Large Loss and Shock Loss Development
Techniques for handling large and shock losses in actuarial reserving and pricing analyses.
Identifying Large Losses
Large losses can distort standard actuarial analyses if not handled properly. Common identification methods include setting dollar thresholds (often the 95th or 99th percentile of individual claims), using statistical tests for outliers, or applying industry benchmarks. The threshold should balance including enough data for credibility against excluding claims that would skew development patterns. Different lines of business require different large loss definitions.
For example, in personal auto liability, a large loss threshold might be $250,000 per claim, while in commercial general liability it could be $1 million or more. In workers compensation, catastrophic claims involving permanent total disability can exceed $5 million. The threshold should be set high enough to capture genuinely unusual claims but low enough that there are still enough large claims above the threshold to support meaningful analysis. A common guideline is that the threshold should produce at least 10 to 20 large claims per accident year for credible frequency analysis.
Why Large Losses Need Special Treatment
Standard actuarial methods like the chain ladder assume relatively homogeneous data. When a single large claim dominates an accident year or development period, it can create misleading development patterns. For instance, if a $5 million claim is reported three years after the accident date, the resulting development factor for that link ratio will be much higher than normal, potentially distorting the projection for every accident year to which that factor is applied.
Large losses also tend to develop differently than typical claims. They are more likely to involve litigation, which extends the settlement timeline. They may have more volatile case reserve estimates as information about the claim's severity evolves. And their ultimate value is more uncertain, making point estimates less reliable. For all these reasons, treating large losses separately from the bulk of claims improves the accuracy of the overall reserve estimate.
Treatment Methods
Actuaries use several approaches for large losses. Truncation caps individual claims at a threshold and adds a separate load for excess amounts. Separate development analyzes large and small claims independently, then combines results. Frequency-severity models estimate the number and size of large claims separately using distributions like Pareto or lognormal. Shock losses from catastrophes or one-time events may be excluded from trend analysis entirely and priced through catastrophe loads or reinsurance cost allocations.
Truncation Method in Detail
The truncation (or capping) method is the simplest approach. Each individual claim is capped at the large loss threshold. For example, if the threshold is $500,000 and a claim has an incurred value of $2 million, only $500,000 is included in the capped loss triangle. The capped triangle is then developed using standard chain ladder methods. The excess portion ($1.5 million in this example) is handled separately, either by adding a loading factor based on historical excess loss experience or by estimating excess losses using a severity distribution.
The advantage of truncation is its simplicity and the fact that it produces more stable development patterns. The disadvantage is that it requires a reliable method for estimating the excess layer, which may itself be volatile due to the small number of claims involved.
Frequency-Severity Models
Frequency-severity models provide a more sophisticated treatment of large losses. The actuary separately estimates the expected number of large claims (frequency) and the expected size of each large claim (severity). Frequency is often modeled using a Poisson distribution, while severity for claims above the large loss threshold is commonly modeled using a Pareto, lognormal, or generalized Pareto distribution.
The Pareto distribution is particularly popular for modeling the tail of loss distributions because it has a mathematically convenient form and captures the heavy-tailed nature of large insurance claims. If the Pareto shape parameter (alpha) is 1.5 and the threshold is $500,000, the expected value of a claim given that it exceeds the threshold is $500,000 times alpha divided by (alpha minus 1), which equals $1,500,000. This expected severity, combined with the expected frequency, gives the total expected large loss cost.
Shock Losses and Catastrophes
Shock losses are one-time events that produce unusually large losses, such as natural catastrophes, industrial accidents, or mass tort events. These losses should generally be excluded from the base analysis and priced through a separate catastrophe or shock loss load. Including them in standard development triangles would inflate development factors and distort the projection of normal losses.
When excluding shock losses, the actuary must clearly document which events were removed and the rationale for removal. The excluded losses should still be accounted for in the total reserve or pricing indication through an appropriate loading mechanism. Catastrophe models (such as those from AIR, RMS, or CoreLogic) can provide estimates of expected annual catastrophe losses to support this loading.
Common Pitfalls
One common mistake is failing to handle large losses at all, allowing a few extreme claims to drive the entire analysis. Another pitfall is setting the large loss threshold too low, which removes too much data from the base analysis and reduces the credibility of the remaining development factors. Conversely, setting the threshold too high allows large claims to contaminate the base data. The threshold should be reviewed periodically as the portfolio and claim cost levels evolve.
Study Tips
For exam preparation, understand the rationale for separating large losses from the base analysis. Practice applying the truncation method and calculating excess loss loads. Be familiar with the Pareto distribution and its application to large loss severity modeling. Know how to estimate expected large loss costs using a frequency-severity approach. Review how catastrophe loads are determined and incorporated into pricing and reserving analyses.