The insurance industry relies heavily on catastrophe modeling to set capital adequacy, adhere and respond to evolving regulatory requirements and stress test portfolios. The same is now increasingly true of the cyber catastrophe sphere, in which key areas of focus include how models can help with capital allocation, stress testing and informing development of underwriting guidelines and insurance products. Parallels can be drawn from the cyber catastrophe and natural catastrophe risk management sectors when modeling these risks.
The introduction of models provided critical insight into the potential for catastrophic claims for all risk policies or policies without clear exclusionary language. Historical events such as the April 1906 San Francisco earthquake (leading to unanticipated claims for fire policies), 2005 Hurricane Katrina flooding (resulting in unanticipated claims for homeowners wind policies) or the 9/11 U.S. terrorist attacks (experiencing unanticipated war exclusion interpretation and definition of a single event), and the current unfolding of the coronavirus pandemic crisis highlight the criticality of understanding the triggers and correlation of potential loss due to a single event.
In many cases, insurers paid losses to avert “reputational risk” and have since used models to provide insight into realistic structuring of policy, reinsurance and other risk transfer vehicles. Clear exclusionary language, endorsements and coverage-specific terms evolved over the decades in concert with evolving scientific knowledge of the risks and modeled loss potential.
Today, we are seeing the same evolution with respect to insuring cyber risk, but over a highly compressed period, without the decades of experience of systemic insured loss events. Many cyber catastrophe risk managers attempt to apply the same lens of current natural catastrophe model availability of data resolution, data quality, catastrophic event knowledge and model validation expectations. But by embracing the commonality of lessons learned from the evolution of the property catastrophe insurance market, we can prepare for an event considered to be a case of not “if” but “when.”
The role of data in models
A first common theme is to recognize that the understanding and availability of information for a rapidly evolving risk means that there is value in aggregate data in the absence of detailed data. This has been and is still the case for property catastrophes and is also the case for cyber catastrophe risk models. Confidentiality obligations in portfolio data as well as the lack of high-quality data is an issue for all models. However, new sources of data as well as sophisticated data science and artificial intelligence analytics are being incorporated into models that provide an increased confidence in assessing the potential risk to an individual company or entity.
See also: Coronavirus Boosts Cyber Risk
A second related common theme is the ability of catastrophe risk models to augment lack of risk-specific data capture at the time of underwriting. This is where all catastrophe risk models add significant value, where context for what should be captured as well as what can be captured is provided. In the case of cyber, this can include access to both inside-out (behind the firewall) and outside-in (outside the firewall) data. Inside-out data refers to aggregate data for segments of the economy, measuring the anonymized trends of security behaviors (such as frequency of software patching). Outside-in data is made up of specific signals that can be identified from outside an organization and that give indications of overall cybersecurity maturity (such as the use of unsupported end-of-life products).
A third commonality is the value in extrapolating the impact of past events into the future given evolving available data on the changing causes of frequency and severity of cyber events. The property catastrophe arena is grappling with very similar issues relative to the rapid and uncertain evolution of climate models. For cyber risks, history is not a predictor of the future in terms of modeling threat actors, the methods they deploy and the vulnerabilities they exploit. However, it is possible to examine historic data and the types of cyber incidents that have occurred while addressing the challenges in the way that information is collected, curated and used. This historic data is used against the backdrop of a near-term threat actor and technological trends to understand future potential systemic losses due to large-scale attacks on bigger and more interconnected entities.
The role of probabilistic models
At the enterprise level, the market is struggling with how to assess potential aggregations within and across business lines. Event clash due to a single event causing multiple loss triggers to policies and reinsurance treaties is a key concern across all lines of business. Use of common cyber and other catastrophe risk loss metrics that can be combined across perils and lines of business are being explored. In addition, regulatory groups are considering requirements similar to property catastrophe risk to address solvency requirements relative to cyber risk.
In this environment, consistent and structured definitions of risk measures are critical for assessing and communicating potential systemic catastrophic loss. Both deterministic cyber scenario event analyses as well as probabilistic stochastic cyber event analyses are required. Given this context, cyber catastrophe risk models that can withstand validation scrutiny similar to property catastrophe risk models require the same level of rigorous attention to transparency in communication of model methodology.
Similarities… but some differences
There are some key differences between the systemic risks of natural disasters and cyber events. One material contrast is that cyber perils manifest with active adversaries seeking to cause malicious damage to individuals and companies globally. The factors affecting modeling include the changing nature of geopolitical threats, the dramatic increase in the use of digital means for criminal enterprises, the hyperconnectivity of developed economies and an ever-increasing reliance on networked technologies. Cyber event scenarios are developed to represent a range of potential systemic events in which technological dependencies affect individual insured companies, due to a common vulnerability or a “single point of failure.” Examples include common cloud service providers, payment systems, mobile phone networks, operating systems and other connected technologies.
See also: Risks, Opportunities in the Next Wave
There are limitations in any model relating to cyber risk, given the inherent uncertainties. Nevertheless, these models provide valuable insights to better decision-making relating to capital planning, reinsurance and addressing regulatory issues. By learning from previous insurance shocks, we can support a more stable and resilient cyber risk insurance market.