The value of CAT models for measuring climate risk

There is an interesting debate simmering at the moment as to whether catastrophe loss (CAT) models or climate risk models are best for modelling the financial impacts of physical climate risk. Presently there is a strong dose of marketing involved by both camps, but an objective discussion on this topic is needed.

CAT models have been developed over the past 30-plus years with the explicit aim of modelling the financial impacts of extreme weather and associated asset damage. They have traditionally not included a climate change component and look to model the ‘baseline’ view of risk, asking the question – what is my risk profile over the coming financial year?

Over a similar period of time, there has been a gradual realisation that there may be longer term financial risks associated with climate change. The Intergovernmental Panel on Climate Change (IPCC) has been warning of the dangers of anthropogenic global warming since the 1990s¹, but more recently the Taskforce of Climate Related Financial Disclosures (TCFD) has provided a framework for companies to disclose the financial risks (and opportunities) posed to them by climate change.

As a result, a new generation of CAT models is now emerging that quantifies the financial impacts of future climate risk. These models try to take the best bits of GCMs (General Circulation Models, an acronym for climate models) and the existing CAT model framework to look at future risk.

The opportunities for using CAT models to measure the impacts of climate change for a wide range of industries are large. Yet, CAT models are still a relatively unknown quantity outside of insurance and have been slow to capitalise on this new market. Here are some of the reasons why CAT models are well placed to measure climate risk.

To understand the impacts of climate change in the future, you need to have a good understanding of your exposure to weather and climate today. This is where a baseline view of risk is crucial.

Historical disaster losses show that extreme weather impacts are as much about where and how we choose to live, as they are about underlying changes to the hazard². In other words, climate risk cannot be assessed without consideration for the built environment.

CAT models have a three-dimensional view of risk at their core, where risk is a function of not only the hazard, but also where the assets are (exposure) and what they are made of (vulnerability). CAT models also talk in terms of natural hazards, not climate parameters. For example, extreme heat does not equate to bushfire risk, nor does heavy rainfall equate to flood risk.

By comparison, some climate risk models use climate parameters as proxies for natural hazards, they cannot always distinguish the relative importance between perils, and often take hazard to equal ‘risk’. These assumptions can all lead to an inflation of projected climate risk.

While knowing the percentage change in the frequency of extreme rainfall by 2050 may be useful for some applications, it tells us nothing about the financial implications of this change. For example, how does this translate to property damage, insurance affordability or the probability of default on a home loan?

CAT models provide a vast amount of information on the probability of a certain level of dollar loss from extreme weather events being exceeded. Fortunately, this can be condensed down into a very powerful physical climate risk metric - the Average Annual Loss (AAL).

The AAL describes the dollar loss that can be expected in any given year, under a set of stable climate conditions. The AAL is the basis for the technical cost of insurance (or the break-even cost for an insurer) and is therefore a directly relevant metric for insurance and banking, where much of the climate risk is related to insurance affordability.

A CAT model can be ‘climate-conditioned’ in a number of ways, so that the event set in the model reflects projected future climate states, rather than present day conditions.

The result is an AAL for today, and an AAL for a range of climate futures. The difference between the present and future AAL is typically expressed as a median percentage change and is called ‘delta AAL’.

Delta AAL includes what we know about the relative contribution of each peril to a portfolio today, the climate-conditioning of each peril-specific CAT model, and the aggregation of these results into a single, financially actionable climate risk metric.

The bottom line for most boards when presented with a climate risk assessment is whether any change is material to their business. In other words, could the magnitude of the projected change be expected year-on-year due to natural variability (in the climate, or financial markets), irrespective of underlying climate change?

This is a difficult question to answer and is often left in the too-hard basket. However, by leveraging the probabilities in the CAT model event set, the climate change signal (delta AAL) can be presented in the context of expected year-on-year variability.

In a recent study³, Aon developed a method to determine whether a delta AAL was in fact significant or not, for a range of climate futures. This provides some realism around the materiality of projected physical climate risk and allows the client’s risk appetite to be considered in what is deemed as material.

While CAT models have a lot to offer for assessing climate risk, there are still several gaps and opportunities for improvements. To provide a more complete TCFD solution, CAT modellers need to also consider how we assess chronic climate hazards, supply chain risk, and peril correlations.

Physical climate risk does not stop at extreme weather perils. Chronic climate hazards, such as drought, heatwaves, and water availability, can also have significant impacts on business operations but for which there is no dedicated CAT model solution.

In addition, CAT models assess risk on an in-situ basis, but physical climate risk can also be inherited upstream, and propagated downstream, through a supply chain. The interconnectivity of assets is not considered in the traditional CAT modelling approach.

Finally, the reactionary nature of CAT model development means the coverage of available models – across perils and regions – remains incomplete. This means the frequency correlation between different perils modelled across the same region is often not considered, and the loss correlation between global regions is difficult to quantify. This is a problem for those with a global portfolio today and compounded by the fact that these correlations are likely to change in the future with climate change.

CAT models can put a dollar loss value on extreme weather risk and have been doing so for the past several decades. Their holistic view of risk, financial applications and reasonable handling of tail risk and uncertainty, makes them well placed to measure the impacts of future climate risk for a large range of industries beyond insurance. However, they do not (yet) offer a seamless global coverage, do not include climate events such as heatwaves or water availability, and struggle with supply chain factors. If these gaps can be supplemented by other data sources, or incorporated by CAT model vendors, this would represent a truly holistic physical climate risk solution.