David Rule, Executive Director of Insurance Supervision at the Prudential Regulation Authority (PRA), has delivered a speech on ‘Model use and misuse’ at the Association of British Insurers Prudential Seminar. The speech highlights PRA concerns about some uses of models. Models are used widely in insurance to mimic various scenarios so that insurers can estimate capital to provide a buffer against worst case scenario occurrences (in Solvency II, a ‘1 in 200’ scenario).
Solvency II firms can apply for permission to use an internal model to calculate their capital requirements. These models are typically highly complex. The insurance capital model framework encompasses the whole balance sheet and estimates overall capital requirements, taking into account assumed diversification benefits across different risks. Internal models under Solvency II include both quantitative and qualitative elements.
In his speech, David Rule comments that the interests of shareholders and regulators do not necessarily align. Lowering regulatory capital requirements can be seen as an easier means to boost return of equity than increasing underlying profits. Rule expresses a concern that in some instances genuine risk reduction has tipped into a weakening of assumptions or more aggressive risk modelling.
Another area mentioned in the speech is model drift. Model drift occurs when an internal model gradually no longer reflects the risks to which the firm is exposed. The PRA measures model drift by comparing an internal model against the Solvency II standard formula and net best estimate of liabilities, as well as comparing movements in the model against the balance sheet. In 2016/17, the aggregate capital requirements of general insurers using internal models was in line with the standard formula and best estimate of liabilities. But for life insurers the standard formula and best estimate of liabilities rose in contrast to internal model capital. Rule explains that there may be good reasons to explain the drift but the PRA will continue to monitor these movements.
David Rule discusses ‘proxy modelling’ and the use of management actions in internal models. Proxy models are designed as an approximation of complex underlying models. If a proxy model is mis-calculated, the amount needed for regulatory capital will be wrong. The PRA has found that the standards of proxy models in firms varies. The PRA has also reviewed assumptions about future management actions in the calculation of the best estimate and internal model capital. The PRA considers it good practice to include all future management actions in a single document that the board can review.
Insurers may be beginning to use new technology in their pricing models, using artificial intelligence to analyse data. Use of such technology, for example in pricing algorithms, raises ethical questions. For example, are such algorithms transparent for customers? How far should such data collection go in terms of privacy? Is it right to charge more to those whose behaviour indicates they are willing to pay more?
Model risk includes being aware that there may be flaws in the model and that models can be misused. Although the board do not need to understand how the model works in detail they should have the confidence to understand how the model is expected to work and to understand whether the output is credible. Management need to have confidence in the models that are being used, tempered with an understanding of the limitations of models and underlying assumptions. Boards of insurance companies should give sufficient attention to model risk management, especially as models are drawing on new sources of data and analytical techniques.