Three years ago, Maximum Information published an article, Why Decision Theory?, in which our founder, Tom Philp, described the origins of our collaboration with two Professors of Philosophy, Roman Frigg and Richard Bradley, at the London School of Economics and Political Science (LSE). 

The origins of the tool go back to 2015, when Tom was working in catastrophe risk at XL Catlin (later AXA XL). Building hurricane risk views involved combining catastrophe models, historical datasets and seasonal forecasts that were all scientifically credible, but often produced conflicting answers, and there was no objective way to decide how much weight it should carry when making insurance decisions. He recognised that the challenge was not simply a modelling problem, but a question of how to make decisions under uncertainty. 

Looking for a way to tackle this challenge, Tom approached Profs. Frigg and Bradley, whose research focused on decision theory and reasoning under uncertainty. What began as a series of discussions between industry and academia evolved into a long-term research collaboration, exploring how decision theory could be applied to catastrophe risk management. 

Those early conversations led to the publication of the paper Making Confident Decisions with Model Ensembles in 2021, which applied the principles behind the decision theory called the confidence approach to hurricane risk. With the founding of Maximum Information and support from the Lighthill Risk Network, the research was developed into a practical decision-support framework for insurance. 

You can access the live tool here. 

What is the decision theoretic tool for the North Atlantic Hurricane Season? 

It is a free, interactive tool which applies a decision-theoretic framework to seasonal North Atlantic hurricane forecasts, to support more transparent and defensible insurance decisions. 

Rather than asking, "Which model is right?", the confidence approach asks, "Given the decision we're making, how much confidence do we need, and what evidence gives us that confidence?" 

The tool lets users explore how to account for the uncertainty in different seasonal hurricane forecasts, combining it with their attitude to risk and volatility to make insurance decisions. 

The tool has the following sections: 

Input  

Data selection 

Decision settings  

Different decisions require different levels of confidence. A high-value underwriting decision may require evidence from multiple independent sources, while a lower-risk decision may require less supporting evidence. 

This section contains two settings: 

The exhibits show hurricane rates from various historical climatologies and seasonal forecast rates from various forecasting centers for the selected season. The historical climatologies are different ways to represent historical hurricane rates, reflecting different climate phases or scientific debates. 

This information is for the user to explore how the rates are “coarse-grained” into different nested intervals, depending on the decision settings. For the selected level of confidence, the upper bound of the ensemble is used as the confidence adjusted rate. More information on how the confidence adjusted rates are estimated is available here.

Pricing settings 

Here, users can adjust pricing inputs to reflect different stages of the insurance cycle. 

The landfall rate implied by the market premium / payout is reported together with the actual landfall rate for past seasons. 

Overlaying stakes and cautiousness, that is identifying which of the three cautiousness ranges (low, medium or high) the stakes fall into, determines the level of confidence required (low, medium or high). 

These settings are then passed into the analysis. The framework evaluates the available evidence and helps answer two questions: 

Analysis 

Once you have set your conditions, two tables will appear with the results, specifically focused on helping you decide whether you should you write the risk; And for past seasons, would you have made or lost money had you decided to underwrite it?  

The purpose of this is to help you to explore how different attitudes to uncertainty can influence your decisions.  

Within the tables, you can consider historical hurricane landfall information (View From IBTrACS Climatology) and seasonal forecasts for the selected year (View From Seasonal Forecasts From Seasonal Hurricane Predictions). Both tables show results derived from hurricane rate statistics (mean or median) and from hurricane rates loaded with the confidence approach. Meaning that you can compare underwriting outcomes using historical information versus seasonal forecasts, and the added value provided by the confidence approach. 

The IBTrACS Climatology table  shows both the Long Term Rate, the median of the various climatologies used, and applies a loading to the climatology median, based on the selected level of confidence, to obtain the historical confidence adjusted rate. Further information on the climatologies used is available in the Reference tab. 

The Seasonal Forecasts From Seasonal Hurricane Predictions table instead shows the median of the seasonal forecasts from different centers, and applies a loading to the median , based on the selected level of confidence, to obtain the seasonal confidence adjusted rate. 

More information on the analysis computed can be found in Glossary in the Reference tab.

What has been updated? 

The latest version of the tool includes several significant enhancements: 

How to use the framework 

The user is encouraged to explore the tool and see how various risk appetites (e.g. changing the stakes and cautiousness settings) affect the resulting decision. For example, would you have underwritten a policy during a highly activity year such as 2020 using the historical long-term rate? Would a more cautious risk appetite have changed that decision? Could the seasonal forecasts available at the time have helped avoid a loss? 

The aim is not to prescribe the “right” answer, but to make the judgement behind a decision more explicit, traceable and open to challenge or revision. We hope the tool encourages conversations among risk takers about how risk appetite is defined, aligned across business units and translated into action. 

And, should you find the approach useful, it can provide a starting point for considering how the framework might be adapted and applied within your own organisation and decision-making context.