Liability Risk Modelling: What does a 1-in-100 Liability Cat look like and other key challenges for evaluating these complex risks

The ISCM in London gathered on Friday 22 November 2019 in the Lloyd’s Old Library to discuss the challenges of understanding and managing liability cat risk. The first part of the event presented two perspectives on the topic, one from the regulator presented by Giorgis Hadzilacos, PRA and another personal perspective from Matt Harrison, Hiscox. The second part of the event saw a panel of liability modelling experts from across the market discuss the challenges in modelling liability risk, and some of the ways we can work to tackle these today.
 
Part I - The Regulator’s Perspective
Giorgis Hadzilacos, PRA
 
Giorgis started the presentation reminding us that both in terms of gross written premium and in terms of volume, casualty risk is by no means small. Likewise, there is correlation within both casualty classes of business, and with other classes of business which justifies paying it more attention.
 
Aside from the size of the risk, one of the key challenges for understanding liability risk and reason it is a regulatory concern is that historical claims which are typically used for pricing and reserving models, are not necessarily good predictors of the future. There is also huge variability in the perception of the risks being underwritten with examples of some casualty writers dropping books of business due to unprofitability with other firms writing those same risks at a lower price and perceived higher rate adequacy than renewed business. Even with the risks that are being written today, the view of risk between underwriting, exposure management and capital is often different leading to inconsistent views of the risk within organisations. To add to these challenges, clash across lines of business which remain largely unassessed, and a lack of proactive reserving are some of the other reasons we don’t currently have the confidence that we are on top of casualty cat.
 
There is however progression towards a more robust liability cat assessment environment and there are a range of methodologies used to help quantify these risks being put into practice in the market today. Different methodologies each have their own merits and Giorgis emphasised the importance of recognising exposure and applying different thinking to the approaches used within the property cat space. We also need to have some way of validating outputs from these methodologies and compare against them to give the industry some comfort.
 
So how we achieve this? We need to keep striving for better data, better tools, better methodologies and most importantly share this knowledge. There’s a lot of space for non – competitive collaboration showcasing clash scenarios, emerging risk and stimulating discussions to help deal with the problem ideally well before it has become the ‘next asbestos’. Giorgis hopes that some of the work on the PRA’s agenda in 2020, will provide some indirect pressure which can accelerate learning and provide feedback to the community.
 
Part 2 
Why is Liability Cat Different? A personal perspective
Matt Harrison, Hiscox
 
Matt started off his presentation asking the critical question: what is exposure management for liability risk? Often people talk about exposure management through the lens of a cat model however you can’t manage a risk if we don’t know what it is and how it’s changing. He made the important point that if we can’t understand the risk yet, why are we even talking about a 1 in 100 year event. We need to focus on what we have and where it is.
 
Whilst the uncertainty with casualty cats is huge, so is the uncertainty in nat cat. Matt asked the question if we had a view of when a hurricane would make landfall, weeks or months ahead of time, would we do anything differently to what we do today? Would we stop writing cat risk in Miami if we knew a hurricane was going to make landfall there next year? Whilst there is inherent uncertainty, we’ve got to get better as an industry at thinking about the potential forward risk. If we take for example the Opioids crisis, there were three waves of signals in the U.S. In 1999 there was little noise or signal and not enough information to not write this risk. Looking ahead to 2016, there was lot more noise and already by 2019 it’s too late and risks which are susceptible to claims are already on the book.  At some point we’ve got to get better at recognising this and start making decisions when we see the signals vs when it’s too late. Similarly, we need to think about the peril in the timeframe of the policy. Whilst we have long policy periods, we need to think about mitigating it on a shorter scale.
 
The key reasons casualty risk is so different to property risk is quite simply the risks that you measure won’t happen, the scenarios you create probably won’t happen and the historical events you use definitely won’t happen again. Similarly, one of the difficulties is we are used to as a community using outputs of models and understanding the exposure in a geospatial way. We need to focus on understanding the policy forms, wordings and focus on what it is and where it is to help understand and manage this complex landscape.  
  
Part 3
An Expert Industry Panel
Moderator: Kirsten Mitchell – Wallace, Lloyd’s
Robin Wilkinson, AIR
Adhiraj Maitra, Willis Towers Watson
Matt Harrison, Hiscox
David Loughran, Praedicat
 
The panel opened with what we can learn from property cat and what we can take across to ensure that liability exposure management benefits from where there has been progress elsewhere. The general agreement was the need for better exposure capture and better understanding the nature and size of the business that is being written. We need the data to run any model and as demonstrated with property cat, the better the data, the better the results.
 
We also need to understand the potential footprint for a casualty cat event and what policies are going to be brought into the same event. However more importantly this all requires internal buy in from organisations, and we can certainly leverage the battles we have fought with nat cat models to help on this.  
 
The panel then discussed what makes liability exposure management so different to nat cat and what we need to watch out for. Whilst liability isn’t a new risk to the market, there was a consensus that the complex nature of risk influenced by social and technological change, and also the non-repeating nature of events make both estimating future events and validating any approach challenging. As the nature of the risk is changing, we have the issue of identifying the next emerging liability risk to focus on which also requires us to understand why some historical events didn’t materialise. In the liability world, everyone is asking what is the next asbestos but conversely in nat cat, people aren’t asking what’s the next Hurricane Andrew.  The latency of the events themselves also present a challenge as the exposures occur over decades and litigation unfolds over decades, and the potential correlation between multiple lines of business is enormous. If we want to be effective in helping understand this risk, we need to understand that things that may happen in decades to come. Some of the challenges are also cultural, with a lack of synergy between teams in a company cited as huge barrier as well. Likewise, when you have 30 years of nat cat models in practice, people expect the resolution and accuracy of a nat cat model which simply isn’t possible.
 
Given all of the above, what type of exposure management framework can be put in place? As we don’t know for certain what events will unfold, we don’t want to get too prescriptive and fixed to a specific scenario. Instead of prescriptive scenarios, we want these to be thematic which is representative of our current thinking. In doing so we can tease out the types of risks, and interrelationships between classes of business. We want to be able to turn around and say we were talking about the right things, even if we didn’t get the specific scenarios and numbers right. If we get too prescriptive we may end up missing important signals.
 
The panel then moved to discuss the different approaches used by each of the modelling vendors to tackle the problem. Robin discussed how Arium’s solution tackles the proximity question by using the supply chain. Whatever the event is in the future, you can understand the supply chain for a product and the liability will be in that supply chain.  The scenario-based model helps you analyse and quantify exposure to historical liability events and understand what types of future events could cause loss to a portfolio. Robin emphasised the need for models to be flexible and transparent, so you can stress test them and adjust parameters as needed.
 
Adhiraj highlighted that a key objective of any model is to understand the underlying risk and exposure. Given nobody knows what the next asbestos is, understanding your exposure better prepares you for when events happen. Their approach replicates that of property cat, looking at historical events and considering the probability and severity of the event. The approach also considers segmentation of the market to aid identification of exposures which is an iterative process.
 
David shared how the focus for Praedicat has been on identifying the underlying events within the scientific literature. In this respect the science is the risk and Praedicat are studying the effects that are already in commerce or causing damage. Praedicat’s approach reads the science in scale, picks out distinct events to track over time to produce a granular model of these events. A probabilistic model is then built on top of these distinct events. This approach focuses deeply on risk identification, and one of the key challenges faced in doing this is the scale of the science and linking the science to commercial activity and specific insureds.  
 
The panel then moved on to the key question in the title of seminar, what could a 1-in-100 liability cat looks like. There were mixed opinions about use of the term “1 in 100” for liability cats and commentary that this terminology could be more harmful than helpful. Most agreed it aids communication by providing a reference point and a lexicon that everyone else in nat cat can use, however the risk is over confidence in the numbers that we have by assigning specific return periods, and a suggestion of inevitability of these events. In liability there is no inevitability as it’s a human driven process. If we are to assign probabilities to events, they should be seen more as guide than a given and to aid communication over anything else.
 
Considering all of the above, what can exposure management professionals do to help today? One focus should be on monitoring the risks and identifying where the vulnerabilities are rather than waiting until the end to take action. Given there are long timeframes for events to happen, educating clients on understanding the risk and how they can manage it can only be helpful. Similarly, it comes down to how you write the risk. If it’s perceived to be a big risk, it should be the main peril on the policy. The key isn’t to exclude the risk but to be affirmative on it and work collaboratively with underwriters to force their understanding of the risk.
 
To conclude the seminar, the group gave their view on what additional skills and knowledge individuals need to develop in order to tackle the challenges discussed. There was agreement that modelling these types of events requires a lot of different expertise to understand coverages, legislation but also more importantly how things are used in products. As well as experts, we need to bridge the gap between exposure management and underwriting through better communication as we’ve seen with property and nat cat. There was also a suggestion that we need to move to a world where we’re not writing all perils policies with a list of exclusions for hazards you don’t understand but looking towards writing more granular policies with an understanding of this risk itself. Different models give different outputs for different purposes and it’s important not to be wedded to any one approach. Similarly, whilst a model may help, it doesn’t take away the responsibility of understanding the underlying risk and the policy wording. The good news is a lot of the required expertise exists within insurance companies today, it’s really about bringing this together.
 
 
ISCM will continue to support cross-industry initiatives that support knowledge dissemination on this topic. Please get in contact with Alan Godfrey if you would like to contribute further in this space.