Taking a house view: Risk analysis in catastrophe bonds
The ILS market benefits from advanced, independent risk models that have been developed over decades – but that also have inherent limitations. Developing an ‘own’ view of risk can enable managers to make better investment decisions.
Autheurs
An essential component of the offering documents for a catastrophe bond1 (cat bond) is the Risk Analysis section, which contains model output generated from assessments conducted by a modelling agent. As well as a detailed description of the modelling firm, the model used and the natural catastrophe hazards to which the cat bond is exposed, this section includes key metrics such as the bond’s expected loss (EL) and the probabilities of ‘attachment’ and ‘exhaustion’.
These metrics provide a high degree of transparency, possibly higher than that found in the offering documents for other asset classes. As a result, the insurance-linked securities (ILS) market typically does not rely on third-party ratings agencies to assign a credit rating, as is often the case for other asset classes. Furthermore, the scientific advancement of natural catastrophe models over many years has enhanced the reliability of the models used.
However, it is important to acknowledge that models have inherent limitations. A thorough understanding of these limitations can significantly aid a cat bond manager to make better-informed investment decisions.
Assessing cat bond risk
As we have outlined in our previous papers on the ILS market, the ‘coupon’ of a cat bond is made up of interest that is paid from the underlying insurance premium, plus a money market return on the capital paid in by end investors.
The difference between the interest spread and the expected loss for the bond is a measure of the expected return in excess of the money market returns. It thus represents the compensation for accepting the associated risks.
In addition to the EL, other important metrics for the risk profile of the investment opportunity are the ‘attachment’ (the likelihood of an insurance loss event exceeding the threshold to trigger reinsurance coverage) and ‘exhaustion’ (the likelihood that all collateral in a bond will be lost) probabilities. In short, these metrics refer to the probability of losses affecting the returns of the ILS, or wiping out the available capital altogether.
It is important to note that all of these metrics, and thus the implied compensation for the risk being taken, are model-dependent.
Typically, cat bond models are constructed using one of the vendor-specific platforms that provide catastrophe models for various natural perils and across regions. In these cases, the vendor’s catastrophe models are applied to the underlying exposure which, for bonds with an indemnity trigger, is specified by the sponsor (the ceding insurer or reinsurer buying protection via the bond issuance) to produce a stochastic set of scenarios for the underlying losses. The bond structure is configured within this platform and applied to the underlying losses, resulting in a comprehensive loss model for the bond.
It is worth reiterating that the primary drivers for the attachment of a cat bond are natural disasters, rather than human activities. Earth science studies the hazard potential for various perils by region, such as the landfall frequency of hurricanes (the first commercial hurricane model was published in 19872). Structural engineers can determine the damage inflicted by a certain hazard impact, for example damage to a residential building as a function of wind speed. Natural catastrophe models are developed through collaboration among numerous scientists, engineers, and insurance experts.
These models are licensed by insurers, reinsurers and ILS managers. Many stakeholders scrutinise these models, engage in regular dialogue with the vendor, and discuss assumptions and methodologies at conferences. This longstanding dialogue, coupled with competition among vendors, has contributed to the development of credible and useful models for natural catastrophe risks, facilitating the growth of the ILS market.
Furthermore, the availability of these models distinguishes natural catastrophe bonds from many other asset classes, which need to rely on the rating agencies to assess the risk.
Evolution of cat bond models
To reflect the risk associated with cat bonds appropriately, catastrophe models must evolve over time. For example, models need to stay current with scientific findings concerning hazard potential and changes to building standards. They also need to incorporate lessons learned from recent events.
Additionally, climate change is gradually impacting hazard potential, a topic discussed further in our recent climate change briefing paper, and this also needs to be factored into catastrophe models.
Model limitations
All models have inherent limitations, and it is crucial to understand the limitations of the catastrophe bond model used in a bond issuance to evaluate the investment opportunity effectively. Such evaluation may require the cat bond manager to adjust the model output for a proposed cat bond.
One common adjustment relates to the freshness of the underlying exposure data. The input data describe the exposure of the underlying portfolio at a specific point in time, and are subject to change as the portfolio of insurance policies evolves after that specific point to reflect the growth of the sponsor company’s business. Furthermore, this exposure is vulnerable to the effects of inflation, for example on building materials and labour costs.
Another consideration is the fact that specific model updates occur every few years, with some updates being more than a decade apart. During these interim periods the accumulated effects of climate change may not be reflected in the models, which can be significant.
Recent years have witnessed numerous secondary peril3 events, such as winter freezes, floods, severe convective storms, and wildfires, whose modelled 'return periods' for insured losses in their specific perils and regions (the average modelled time for an event of the same type and with the same or greater scale of losses to occur again, everything else being equal) exceed one hundred years.
The most recent of these events, the California wildfires in January 2025, set a new record for the largest insured fire loss in history. Based on our model validation process, this trend cannot be explained solely by the changing environment since the most recent release of the respective catastrophe model.
Generally, we know of fewer limitations of catastrophe models focused on the primary perils, such as hurricanes and earthquakes, in peak zones compared to those for secondary perils.
It is worth noting that many sponsors have their own natural catastrophe experts who are well aware of the limitations of the models. From a cat bond sponsor’s perspective, there is nothing objectionable about designing structures with consideration of model limitations, thereby enhancing the attractiveness of the placement for investors who accept the agent’s model view. It is the responsibility of the manager to be aware of the ways that a transaction has been structured, especially if it may benefit from a weakness in the chosen vendor model.
Model adjustments
Within Schroders Capital’s Insurance-Linked Securities team, we not only validate the vendor models and understand their limitations, but we also develop methods to adjust the modelling agents’ cat bond models where possible, based on in-house expertise. Sometimes, this means replacing certain components of the model with proprietary developments, thereby establishing our own view of risk (see chart) to assess the investment opportunity.
ILS: Own view of risk
Figure 1: The plot compares the return periods, set out in years on a logarithmic scale, for the same perils from two models. The vendor model shown did not pass our internal validation and was hence replaced by a Schroders Capital ILS in-house model, developed with methods of extreme value theory. Observe that a 100-year loss in the vendor model is a 50-year event when assessed with our model.
The use of an own view of risk has some important implications for investors who are already invested, or are thinking about investing, in the ILS market.
For example, when comparing the prospective estimated performance of different managers’ portfolios, an investor should ask on which basis the performance has been calculated: using an unadjusted, vendor model view or an adjusted, manager’s own view of risk? Even if the two portfolios are identical, the stated expected return of a portfolio modelled using an unadjusted view will be higher than the return of a portfolio modelled using an adjusted view.
However – and critically – despite the apparent higher return expectation suggested by the unadjusted model view, the actual return will be the same. The difference is a more conservative view of the risk of losses, which, all else being equal, should translate into a higher confidence in the model and so the modelled outcomes being achieved.
Given that most cat bonds have remote attachment levels, there are few opportunities to demonstrate an outperformance that resulted from the superior risk selection afforded by an own view of risk. The California wildfires afforded us that opportunity. As a result of managing our portfolios using our view of risk, we were heavily underweight in bonds exposed to California wildfire and we are generally underweight in secondary perils overall.
Conclusion
The ILS market benefits from the existence of advanced, independent risk models that have been developed over several decades. They provide a great starting point for managers and investors who are focused on understanding the risk and return possibilities in the ILS market and making the most informed investment decisions. The models are technically complex, and are subject to ongoing development.
Accepting that there is always room for improvement, and developing an own view of risk, improves the quality of our investment decisions, as offerings which may appear attractive using the agent’s model can become unattractive using our view of risk. Notably, by being underweight secondary perils our portfolios were less impacted by the California wildfires that led to large insured losses across the market earlier this year.
1 In this article, cat bond refers to the “classical” cat bond covering only natural perils.
2 Published by AIR Worldwide (now Verisk), see AIR Worldwide - Wikipedia.
3 ILS market participants often refer to primary and secondary perils. Primary perils typically include US earthquake and hurricane, Europe winter storm and Japan earthquake and typhoon. Secondary perils are those perils which aren’t considered to be primary perils, and include the perils listed in this paragraph.
Subscribe to our Insights
Visit our preference center, where you can choose which Schroders Insights you would like to receive.
Autheurs
Topics