Basic Research Provides Data for Tomorrow's Insurance Risks

"Without basic research, our level of knowledge would be entirely different - and new insights are also central to understanding future risks," emphasizes Iwan Stalder, Head of Group Accumulation Management at Zurich Insurance Group. Stalder and his team identify, quantify, and aggregate risks. This enables them to demonstrate how exposed their company is across the entire portfolio (excluding life insurances). All risks from natural hazards to man-made events are included.

The team originally focused on natural disasters. However, the explosion in the Chinese port city of Tianjin in 2015 showed that industrial accidents can also lead to a catastrophe. Since then, scenarios such as pandemics, industrial accidents, or questions about where "the next asbestos scenario" might loom have become part of its portfolio.

Due to working with complex risks, insurance companies are among the actors who particularly rely on well-founded scientific insights. When assessing climate change, natural disasters, pandemics, or cyber incidents, they require models from mathematics, physics, and climate sciences. These models form the foundation for risk quantification, capital allocation, and insurance products that provide protection and stability for society and the economy.

Data, Models, and Uncertainties

The heart of any risk assessment is data. Among the most valuable sources are the loss histories with reported damages by policyholders and paid claims. In the USA, for example, there are decades of experience with hurricanes, tornadoes, and hailstorms. Data about these events provides a detailed picture of how frequently such storms occur and the potential for damage. Using this loss data, vulnerabilities can be identified, and risk models optimized.

When using different external models, the results sometimes differ significantly. If an insurer instead feeds the models with its own data on storm damages, it quickly becomes clear which closest reflects reality. No model is perfect, as Stalder emphasizes, but by comparing with real experiences, the most accurate one can be determined. In a further step, all models need regular updates. If data on loss events is available, Stalder's team uses it to recalibrate assumptions; otherwise, it relies on other models, scientific insights, and expert opinions. Catastrophe models are thus not static but continuously evolve as new scientific findings emerge, building codes change, new risks arise, and the climate itself changes. Even the hurricane models for the USA, spanning an entire generation, need constant updating. "Everything is in motion; we are always learning," summarizes Stalder.

Certain uncertainties always remain, yet they are greater when there are few data points. Natural hazards are relatively well-researched, but accurately predicting the impacts of climate change on the frequency and intensity of specific events is challenging. Even greater is the uncertainty with cyber incidents: there is a lack of extreme events that could serve as reference points. The "CrowdStrike Event" in July 2024, where computer systems worldwide failed, was due to a faulty software update. Although it was not a cyber-attack, the event highlighted how millions of systems can be incapacitated at once.

Role of Basic Research

For the insurance industry, basic research is central. For instance, climate science provides scenarios for capturing current data in catastrophe models: rising sea levels, more intense rainfall, or new storm trajectories. Toxicological studies on substances like PFAS highlight new liability risks. These "forever chemicals" were recently responsible for a sales ban on agricultural products in Switzerland. "Such research work, often supported by institutions like the SNF, forms the scientific foundation for our practical risk models," explains Stalder.

The connection between science and practice often runs via data, methods, and scenarios from basic research. An example is the SNF-supported project scClim: the University of Bern, ETH Zurich, and Agroscope are jointly developing high-resolution simulations of supercells in the Alpine region. The goal is better forecasts of locally confined, intense thunderstorms.

Technological research also plays a key role. Artificial intelligence, particularly neural networks, are increasingly used today to detect patterns in vast datasets. The roots of these methods trace back to foundational research in the 1950s and 1960s. Initially, such applications were neither extensive nor efficient. After immense progress in computing power and algorithms in the past decade, these systems now work much faster and more precisely. "Certain insights into current risks would be unthinkable without neural networks. They are the direct result of decades of basic research," emphasizes Stalder. In a project with SNF contributions, the University of Basel and IBM are developing optimized neural networks using quantum computers.

From Model Validation to In-house Development

The connection between science and practice is also evident in the application of insurance models. For many years, Stalder's team has been licensing, validating, and calibrating models—a practice Zurich pioneered in 2004, which has since become the standard. Over time, the team progressed further by developing its own scenarios or probability models for terrorism, liability cases of catastrophic magnitude, crop failures, pandemics, and cyber-attacks. These in-house models ensure transparent assumptions and calculations, which is particularly important from a regulatory perspective.

The exchange with the scientific community is a central part of this work. Zurich has created the Advisory Council for Catastrophes, where leading researchers discuss the latest findings, from climate change to seasonal cyclone forecasts and earthquake early warning systems to predictive models. Insurance experts also participate in the sessions. They specialize in risk management, underwriting, risk engineering, and claims, ensuring that scientific insights are directly incorporated into business practices. "We want to understand the state of research and use these insights to result in better risk assessments and products," explains Stalder.

A Contribution to Resilience

Ultimately, it is not only about the stability of a single company but the functioning of entire economies. Insurances enable investments even in uncertain environments, for example, the construction of a multi-billion-dollar factory or the financing of renewable energy projects. They create predictability by making risks quantifiable and distributable.

Thus, basic research is much more than a purely academic endeavor: it lays the foundation for models of complex risks and for innovations that contribute to the resilience of the economy and society. Or, in Stalder's words: "Thanks to basic research, we may not eliminate uncertainties, but we can make informed decisions in an uncertain environment."

The text of this news, a downloadable image, and further information are available on the Swiss National Science Foundation's website.

Press contact:
Swiss National Science Foundation
Communications Division
Tel.: +41 31 308 23 87
Email: com@snf.ch

Editor's note: Image rights belong to the respective publisher.