Skip to main content
main content, press tab to continue
Article | WTW Research Network Newsletter

Why relying on frequency-severity adjustments could underestimate your tail risk from climate change

By Cameron Rye | June 20, 2023

When modifying insurance catastrophe models for current and future climates, it’s vital to consider whether changes to losses are consistent with expectations and if there could be unquantified risks lurking in the tail.
Environmental Risks|Insurance Consulting and Technology|Risk and Analytics|Willis Research Network
Climate Risk and Resilience|Insurer Solutions

Climate change is altering the frequency and severity of extreme weather events such as wildfires, floods, and windstorms. Over the last few years, it has become increasingly common for insurers to adjust catastrophe model output to reflect these changes, driven in part by new regulatory requirements[1]. While most of the focus has been placed on methods for adjusting frequency-severity relationships, little attention has been given to scenario completeness, particularly in the tail of the distribution where some of the most severe impacts are expected to materialise. As a result, insurers could be underestimating the physical risks to which they are exposed.

The dominance of short return periods

When it comes to physical climate change impacts on the insurance industry, one of the most cited research papers is Knutson et al. (2020)[2], which presented a synthesis of the expected changes in global tropical cyclone activity for a 2°C warming scenario. This work has been translated into catastrophe model frequency-severity adjustments[3], which allows insurers to estimate how changes in the frequency and severity of tropical cyclones could affect insurance claims. These types of adjustments have been widely used by insurers for regulatory stress tests such as the UK’s Climate Biennial Exploratory Scenario (CBES)[4].

But if you have ever applied the Knutson et al. frequency-severity adjustments to a catastrophe model, then you might have noticed something that at first seems counterintuitive: after the correction, short return period losses tend to increase more percentage-wise than those in the tail of the distribution. Figure 1 below shows the impact of a 20% increase in the number of Category 4 and 5 landfalling hurricanes in a U.S. tropical cyclone model. The largest effect is seen near the bottom of the exceedance probability (EP) curve, with a 15% increase at the 1-in-2 year return period loss. In comparison, tail losses around the 1-in-200 year return period only increase by 5.5%.

The reason for this inconsistency is simple: short return periods comprise most of the loss distribution, so adding in new events fattens this portion of the EP curve the most. In a 100,000-year simulation, for example, 90% of the years are at or below the 1-in-10 return period. This means that if a new event is included at random, it has a 90% chance of being added to a year at or below the 1-in-10.

This effect is counterintuitive because we know that the upper tail of the distribution is likely to contain some of the most severe physical consequences of climate change, particularly under higher emission pathways. Does it, therefore, stand to reason that the percentage increase in tail return periods should be lower compared to shorter return periods? What could we be missing from our scenarios that explains this?

Unquantified tail risks

As Nassim Taleb has written about in relation to financial markets, traditional models do not handle fat-tailed events well[5]. This limitation means that important things are likely to be missing from the view of risk. The same is true for traditional catastrophe models when it comes to climate change: while frequency-severity distributions can be conditioned for various climate states, they underestimate the true tail risk, especially when we start to think about things like tipping points, feedback loops, and systemic risks.

This is not to diminish the usefulness of catastrophe models. They combine detailed hazard modelling, engineering knowledge, and information on exposures in ways that other tools, such as climate models, cannot. However, just as insurers evaluate and quantify non-modelled risks today (e.g., under Solvency II), they need to deploy the same thinking and methodologies when it comes to climate change adjustments and scenarios.

For example, traditional risk assessment methods often focus on a single hazard at a time, but research indicates that the likelihood of hazards co-occurring – such as extreme winds and precipitation – will increase in a warmer world[6]. An increase in cross-peril correlation will lead to increased tail risks, but most insurers are not currently considering this possibility in their modelling.

There is also mounting evidence that some tipping points, such as the rapid collapse of ice sheets or the melting of Arctic permafrost, may be triggered at a global mean temperature of 1.5°C[7]. Such events would have far-reaching ramifications, with severe consequences not just for the insurance industry, but society as a whole. The world is expected to reach 1.5°C at some point in the 2030s[8], meaning some of these fat tail outcomes could be closer than many realize.

In addition to these direct physical risks, there are several indirect effects that are frequently overlooked, including supply chain disruption, food insecurity, geopolitical conflict, and infrastructure failure. All of these have the potential to manifest as systemic effects, which will stress local, regional, and national economies.

The WTW Research Network is exploring and quantifying many of these tail and emerging risks to offer broader perspectives for our clients, and some of those themes are highlighted in our annual review.

Climate Amplification Factors

The breadth and complexity of climate change tail risks mean that careful consideration is required when incorporating them into our modelling. In some situations, it will be possible to explicitly simulate the effects within existing frameworks – for example, cross-peril correlations can be included in simulation-based capital models. However, it will be more challenging for other risks, particularly those with socio-economic and systemic components.

Fortunately, there is an analogue in current catastrophe modelling frameworks when it comes to representing socio-economic factors that are difficult to quantify. For nearly two decades, catastrophe models have used post-event loss amplification (PLA) to account for a range of complex tail sources of loss including economic demand surge, long-term evacuation, and systemic economic problems. PLA is applied to the most severe events, such as Hurricane Katrina in 2005, where traditional hazard and vulnerability modelling struggle to capture the full extent of socio-economic factors that lead to elevated losses.

The same approach could be used to model climate change tail risk using “climate loss amplification” (CLA). The more severe an event, then the larger the CLA factor, reflecting the increasing likelihood of socio-economic effects materialising such as geopolitical conflict and infrastructure failure.

So, when you next think about building or updating your climate change scenarios, remember to consider not only how to best adjust frequencies and severities, but also the completeness of your risk assessment.


  1. e.g. Supervisory Statement SS3/19: Enhancing banks’ and insurers’ approaches to managing the financial risks from climate change (Prudential Regulatory Authority, 2019). Return to article
  2. Knutson T. et al (2020), Tropical cyclones and climate change assessment: Part II: projected response to anthropogenic warming, BAMS 101(3):E303–E322. Return to article
  3. e.g. Jewson, S., Conversion of the Knutson et al. Tropical Cyclone Climate Change Projections to Risk Model Baselines. Journal of Applied Meterology and Climatology, 60(11):1517-1530. Return to article
  4. Key elements of the 2021 Biennial Exploratory Scenario: Financial risks from climate change (Prudential Regulatory Authority, 2021). Return to article
  5. Taleb, N. (2022), Statistical Consequences of Fat Tails: Real World Preasymptotics, Epistemology, and Applications (Technical Incerto), STEM Academic Press, ISBN-10 1544508050. Return to article
  6. Ridder, N. et al. (2022), Increased occurrence of high impact compound events under climate change, Njp Climate and Atmospheric Science, 5(3). Return to article
  7. OECD (2022), Climate Tipping Points: Insights for Effective Policy Action. Return to article
  8. Marotzke, J. et al. (2022), How close are we to 1.5 degC or 2 degC of global warming? Weather, 77(4). Return to article

Head of Modelling Research and Innovation
email Email

Contact us