Solvency 2
Speech by John Tiner, Chief Executive, FSA
CEA/Geneva Association Seminar on Solvency 2
14 November 2005
I last had the opportunity to speak on Solvency 2 at a Geneva Association Conference in November 2004. On that occasion I drew attention to the need for radical new thinking on the reform of insurance solvency rules within Europe. The existing EU arrangements – already supplemented and in effect superseded on a national basis by growing number of member states – rely upon an outdated mixture of opaque prudence in the quantification of the technical provisions, inflexible asset allocation limits and a simplistic volume-based capital requirements formula. This recipe of the outdated, the inflexible and the inadequate is not a sound basis on which to take insurance prudential regulation into the 21st Century. I believe strongly that an insurance company’s balance sheet should be prepared on the basis of transparent market, economic and historical data which is consistent with how management run the business. Insurance Supervisors can then have sensible conversations about the level of prudence built into the balance sheet and the capital needed to support all the risks of the company.
I am pleased to say that the fact that the existing EU solvency rules are inadequate is now widely recognised and since my previous speech there appears to be a growing convergence of view as to how reform should now proceed. In my brief remarks today I aim to explain why and how I believe views are converging, describe some obstacles to reform that still remain and set out my vision as to how those obstacle might be overcome. I will organise my remarks under the headings technical provisions, asset allocation and capital requirements.
Technical provisions
In my speech in November 2004, I argued against the inclusion of opaque prudence in the quantification of technical provisions. I was not of course arguing against the prudent need for insurance companies to hold financial resources in excess of their liabilities but that unless one first measured liabilities one could not be confident that financial resources did indeed exceed liabilities, nor know the amount the of the excess and so judge whether it was adequate. I argued that measuring liabilities meant measuring their market-consistent value. However, I acknowledged the significant practical difficulties that still remained in implementing a market-consistent standard and argued that as an interim measure thought should be given to setting the soundness standard for technical provisions by reference to a defined confidence level. For example a 75% confidence level would mean that the aim should be to quantify technical provisions at an amount that is expected with a probability of 3 in 4 to at least equal the outcome on the run off of liabilities. This suggestion – which was based on the precedent set in Australia – was at the same time being made by several others and was taken up by the EU Commission and included in their Solvency 2 Roadmap and calls to advice from CEIOPS as a 'working hypothesis'.
CEIOPS in response to the EU Commission 'second wave' calls for advice has now examined this idea thoroughly and concurred that as a working hypothesis it and alternatives should be tested in the next Solvency 2 Quantitative Impact Study (QIS 1). The quantitative impact studies test ideas put forward for the design of Solvency 2. They are co-ordinated by CEIOPS which drafts the specification for the study and collates the results, but are actually executed on a voluntary basis by insurance companies under the direction of their national supervisors.
The QIS 1 specification recently issued by CEIOPS includes testing of not only the 75th percentile but also the points at plus or minus 15%, i.e. the 60th and 90th percentile. It also allows firms to submit results from other methods such as the cost of capital approach articulated by the Chief Risk Officers Forum. CEIOPS will use the data collected for the different percentiles to interpolate between the 60th, 75th and 90th percentiles to understand how the quantum of technical provisions changes as the defined percentile confidence level is varied. It will also compare the results of other methods to the percentile approaches. This and the other detailed information from the QIS 1 results will help CEIOPS answer the big remaining questions on design of the Solvency 2 soundness standard for technical provisions. As I see it these big questions include:
- Should a defined-percentile approach be used?
- If so, what percentile should be set as the standard?
- Which risks should be included in the scope of the percentile?
- And should other limits, floors or restrictions be imposed on the quantification of technical provisions?
On the first of these questions it has to be said that the defined-percentile approach is susceptible to criticism from two different directions. First those insurance companies who are more advanced in quantitative risk management such as the companies represented in the Chief Risk Officers Forum argue for a soundness standard such as the cost of capital approach that is more explicitly based on market consistent valuation. They argue that they are – or within the timeframe of Solvency 2 will be – ready to implement such an approach. From the other direction some are expressing well-founded concerns that firms who lack rich data histories (perhaps because they are small, or are new start ups or have newly diversified or simply haven’t kept long data histories in accessible form) will be unable to project the 75th percentile from their own data alone in a statistically robust manner.
I believe the CRO Forum are right to raise the question of whether at least for some insurance companies it might be possible to use a soundness standard for technical provisions that is more explicitly based on market consistent valuation. However, I continue to believe at least for the time being that it is premature to use such a standard for all insurance companies. For that we need to await the outcome of the IASB's Phase 2 project on the measurement of insurance contracts' liabilities. At the other extreme, however, I do believe that with appropriate safeguards some insurance companies should be allowed to depart from a percentile based approach and adopt a purer market consistent standard. I note that insurance companies that are likely to be best able to do this are precisely those companies that under Solvency 2 will be applying to their supervisory authority for permission to use an internal model to calculate their solvency capital requirement. I would propose that simultaneous with applying to their supervisory authority for permission to use an internal model to calculate their solvency capital requirement they should be allowed to apply for permission to use a market consistent approach, such as a cost of capital approach, for the measurement of their technical provisions. Permission would of course only be granted where the supervisory authority is satisfied that the applicant has demonstrated that they are able to implement a statistically robust and economically valid methodology. The application process, and resultant grant or refusal of permission, would thus dovetail with the parallel application in respect of the use of an internal model for the solvency capital requirement.
The problem of insurance companies that do not have rich data histories is more difficult to address but can be solved. I would draw attention to two complementary ways forward. First a defined-percentile soundness standard should be framed so that insurance companies are not required to use a stochastic methodology where there is insufficient data adequately to support that methodology. Reasonable approximations using deterministic methods should be permitted. There are of course risks with this approach. The principle risk is that a diverse range of different approximations and methods may produce outcomes that are inconsistent as between different insurance companies that use them and as compared to stochastic methods. I believe that the best way to mitigate this risk is for clear actuarial standards to be put in place defining, limiting and describing appropriate deterministic methods, how they should be calibrated and how and when they should be used. In this respect I welcome the recent initiative by the EU Actuarial Groupe Consultatif to investigate how this might be done. The second way to help solve the problem is for industry associations – in so far as they are permitted by competition laws – to collect and collate data to enable insurance companies to use industry data as well as own data. I would emphasise these two ways of helping to solve the problem are complementary. They should both be pursued and can each help reinforce the other.
Next on the question of what percentile should be used in the defined percentile approach I would say simply that in my view the purpose of the defined-percentile approach is to serve as a temporary proxy for a market consistent approach. So, the defined percentile should be set at a level that is roughly equivalent to the market consistent value.
Which risks, then, should be included with the defined-percentile standard. As I see the defined-percentile approach simply as a proxy for market consistent valuation, I only see the need to include within the scope of the defined-percentile those risks such as non-life underwriting or life mortality and morbidity risks for which market consistent prices are hard to obtain. I see no need to include financial risks such as interest rate risk, asset-price risk or other forms of market risk which can be hedged in financial markets and for which market consistent values can be obtained. These risks where they occur within technical provisions – just as where they occur within assets held matching technical provisions – should be marked to market-consistent values. This of course will help ensure the consistency of the valuation of assets and liabilities for the purpose of setting prudential regulatory requirements. Of course, not necessarily all market risks in technical provisions are always hedged by offsetting market risks within assets. However I would not – as some argue – limit the use of marking to market consistent values solely to those cases where risks are hedged. It is sufficient that an insurance company can demonstrate an ability to hedge the risk within the financial markets and therefore can ascertain a market consistent price. The risk that arises from unhedged financial risks or the mismatch of assets and liabilities must be taken into account in setting the solvency capital requirement.
Turning now to the last question on the quantification of technical provisions. I am aware that some would like to see additional limits, floors or restrictions – beyond merely the defined-percentile soundness standard. These might include - floors equivalent to the aggregate of surrender values for life insurance policies and the aggregate of 'prudent' case estimates for non-life insurance; limits or restrictions over claims discounting, the extent to which diversification benefit within and between lines of business may be recognised and on the recognition of reinsurance and other risk mitigation or transfer. As a general view I would reject any such impenetrable pre-set limits, floors or restrictions. That is not to say that if, for example, an insurance company wishes to claim diversification benefit it should not be required to provide good evidence that it is present. Similarly, if it wishes to recognise reinsurance it needs to demonstrate that the reinsurance is present in the amount and type claimed and if necessary set up an adequate provision for any residual risks such as reinsurance credit risk. Policy case estimates and surrender value floors provide information that is often essential in projecting the defined-percentile value of technical provisions. They should not be ignored, but that does not mean that simply aggregating their policy values and imposing the aggregate as a regulatory floor is the best approach. If case estimates are probabilistic best estimates then one would expect the quantum of technical provisions to exceed their aggregate value. However, the same is not necessarily true for so-called prudent case estimates. The prudence included at a policy level when aggregated for all policies in a portfolio may exceed the prudence imposed solely by the defined-percentile standard applied at the level of the portfolio. If it does it should not be imposed as an additional prudential requirement. The defined-percentile standard is sufficient. Setting an aggregate surrender value floor ignores actual and expected policyholder behaviour. Not all policyholders surrender at the point in time that is most disadvantageous to the insurance company (which of course is not necessarily always the point in time that is most advantageous to them). For the purposes of quantifying technical provisions insurance companies that have good evidence as to policyholder behaviour should be allowed to factor this into their calculation. Of course the risk of very high surrender rates in extreme circumstances should not be ignored but as with other extreme or tail risks the appropriate place for dealing with this is by placing an appropriate weight on it in the quantification of the solvency required capital.
Asset allocation
Technical provisions of course need to be matched by assets and this leads me to my next topic, asset allocation. At present the EU directives impose a few asset admissibility limits on insurance companies and authorise supervisory authorities to set more detailed limits. The question I would like now to address is what place if any should asset limits have in Solvency 2.
Assets limits serve a variety of purposes. By restricting the amount and quantity of admissible assets they limit market, credit and liquidity risk. However they can give a false sense of security and, as they tend to force diversification, operate as a blunt instrument. Risk below a certain level is ignored. Risk above that level is treated as fully realised. This is an all or nothing approach to setting prudential requirements which is both excessively prudent – by treating some risks as if they were fully realised – and at the same time excessively imprudent – by ignoring other risks. I believe that a far better way of dealing with market and credit risk is by designing a solvency capital requirement that includes an appropriate weighting for all types of market and credit risk. Only to the extent to which this proves not possible should we contemplate including some asset admissibility limits in Solvency 2. And even then the limits should be high level and focus solely on large exposures or excessive concentrations analogous to the limits that apply in the EU banking directives.
Capital requirements
I turn now to my last topic that of the solvency capital requirement. This is a vast topic and today I will touch on only one aspect – that is the critical need to avoid under or overlaps in the design of the soundness standard for technical provisions and the soundness standard for the solvency capital requirement. It is important that together the assets that match liabilities (and in particular the technical provisions) and the assets that represent the solvency capital requirement are adequate to meet the risks of the business. Equally I know that the insurance industry is concerned that there should not be double counting of risks. To the extent that margins are held in the technical provisions to deal with a risk they should be allowed to offset any capital requirement also demanded in respect of that risk. I share this view but with one important point of clarification as to how I think this should work. My vision is that technical provisions should be set at the market consistent value, or by some close proxy to the market consistent value such as the defined-percentile approach appropriately calibrated. The solvency capital requirement should then measure the potential variability of the near-market-consistent technical provisions over the 12 month time horizon to a confidence level of 99.5%. If, and only if – contrary to my opinion – the defined-percentile approach is calibrated so that technical provisions are materially in excess of the market consistent value, is there a need to allow the margins in the technical provisions to be used to offset the solvency capital requirement. In that case the relevant margins would of course be the amount of the excess relative to the market consistent value.
I have now touched upon the three important areas of technical provisions, asset allocation and the solvency capital requirement and I hope you can see a consistent theme in my comments. The best way forward in designing Solvency 2 is to follow a broadly market-consistent approach making appropriate allowances for the difficulties in implementing that approach in some aspects. I note the strong support this view appears to be gaining from the insurance industry itself, many insurance supervisors and other stakeholders and I would like to add my voice to that view.