Insurance has always been built on a trade: a certain payment exchanged for an uncertain outcome. For much of the twentieth century, insurers modelled that uncertainty at an aggregate level, treating a company’s exposure as a single pool of risk. Over the past three decades, the field has shifted toward micro-level modelling: individual claims, policy lapses, and the dependence between losses when disasters strike many policyholders at once. That shift, powered by advances in statistical computing and now by AI, has reshaped how insurers assess, price, and manage risk under pressures ranging from climate change to questions of fairness in algorithmic decision-making.

Edward “Jed” Frees, Emeritus Professor and former Hickman-Larson Chair of Actuarial Science at the University of Wisconsin-Madison, has spent his career at the centre of that shift. In a conversation with UNSW Business School’s Senior Deputy Dean Research & Enterprise, Professor Karin Sanders, Prof. Frees reflected on the rise of micro-level and AI-driven risk modelling, the unresolved questions climate risk poses for private insurance, and the research habits and mentoring philosophy behind a career that has shaped generations of actuarial scholars.

Prof. Sanders: Your research has transformed how actuaries and insurers model dependent risks (particularly through your pioneering work on copulas and risk portfolios). Looking back, what do you see as the most important conceptual breakthrough in understanding risk over the past three decades?

Prof. Frees: Risk and uncertainty lie at the heart of insurance – by definition, this is a contract that trades a certain payment for an uncertain outcome. However, for much of the 20^th century, the modelling of insurable outcomes was at a macro, or aggregate, level, such as a collection of policies held by an insurance company. In the past three decades, analysts have focused more on micro-level outcomes such as specific claims outcomes, policyholders not renewing contracts, decisions about the resources required to manage a loss when a claim occurs, and so forth. Conceptually, we can now focus more on micro-level outcomes. From this perspective, analysts can provide practical tools insurers can use to deliver more efficient services to their customers. However, such models are necessarily very, very complex.

Learn more: When AI meets insurance: Fairness, climate and the future of cover

Prof. Sanders: You have been at the forefront of predictive modelling in actuarial science, well before the current excitement surrounding big data and AI. How has the role of prediction evolved in insurance, and what opportunities and limitations do you see in the growing use of AI-driven risk assessment?

Prof. Frees: Much of my career has been based on porting disciplined statistical thinking into the insurance arena. Statistical thinking has been well supported by the medical and social sciences, and the field has steadily evolved as a result. Certain aspects, such as the predictive nature of statistical inference and non-standard (not bell-shaped or “normal”), have been very useful in the insurance sector.

More recently, researchers in computer science have made significant contributions by developing algorithms and approaches for handling data sets that are massive compared to historical standards. This has allowed us to incorporate textual information (such as underwriters' notes) and graphical information (such as photos of cars in accidents) into our predictive models. This approach to modelling, often referred to as “machine learning,” provides more precise methods of risk assessment.

In addition, we are also seeing these AI-inspired tools to not only help us assess but also manage risk. For example, automobile drivers are beginning to receive near-instantaneous information about their driving behaviour (car velocity, number of hard braking events, and so forth) that not only assesses the riskiness of their behaviour but also suggests ways to become safer drivers.

"A severe earthquake in Tokyo may not be globally diversifiable even through security markets and requires government intervention"

EDWARD (JED) FREES

Prof. Sanders: Actuarial science increasingly intersects with fields such as machine learning, climate risk, and population ageing. Which emerging research questions do you believe deserve much greater attention from scholars and practitioners?

Prof. Frees: To me, the most interesting research questions are the ones at the intersection of important societal needs and the foundations of science. Climate risks are an excellent example. Climate risks are relevant due to their impact on the global economy. Insurance is viewed as a mechanism that can help hard-hit economies recover from disasters and serves to enhance risk management practices, both before and after an extreme event.

The context of climate risks leads us to question the foundation of modelling insurance risks. To illustrate, the potential for increased short-term volatility and severe long-term consequences of climate change has led some analysts to question whether private insurance marketplaces can exist in the future. For example, a severe earthquake in Tokyo may not be globally diversifiable even through security markets and requires government intervention. I enjoy working on these core questions that have potential societal impact.

Prof. Sanders: You are one of the most highly cited scholars in actuarial science and have influenced generations of researchers and practitioners. Looking back on your career, what decisions or habits had the greatest impact on your success?

Prof. Frees: One theme that I seek to maintain is to investigate problems of interest to practitioners. I started my research journey in mathematical statistics and did my PhD in sequential analysis. I stayed with that field for a couple of years, but quickly gave up on it. At that time, people did not analyse data sequentially. In business, one could not make interventions the way that we typically thought about in sequential analysis. Switching research fields so quickly after graduation was a big deal, and this switch was motivated by my interest in applications. I believe that researchers who focus on business contexts, such as actuarial science, should recognise that their work must be of interest to the business community. 

Learn more: Geopolitical risk emerges as the next frontier for actuaries

But the one caveat is that the project needs to be interesting eventually. Businesses operate on shorter timeframes than academia, and we can afford to support long-term contributions. I believe applications should be the driving force in actuarial science research, but one could consider using a predictive model to forecast whether research conducted now will still be relevant in, for example, ten years’ time.

Prof. Sanders: Throughout your career, you have collaborated extensively across disciplines and countries. What advice would you give to early-career researchers who wish to build a meaningful and internationally recognised research program?

Prof. Frees: That’s a tough question; I can tell you some of the things not to do. Do not seek breadth early in your research career by working in several areas. At many universities, the PhD requirement is three papers. This tempts one to try to contribute to different topics without an underlying theme. Especially at an early stage, you cannot become an expert in the field when working on disparate topics. When advising (and hiring), I know someone is an independent researcher when they have developed expertise and a reputation in a certain area, so I advise people starting a research program to stay focused.

The other thing not to do is to focus on extensions of others’ work. It is common in mathematics research to generalise this theorem or extend a result in a minor, incremental manner. It is much more productive for society if you can take a problem, characterise it mathematically, and develop a model that helps us understand the problem at hand. If you can do 90% of the work, that’s plenty. Let other people do the extensions (a good way for you to garner citations, as well).

"It is much more productive for society if you can take a problem, characterise it mathematically, and develop a model that helps us understand the problem at hand"

EDWARD (JED) FREES

The ideal situation is to use industry connections to seek interesting problems to work on. People in the industry, in particular the actuarial community, are very smart. They know how to address problems and are very intelligent.

Prof. Sanders: People know you for your contributions to actuarial science, but what continues to excite and motivate you about research after so many years in academia?

Prof. Frees: To explain my passion for research and education, let me start with a quote from a renowned statistician, David Blackwell, who said: “Basically, I’m not interested in doing research and I never have been.... I’m interested in understanding, which is quite a different thing.” I am also interested in understanding; unlike Blackwell, I find it particularly fun to consider ideas no one else has explored – the research aspect.

I believe that the true test of understanding is the ability to explain a concept to someone. And this is where education enters the picture – I learn so much when I teach. In general, I enjoy teaching at all levels: undergraduate, master's, doctoral, and adult education. To me, doctoral advising is especially enjoyable because I often stay in contact with students throughout their lives. This long-term relationship is rewarding.

Prof. Sanders: Many people associate insurance with financial products, but insurance also plays a critical role in creating resilient societies. How do you think your research has contributed to helping individuals, organisations, and communities better manage uncertainty and risk?

Prof. Frees: In an insurance context, one can partition the concept of resiliency into before and after the occurrence of an insured event. Before an event, the goal is to enhance our ability to anticipate and prepare for the impact of a risk. Much of actuarial work involves establishing a premium structure that reflects the costs associated with risk and encourages sound risk management. For example, in a climate risk setting, one might look to building codes for a property near a hurricane-prone area. Compliance with current codes for a property yields a lower premium than non-compliance; in this way, the premium structure encourages sound risk management. Along with many other colleagues in the field, I seek to establish foundations for appropriate premium structures that not only encourage sound risk management but also facilitate fair treatment of consumers.

Subscribe for the latest research, analysis and insights from UNSW Business School

After an insured event, consumers focus on recovering from the impact of the risk and are concerned about insurers' solvency and ability to pay claims. To understand the difficulties, let us again think about climate risks, extreme weather events such as heatwaves and droughts, heavy precipitation and flooding, and tropical cyclones and hurricanes, and so forth. These extreme events can simultaneously affect many individuals and businesses, thus naturally inducing dependence. Understanding this dependence is a focus of my research.

Prof. Sanders: As insurers gain access to increasingly detailed data, questions about fairness, transparency, and inclusion become more important. How can researchers help ensure that advances in risk analytics benefit society while avoiding unintended inequities?

Prof. Frees: Questions of fair and responsible risk assessment have long been critical to the insurance sector. In my work with UNSW Associate Professor Fei Huang, we trace some of the history and international perspectives around the globe. Like all financial institutions, the insurance industry is built on trust. For example, when a 20-year-old enters the workforce, starts a family, and purchases a life insurance policy, that person may not see proceeds for another 80 years, until death; trusting that the insurer will fulfil its obligations so far in the future is critical to the industry. Unlike other financial industries, the insurance sector is built on risk classification, that is, discrimination. For example, auto insurers often charge younger (presumably riskier) drivers more than older (presumably safer) drivers, but do not distinguish between brown-haired and red-haired drivers (presumably because the two groups are equally risky). So, discrimination based on age is done routinely, whereas discrimination based on hair colour is not.

Learn more: A four-step framework: improving fairness in insurance pricing

These questions have taken on greater prominence with the increasing availability of big data and AI algorithms for risk assessment. A very active research field is concerned with developing procedures that are transparent to the company, regulators, and policyholders (sometimes this field is referred to xAI for “explainable AI”). Transparency is critical to maintaining trust among all market participants, and to achieve this, we need to continue educating them.

Prof. Sanders: If you could change one thing about how society thinks about risk, uncertainty, and the future, what would it be?

Prof. Frees: Changing how society thinks about anything is certainly “beyond my pay grade.” However, I do think we can educate people to think carefully about our changing world; we are not certain how or when these changes will occur, and this is where uncertainty enters the picture. Uncertainty is not necessarily a bad thing (to underscore this point, the North American Society of Actuaries has adopted the slogan “risk is opportunity”). The important attitude to adopt is one of resilience; we need to educate ourselves to anticipate, prepare for, respond to, and recover from the impacts of risks.