Uncovering assumptions and evaluating information quality in reinsurance modelling of weather and climate events.
State of the field
(Re)insurance is increasingly taking up the challenge that climate change poses to the financial stability of the sector and the role it can play in mitigating and adapting to the impacts of a changing climate (Collier et al. 2021). Ensuring financial stability (capital modelling) in the (re)insurance sector involves a thorough analysis of risk, i.e. the intersection of hazard, exposure and vulnerability. To do so, however, companies must rely on scientific assessment of current and future impacts of climate change, as well as how these impacts propagate across financial systems and material assets that fall within the remit of a (re)insurance company.[1] Each element of risk assessment requires an expert team of modelers, such as catastrophe modellers, risk managers, actuaries, etc. who develop and/or use models each with its own assumptions and limitations.
Actuaries play a key role within the insurance ecosystem of experts: they evaluate risk and associated uncertainties for pricing insurance premiums – which is primarily dependent on the average annual loss - and ensuring solvency - which is primarily dependent on the characteristics of the tails of a distribution, i.e. of extreme events which cause peak in losses - of the insurance company. The World Bank Group[2] describes actuaries as expert modelers who use mathematical models to provide assessments of the impact of current future events on the financial stability and success of the company, and to do so they need to have expertise in statistical modelling, business and communication – as the output of their modelling affects many stakeholders in the sector. So, actuaries engage in extensive mathematical modelling and expert knowledge, which in turn relies on modelling done by other experts, such as catastrophe modellers and scientists evaluating the impacts of climate change on assets and/or systems of interest to an insurance company.
Given the actuaries’ central role in (re)insurance companies, the Financial Reporting Council has developed a Technical Actuarial Standard that specifically targets insurance.[3] In particular, the standard has the purpose of supporting the so-called “reliability objective”:
“To allow the intended user to place a high degree of reliance on actuarial information, practitioners must ensure the actuarial information, including the communication of any inherent uncertainty, is relevant, based on transparent assumptions, complete and comprehensible.”[4]
This standard emphasizes the evaluation and communication of risks, and, in particular, the assumptions and uncertainties involved in modelling risk – which, for the case of climate can pose serious difficulties, as has indeed been recognized by the Institute and Faculty of Actuaries itself (Trust et al. 2023)[5]. To tackle these difficulties, their recent publication on climate scenarios (Ibid.) stresses that climate scenario modelling for actuarial purposes should clearly evaluate and communicate modelling limitations. To do so, modelers should be explicit about assumptions involved in developing and using models as well as about assumptions involved in advising and developing regulations on the basis of model output. Making such assumptions explicit can help reduce the discrepancy between different types of expertise involved in evaluating a complex system like the climate. This, in turn, can help achieve the reliability objective.
Epistemology and values
It has already been argued extensively that it is important to individuate assumptions as they can help mediate between experts, especially when there is large uncertainty (Stainforth et al. 2007) and bridge between model land and the real world (Thompson, 2022). These considerations become particularly important when dealing with risk coming from complex systems, where evaluating and communicating risk is a non-trivial endeavour that necessitates continuous engagement with all the stakeholders involved in developing and exchanging information on the risks under consideration (Morgan et al. 1992, 2002).
It is also widely recognized by philosophers of science (and increasingly in other disciplines), that different epistemic and non-epistemic values influence all aspects of modelling, including the above mentioned assumptions (Douglas, 2009, Pulkkinen 2022). For example, different values are reflected in different approaches to modelling the climate (Baldissera Pacchetti, Jebeile and Thompson 2024), or in estimating the Equilibrium Climate Sensitivity (Undorf et al. 2022).
Moreover, inter- and trans-disciplinary groups of scientists are starting to evaluate how different value judgements affect their modelling of climate risks (Bessette et al. 2016, Mayer et al. 2017). This work has been motivated by Morgan et al.’s (2002) work on developing mental models to study how experts conceptualize risk, and expanded to include how values define different aspects of this conceptualization as well as its communication.
How model assumptions and values enter the modelling development, evaluation and communication processes is not only important when data and information travels across different types of expertise and models, but also when different disciplines are drawn upon to develop one model, such as integrated assessment models (IAMS). These have attracted criticism lately in so far as they are not adequate for purpose, as they fail to capture mechanisms that may lead to catastrophic outcomes (Stern et al. 2022, see also Condon 2022). As such, they fail to produce information to catalyse the action that is needed to build (financial, social) resilience. Catastrophic climate change scenarios are important for prudent risk management (Kemp et al. 2022) – a type of risk management that requires the explicit consideration of social and ethical values by definition.
The epistemic problem of prediction
Climate change science requires explicit consideration of epistemic values particularly in when generating and evaluating forward looking information (e.g. climate scenarios for resilience building and/or reporting). Climate “predictions” are impossible to verify as they are out of sample, i.e. they are predictions of a state of the system that has never been observed before (and even after the fact, are conditional on specific forcing scenarios). Moreover, while climate models are based on scientists’ understanding of physical theory, there are many aspects of the climate system that are either not included in the model or not well understood (such as clouds), leading to gaps in assessments of uncertainty.
While these issues are not necessarily a problem for exploratory climate change research, they become so when climate change research is used as the basis for information for climate change adaptation decisions: the consequences of relying of information that may not be accurate may in fact have much more wide-ranging implications (e.g. maladaptation, erosion of trust, etc.). To address this problem, researchers are discussing what it means to define quality of climate change information for decision making,[6] and developed and tested a framework for assessing the quality of such information (Baldissera Pacchetti et al. 2021a, 2021b). The quality criteria identified in the framework proposed are the epistemic values that are considered necessary for developing information that can be relied upon and shares some similarities with the standard for actuaries described above. However, this framework does not include any guidance or discussion on the role of non-epistemic values can play in this context, and the present study plans to complement this work by taking a broader outlook on how values shape model building and model use.
Research aims
This research will provide a comprehensive picture of how model-based information is developed, used, communicated and regulated in the (re)insurance sector. The starting assumption here is that identifying values and assumptions made by experts when developing and using models and communicating their result, we can improve the transparency of model use and support better communication across different domains in the (re)insurance sector, as well as the values and assumptions of regulators.
To achieve this aim, we will work with modelers in (re)insurance companies to build a mental model (see Morgan et al. 1992, 2002 for an application of the use of mental models to analyse information about complex risk) of how they develop, use, and transfer information across different domains of expertise. These mental models will then be used as a basis for capturing, through interviews, what assumptions and values occur at key steps of the information chain.
The data collected will be analysed to study how models are used in the sector (both within a company and across companies) and the extent the information production and exchange meet the standard for actuaries in insurance. We will further evaluate the purposes for which the models are indeed fit for purpose.
Footnotes
[1] This is because not only the direct (compound and cascading) impacts matter, but indirect impacts of CC are important for risk transfer considerations.
[2] https://documents.worldbank.org/en/publication/documents-reports/documentdetail/158161468157173893/the-role-of-the-actuary-in-insurance#:~:text=The%20role%20of%20the%20actuary%20in%20insurance%20(English)&text=They%20evaluate%20the%20financial%20implications,the%20likelihood%20of%20future%20events.
References
- Bessette, D.L., Mayer, L.A., Cwik, B., Vezér, M., Keller, K., Lempert, R.J. and Tuana, N. (2017), Building a Values-Informed Mental Model for New Orleans Climate Risk Management. Risk Analysis, 37: 1993-2004. https://doi.org/10.1111/risa.1274
- Collier, S. J., Elliott, R., & Lehtonen, T. K. (2021). Climate change and insurance. Economy and Society, 50(2), 158–172. https://doi.org/10.1080/03085147.2021.1903771
- Condon, M. (2021), Market Myopia's Climate Bubble, 2022 Utah Law Review 63. Available at: https://scholarship.law.bu.edu/faculty_scholarship/1087
- Douglas, H. (2009). Science, policy, and the value-free ideal. University of Pittsburgh Press.
- Kemp, Luke, Chi Xu, Joanna Depledge, Kristie L. Ebi, Goodwin Gibbins, Timothy A. Kohler, Johan Rockström et al. "Climate Endgame: Exploring catastrophic climate change scenarios." Proceedings of the National Academy of Sciences119, no. 34 (2022): e2108146119.
- Mayer, Lauren A., Kathleen Loa, Bryan Cwik, Nancy Tuana, Klaus Keller, Chad Gonnerman, Andrew M. Parker, and Robert J. Lempert. "Understanding scientists’ computational modeling decisions about climate risk management strategies using values-informed mental models." Global Environmental Change 42 (2017): 107-116.
- Morgan, G. "Communicating risk to the public-First learn what people know and believe." Environmental Science and Technology 26 (1992): 2048-2056.
- Morgan, M. G., Fischhoff, B., Bostrom, A., & Atman, C. J. (2001). Risk Communication: A Mental Models Approach. Cambridge University Press.
- Pulkkinen, K., Undorf, S., Bender, F. et al. The value of values in climate science. Nat. Clim. Chang. 12, 4–6 (2022). https://doi.org/10.1038/s41558-021-01238-9
- Stainforth, D. A., Allen, M. R., Tredger, E. R., & Smith, L. A. (2007). Confidence, uncertainty and decision-support relevance in climate predictions. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 365(1857), 2145-2161.
- Stern, N., Stiglitz, J., & Taylor, C. (2022). The economics of immense risk, urgent action and radical change: towards new approaches to the economics of climate change. Journal of Economic Methodology, 29(3), 181-216.
- Trust, S., Joshi, S., Lenton, T., Oliver, J. (2023). The Emperor’s New Climate Scenarios: Limitations and assumptions of commonly used climate-change scenarios in financial services. Available at: https://actuaries.org.uk/media/qeydewmk/the-emperor-s-new-climate-scenarios.pdf
- Undorf, S., Pulkkinen, K., Wikman-Svahn, P., & Bender, F. A. M. (2022). How do value-judgements enter model-based assessments of climate sensitivity?. Climatic Change, 174(3), 19.