Probabilistic modelling is only as good as the experts interpreting it

Sophisticated models and high-tech optimisation tools have become the order of the day for accurate risk modelling. But as precise as these carefully calculated predictions are, they mean little without the human element; that is, the expertise and knowledge that comes from years of experience, which is needed to match the right predictions to the right solutions.

Predictive modelling is considered the industry standard, and no self-respecting (re)insurer, or their broker, would dare to take on any risk without first modelling it.

In our own business we use an array of actuarial modelling tools that allows us to accurately assess the impact of reinsurance structures on a client's gross portfolio to identify the best reinsurance solution.

For catastrophe modelling across the African continent, we have partnered with industry leading experts such as Cat Risk Solutions for earthquake risk; JBA Risk Management for flood risk; CoreLogic for cyclone risk; and QED for hail risk, and use their predictive models in our value proposition.

However, no amount of modelling that we throw at a scenario will ensure the risk is adequately covered, because the truth is, this technology cannot work in isolation. In my experience, that is when costly mistakes can creep in. That is why, to all this technology we add a further tool - one that is invaluable, difficult to replicate and that helps to maximise the full benefit provided by any sophisticated modelling that we do - that of human experience.

It is the human element that adds context around these tools at input stage, and that uses the outputs from these models to provide sensible, usable conclusions. It is the human expertise that ultimately decides what risks can be placed and at what price or level.

Understanding the models

Using our DFA modelling tool, we are able to fully model the expected holistic experience of an insurance and reinsurance entity's performance from one year to the next, including capital impact, claims and profitability. However there is nuance around understanding the prediction of this performance, such as any changes in the underlying portfolios, including underwriting guidelines and approaches, and risk appetite, as well as the market context.

For example, are we in a hardening reinsurance market where certain reinsurance terms are unlikely to be achieved and reinsurance costs are expected to increase? Or perhaps underlying growth targets are expected to be challenged due to increased inflation and/or increased litigation and sentiment?

Other considerations might be understanding what the leadership team is trying to achieve with the business and how we assist it in getting there, which is key to the models' outputs. An example of this would be managing the underwriting performance volatility in the underlying business, or an expected capital injection from shareholders.

The catastrophe models all have multiple years' worth of loss experience for each peril considered (i.e. earthquake, flood, cyclone and hail), and all require interpretation.

The human team analysing the models needs to know what the underlying portfolio actually looks like, as this has an impact in understanding where the risks are situated (compared to the underlying model data set) as well as their vulnerability to each of the perils (places that are below sea level are prone to disastrous flooding, places that are near mining towns are vulnerable to seismic movements due to tremors, properties that have a braced steel frame have a different sensitivity to earthquake etc.)

These scenarios all need human interpretation and context in order to allow the models themselves to perform at their best.

Predictive models use a variety of inherent factors to try to predict an outcome i.e. a cedant's underlying data, to understand where the biggest losses are likely to occur in the portfolio in future. Data still needs to be properly contextualised to interpret events that accurately influence the outcome of the said portfolio. For instance, it's possible that the data may indicate losses driven by policyholders in a specific region. However, this could have been due to circumstantial events that would unlikely happen again.

A case study

As part of our service offering to clients, we assess the capital requirements and impact of the various reinsurance structures we propose. This, coupled with key metrics vital to the client, allows us to recommend and execute the most efficient reinsurance structure - one that allows the client to meet their strategic objectives. We use a bespoke DFA modelling tool to assist with modelling and testing various scenarios to arrive at an optimal structure.

One of our clients had a portfolio exposed to a high frequency of low severity claims in one of the lines of business that they write. This line of business contributed a significant amount of the income to the business. This was coupled with a line of business with relatively large single risks with low frequency that made up a small proportion of their overall portfolio. These lines of business were to be placed through the same reinsurance treaty. Due to the nature of the portfolio mix and expected modelled claims experience, the ideal structure with the highest capital efficiency as recommended by the modelling necessitate a reduced deductible on their Excess of Loss reinsurance treaty.

Given our extensive knowledge of the reinsurance market and underwriting expertise, it became clear that the recommended technical structure would not be practically achievable and required alteration.

The appreciation of quantitative optimisation from the modelling coupled with our qualitative prowess allowed us to deliver a solution that was practically effective with reduced execution risk.

This perfectly illustrates the integral relationship between tools and human experience, and the balance in value of both.

It's clear that the key to maximising cutting-edge risk modelling is understanding what the numbers and outputs mean, where they come from and, in a real business context, what impact they might have on any proposed solution and the implementation thereof.

The (re)insurers and brokers that get this balance right are best placed to offer the most appropriate risk assessment and related cover.