7 Stress testing

Banks and insurers generally make money in two ways, investing the capital that they are holding and retaining payments from their customers. A major economic shift, such as a transition to a low-carbon economy, could have major implications on their investments. Major physical changes in the world’s climate system could increase the frequency and severity of weather-related catastrophes. The frequency and severity of these catastrophes has important implications for the financial sector, in particular for the underwriting of insurance premiums and for the lending of capital for residential and commercial property. If not otherwise addressed, the increased frequency and severity of weather-related catastrophes is likely to result in consumers making more claims on policies and property owners defaulting on their loans, thereby impacting the ability to collect and retain payments from customers.

7.1 Weather-related catastrophes

In fact, the impact of physical changes in the world’s climate system is already becoming observable. Temperature increases in the world’s atmosphere are leading to increases in the frequency and severity of certain weather-related catastrophes. For instance, an IPCC study showed that climate change will result in weather-related catastrophes becoming normal occurrences (IPCC 2001; Mills et al. 2018, 2). The World Economic Forum further identified extreme weather events as their top concern (WEF 2018; Mills et al. 2018, 4). Moreover, according to the TCFD, “The scientific consensus is that a continued rise in average global temperatures will have a significant effect on weather-related natural catastrophes and will account for an increasingly large share of natural catastrophe losses” (TCFD 2017, 28). Figure 7.1 shows how temperature increases can exacerbate the frequency and severity of weather-related catastrophes.

Wildfires, hurricanes, windstorms, floods, and droughts are among the weather-related catastrophes that are likely to increase in frequency and severity as a result of climate change (IPCC 2001). Significant weather-related losses are already being observed. California has experienced fifteen of its twenty largest wildfires in state history since 2000, with devastating wildfires in the last three consecutive years. The larger and more common wildfires have coincided with warmer, drier, and longer-lasting wildfire seasons. CalFire has said that “climate change is considered a key driver of this trend” (Samenow and Freedman 2019). Total financial losses have amounted to tens of billions of dollars (Samenow and Freedman 2019), and the insurance industry has paid many billions on dollars in claims related to the wildfires in the last few years alone, with over twelve billion dollars in insured losses in a single month (CDI 2019).

Meanwhile, some of the most destructive storms have recently occurred. Harvey, Irma, and Maria hit the Southeastern United States and Caribbean in 2017. In August, Harvey was a Category 4 storm that struck Texas with torrential rainfall and flooding. In September, Irma hit the Virgin Islands and then Florida. Maria hit Puerto Rico, disabling large portions of the island. This set of storms coincided with warmer ocean waters and a conductive Atlantic weather pattern. Harvey, Irma, and Maria are each among the top five costliest storms (Fritz 2018). Total financial losses have accounted to over two hundred billion dollars (Fritz 2018), and the insurance industry has paid between fifty and one hundred billion dollars in claims related to these three storms alone (III n.d.).

7.2 Metrics

Stress testing is an analytical tool used to evaluate the impact of severe events on the profitability or solvency of companies. For stress testing, catastrophe modeling is typically used for evaluating losses connected with weather-related catastrophes. Catastrophe models are computerized systems that estimate potential losses from extreme weather event simulations. Insurers and banks have been using catastrophe models to evaluate their exposure to risks. For insurers, it is important to ensure that they set capital requirements appropriately so that they have sufficient available reserves to pay claims made on policies. In fact, according to the professional service firm PricewaterhouseCoopers, United States insurance regulatory best practices demand solvency assessments to determine that insurance companies have sufficient capital reserves under both normal and stressed conditions (PwC 2016; Mills et al. 2018, 64). For banks, it is also important to ensure that they set capital requirements appropriately so that they make loans in a manner where they are not overleveraged. Catastrophe models can inform these decisions.

7.2.1 Overview

The Bank of England has long advocated for stress testing, specifying best practices and conducting its own stress testing in certain circumstances. This stress testing helps identify events that are especially likely to adversely impact capital requirements and the companies that are likely most exposed to these events (Dent, Westwood, and Segoviano 2016). However, outside of the United Kingdom, insurers and banks have largely been left to decide on their own on when and how to apply stress testing. Financial supervisors and regulators are generally neither specific in how stress testing should take place, nor do they conduct their own stress testing to verify or otherwise assess private-sector practices.

There is very little data on how the private sector is performing stress testing related to extreme weather events. With respect to what types of weather-related stress testing efforts the insurance industry is performing on their own underwriting risks, not much is known on which companies are even conducting such an analysis. In the results of the only mandatory insurance industry–specific climate-related reporting framework, the NAIC Climate Risk Disclosure Survey, not that many insurers state that they are conducing such an analysis. Those insurers that state that they are conducting such an analysis typically do not elaborate on what the analysis entails (Mills et al. 2018, 63–65). Although unfortunately a comprehensive understanding of stress testing practices is not available, insurers and banks do appear to be incorporating catastrophe modeling into business practices in several ways, albeit generally only for single events in specific geographic regions rather than multiple events that could occur in combination (Mills et al. 2018, 14–19, 60–65).

First, insurers and banks retain third-party catastrophe modeling firms to assess their particular portfolios of liabilities for specific events. Doing so is likely less expensive than developing their own tools, but they may not be as precisely calibrated to portfolios of liabilities, and they would not provide a competitive advantage over other companies that are purchasing the same or similar tools. A number of commercial vendors provide catastrophe models, and the modeling of specific simulations can be performed by the vendor or subcontracted to another party.

Second, some insurers and banks are developing their own catastrophe models or running their own simulations on open source catastrophe models. For example, Metropolitan Life Insurance Company tests for storm conditions ranging from 1-in-100 to 1-in-2,000-year probabilities (MetLife 2016; Mills et al. 2018, 64–65). Companies that develop their own catastrophe models or simulations can specifically assess risks that they believe are most pertinent to their portfolios of liabilities and can potentially gain competitive advantages over other companies. However, developing catastrophe models and simulations can be expensive, and the ability of such tools to accurately gauge risk is not certain, especially when compared to catastrophe models and simulations developed by companies specializing in creating catastrophe models and simulations.

Third, some insurers are relying on reinsurers to use catastrophe models when the reinsurers decide whether they will ultimately hold selected layers of insurers’ risks for losses related to extreme weather events. Catastrophe models generate pricing for specific loss events using simulations. Insurers and reinsurers then negotiate pricing for specific perils using what catastrophe models reveal about the likelihood of the occurrence of events and related damage amounts. Because reinsurers have worldwide portfolios, this allows for the spreading of risk and less concentrated exposure (American Academy of Actuaries 2018, 26–28).

Fourth, some insurers are providing their portfolio of liabilities to rating agencies that rate companies at least in part on information from catastrophe models (S&P Global 2018; Lavakare and Mawk 2008). A rating using input from catastrophe models allows investors to view a comprehensive evaluation of solvency that takes into account exposure to extreme weather events without insurers or banks having to publicly reveal data that it deems private or proprietary. For example, a rating agency can seek catastrophe model information for financial strength ratings, such as information on exposure to different perils in different geographic regions. However, the majority of insurers and banks are not necessarily providing their underwriting and lending information to rating agencies for evaluation. How such ratings are being composed using such information is not entirely clear either.

There has also not been a lot of sectorwide catastrophe modeling that explores how large catastrophic events could impact sectors as a whole. For a very long time, there had been only limited attempts to test sectorwide impacts of major catastrophes. The most well-known insurance industry stress testing effort was performed over thirty years ago when an insurance research organization financed a report examining how two seven-billion-dollar hurricanes would impact the underwriting of the insurance sector as a whole. The analysis revealed that a number of smaller insurers would likely become insolvent under this simulation being stress tested (All-Industry Research Advisory Council 1986). At that time, such an occurrence of events seemed exceedingly unlikely. However, the likelihood of an occurrence of such events is no longer questioned, but not that much more has been done to test for them on a sectorwide basis (Mills et al. 2018, 63).

Only recently has more sectorwide analysis started to occur. The most significant of this is likely the analysis that the Bank of England has been including in its annual insurer stress tests. In addition to examining the impact of a deterioration in economic markets, the Bank of England has incorporated stress tests for liability shock into its annual insurance sector stress testing. This includes extreme weather event simulations involving major hurricanes, windstorms, and floods. These simulations are to be applied while at the same time taking into consideration a deterioration in market conditions that includes a downward shift in risk-free interest rates, a widening in corporate bond spreads, and a decrease in the value of other assets (Bank of England 2019, 4).

Among the simulations is one where there is a cluster of three hurricanes making landfall in the continental United States with approximately one hundred and eighty-one billion dollars in aggregate insurance sector losses. This stress simulation involves a situation similar to when Harvey, Irma, and Maria occurred in the same year. Here an Irma-like hurricane makes two landfalls in Florida, a Harvey-like hurricane makes landfall in Houston, and a third hurricane dissimilar from Maria makes landfall on the East Coast of the United States. The simulation provides separate modeled hurricane tracks. There is also data provided on how different vendors’ catastrophe models display storm severity and gross losses. Insurers are asked to provide loss information specific to types of perils and their lines of business and coverage (Bank of England 2019, 4, 13–16).

Among the extreme weather event simulations is also one where there is a large United Kingdom windstorm and a United Kingdom flood, leading to approximately twenty-five billion dollars of aggregate United Kingdom insurance sector losses. This simulation includes a set of two events. First, there is a large United Kingdom windstorm, resulting in sizable storm surge losses along the East Coast of England. Second, there is significant flooding in England and Wales. There is also data provided on how different vendor catastrophe models display event severity and gross losses. Insurers are asked to provide loss information specific to types of perils and their lines of business and coverage (Bank of England 2019, 23–25).

Some significant sectorwide stress testing efforts have also been conducted in the United States. In 2015, Florida’s insurance regulator asked insurers for catastrophe model results that evaluated their ability to absorb one or more historical hurricane simulations. Insurers needed to demonstrate how the simulations would impact their reserves in terms of capital and surplus (FLOIR 2019; Mills et al. 2018, 65). With respect to what types of climate-related catastrophes are being modeled on lending, a study involving Risk Management Solutions – a major international risk modeler – was conducted on the impact of droughts. After applying a drought stress-test model to over a dozen industry sectors in countries around the world, the study determined that there was the possibility of the erosion of creditworthiness with bank loans and increased default risks in certain industries such as water supply, agriculture, food production, and power generation (Carter and Moss 2017, 8–9; Mills et al. 2018, 63–64).

In recent years, there have also been calls to update stress testing to include climate-related projections that can be incorporated into catastrophe models. As far as what is understood from publicly available information, most catastrophe models have incorporated weather events with behaviors based on historic events, but financial supervisors and regulators, academic institutions, companies, and investors are increasingly seeking the inclusion of weather events based on climate projections.

7.2.2 Bank of England

The Bank of England has been leading the development of forward-looking stress tests for liabilities. In 2019, the Bank of England asked insurers to conduct what it labeled as “exploratory exercise” with their annual insurance sector stress testing. The Bank of England explained that the exercise was “designed to capture information about how different firms are exposed to difficult-to-assess risks, eg climate change” (Rule 2019).

The Bank of England outlined several potential climate change scenarios for insurers to evaluate. First, a scenario where there is a sudden transition to a low-carbon economy with global temperature targets being aligned with the Paris Agreement by 2022. Second, a scenario where there are disorderly transitions that are identified in the IPCC Fifth Assessment Report published in 2014. This scenario envisions a long-term transition to a low-carbon economy, with global temperature targets being aligned with the Paris Agreement by 2050. This scenario is based on several decade transition scenarios that are identified in the IPCC’s recent report emphasizing the importance of limiting global warming to 1.5 degrees Celsius. Third, a scenario where there is a failed transition to a low-carbon economy, where there is global temperature increase in excess of 4 degrees Celsius by 2100 (Bank of England 2019, 28).

For each potential climate change scenario, the Bank of England outlined likely changes in weather-related occurrences and economic activities. This includes information on expected increases in the frequency and severity of certain weather-related catastrophes, including hurricanes, storms, and floods. This also includes information on expected changes in fossil fuel extraction and power generation using both fossil fuels and non-fossil fuel sources such as renewable energy and nuclear power. Using this information, the Bank of England asked insurers to conduct a stress test for their liabilities in combination with a scenario analysis for their assets (Bank of England 2019, 29–37).

The Bank of England recognized that this exploratory exercise is not an exhaustive assessment and that it may be more beneficial for insurers to develop exercises that more closely meet the challenges that they believe that they are likely to encounter. Indeed, the Bank of England encouraged insurers to provide it with information from separate exercises that they have performed (Bank of England 2019, 29). Nevertheless, the Bank of England still saw the value in insurers conducting the Bank of England’s forward-looking stress testing exercise (Rule 2019).

The Bank of England also appears positioned to lead the development of forward-looking climate-related stress tests in the banking sector. The Bank of England recently announced that it will include climate simulations with its bank stress testing starting in 2021. Governor Carney stated that “banks will need to establish how their borrowers are managing current and future climate-related risks and opportunities” (Carney 2019; Horton 2019). According to Governor Carney, banks that align themselves with a net zero world “will be rewarded handsomely” and “[t]hose that fail to adapt will cease to exist” (Carney 2019; Horton 2019). Climate simulations will include one for business as usual and another for the ideal transition to a net zero emissions by 2050 that is consistent with United Kingdom and European Union’s legislative objectives (Horton 2019).

Perhaps work already conducted by the University of Cambridge could inform Bank of England forward-looking climate-related stress tests in the banking sector. The University of Cambridge recently released a study showing how climate simulations can be incorporated into catastrophe models used by banks in evaluating their real estate lending portfolios. Specifically, the University of Cambridge identified expected severity increases in European winter wind storm and tropical cyclone risk, uploaded that into an open source catastrophe model, and ran that against a dozen portfolios of assets with a market value of over two trillion dollars. The results revealed that under a 4 degrees Celsius temperature increase, United Kingdom residential mortgage portfolios could see an increase in average annual loss from floods of one hundred and thirty percent, and the number of properties at considerable risk of flood could increase by forty percent (Bronwyn et al. 2019, 7, 44–46).

7.2.3 Chinese efforts

The Industrial and Commercial Bank of China was involved with the Risk Management Solutions effort to model the impact of droughts on the banking industry, which determined that there was the possibility of erosion of creditworthiness of bank loans and increased default risks in certain industries such as water supply, agriculture, food production, and power generation (Carter and Moss 2017; Mills et al. 2018, 63–64). The effort included uploading a Chinese portfolio with approximately 2,500 companies in eleven different industries (Carter and Moss 2017, 30). The effort then ran several drought simulations against this portfolio, with droughts lasting two and five years (Carter and Moss 2017, 16). Credit rating models were used to interpret results (Carter and Moss 2017, 24).

The effort revealed the overall impact of projected droughts on the Chinese economy and impacts of droughts on specific industries and geographic regions. Certain simulations revealed only moderate credit impact in most regions but a material risk for individual companies and potentially a systemic risk to corporate lenders. Smaller companies faced a greater impact than larger companies. This was because smaller firms are more sensitive to revenue reductions and cost increases, and their operating sites are more concentrated in highly industrialized areas near major cities, which face the greatest risks among regions (Carter and Moss 2017, 30).

7.2.4 California Climate Assessment

Some significant forward-looking stress tests for liabilities have also been conducted in the insurance sector within the United States. As part of the California’s Fourth Climate Change Assessment, RAND Corporation released a report on the impact of changing wildfire risks on California’s insurance market. For this report, the California Department of Insurance provided data and otherwise supported the work through the provision of advice. Private insurers provided information on claims. The University of California, Merced provided wildfire simulations (Dixon, Tsang, and Fitts 2018, ii) through a wildfire risk tool that was developed with a Cal-Adapt effort organized by the University of California, Berkeley, and California Energy Commission also as part of California’s Fourth Climate Change Assessment (Cal-Adapt n.d.; Westerling 2018, 2020). Pairing historical data on climate, vegetation, fire, and population density with regionally downscaled localized constructed analogs for climate projections, the tool provides California wildfire burn simulations through the end of this century using Scripps Institution of Oceanography downscaled climate model outputs for the IPCC’s RCP 4.5 Scenario and RCP 8.5 Scenario (Cal-Adapt n.d.; Westerling 2018, 2020).

The RAND Corporation analysis used this information to evaluate the projected impact of wildfires for structures within dozens of zip codes through the middle and end of this century (Dixon, Tsang, and Fitts 2018, 6–14). This showed that structures in certain zip codes were likely to face a significant increase in risk (Dixon, Tsang, and Fitts 2018, 15–29). The analysis then evaluated the likely increase in cost of insurance premium by zip code using wildfire simulations composed from downscaled climate model outputs for the IPCC’s RCP 4.5 Scenario and IPCC’s RCP 8.5 Scenario (Dixon, Tsang, and Fitts 2018, 55–64). Of note, in a particularly vulnerable area of California, wildfire simulations composed using downscaled climate model outputs for the IPCC’s RCP 8.5 Scenario through the middle of the century showed little or no projected change in insurance premium for about one-third of the zip codes in the area, premium increases ranging from twenty to forty percent for approximately twenty percent of the zip codes in the area, and premium increases somewhere between for the remaining zip codes in the area (Dixon, Tsang, and Fitts 2018, 56).

S&P Global further indicated the importance of forward-looking modeling efforts, such as RAND Corporation’s analysis, because existing insurer catastrophe modeling was too limited in its ability to fully evaluate changing wildfire conditions (Wilkinson 2019). This area of modeling is rapidly developing. In particular, California has experienced significant wildfire loss since this RAND Corporation analysis was conducted, and the underlying wildfire simulations have further developed. It is unclear how exactly the additional loss and updated simulations would impact similar analysis, but a forward-looking model could take this into account.

7.2.5 Emerging efforts

The TCFD advised that insurers assess aggregate exposure to weather-related catastrophes of their property business by jurisdiction (TCFD 2017, 31). Moody’s concluded that multiple simultaneous and correlated carbon risks could exacerbate current volatility levels and adversely impact credit within the insurance industry, especially for smaller or more geographically concentrated insurers (Moody’s 2018; Mills et al. 2018, 2–3). In fact, as a result of fire that destroyed a significant portion of small California mountain community, a small insurer that had written a number of policies in that area became insolvent (Koren 2018). Forward-looking climate-related stress testing has the potential to more precisely address these types of challenges, yielding information that would be valuable in informing companies, supervisors and regulators, and investors on what companies are at risk of insolvency when confronted with weather-related catastrophes in certain areas where there are concentrated risks.

Better projections will become available as forward-looking modeling continues to improve. New climate models are increasing their resolution, and the ability to downscale climate model outputs is advancing. Greater resolution can potentially allow for incorporating individual and community-level remediation efforts, such as firebreaks, vegetation management, and home fortifications. Some modelers already assert that they have such capabilities (Zesty.ai n.d.). Further, using downscaled climate model outputs, new weather simulations allow for the inclusion of additional features. This can include, for instance, wind calculations. Wind can exacerbate or reduce the impact of wildfires depending on the wind’s intensity and temperature.

Because of the lack of available data with respect to how proprietary catastrophe models operate, it is difficult for anyone to explain with a high degree of accuracy what the best approach is for catastrophe models to incorporate climate-related risks. This could depend on what type of risk is being evaluated. There may be more comfort with incorporating simulations for weather events that seem most likely to occur in the future and that use models or model outputs with the highest resolution. For instance, while certain perils currently may be modeled with low resolutions, it currently may be possible to model other perils, such as floods, with much better resolutions (Bronwyn et al. 2019, 23). Still, there are remaining decisions to be made on how to best incorporate climate-related risks into property-specific hazard scores and whether to allow overall insurance and lending rates to be influenced by climate-related projections rather than historic loss data. If more data becomes transparent, opinions will likely be influenced by the prevalence of climate-related weather disasters moving forward, the transparency and harmonization of tools, the reliability of predictions, and the holistic incorporation of factors that fairly reflect carbon risks.