04 March 2009

Published in: Capital - models, Regulation - supervision

Prepare your models for the six tests now

Graham Fulcher reviews the Solvency II test criteria for internal models and offers practical advice on dealing with them in advance of further guidance from CEIOPS in June.

Many businesses will seek to take advantage of the opportunity under Solvency II to adopt an internal model that better reflects the nature of their business than the standard formula, but this is not an option to be undertaken lightly. In addition to the challenges of developing and maintaining such a model, the draft regulation contains requirements whereby the efficacy and robustness of an internal model is challenged and must be shown to be adequate.

Articles 118-123 of the draft Solvency II directive set out the six key tests and standards that the internal capital models of (re)insurers will have to pass if they are to be permitted to set the capital for Pillar I of the regime (the solvency capital requirement -- SCR) using their internal model rather than the standard formula.

Further Level 2 guidance is expected from CEIOPS in June this year and should provide further detail on how these tests will be interpreted.

However for UK insurers, the FSA's discussion paper DP08/4: Insurance Risk Management: The Path to Solvency II from September 2008 sets out the FSA's initial observations on the tests. Companies have between March 2009 and June 2009 to indicate to the FSA if they are likely to apply for internal model approval, and such an assessment will need to be based on the current draft irective and on the FSA's current interpretation of that guidance (in advance of the CEIOPS paper).

The FSA has indicated a number of times that in its view, the UK's current level of compliance with all six tests and standards is low. Given that the UK has had the benefit of several years of using internal models to set regulatory capital under the ICAS regime, the obvious conclusion has to be that the challenges for the non-UK industry are even greater. Non-UK insurers therefore need to be preparing now for the challenges of the six tests and standards.

This article sets out some selective observations on the tests and the challenges they are likely to pose. While it has been written with the viewpoint of a London market insurer, the descriptions of each test apply to all insurance undertakings and many of the observations apply much wider than the London market.

As this article is a basic introduction to the topic, it does not consider either any potential mitigation from the application of the proportionality principle or the possibility of the use of partial internal models.

There are six key tests or standards:

• Use test
• Statistical quality standards
• Calibration standards
• Profit and loss attribution
• Validation standards
• Documentation standards

I'll describe and comment briefly on each of these in turn.

Use test

Description

The key requirement of the use test is simply that the internal model is demonstrably and widely used and plays an important role in risk management, decision-making and setting and allocating economic capital. Two additional requirements are that the SCR is recalculated at the same frequency as the model is run for the other purposes, and that the model is regularly updated to properly reflect changes in the insurer.

Observation

The FSA's discussion paper sets out a long list of potential examples of how an internal model might be used including reinsurance purchase, capital allocation for pricing and assessing different strategies. It is a sign of when the paper was written that there is at best an implicit mention of using the model to assess remuneration. Given recent events one can only assume that this will move towards the top of the list in future. As the FSA commented in a draft code of practice on remuneration in late February 2009 "significant doubt has recently arisen over the efficacy of other risk management mechanisms to override the specific risks created by extravagant or excessive incentive structures which in turn can lead to excessive risk taking".

Statistical quality standards

Description

This is the lengthiest of the standards set out in the directive with a large number of requirements around the probability distribution forecast underlying the internal model including:

• Methods to be based on adequate actuarial and statistical techniques
• Methods to be consistent with assessment of technical provisions
• Methods to be based on current and credible information
• Assumptions to be justifiable to supervisors
• Data to be accurate, complete, appropriate and updated at least annually
• It covers all material risks and has sufficient ability to rank risks that the use test can be satisfied

Further the standard sets criteria around items that would naturally reduce capital

• That any diversification effects taken in account are adequately measured
• That any allowance for risk mitigation techniques fully allow for the associated credit risk

Observation

Some of these requirements are particularly challenging especially in the London market where data has been notoriously poor for many years. Many companies still have issues with the reliability of data for assessing their net claims on a deterministic basis. Having credible data on which to base assumptions about the tail of claims distributions or the ability to demonstrate adequate measurement of diversification effects will be far more challenging.

One possible effect of the standard could be a move to more pooling and sharing of data to enable benchmarking and testing of assumptions.

Calibration standards

Description

This standard allows for the possibility that insurers may make their own choice of both time horizon and capital measure in setting internal economic capital measures. This standard requires insurers to demonstrate how they can then use the model to provide a level of capital which is equivalent to the Solvency II standard of 99.5% value-at-risk over a one-year time horizon.

Observation

This test is likely to be most onerous for companies which have developed models to be used primarily for internal economic rather than external regulatory purposes and which are primarily based around different risk measures and time horizons which could range from: volatility of quarterly earnings; through to probability of ratings downgrade; through to some form of tail value-at-risk (TVaR) for run-off to ultimate. The first of these three examples would be particularly useful for a listed entity. The last of them may apply more for a small subsidiary of a larger group (domiciled outside the EU or equivalent regimes) and with a parental guarantee where the parent's concern may be more focused on the ultimate risk of the EU subsidiary producing claims which would impact on the overall capital of the group.

Given the requirements of the use test (and other Solvency II requirements around demonstrating the embedding of risk management in the company) it is likely that more companies will move to having models primarily used for internal economic purposes, and therefore will have to demonstrate compliance with this standard.

A further requirement of Solvency II which may aid companies with this standard is that for two years after receiving internal model approval companies calculating SCR using an internal model approach must also estimate SCR using the standard formula and provide a reconciliation of the differences.

Profit and loss attribution

Description

This standard is the shortest in the directive and requires insurers to provide at least annually a review of the causes of profit and loss for each major business unit and to demonstrate how these correspond to the categorisation of risk within the internal model.

Observation

This test appears to be based around back-testing of the model and in that respect co-incides with some of the requirements of the validation standard (see below)

It further emphasises the requirement for models to not simply be focused on the tail of the distribution of possible outcomes. Models should also accurately reflect experience in normal years, and capture the general annual volatility of profit and the key drivers of that profit. A model which has robust representation of a company's P&L and balance sheet within it and which can easily be reconciled to planned and actual results as they emerge will most easily meet this test as opposed to one which concentrates on detailed modelling of tail correlations (particularly when that modelling cannot be statistically measured).

Additionally this test challenges the granularity of risk modelling. If modelling is at too high a level, then it will be difficult to explain sources of emerging profit and loss.

Validation standards

Description

This standard requires a regular and statistically effective cycle of validation. Validation should be both backward-looking (testing the model against experience) and forward- looking (assessing how the model validates against newly emerging information). Validation should include sensitivity testing and an assessment of data.

Observation

The FSA's take on this requirement in its discussion paper was that validation required an objective and demonstrably independent validation function. This could be internal but would require strict separation of responsibilities. For many firms some form of external sign-off may well be appropriate and an easier way to demonstrate compliance with this standard.

Documentation standards

Description

This standard requires insurers to document their internal models including: design; operational details; assumptions; underlying theory and mathematical and empirical bases; and compliance with the other tests and standards. In addition some form of change control documentation is required.

Observation

The FSA has suggested a suitable test may be that documentation "is extensive enough for a firm to replicate its model in a different platform, and in the absence of original developers".

Many insurers are assuming that the requirements of Solvency II may require them to build more complex models. The documentation test can pull in the opposite direction: a model built on a platform where the structure and dynamics of an insurance business as well as the underlying modelling theory and approach is already established and fully documented by the model provider would go a long way to assisting insurers in meeting this standard.

What to do next

Each of the tests sets its own challenges to existing practices. The discussion above is by no means exhaustive and any company looking for internal model approval should be already starting a gap analysis between their existing model and the requirements of the directive.

Graham Fulcher is Principal Actuary, Watson Wyatt, UK
graham.fulcher@watsonwyatt.com
www.watsonwyatt.com

Editor's note: for more on the use test, see our article, "Is the test of use too loose?", Regulation/supervision, 28 November 2008.

Back to top