PRA Consultation Paper CP7/14

Solvency II: calculation of technical provisions and the use of internal models for general insurers

This consultation paper was issued by the PRA in March 2014.  It is a short, accessible read and contains several helpful pointers to firms as they prepare for Solvency II. 

The scope of the note is the set of general insurance actuarial and risk models that firms use for technical provision and SCR calculations.  Essentially it reads like a top-10 list of points that firms need to be able to answer when undergoing a PRA review.  Here is our view on the main issues.

Area of PRA concern What we think
1. Firms should not use the proportional proxy approach in their risk margin calculations without justification.  Equally, internal model approaches that assume that reserve risk will emerge in line with paid or incurred claims patterns may be unacceptable. We are pleased that the PRA has come out strongly on these points.  It has been a matter of concern for us (and many others) that there has been an implicit acceptance of naïve, statistically weak and mechanically applied approaches. What this means is that if your model relies on any of the following approaches, you may need to reflect on whether they will continue to withstand regulatory scrutiny:

  • Bootstrap models, with no adjustment for calendar year effects – used to obtain estimates of ultimate reserve risk.
  • “Actuary in the box” – used to translate ultimate reserve risk to one year reserve risk.
  • Proportional proxy methods – used to translate ultimate reserve risk to one-year reserve risk.
2. Internal model end-of-year technical provisions calculations need to allow for the gearing effect of changes to the reserving basis used following adverse experience.  Specifically, “actuary in the box” chain-ladder methods are highlighted as potentially inadequate here.
 3. Firms should only use approaches that ignore calendar year effects, such as claims inflation, where they have shown that the result is appropriate.
 4. Technical provisions and internal models must reflect ENIDs (Events Not In Data). We are pleased that the term “binary events” has (hopefully) been consigned to the dustbin of terms that became unhelpful through being stretched from one purpose to another. In our view, an allowance for ENID is something that should already have been taken into account by firms.  We see this restatement by the PRA as a natural complement to the points immediately above.  Firms using the more mechanical approaches listed will inevitably fail to be allowing for ENIDs. 

A small reminder of the original (useful) intent of the term binary events: low likelihood / high severity scenarios that contribute to translate 50/50 reserve scenarios to mean reserves.  It is important for such elements to be robustly determined as, in marginal profitability years, they can come under considerable pressure.



5. Internal model parameters should allow for estimation error. These two points may be helpfully considered together.  In reality many firms will have a number of assumptions (particularly correlation matrices) where it will be difficult, if not impossible, to provide a hard basis for the selected assumption.  We think the PRA is simply saying that they expect you to:

  • carry out sensitivity tests for those parameters which are most uncertain, and then
  • take an appropriate margin in the selected parameter that reflects the scope for its mis-estimation.
6. Sole reliance on “industry standard”, “accepted good practice” or “default” assumptions without further testing will not be acceptable.  Firms need to consider their appropriateness relative to their own risk profile.
7. Profitability improvements (both in premium provisions and internal models) must be justified.  Relying on their inclusions in business plans will not be considered sufficient. Care is needed here: aside from the behavioural psychology associated with planning for improved profitability, this can present real practical challenges for firms. Often business plans will include profitability improvements, possibly to provide a stretch element in objectives, or for other practical purposes.  It goes without saying that this provides plenty of scope to tie yourself in knots.  Is it better to have a business plan that is not set at the mean, or to have two sets of planning numbers: a true best estimate, and an ambitious target for planning purposes?
8. Third party models (such as economic scenario generators and catastrophe models) need to be subject to the same level of scrutiny and testing that models developed in-house receive. This requirement should come as little surprise for firms.  In our experience they have struggled in two areas:

  • Presenting the testing performed in a manner that provides an effective holistic picture of the work performed.  Firms have often found themselves stuck in silos of tests performed for validation and not communicated a clear overarching narrative regarding their external model usage.
  • Demonstrating ownership of the model testing performed by providers.  A simple regurgitation of third party material is, in our experience, a good way to draw attention to yourself.
9. Models must allow for risk mitigant failure (most notably reinsurance default, dispute and exhaustion). In our experience this has a relatively minor impact on technical provisions.  That said, you do need to take care in situations where the mean reserves approach points where the reinsurance programme changes.  Here the upside and downside will no longer cancel one another out. This issue is particularly important in technical provisions calculations and capital calculations where a deterministic approach is being used.
10. Firms need to consider the cumulative effect of approximations that individually are found to be immaterial. This can be important; however for many firms some simple checks will suffice to confirm no further work will be required.