The future of cat models
Modelling natural catastrophes is at the heart of what Bermuda’s re/insurers do. It is an imprecise business, but the industry has a vested interest in achieving precise results. Welcome to the Bermuda:Re+ILS cat models survey.
“The cat events we are trying to model are not stationary; we are applying past experience to a changing environment.”
Given the highly complex systems that drive weather patterns, modelling catastrophe (cat) risk has always been a particular challenge. In recent years, however, that challenge appears to have been getting increasingly difficult, due to the impact of climate change.
Whether it is rampant wildfires, record-breaking Atlantic hurricane seasons, or rivers running through city centres, extreme weather has shocked politicians and the public in recent years.
Re/insurers, on the other hand, cannot afford to be surprised, and must have an accurate sense of the risk generated by natural catastrophes (nat cats) if they are to remain relevant—and solvent.
Bermuda:Re+ILS wanted to understand better how readers see this challenge, and how confident they are in their ability to overcome it.
How confident are you in the industry’s ability to improve its cat models in the next five years?
Models for the future
The industry is cautiously optimistic about its ability to improve cat models in the medium-term horizon of five years, with 56 percent of respondents describing themselves as “a little confident”, and a further 23 percent feeling “quite confident” (Figure 1). Just under one in five (19 percent) said they were “extremely confident”, while only 2 percent admitted they were not at all confident in the industry’s ability to improve its cat models.
“They put great energy into veracity and there’s no reason to believe they’ll not continue to improve,” noted one of the more upbeat respondents
Another had a different perspective. “I think the models will begin to fold in the effects of climate change in more advanced ways. I also think that the modellers will be improving some of their non-peak peril models significantly over that time,” said the respondent.
Some felt that the attention of the modellers might be elsewhere over the given timeframe. “The pandemic will be hot and very fluid as well. The crisis of the moment will be the overriding priority, while cyber will remain vague and reactive at best.”
Others argued that the challenges around modelling a risk which is being impacted by climate change will limit improvements. “Chaotic systems are notoriously difficult to replicate, and hidden correlations mask significant relationships,” said one respondent.
“The cat events we are trying to model are not stationary; we are applying past experience to a changing environment. Events that never happened before are now happening due to global warming.
“I do not see a blending of new global-warming science with old statistics modelling. Therefore, I am not confident that modelling will always be at least 90 percent accurate,” commented another pessimistic respondent.
Who is driving the most improvements in cat models?
New players on the scene
Having established that the industry will likely be enjoying the benefits of improved cat models over the next five years (however modest), the next question is: who will be delivering those improvements?
The dedicated modelling companies are still the most trusted businesses to lead the way in terms of improving cat models, according to 42 percent of respondents (Figure 2). This did, however, leave more than half of the respondents backing other types of companies to beat the specialists at their own game and do more to drive improvements in cat modelling.
The second largest group expected to drive improvements were the fintechs, who have already done much to shake up other parts of the industry. In all, 29 percent of respondents backed them to improve the predictive power of the models.
Only 11 percent of respondents backed the re/insurance companies themselves, but 18 percent said that other types of companies not mentioned in the survey would play the biggest role, including academics, re/insurance brokers and “all of the above”.
Which catastrophes are the most difficult to model accurately?
Cyber and other threats
Not all perils are created equal, in terms of their impact or, more importantly, in the context of risk modelling their predictability. So which perils do re/insurers have most confidence in their ability to predict, and which ones pose the greatest challenges?
According to our survey, by far the most challenging peril to model at this stage is cyber, with 44 percent of respondents citing it as the most problematic (Figure 3). Cyber is the newest threat on the list, hence it is also the peril for which there is least historical data on which to base models. It will be interesting to observe how this perception evolves over time, as the amount of data that re/insurers have about cyber risk increases.
One respondent argued that cyber should be seen less as a catastrophe and more as an omnipresent threat that businesses will have to learn to manage. “Cyber is not just an ‘event-driven’ situation, it’s an ongoing obligation now. Be aware,” the respondent warned.
Meanwhile, 30 percent of respondents cited wildfires, a peril that has revealed its devastating potential in recent years, especially in areas such as California and Australia. Wildfires are not a new risk, but they have certainly caught re/insurers off-guard with their ferocity in the years since 2017—before that, they were not one of the leading causes of claims for nat cat insurance.
Earthquakes came in as the third most challenging peril to model, chosen by 14 percent of respondents. They are not thought to be affected by climate change, so the challenges modelling this peril are much the same as they have always been.
Only 5 percent of respondents cited hurricanes as the most difficult peril to model, which was still ahead of terrorism on 4 percent. None of the respondents to the survey listed floods as the most challenging peril to model.
Which cat models are being most severely tested by climate change?*
*Respondents could choose more than one answer
Modelling natural disasters
Having established which perils are the hardest to model, we wanted to know which were being most severely affected by climate change. The answer from our respondents was emphatic, with 63 percent citing wildfire (Figure 4).
The interplay between hotter and drier weather has pushed wildfires to the top of the agenda for nat cat modellers, while demographic transition and urbanisation have also made its impact more costly. Respondents were free to pick multiple answers, but floods and hurricanes came out with relatively modest figures of 26 percent and 23 percent, respectively.
One respondent noted that “climate change, insofar as it is real, is a slow process, moving means and perhaps increasing variances, but not altering operating pathways—at least not yet.”
Are re/insurers investing enough in developing and improving their cat models?
Could do better?
Modelling catastrophes is a challenging business, and one that climate change is making even more difficult. It is also clear that re/insurers need to overcome these challenges and model these risks as accurately as possible if they are to remain profitable and retain the trust of their customers.
Given the importance of this task, how much money are re/insurers throwing at this problem? Is it enough?
Respondents were relatively evenly distributed in their answers to this question (Figure 5). Nearly one-third (32 percent) admitted they could perhaps afford to invest a little more, while other results were almost flat: 28 percent felt that a lot more investment was needed, whereas 27 percent said that an appropriate amount was already being invested.
Only 13 percent argued that re/insurers already spend too much on models, which is a high number—given the importance the industry places on being accurate in its forecasts.
Several respondents pointed out that the purpose of a model is not to provide a definitive number to quantify the risk of a peril. “Models do not present the answer to the question being asked, they present an assessment of the effect/impact of a cat event. They offer a range, not an answer,” said one respondent.
Another noted that the value of the models is as much in mitigation as in prediction, saying “Remember that the fundamental purpose of insurance is to incentivise prevention.”