A Primer on Climate Physical Risk

With extreme weather events dominating the headlines, management of physical risks is becoming a rapidly increasing focus of financial institutions and regulatory authorities. They define physical risk as the risk of financial loss driven by acute events, such as extreme hurricanes or floods, or chronic events, such as permanently elevated temperatures or sea levels. Banks are forming climate risk management functions that include a physical risk component and are beginning to integrate climate physical risk considerations into risk management processes. At the same time, regulators are developing climate risk stress tests, formulating climate risk management expectations and proposing climate risk disclosure standards. The Federal Reserve recently conducted a climate risk stress test[1] that included an extreme hurricane. Other regulatory authorities, such as the European Central Bank (ECB) and the Bank of England (BOE), have included physical risk in their stress tests. The ECB has also formulated supervisory expectations on climate risk management that includes both acute and chronic physical risks.[2] A recent BCBS consultative document[3] proposes that banks disclose their exposures to physical risk by geographical region.

Despite all the interest in physical risk, progress in developing physical risk models and methods has been hampered by three difficulties that are not present in other risk areas. First, physical risk is inherently interdisciplinary, requiring results from physics, economics, hydrology, meteorology, fire science, industrial health, biometeorology and many other disciplines. Bank risk managers as well as regulators are unfamiliar with all these areas. Second, although climate science is able to make general statements about future changes in the climate, it is not sufficiently advanced at present to make specific geographical predictions that are accurate enough for bank risk management. Third, there are still crucial outstanding questions in the academic literature in many areas that must be resolved before physical risk management can become mature. These difficulties create considerable uncertainty for both bank risk managers and regulators.

In this post, we offer some observations on these difficulties, ultimately suggesting that stress testing, as opposed to climate-adjusted risk modeling, is the appropriate risk management tool for banks to employ to understand physical risk. Bank risk managers must also understand what is known and unknown and should design stress tests to probe the unknown. The regulatory community should follow suit by encouraging the development of stress testing. Regulators should dispense with physical risk expectations that require the use of geographically imprecise metrics or the incorporation of physical risk into risk management models or parameters.

Although physical risk is a vast subject that cannot be thoroughly covered in a short note, we illustrate these difficulties with two case studies. The first case study examines the effect of the increased frequency and severity of hurricanes on banks’ financial losses, showing that hurricane science is not sufficiently developed to make accurate geographical predictions. The second case study takes a deep dive into what is known and unknown about the problem of measuring the effect of higher future temperatures on bank exposures and local economies. In both cases, stress tests are the right risk management tool for the job.

Case Study 1: The Effect of Increased Frequency and Severity of Hurricanes

Knutson et al.[4] review current evidence on the effects of climate change on the frequency and severity of hurricanes. The most confident prediction of hurricane scientists is that sea level rise will increase damage from coastal flooding that accompanies hurricanes. Beyond that, there is medium to high confidence that 1) the frequency of category 4 and category 5 storms will increase, with the proportion of the total rising by a median estimate of 13 percent; 2) hurricane precipitation will increase, with a typical increase of 14 percent for 2°C of warming; and 3) hurricane wind speed will rise, with a median estimate of 5 percent[5]. It is important to emphasize that these are undifferentiated, global predictions of future hurricane characteristics.

Predictions about what will happen in specific hurricane basins are far less certain. Not only is it not clear which of the global predictions will apply in which geographic areas but also there is limited consensus on the effects of climate change on hurricane tracks and genesis points, even on a global basis. Understanding hurricane track and genesis points is critical to determining whether the more frequent and damaging hurricanes make landfall in the same locations as they did historically or indeed make landfall at all.

For example, Hall et al.[6] propose a hurricane model in the Atlantic Basin that describes the full lifecycle of hurricanes from genesis to landfall, adjusted for projected climate change in the 2030s. The dependence of hurricane genesis points as well as hurricane tracks on sea surface temperature (SST) is explicitly estimated in the paper. Hall et al. find that future projections of SST from climate models would cause genesis points to move farther away from the U.S. East Coast. Hurricane tracks would also curve more northward. The effect of these changes on genesis points and tracks is to reduce the risk of future hurricanes in the Northeast, even though the proportion of more intense storms would be greater. Similarly, the movement of the genesis points and the increased curvature of the tracks would shift more intense hurricanes from the western coast to the eastern coast of Florida.

Academic results also depend on the basin. Nakamura[7] finds in the Western North Pacific a tendency for models to produce tracks that move poleward with climate change as well as move towards the east, potentially increasing hurricane risk in Hawaii. Nonetheless, the weight of this evidence must be tempered by Knutson et al.’s overall conclusion that “Despite the large number of studies that have explored the issue of future projections of TC tracks and occurrence changes, there is currently relatively limited overall confidence in these projections. The principal reasons for this include the difficulty in obtaining a clear model consensus in projected track/occurrence behavior, the lack of a clear detectable anthropogenic influence on such TC metrics in the historical data, and limited confidence in IPCC projections of regional circulation features and future SST pattern changes that could affect tracks.”

Besides the uncertainty in the hurricane models themselves, there is also substantial uncertainty in the many other factors necessary to determine the significance of extreme storm risk for a particular location. Direct physical damage is caused by wind damage and flooding. The level of damage depends on the exact location of the property, the physical characteristics of the property, and what is around it. Wind damage functions depend precisely on construction materials as well as any structures, foliage or trees around a property location that either attenuate the wind speed or provide material for projectiles. Flood damage depends on exactly where a property is located, whether there is drainage nearby and the details of construction. Age of property is also extremely important. In Florida, for example, building codes require building construction from 2020 on to be able to withstand category 5 hurricane wind damage. The economic loss that results from physical damage depends on the details of insurance coverage.

The time horizon assumed in the analysis also creates uncertainty. The changes that are expected in hurricanes are projected to materialize sometime after 2050. For the much shorter horizons that bank assets will typically have, it is much more difficult to decide how much hurricanes would change. Would the proportion of category 4 and 5 hurricanes rise by 5 percent or some other amount before 2025 or 2035? Would wind speed increase before 2025 or 2035? By how much? Unfortunately, there is virtually no academic literature that can be relied on to answer these questions.

Case Study 2: The Chronic Risk of Higher Temperatures

Chronic physical risks suffer from the same uncertainty problems as extreme weather events. Although it is possible to make general statements about rising temperatures and other chronic risks, making reasonably precise statements on the effects of temperature increases on financial losses and the local economy is bedeviled by significant uncertainty.

General Circulation Models

Part of the uncertainty derives from a necessary feature of the underlying climate models. Temperature predictions are made by General Circulation Models (GCM), large-scale physical risk models of the earth’s surface, oceans and atmosphere. Figure 1 shows a schematic of a GCM.

Figure 1

figure 1-primer on climate risk

Credit: David Bice © Penn State University is licensed under CC BY-NC-SA 4.0

A GCM is a physical model that divides the atmosphere, oceans and land into boxes and then uses physical laws to simulate the movement of heat, temperature and other quantities through the boxes over time. GCMs project physical variables such as temperature, precipitation, humidity, wind and speed far into the future, generally until 2100. Because of computational constraints, the size of the boxes in current climate models is about 100km by 100km. Thus, from the models alone, projections of physical quantities in the future are known at a granularity of about 100km or 62 miles. Projections to more precise geographical regions will require some sort of interpolation, statistical downscaling or other statistical projections, introducing some additional uncertainty. Also, many important physical processes, such as cloud microphysics, cannot be simulated at such a coarse model granularity and must be approximated in the GCM. As a consequence, there can be substantial differences in the projections of GCM models.

Climate Scenarios

GCM projections in turn depend on the climate scenario — i.e., the path of assumed greenhouse gas (GHG) emissions –that is fed into the model. The GHG scenario drives the model simulations, with more GHG introduced into the model producing higher temperatures over time. Chart 1 shows examples of the representative concentration scenarios (RCP) used by climate scientists, and Figure 2 provides some background on the underlying assumptions of the scenarios.

Chart 1

Chart 1 - primer on climate risk

Figure 2

fig 2-primer climate

Equilibrium Climate Sensitivity

Because many underlying physical processes must be approximated in a GCM, the projections of temperature, humidity and other meteorological quantities depend on the GCM model used. GCMs can differ substantially in the projections they make. For temperature projections, the critical feature of a GCM is its equilibrium climate sensitivity (ECS). The ECS is the amount that temperature would rise in long-run equilibrium for a doubling of atmospheric CO2 concentrations. The average ECS of GCMs is around 3°C but the models vary substantially.  Figure 3 shows estimates of ECS for GCM models from the older collection of CMIP5 (Coupled Model Intercomparison Project) models and the more recent collection of CMIP6 models.

Figure 3

fig 3- primer climate

How Should Temperature Be Measured?

Academic economic studies of the effects of chronic heat stress have generally used the increase in temperature as the metric, often statistically downscaled. Vendor models will typically use temperature as well. The physical risk component of the 2021 ECB/ESRB climate stress test, which relied on a vendor model, used increases in temperature that were statistically downscaled from CMIP5 GCMs to a granularity of 25km by 25km under the RCP 8.5 scenario.[8] However temperature is not the best measure of heat stress. Heat stress depends not only on temperature, but also on humidity, wind speed, cloud cover, the time of day, the angle of the sun and whether it is experienced in the shade or indoors. Researchers working in the intersection of health and climate have generally used more biologically based measures of heat stress. Multiple heat stress measures that have been proposed. One such measure is the Heat Index, which depends on the air temperature and relative humidity. Another is the Universal Thermal Climate Index (UTCI), which is founded on the human body’s physiological response to heat stress. Perhaps the most studied metric is the Wet Bulb Globe Temperature (WBGT), recommended by the Occupational Safety and Health Administration (OSHA)[9] as well as the International Organization for Standardization (ISO).[10]

WBGT can be directly measured by instruments, but if it is to be projected into the future, it must be approximated from meteorological quantities predicted by GCMs. There are many proposed algorithms to calculate WBGT, with some methods more suitable for predicting WBGT indoors and others better for outdoors. An analysis by Lemke et al[11] shows that Bernard and Pourmoghani[12] is at present the best method for predicting indoor WBGT while Liljegren et al.[13] is the best method for outdoor predictions.

Simplified methods that rely on only temperature and relative humidity, such as the Australian Bureau of Meteorology (ABM) method,[14] are not as accurate, but are commonly used. The Network for the Greening of the Financial System (NGFS) has recently developed[15] a heatwave analysis that is based on the wet bulb temperature calculated from temperature and relative humidity. It is important to note that the wet bulb temperature used by the NGFS is not the same as the WBGT, but is rather a component of it. 

Uncertainties in Projecting Future Heat Stress

To illustrate the current uncertainties in performing a heat stress test analysis, we compare the effects of GCM models with different ECSs in July of 2035. Although most bank risks are shorter-term, we use 2035 to cover cases in which assets may be held for longer periods. The specifications of the heat stress test are:

  • Select GCMs CESM2[16] (ECS 5.2C) and MPI-ESM1-2-LR[17] (ECS 3C) from the CMIP6 model group
  • Use native model granularity with no statistical downscaling
  • Select scenario ssp585, the successor to RCP 8.5 for CMIP6 models
  • WBGT estimated using the Australian Bureau of Meteorology Method (ABM) for simplicity of calculation
  • Focus on Texas

We download GCM model simulations from the National Center for Atmospheric Research (NCAR) hosted by Pangeo[18] and construct average estimates using the ABM method for WBGT in Texas in July 2035, as shown in Figures 4 and 5.

Figure 4

fig 4 - primer climate risk

Figure 5

fig 5 - primer climate risk

Figure 4 shows that WBGT under the CESM model (with a high ECS) is projected to be between 33°C and 34°C in Dallas and Austin and between 34°C and 35°C in Houston. In contrast, WBGT under the MPI-ESM1-2-LR (mean ECS) model is projected to be between 32°C and 33°C in Dallas and Austin and between 33°C and 34°C in Houston. Thus, even at a short horizon of 11 years, the underlying GCM models can differ by as much as a degree, which could be significant as we will see later. Figures 4 and 5 also show that estimated WBGT can vary substantially around Texas and other states.

Figures 6 and 7 show that WBGT also varies substantially by season, being lower in the spring and fall.

Figure 6

fig 6 - primer climate risk

Figure 7

fig 7 - primer climate risk

Economic Impact of Heat Stress

To use WBGT projections for risk management purposes, we need to translate them into economic effects. One important consequence of heightened heat stress is labor productivity, since declines in labor productivity could be fed into local economic growth estimates, or, alternatively, into cost increase estimates for particular companies or industries. The economics literature tends to look at temperature rather than WBGT to study productivity loss.[19] For WBGT studies, we must refer to the industrial health literature. To economically interpret these WBGT projections, we use productivity functions estimated by Brode et al.[20] that show the relationship between percent of labor productivity lost and mean WBGT for various levels of work. These productivity curves, depicted in Chart 2, were based on a 1969 epidemiological study of gold mine workers in South Africa and a 2013 epidemiological study of rice harvesting workers in India.

Chart 2

chart 2 - primer climate risk

The epidemiological estimates in Chart 2 show that when WBGT gets to levels above 28°C, an additional 1°C of WBGT can have large impacts on labor productivity, especially for heavy work. In contrast, in another paper Kjellstrom et al[21] provide a productivity curve based on the ISO 7245 (1989) standard, depicted in Chart 3.

Chart 3

chart 3 - primer climate risk

Chart 3 shows even steeper productivity losses for an additional 1°C above 26-27°C.

Because of the sensitivity to labor productivity to increases in WBGT, it is necessary to be more careful in its estimation. The ABM method is simple to calculate but it is also generally inaccurate, tending to overestimate WBGT. Accordingly, we estimate both indoor and outdoor WBGT using the Bernard and Liljegren methods, respectively, and compare to the ABM method. We focus on a particular city in Texas by way of example, Dallas, and switch to another GCM, the ACCESS-CM2 model, since it is on the high side of ECS (ECS = 4.7°C) but does not have the highest ECS in CMIP6. We use model simulation data downscaled to a 25km-by-25km resolution.[22] Figure 8 shows the results for Dallas in mid-July 2035.

Figure 8

fig 8 - primer climate

Figure 8 confirms the importance of using the most accurate calculation methods as well as distinguishing between indoor and outdoor WBGT. An analyst would obtain substantially different productivity loss estimates from Chart 2 or Chart 3 using the more accurate estimates of WBGT. It should also be noted that indoor WBGT would be lower if air conditioning or fans were used. Similarly, outdoor WBGT would be lower if measured in shade or with outdoor fan use.

Additional Uncertainties

Outdoor WBGT depends on relative humidity but also on the amount of solar radiation and wind speed. The calculation of outdoor WBGT used the GCM model inputs. Figure 9 shows that WBGT can vary substantially if there is no wind on a particular day (all other factors equal) or if workers have shade that shields them from direct solar radiation.

Figure 9

fig 9 - primer climate

The analysis of productivity loss is further complicated by time-of-day considerations. The calculation of outdoor WBGT in Figure 8 assumed the time was 12 noon. However, the GCM predictions for solar radiation and other quantities used to estimate WBGT are daily averages. In reality, WBGT varies over the day, starting out much lower in the morning, climbing to a maximum mid-afternoon, and then declining back to a lower level upon sunset. The temporal variation of WBGT is caused by increasing levels of solar radiation as the sun moves across the sky and also by the changing angle the sun makes with the earth. Thus, outdoor calculations of WBGT must include the precise location and time of the measurement during the day, since the amount of solar radiation and the angle of the sun depend on them. Chart 4 shows how WBGT can vary over the day with respect to solar radiation and wind speed. Productivity estimates, then, depend on the time of day.

Chart 4

chart 4 - primer on climate risk

Summarizing, the climate risk analyst is confronted with the following uncertainties in projecting the effect of heat stress on any location:

  • Limited academic work on labor productivity functions that depend on WBGT
  • Sensitivity of labor productivity functions to small changes in WBGT
  • Labor productivity functions likely vary substantially by country and industry
  • Labor productivity functions depend on the amount of mechanization available, likely varying with the income of the country
  • High sensitivity of WBGT to meteorological inputs that can vary throughout the day and cannot be projected reasonably by GCMs
  • Sensitivity of results to simple heat stress mitigation strategies such as providing shade and/or fans outdoors, fans and air conditioning indoors, working at different times of day or at night, etc.

With these uncertainties, a heat stress analysis could produce a very wide range of possible increases in costs to particular businesses as well as effects on the local economy. A stress test is a good risk management tool in this situation since an analyst can make a range of assumptions, calculate risk parameters conditional on the assumptions, and then bound the size of potential losses. However, given the considerable uncertainty of estimates and wide variability of potential outcomes, incorporating physical risk directly into risk management models and parameters would not be credible and therefore would not be useful, given that almost any result could be obtained with little empirical evidence to guide the analysis.

If including physical risk into risk management models is not useful for banks at this time given the uncertainties, it would be counterproductive for regulatory policy to require banks to do it. Although the ECB has commendably emphasized the development of climate stress testing, it has proposed some regulatory expectations that are not feasible for banks to comply with given the current uncertainties in the academic research. For example, for physical risk the ECB “…expects banks to also account for acute physical risk within their credit models…”[23] However as we have seen, there is generally no scientific consensus on where acute physical risks are getting worse (or better) and substantial uncertainty on their magnitudes, making accurate adjustment of parameters of credit risk models infeasible at this stage.  

Similarly, the BCBS consultation document suggests rules that would require banks to include geographical area in physical risk exposure disclosures. The motivation for this disclosure is to give market participants better information to understand a bank’s physical climate change risk. The example template, CRFR2, included in the BCBS consultative document, would require banks to disclose both corporate and real estate exposures in geographical regions subject to physical risk along with the bank’s selection methodology. As we have seen, this disclosure would not be credible since geographically precise statements at present cannot be made with any accuracy. 

The requirement to disclose physical risk by geographical area is not without potential costs: it may also increase litigation risk. Banks could be accused of failing to mitigate, adapt to climate change or report material risks when they have excluded particular locations from Pillar 3 reporting that external actors believe are subject to physical risk, given their own reading of the highly uncertain evidence. Moreover, the potential increase in risk would not be accompanied by the benefit of increased market clarity from the disclosures, since geographically specific physical risk exposures would have high uncertainty.


Given the uncertainties reviewed in this note, climate risk managers at banks should focus their risk management efforts on developing robust physical risk stress-testing capability. Stress tests are ideal for answering questions such as: What if hurricane risk decreases in the Northeast and shifts to the east coast of Florida? Is there a concentration of physical risk that is too large there? What would happen if the insurance market failed in the face of a giant storm? Credit ratings, probabilities of default (PD), losses given default (LGD) and other quantities could be modified conditional on the stress tests to quantify hypothetical losses. Stress tests allow bank risk managers to bound the order of magnitude of physical risk losses under hypothetical conditions, which can be very useful for risk management. Although it is tempting for banks to go farther, perhaps adjusting metrics such as PD for physical risk or setting geographic limits, the current uncertainties would make these methodologies unusable. 

Climate regulatory policy should follow the same philosophy, encouraging banks to develop robust stress testing capabilities. Regulatory policy will be counterproductive if it gets ahead of the science by requiring banks to implement models and methodologies for climate that are not possible given the current uncertainties. If regulators require banks to disclose climate metrics that are highly uncertain, they may be creating significant legal risks for banks.

[1] Board of Governors of the Federal Reserve System, “Pilot Climate Scenario Analysis Exercise: Participant Instructions,” (2023), available athttps://www.federalreserve.gov/publications/files/csa-instructions-20230117.pdf 

[2] European Central Bank, “Guide on climate-related and environmental risks: Supervisory expectations relating to risk management and disclosure,” (2020), available at https://www.bankingsupervision.europa.eu/legalframework/publiccons/pdf/climate-related_risks/ssm.202005_draft_guide_on_climate-related_and_environmental_risks.en.pdf

[3] Basel Committee on Banking Supervision, “Disclosure of climate-related financial risks,” (2023), available at https://www.bis.org/bcbs/publ/d560.pdf

[4] Knutson, T, Camargo, S, Chan, J, Emanuel,K, Ho, C, Kossin, J, Mohapatra, M, Satoh, M, Sugi, M, Walsh, K and Wu, L. “Tropical Cyclones and Climate Change Assessment Part II: Projected Response to Anthropogenic Warming,” Bulletin of the American Meteorological Society, (2020)

[5] Economic damage is non-linearly related to hurricane wind speed, so damage would be expected to rise much more than 5 percent. For further analysis, 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/

[6] Hall, T, Kossen, J, Thompson, T, McMahon, J, “U.S. Tropical Cyclone Activity in the 2030s Based on Projected Changes in Tropical Sea Surface Temperature,” (2020), Journal of Climate

[7] Nakamura, J et al, “Western Northern Pacific Tropical Cyclone Model Tracks in Present and Future Climates,” Journal of Geophysical Research Atmospheres, (2017)

[8] ECB/ESRB, “Climate-related risk and financial stability Data Supplement,” (2021), available at https://www.ecb.europa.eu/pub/pdf/other/ecb.climateriskfinancialstability202107_annex~4bfc2dbc5e.en.pdf?ac86a11e85cb0b7efc726e234831412b

[9] See https://www.osha.gov/heat-exposure/hazards

[10] WBGT has been recommended as the preferred heat stress measure by ISO 7243 since 1989. For the most current version of ISO 7243, see https://www.iso.org/obp/ui/#iso:std:iso:7243:ed-3:v1:en

[11] Lemke, B and Kjellstrom, T, “Calculating Workplace WBGT from Meteorological Data: A Tool for Climate Change Assessment,” Industrial Health, (2012)

[12] Bernard, T and Pourmoghani, M, “Prediction of workplace wet bulb globe temperature,” Applied Occupational Environmental Hygiene, (1999)

[13] Liljegren, J, Carhart, R, Lawdy P, Tschopp S, and Sharp R, “Modeling wet bulb globe temperature using standard meteorological measurements,” Journal of Occupational and Environmental Hygiene, (2008)

[14] See http://www.bom.gov.au/info/thermal_stress/#approximation

[15] Network For the Greening of the Financial System, “NGFS Climate Scenarios Technical Documentation,” (2023), available at https://www.ngfs.net/sites/default/files/media/2024/01/16/ngfs_scenarios_technical_documentation_phase_iv_2023.pdf

[16] See https://www.cesm.ucar.edu/models/cesm2 for more information

[17] See https://gmd.copernicus.org/articles/12/3241/2019/ for more information

[18] https://pangeo.io/

[19] See Cai, X, Lu, Y and Wang, J, “The Impact of Temperature on Manufacturing Worker Productivity: Evidence from Personnel Data,” Journal of Comparative Economics, (2018) and Somanathan, E, Somanathan, R, Sudarshan, A and Tewari, M, “The Impact of Temperature on Productivity and Labor Supply: Evidence from Indian Manufacturing,” Journal of Political Economy, (2021) for a few recent examples.

[20] Brode, P, Fiala, D, Lemke, B, and Kjellstrom, T, “Estimated work ability in warm outdoor environments depends on the chosen heat stress assessment metrics,” International Journal of Biometeorology, (2017)

[21] Kjellstrom, T, Freyberg, C, Lemke, B, Otto, M and Briggs, D, “Estimating population heat exposure and impacts on working people in conjunction with climate change,” International Journal of Biometeorology, (2017)

[22] Data from Nasa Earth Exchange Global Daily Downscaled Projections (NEX-GDDP-CMIP6) available at https://www.nccs.nasa.gov/services/data-collections/land-based-products/nex-gddp-cmip6

[23] European Central Bank, “ECB report on good practices for climate risk stress testing,” (2022), pg 42