Reinsurance or capital - which is best? Kathryn Morgan looks at the options
The latest consultation paper on Solvency II talks of reinsurance as the key risk management tool in insurance. CEIOPS, who are leading the development of the solvency framework, recognise that the assessment of solvency needs to allow for the mitigating effects of reinsurance.
This means that both the standard formula and internal models need to allow for an insurer's reinsurance programme in estimating the amount of capital required.
At present, the European solvency regime does not allow for reinsurance in a way that reflects the structure of reinsurance programmes or the security of the providers. Solvency I, the current regime, is a broad-brush approach to solvency assessment that provides a 'one-size-fits-all' solvency margin.
The aim of Solvency II is to provide a more sophisticated framework for insurers and regulators to work in, and, as part of this, allow for reinsurance programme structure and reinsurer security explicitly.
So how would an insurer do this using an internal model? Fortunately, the UK regulatory regime is ahead of the European Union (EU) and insurers have been working on their individual capital assessments using models of their business from 31 December 2004. And the good models explicitly allow for reinsurance.
The models are simple in concept, and complicated in practice. Basically, a general insurer starts with a pot of money held in provisions, and receives money from premiums and investment income, pays out money in claims and expenses, and has assets to back the business. Money is paid in reinsurance premiums, and received from reinsurance recoveries. The models generally look at one year of future business and project the profit and loss account and balance sheet.
Complications arise from setting assumptions about levels of new business, levels of retention at renewal, levels of cancellations and average premiums.
There are assumptions about claims - frequency and severity could be used, or loss ratios, split by claim type. Investment returns are also uncertain.
Overlying this part of the model are assumptions about the frequency and severity of large claims, which feed into the reinsurance recovery calculations. Another assumption needed is the frequency and severity of catastrophe events, which again directly affects reinsurance recoveries.
This will usually be based on as assessment of the insurer's exposure, probably using external data on large events such as storm and flood applied to the policy base to estimate the likelihood and cost of these events to the insurer.
What happens is that the model is set up and different scenarios are run through it.
In a stress and scenario testing model, these could be a set of different scenarios based on management experience and what are expected to be extreme values for the assumptions. The results give an insight into what might happen to the company's profitability and solvency under these conditions.
In a stochastic model a statistical distribution is applied to each assumption.
This means that the model knows how uncertain each assumption is and factors this in to the profit and loss, and balance sheet projections.
A stochastic model is run several thousand times. Parameters are taken randomly from the statistical distributions, so the claim frequency for theft claims, for example, will be better than average in some years and worse in others. Linked with all the other assumptions, the insurer can see what would happen in a variety of scenarios. Some will be the stuff of nightmares, where all the parameters go wrong together. Some will be wonderful years, where all the assumptions go well. The insurer will choose a risk level that they are comfortable with and look at the capital needed to prevent insolvency to that level.
The models will store the parameters and results from each different run of the model, so that the extreme years can be looked at in detail.
This helps management to understand the risks in the business and think about what governance they need to mitigate them.
This approach allows for reinsurance programmes in terms of the premiums payable to reinsurers, the recoveries on large claims and severe events and the reinstatement premiums payable. The security of the reinsurers can also be allowed for by attaching a probability of insolvency to each one, and allowing for failures or reduced recoveries according to this.
Using these kinds of model allows insurers to assess their overall capital according to their appetite for risk. It allows capital to be allocated by class of business, with the most risky classes getting most capital.
And it allows different management actions to be tested - this is particularly useful when deciding what reinsurance to buy.
Different programmes, different retention levels and different retained proportions can be tested to see what is best for the company from a value for money point of view and from a best use of capital point of view.
They will be able to see how volatile their profits might be if no reinsurance is bought.
The insurer will also be able to see the risk from buying no reinsurance and assess the amount of capital freed by having a good reinsurance programme.
They will be able to compare the amount of capital saved with the reinsurance premium spent.
And the impact of reinsurer default can be assessed - if an insurer pays a lot of money for reinsurance and the reinsurer does not pay, then the effect on solvency can be seen.
UK companies are now working on extending their models to cover more than one year of business. This increases the uncertainty as there are more assumptions needed on business volumes. These depend on the market climate and modelling underwriting cycles is not straightforward. For reinsurance, the capital available in the market will affect prices, and this will depend on recent claims worldwide, not just those affecting the insurer.
Allowing for the future
Another big issue is allowing for future management actions. In a multi-year model there will be scenarios where management will do something in response to circumstances. An example is putting up premium rates if the underwriting cycle turns, even if claims experience does not justify it. Or buying more reinsurance after a bad year.
The model should allow for these actions, but the business needs a framework to show that the decisions would actually be taken. Otherwise, the model will always show that the business is doing well, thanks to its far-sighted management.
Assuming a reliable, realistic multi-year model can be produced, this model provides the business with a good framework for decision-making and risk assessment. This could form the basis of long-term reinsurance deals, which benefit insurers and reinsurers. The most useful output may be the conversations about the extremely bad and extremely good scenarios, as they will help management to think through what they really would do in those circumstances - and they will be able to see the effects of their actions.
Developing an internal model requires a good understanding of the business, an understanding of the key risks in the business and access to data on claims and premiums so that parameters can be set that reflect the company's experience. Models are really helpful to general insurers - but they are only as good as the data that goes into them - garbage in means garbage out.
So, insurers will have a good handle on the relationship between capital and reinsurance, and will be able to use these models to inform their decisions on reinsurance buying.
- Kathryn Morgan is a member of the General Insurance Board of the Actuarial Profession.