Shoib Khan the Bank of England’s director of Insurance Supervision has told (re)insurers the industry must take a more pragmatic approach to the benefits that models deliver in a rapidly changing world and that the UK regulators are keen to ensure that they remain fit for purpose.
“Every insurer should have a robust and embedded approach to risk management,” he explains. “This allows boards and executive teams to be confident in their ability to manage the business safely in a variety of operating conditions. As we have seen over the past year, operating conditions can change rapidly, driven by combinations of internal and external variables. “
He adds that models play a critical role in the way insurers manage their risk.
“They take a vast array of inputs, indicators and expert judgement to produce quantitative narratives which articulate how the future might play out.”
Khan adds the PRA encourages firms to develop their own internal models, particularly where firms have bespoke risks which are not captured by the standard formula.
“That’s why we plan to implement a significant streamlining of the rules for internal model approvals, which currently require nearly 200 tests; and instead shift the focus to key principal-based requirements. Also, a streamlined process for considering model changes will help firms keep these models up-to-date and more dynamic.”
However he says his message on models applies much more broadly.
“Common examples of models used by insurers include third-party natural catastrophe models used in pricing and reserving; economic capital models used to make business decisions; and models used to carry out stress testing. What these models have in common is their ability to drive business decisions made by boards and executives.
“The capability, limitations and intended use of models must, however, be well-understood by those relying on them. In our January 2023 letter on supervisory priorities, we reminded insurers to reassure themselves of the continued validity of their models, including the extent to which model risk management principles for banks applies and whether current validation remains robust in the face of multiple concurrent stresses.
“To put it another way, we would never expect our insurers to rely solely on autopilot.”
He explains: “Firstly, users of a model should consider whether the factors inside and outside of the business have moved on since the model was first designed. As operating conditions move closer to the boundary of what models were originally designed for, their outputs become less reliable and less informative. I would assert that even the most experienced industry practitioners amongst us have not experienced the combination of stresses that firms face today.
“Whilst I’m not saying that firms should rewrite their models, observed step changes in factors such as longevity, inflation and the impacts of climate change on natural catastrophe events are cause to reassess key assumptions and give greater weight to other non-modelled factors.
“Secondly, users of the model should understand that events in the future may not be reflected in historical experience. We don’t have to look back too far to see an example of this.”
He highlights last year when the UK’s financial sector was shaken by the rapid repricing of long-term gilts, requiring the Bank of England to take action to reduce risks of contagion to credit conditions for UK households and businesses.
“Models might be misleading in the midst of a crisis – to use the same example, the fact that you have just witnessed a 1-in-100 year rise in gilt yields does not necessarily mean that you would not see a similar rise the next week,” he explains. “Unprecedented events such as these remind us that the need for action – in this case, the provision of liquidity – can arise suddenly, unexpectedly, and assets may not perform as they traditionally have.
“And thirdly, users of a model must understand the intended use of the model and its outputs. Model use should focus beyond one biting scenario to understand how different combinations of stressed inputs can lead to modelled outputs that are just as severe.
“Whether it is due to a change in weather conditions, a faulty indicator or worse, insurers need to know the limits of their models. That includes understanding how reliable models are in different circumstances; how model output can be validated; and, how to use the information that models tell you about the distribution of risk around a central scenario. Knowing these limits allows insurers to balance the use of autopilot with manual control, and we expect firms and their boards to actively do the same.”