This is part 2 of a 3 part series. Part 1 is here.
Non-life claims reserves are regularly not discounted, for bad reasons and good. This part of the series looks at the related issue of inflation in claims reserving. (You’ll have to wait for part 3 for me to talk about the analysis that prompted this lengthy series.)
In many markets, inflation is low and stable. Until a decade ago, talk of inflation wouldn’t have raised much in the way of deflation either. That’s still sufficiently unusual to put to one side.
Low, stable inflation means that past claims development patterns are mostly about, in approximate descending order of importance (naturally depending on class and peril)
- development of number of claims,
- development of claim estimates due to the reversal of initial conservatism (a typical but not universal practice), and in distance third place
- the impact of inflation of claims.
Similarly to discounting, one of the reasons inflation has such a small impact (especially when inflation itself is low) is that most claims for many classes are reported and paid quickly. This is clearly not true across all classes, and long tailed liability is an obvious counter example.
What the different common claims reserving methods assume about inflation
Basic Chain Ladder
Basic Chain Ladder assumes constant development patterns. Since inflation development contributes to development patterns the implicit assumption is that inflation is stable too. Strictly, we usually talk about “future inflation being a weighted average of past inflation” or similar.
Too many non-life reserving actuaries don’t spend enough time looking at the variability of and trends in development patterns before blindly following the methodology. Inflation isn’t the only variable that can disrupt past development patterns.
Bornhuetter-Ferguson does the same, plus the addition of an initial or a priori estimate with even fuzzier implicit views on implied future inflation.
Most subjective assessments of assumed loss ratios will factor in a subjective view of inflation, most typically assuming it will continue as it has in the past. Practically, the choice of loss ratios to report on (accident year, underwriting year, financial year etc.) may also have an impact here.
Loss Ratio approaches
Simple Loss Ratio methods suffer the same fuzziness. We are assuming something about inflation, but almost certainly not explicitly so and I have never see a robust analysis of the impact of unexpected inflationary differences on loss ratios as feedback into Loss Ratio based reserving.
Average Cost Per Claim
Average Cost Per Claim (ACPC) methods have always appealed to me for their innate separation of development of number of claims from the actual amounts involved. (Don’t worry, I am firmly in the school of “apply many methods and understand the differences in results as part of building up a final estimate” rather than sticking with just ACPC.)
Apart from the ability to separate a pure IBNR from development of case estimates in an OCR (or Reported But Not Settled / RBNS to use a name that is less prone to being misunderstood) it also focuses the mind on the effects of inflation on individual losses.
It isn’t universally applied, but there is no reason not to apply a separate average cost per claim for different calendar years.
When it comes to an analysis of change in reserves, an Actual vs Expected comparison, there are plenty of good reasons to separate out number of claims and actual vs expectation inflation separate from the average underlying severity of claims that drives the final cost per claim.
Inflation Adjusted Chain Ladder
Then on to the Inflation Adjusted Chain Ladder (IACL). As a reminder, since it isn’t that commonly used in markets where inflation is uninteresting, the method is as follows:
- Adjust past claim payments for past actual inflation (ideally using indices of claim costs or at least sub-indices that should closely follow claim costs) to put all claims in current money terms;
- Use traditional triangle completion methods are followed to estimate real ultimate losses;
- The final step is taking those future real claim payments and inflating them for future inflation. This naturally requires an estimate of future inflation.
Side note – impact of currency on inflation
In emerging markets, many parts and components are imported. This is true for motor, but also true for some property damage.
Rapid currency depreciations (Nigeria, I’m not only looking at you, but I am looking at you) can have a predictable impact on future claims inflation that might irk an efficient market proponent. There is no technical challenge here. Inflation is a time series process not a market price one – therefore we should expect inflation to be predictable.
The methodologies that allow for more explicit inflation estimates are better able to factor in the information about future inflation. Currency impacts are a clear example that should be factored into reserves, but cannot be if the inflation allowances are implicit.
Break even inflation, economist forecasts or last year’s inflation to predict future inflation?
This is worth an entire blogpost (or series) in itself. Estimates of future inflation should ideally be accurate, objective, directly related to the types of costs involved (rather than generic CPI), and easy to update.
Last year’s actual inflation isn’t bad for many of these measures. And often it’s surprisingly accurate. It doesn’t have the same actuarial panache that more complex methods have though…. and yes when inflation isn’t stable it can give rise to poor estimates and poor decisions around reserve sufficiency and adjustment to future pricing based on experience analysis of incompletely run off periods.
What many don’t realise is that “break even inflation” or the difference between nominal and real “risk free” government bond yields is not an unbiased estimate of future inflation. It includes an inflation risk premium, typically positive, due to the imbalance of supply and demand for real risk-free investments to match real liabilities of pension funds, life insurers, and small portions of non-life insurers’ balance sheets.
In the same way estimates of CPI may need to be adjusted to reflect systematic differences to specific damage types or perils or just the inflation experienced by specific lines, so break even inflation based estimates also require adjustment before use.
Real estimates of claim reserves?
One methodology I saw adopted recently was a version of the IACL method where past claims were adjusted with inflation, development factors derived to estimate future real claims in current money terms, but then no future inflation and no discounting applied.
The actual face value of a reserve derived this was is not clearly wrong. By implicitly allowing for discounting equal to future inflation, this might be a reasonable proxy for investment returns that could actually be realised given a portfolio with plenty of cash and money market instruments included to bolster liquidity. (See part 1 for a discussion on why a “risk free yield curve” may not always be a reasonable rate at which to discount claims reserves, at least not without adjustment.)
By now, you may gather that I don’t approve of the implicit nature of assuming future inflation and a reasonable, achievable discount rate will naturally offset. But, there is a more significant problem with this approach that I will cover in part 3.
This is part 2 of a 3 part series. Part 3 is here (not yet available).