the Northeast at varying degrees of severity, but each participant determined the precise characteristics of the hurricane, such as the geographic path, wind speed, storm surge, or precipitation level.

Participants took a range of approaches in the design of the physical risk shocks. Approaches were largely a function of the capabilities that participants had onboarded or previously developed internally. The design of the physical risk shock determined the geographic path and footprint of the selected hazard event, such as a hurricane, the severity of the hazard, and its specific characteristics, such as wind speed and storm surge at points along that path. The design determined the level of damages to individual properties and reported credit risk parameters.

Participants that onboarded catastrophe models typically used these models to select specific hazard events (e.g., a hurricane) from a catalogue of simulated events in the relevant region. These participants used the catastrophe models to design hazards consistent with the future climate parameters provided by the Participant Instructions. When designing a hurricane shock, for example, some participants selected a single hurricane event that produced aggregate damages consistent with the prescribed return periods. Other participants selected multiple hurricane events within the region, each with levels of aggregate damages within a narrow range around the specified return period, and then averaged the damage estimates across hurricanes. Participants took this approach to reflect the uncertainty associated with climate modeling and to account for the variation of different hurricane characteristics. Participants noted that hurricanes with similar levels of aggregate damages in a region may have very different footprints and impact different properties within the region.

Participants' generally selected hazard events impacting geographic areas in which they have concentrations of loans. Participants' approaches varied in whether they applied the same or different hazard events across their CRE and RRE portfolios. Practices ranged from selecting a single hazard event (e.g., a hurricane path) based on aggregate exposures across both CRE and RRE portfolios to selecting different hazard events for each portfolio in order to recognize the differences in geographic concentrations of CRE versus RRE.

Participants that did not use catastrophe models to simulate hurricane events relied on other approaches to design a hurricane event with the prescribed characteristics. One approach was to combine historic hurricane paths in the region with an external climate vendor projection of physical risks, such as coastal flooding, river flooding, and wind. In this approach, historic hurricane paths were adjusted so the hypothetical path of the event covered areas where participants had concentrations of RRE and/or CRE loans.

Depending on the approach taken, participants used different levels of granularity of hazard estimates, typically ranging from property-level estimates to zip code-level estimates. The level of granularity affected the participant's ability to consider the local geography in the damage estimation process.

Where participants did not use a catastrophe model, they differed in how they derived flood and wind damages caused by the hurricane. In some cases, participants modeled flood and wind damages separately and aggregated them. In other cases, participants only modeled flood damages and added a scaler for assumed wind damage. The share of damage caused by flood versus wind for selected hurricane events varied significantly across participants.

Idiosyncratic Shock Selection

For the idiosyncratic shock, participants were asked to select a hazard event and an NCA region based on materiality to their business models and exposures.⁶ The Participant Instructions prescribed certain features that determined the severity of the shock.

The number and types of hazards that participants considered for the idiosyncratic shock was largely a function of participants' modeling capacity. Most participants considered multiple hazards, such as hurricanes, floods, wildfires, convective storms, and winter storms, while others limited consideration to fewer hazards. When selecting the hazard and region for the idiosyncratic shock, most participants estimated property-level damages across NCA regions for properties securing RRE and CRE loans and selected the hazard-region combination with the highest aggregate damages.

Some participants based their choice of idiosyncratic hazard on damage estimates for the most severe scenario, while others used a combination of severities. Most participants used estimates of total property damage without insurance protection as the metric by which they chose the idiosyncratic shock. Another approach was to convert estimated property damages into expected loan losses to identify the hazard-region combination that resulted in the highest expected loan losses.

For the idiosyncratic shock, participants chose hurricanes, extensive flooding, or wildfires in either the Southeast or Southwest geographic regions.

Consideration of Future Climate Conditions in Year 2050

As prescribed by the pilot CSA exercise, all participants considered future climate conditions in the year 2050 when modeling their physical risk shocks, although participants employed different methods to estimate these conditions. All methods relied on climate models for future climate estimates based on specific SSP/RCP pathways. In cases where participants utilized catastrophe models, vendors used climate model outputs to generate climate-conditioned simulated events reflecting future climate conditions rather than conditions based on historical data. In other cases, external climate vendors downscaled climate estimates from climate models for specific hazards and regions, or participants used climate model outputs directly in their internal models to project future hazard risk factors without relying on third-party vendors.

In the design of hurricane events consistent with possible future climate conditions in 2050, most participants included global and regional sea level rise, temperature increases, and changes to wind speeds. All participants considered how future climate conditions could impact the severity of physical shocks used for the exercise. For example, approaches included considering the impact of sea level rise on storm surge, permanent inundation of coastal properties outside of the direct path of the storm, or heat stress and precipitation on CRE business interruption.⁷

Estimate Property Damages

After identifying a specific hazard event, participants used different approaches to model the resultant damages (middle left of figure 3). Vendors that provided property-level damage estimates derived those estimates from hazard intensity parameters at the property location and key building characteristics, such as year built, occupancy codes, number of floors, and building materials.

Most participants reported data gaps around building characteristics and largely relied on national or regional "default" property characteristics from their vendors. Where third-party vendors were not used, or where third-party vendor data were limited, participants based the damage estimates on academic studies of historic events.

When estimating damages, participants made different assumptions about the decomposition of property value into building structure value and land value. Some participants estimated land value based on expert judgment or county records, others made implicit assumptions of land value based on the property value, and others used external estimates at the zip code level. When estimating damages to the building structure, some participants assumed that land value depreciated, while others assumed that damages would impact only the building structure. In the latter case, land value could serve as a mitigant to the level of damages applied to property value.

Credit Risk Models

After constructing the common and idiosyncratic shocks and estimating the physical damages to their real estate portfolios, participants calculated climate-adjusted credit risk parameters (right side of figure 3). Participants focused primarily on estimating the impact of damages to properties in the path of the physical hazard with less work on indirect impacts or broader impacts from chronic changes in climate conditions.

Direct Impacts

Most participants largely relied on existing credit risk modeling frameworks used in stress testing and regulatory capital requirements to estimate the direct impacts of acute physical shocks on credit risk parameters, such as PD and LGD, at the loan level. Participants generally modified key inputs into existing credit risk parameter estimation models, rather than changing the models themselves. Participants noted that reliance on existing models assumes that historical relationships between model inputs and outputs continue to hold in future states as the climate and structure of the economy evolve.

Figures 4 and 5 show stylized models for estimating impacts on credit risk parameters for RRE and CRE, respectively. These credit models combine MEVs and loan-level inputs, such as LTV ratios to estimate the PD and LGD of a given exposure. The primary adjustments for the physical risk shocks were through LTVs for RRE, and through LTVs and debt-service-coverage ratios (DSCRs) for CRE.

For both RRE and CRE, most participants translated climate shocks to estimates of PD and LGD by decreasing property values by the amount of uninsured damages from the event, thus increasing LTV ratios. Another approach assumed that obligors take on additional debt to cover uninsured or underinsured structure damage repair costs, increasing loan values and LTVs, and increasing debt-to-income ratios. For CRE, in addition to incorporating the impact from property damage on LTVs, some participants also assumed that climate shocks would impact the amount of time that an income-producing property was out of service. DSCRs were therefore adjusted to reflect declines in net operating income (NOI) due to damage-related business interruption. Another approach estimated the impact of repair costs on borrower cashflows in cases where repair costs were significant due to deductibles or when assuming no insurance coverage.

Indirect Impacts

In addition to estimating direct impacts of physical hazards on in-scope portfolios, participants were encouraged, but not required, to incorporate indirect impacts of the event where possible. Examples of indirect impacts could include, but are not limited to, impacts on the local economy, infrastructure, pricing effects, and supply chains, all of which could impact credit risk parameters.

Participants took varying approaches in considering indirect impacts. Given the optional nature of indirect impacts in the pilot CSA exercise and the challenges around estimating their impact, some participants either did not estimate indirect impacts or considered them in a separate analysis that was not included in the formal submission of credit risk parameters. Other participants incorporated one or more of the indirect impacts listed in table 5 in their PD and LGD submissions. These indirect impacts were captured either through adjustments to state- or metropolitan statistical area (MSA)-level MEVs included in credit models or through loan-level inputs.

Table 5. Indicative examples of indirect impacts in the physical risk module

Insurance Assumptions and Modeling

For the pilot CSA exercise, participants were asked to estimate the credit impact of physical hazards in both the common and idiosyncratic shock with two different assumptions: (1) assuming current insurance coverage, and (2) assuming no public or private insurance coverage, including no coverage from the NFIP. For purposes of the pilot CSA exercise, most participants assumed that insurance markets would operate effectively and that insurance payouts would be sufficient to rebuild damaged properties. Participants noted that the no insurance shocks naturally involved uncertainty, since the available historical data do not contain relevant examples of widespread damage not covered by insurance and/or government assistance.

See box 1 for background information on the types of insurance relevant for banks' RRE and CRE portfolios.

All participants reported data gaps related to insurance, including insurance coverage details, replacement cost value, and deductibles, particularly for CRE (see table 6). While participants have some level of hazard insurance information, it is not always consistently recorded or easily aggregated. As a result, participants generally had to rely on assumptions to estimate the degree of insurance mitigation for physical risk shocks.

Table 6. Participant-identified data gaps and assumptions around insurance

In the scenarios that assumed existing insurance coverage for RRE and CRE portfolios, insurance generally mitigated LTV shocks. Similarly, for CRE, business interruption insurance mitigated the NOI impact from downtime caused by physical damage for some participants. In the no-insurance shocks, this mitigation was removed.

Most participants noted that the pilot CSA exercise did not require that participants capture the impact of an increase in insurance premiums over time. Participants noted that an increase in insurance costs, which some participants voluntarily included as indirect impacts, could lead to increased financial burden for obligors, potentially impacting obligors' disposable income and overall credit profile. For CRE, rising insurance premiums could cause NOI shocks, which could drive rent increases if higher costs were passed on to renters.

Impact Estimates from Participants

This section summarizes estimates of average loan-level PDs across participants for the physical risk module.⁸ Estimates were produced by participants using their own models and assumptions and submitted to supervisors as part of this exercise. Basic quality control checks were performed, but the data were not independently verified by supervisors, and the Federal Reserve did not independently estimate the impact on risk parameters. Given the goals and design of the exercise, the quantitative estimates should be interpreted with an understanding that broad variation in approach makes comparison and aggregation of estimates across participants difficult.

For the common shock of the physical risk module, participants estimated the impact of a hurricane in the Northeast NCA region on in-scope CRE and RRE portfolios. For the idiosyncratic shock, participants were asked to select a hazard event and an NCA region based on materiality to their business models and exposures. See table 4 for a summary of the shocks.

To provide some scale of the portfolio coverage, table 7 shows the aggregate number of CRE and RRE loans across participants in the common shock. A similar table for the idiosyncratic shock is not included given the variation in regions selected by participants. More than 1,000 CRE loans and more than 238,000 RRE loans in the Northeast NCA region were impacted by the most severe common shock (200-year, no insurance). This represents about 20 and 50 percent of participants' total Northeast NCA region CRE and RRE loan counts, respectively.

Table 7. Physical risk common shock summary

As reported in figures 6 and 7, participants estimated that PDs generally increased with the severity of the shocks, e.g., moving from SSP2-4.5 (or RCP 4.5) pathways with a 100-year return period loss to SSP5-8.5 (or RCP 8.5) pathways with a 200-year return period loss and to the no insurance shock.⁹ In the common shock, estimates of average PDs across participants for properties in the Northeast region increased by about 40 basis points (bps) for CRE and about 10 bps for RRE in the most severe iteration (200-year, no insurance) relative to the baseline (figure 6).¹⁰ In the idiosyncratic shock, average PDs across participants for properties in the selected NCA regions increased by about 260 bps for CRE and about 110 bps for RRE in the most severe iteration (200-year, no insurance) relative to the baseline (figure 7).

For most participants, the idiosyncratic shock was more impactful than the common shock. Participants generally found that the SSP/RCP pathway characteristic was less significant than the return periods in terms of severity because the SSP/RCP paths and related physical risks only start to meaningfully diverge after 2050.

Insurance mitigated participants' estimates of the impact of physical risk hazards on credit exposures. Assuming no insurance coverage generally increased PDs across RRE and CRE portfolios for most participants. In some cases, conservative assumptions about the level of insurance coverage in the 200-year shock limited the impact of insurance removal (i.e., where insurance coverage was assumed to be low, its removal did not result in significantly higher estimated PDs).

Assuming no insurance coverage had a more pronounced impact in the idiosyncratic shock relative to the common shock. In general, property damage estimates were lower in the common shock, as less severe hurricanes in the Northeast caused less damage to individual properties than hazards selected in the idiosyncratic shock. When property-level damages are low and do not exceed insurance deductibles, insurance provides minimal protection, and its removal does not have a significant impact on estimated PDs.

Participants reported that physical damages to property and related effects were the primary drivers of increased PDs. Credit risk was primarily concentrated in small pockets of loans with the highest expected damages in geographic areas directly in the paths of physical hazards. CRE damages were particularly sensitive to storm selection given the fewer number of properties and higher value of properties.

The underlying distributions of loan-level PDs in all scenarios were heavily concentrated in the 0-50 bps range with a small fraction of loans with PDs greater than 500 bps. As the severity of the shock increases, the distribution of loan-level PDs shifts to the right relative to the baseline, and the share of loans with PDs greater than 500 bps generally increases (see table 8). A larger percentage of CRE loans was impacted by the idiosyncratic shock.

Table 8. Percent of loans impacted by physical risk shock

Figure 8 shows the distribution of changes in PD for CRE loans between the baseline scenario and the 200-year, no insurance scenarios for both the common and idiosyncratic shocks. The distribution shows that the majority of loans had a 50 bps or less change in PD for both the common and idiosyncratic shocks, and approximately 1 percent and 9 percent of loans had a change in PD of more than 500 bps in the common and idiosyncratic shocks, respectively.

 

References

6. Participants were directed to select a geographic region other than the Northeast region, which was used in the common shock. Return to text

7. Some of the analysis described was not included in the credit risk parameters submitted by participants for the exercise because it was not required as part of the exercise. Return to text

8. See the appendix for more detail. Return to text

9. The majority of participants show an increase in the estimated impact from the 100-year shock to the 200-year shock. Return to text

10. See the appendix for more detail. Return to text