Policyholder Behaviour in Extreme Conditions

Evidence from the past

We researched public domain material on causes of Insurance Company failures around the world. We found no examples in which policyholder behaviour was instrumental in bringing about the demise or near-demise of any insurer. However, policyholder behaviour has sometimes featured as a key factor in accelerating such demise once the process is triggered.

This suggests that events of interest are likely to arise as a result of unique combinations of circumstances. This also seems reasonable, since companies (and regulators) generally learn from past events.

In view of this, we expect that the main benefit of studying policyholder behaviour in extreme situations will be to help companies recognise such potential situations more quickly. This should help them take appropriate action to mitigate risks or to exploit new opportunities.

Scenarios

We are examining a range of scenarios that might provoke significant levels of unexpected policyholder behaviour.

It is clearly impossible to consider every extreme situation that might arise. Indeed, situations tend to become extreme precisely because they are unexpected! Nevertheless, we have tried to make our list of scenarios as varied as possible, and are considering:

• Non-financial catastrophes
• Major dislocations of financial markets
• Medical advances
• State intervention in markets
• Social/anthropological change

We are considering policyholder behaviour under these scenarios using insights from behavioural economics. This should help to identify to what extent our intuitive understanding of behaviour in these scenarios might be misleading.

Modelling

We are investigating ways to model policyholder behaviour in extreme conditions since this will help to:

• Deepen understanding of the dynamics, interactions and key drivers of the situations being studied
• Share understanding of behaviour in as-yet-unobserved situations with a wider audience
• Explore the range of possible outcomes, including unexpected ones
• Identify key drivers of extreme behaviour and any “tipping points"
• Help to recognise warning signs earlier as a scenario unfolds
• Identify and test possible interventions to mitigate or take advantage of an emerging situation.

It will be difficult – if not impossible – to estimate suitable statistical parameters for a stochastic model. It is unclear how any such model’s structure might change under extreme conditions. More fundamentally, we are likely to be interested in variables that stochastic approaches do not usually model explicitly.

We are therefore exploring Agent Based Modelling and Systems Dynamics Modelling – two alternative modelling techniques that might be more helpful.

Outputs

• Life Conference presentation 2013
• Working party notes 2019
• August 2019 article in The Actuary