Occurrence Losses and Catastrophe Loss Classification

Why Loss Classification Matters

A collection of individually ordinary claims can arise from one severe event, while a single unusually large claim can arise from an unrelated, routine cause. For reserving and experience analysis, the distinction matters: event-related claims may develop differently from ordinary claims, and pooling them can conceal that difference. Classification therefore begins by identifying the unit being analyzed and applying a consistent event rule, not by judging each claim’s size in isolation.

The objective here is to distinguish catastrophe losses from attritional losses, define event-based groupings for analysis, and show how classification affects experience summaries and unpaid-loss estimates. A catastrophe label is an analytical classification; it is not automatically the same as the meaning of “occurrence” in a reinsurance contract. Contract wording may define an event for recovery purposes, and that contractual question must be evaluated under the applicable terms.

Definition: An attritional loss is a loss treated as part of ordinary, recurring claim experience under the classification rule being used. A catastrophe loss is a loss assigned to a qualifying catastrophic event under a stated analytical or operational rule. An event group is the set of claim records linked to the same event for a particular analysis. Classification rules should specify what evidence establishes a link, the relevant time and geographic scope, and any threshold used.

These terms depend on context. A catastrophe can produce many small claims, and a very large isolated claim need not be a catastrophe. A company might classify an event as a catastrophe because of its cause, a designated event identifier, an aggregate-loss threshold, or a combination of criteria. Such a rule is an assumption for the analysis, not a universal definition. Classification should also be kept distinct from the claim’s gross, ceded, and net amounts: those are different measures of the same underlying claim experience.

Building an Event Group

An event group is useful only if records are assigned consistently. Relevant evidence can include a common event identifier, reported cause, date range, location, and claims-handling information. A shared location or date alone may not establish that claims arose from the same event. Conversely, one event can affect multiple locations. If the available evidence is incomplete, the classification should be marked uncertain rather than treated as definitive.

For an internal analysis, a rule might classify an event as a catastrophe when its aggregate reported loss reaches a specified threshold. The analyst first groups claims by event, then applies the threshold to each group’s aggregate loss. Applying the threshold separately to each claim would answer a different question and can misclassify events composed of many modest claims.

Specify the rule.
State the event-linkage evidence and any threshold, including the loss measure and valuation date to which the threshold applies.

Assign claims to groups.
Use the available event evidence consistently; keep unrelated claims separate even when they have similar dates or causes.

Aggregate and classify.
Sum the relevant loss amounts within each group, then apply the stated classification rule to the group total.

Reconcile the result.
Confirm that every claim is counted once and that the catastrophe and attritional totals equal the total analyzed loss.

Worked Example: Applying an Aggregate Event Threshold

A hypothetical insurer uses a $900,000 event-loss threshold for an internal catastrophe report. Its event records for a review period are: Event A, claims of $420,000, $310,000, and $190,000; Event B, claims of $600,000 and $350,000; and two claims with no common event evidence, of $275,000 and $180,000. Assume the event groupings are reliable, and classify a group as catastrophe if its aggregate loss is at least $900,000. Find the catastrophe and attritional totals.

Method. Aggregate by event first, then compare each event total with the threshold. The two claims without common event evidence remain separate for this analysis.

Event A totals $420,000 + $310,000 + $190,000 = $920,000. Since $920,000 is at least $900,000, Event A is classified as catastrophe.

Event B totals $600,000 + $350,000 = $950,000. Since $950,000 is at least $900,000, Event B is also classified as catastrophe.

The two unlinked claims are each below the threshold and have no evidence connecting them as one event, so they are treated as attritional: $275,000 + $180,000 = $455,000.

The catastrophe total is $920,000 + $950,000 = $1,870,000. The total loss is $1,870,000 + $455,000 = $2,325,000, which also equals the sum of all seven claim amounts. Check: catastrophe plus attritional equals total loss, with no claim counted twice. A common mistake would be to test each claim against $900,000; that would classify none of these claims as catastrophe and would ignore the aggregate-event rule.

Classification and Unpaid-Loss Estimates

A classification can affect a liability estimate when catastrophe and attritional experience are assigned different development assumptions or selected factors. For a simple illustration, suppose a category’s paid loss is $P$, and a selected paid-to-ultimate factor $f$ is defined as ultimate loss divided by paid loss. Under that assumption, estimated ultimate loss is $Pf$, and estimated unpaid loss is ultimate less paid.

\widehat{U}=Pf,\qquad \widehat{\text{unpaid}}=Pf-P=P(f-1).

The identity follows directly from the definition of the factor: $f=\widehat{U}/P$, so multiplying both sides by $P$ gives $\widehat{U}=Pf$. Subtracting the paid amount gives $Pf-P=P(f-1)$. This calculation assumes the factor applies to the stated paid measure, valuation date, and population. It is not appropriate to apply a paid-to-ultimate factor to incurred loss, or to combine factors from inconsistent definitions without adjustment.

If categories $j$ have paid amounts $P_j$ and corresponding factors $f_j$, the category-based estimate is $\sum_j P_j f_j$. A single factor that reproduces that same total must be paid-weighted, not a simple average:

f_{\text{pooled}} =\frac{\sum_j P_j f_j}{\sum_j P_j}, \qquad \sum_j P_j>0.

This result is obtained by setting the pooled ultimate estimate equal to the sum of category estimates: $P_{\text{total}}f_{\text{pooled}}=\sum_j P_jf_j$, where $P_{\text{total}}=\sum_j P_j$. Dividing by $P_{\text{total}}$ gives the formula. A pooled factor reproduces the category-based ultimate only when it equals this weighted factor; a separately selected pooled factor need not do so.

Worked Example: Estimating Unpaid Loss by Category

For a hypothetical book at one valuation date, catastrophe claims have $200,000 paid and an applicable paid-to-ultimate factor of 1.50. Attritional claims have $400,000 paid and a factor of 1.10. Estimate ultimate and unpaid loss separately by category, then compare with a separately selected pooled factor of 1.20 applied to all paid loss.

Method. Apply each factor to its matching category’s paid loss, then subtract paid loss to obtain unpaid loss.

For catastrophe claims, estimated ultimate is $200,000 × 1.50 = $300,000. Estimated unpaid is $300,000 − $200,000 = $100,000.

For attritional claims, estimated ultimate is $400,000 × 1.10 = $440,000. Estimated unpaid is $440,000 − $400,000 = $40,000.

The category-based ultimate is $300,000 + $440,000 = $740,000. Total paid is $200,000 + $400,000 = $600,000, so category-based unpaid is $740,000 − $600,000 = $140,000.

Under the separately selected pooled factor, pooled ultimate is $600,000 × 1.20 = $720,000, and pooled unpaid is $720,000 − $600,000 = $120,000. The pooled estimate is $20,000 lower than the category-based estimate. Check: The category-based unpaid amounts sum to $100,000 + $40,000 = $140,000. The common mistake is to average 1.50 and 1.10 as though the categories had equal paid amounts; that simple average is 1.30 and would not represent the stated paid-loss mix.

Why the Grouping Rule Changes the Result

A grouping error can move losses between categories even when the total claim dollars do not change. If the categories use different factors, the estimated ultimate and unpaid amounts can change as well. The next example uses a hypothetical event threshold solely to demonstrate this effect; the factors are stipulated assumptions, not empirical claims about catastrophe experience.

Worked Example: The Effect of Splitting One Event

A hypothetical insurer classifies an event as catastrophe when its aggregate paid loss is at least $1,000,000. Three claims of $620,000, $290,000, and $170,000 are confirmed to arise from one event. An unrelated claim is $500,000. Use a catastrophe factor of 1.30 and an attritional factor of 1.08. Estimate unpaid loss under the correct grouping, then calculate the result if the event is incorrectly split into a $620,000 group and a $460,000 group.

Correct grouping. The confirmed event total is $620,000 + $290,000 + $170,000 = $1,080,000. Since $1,080,000 is at least $1,000,000, all $1,080,000 is classified as catastrophe. The unrelated $500,000 claim is attritional.

Catastrophe ultimate is $1,080,000 × 1.30 = $1,404,000, giving catastrophe unpaid of $1,404,000 − $1,080,000 = $324,000. Attritional ultimate is $500,000 × 1.08 = $540,000, giving attritional unpaid of $540,000 − $500,000 = $40,000. Total correctly classified unpaid loss is $324,000 + $40,000 = $364,000.

Incorrect split. The two artificial groups are $620,000 and $290,000 + $170,000 = $460,000. Both are below the threshold, so this erroneous classification treats the full $1,080,000 event loss as attritional, along with the unrelated $500,000. Total attritional paid loss becomes $1,580,000. Estimated ultimate is $1,580,000 × 1.08 = $1,706,400, and estimated unpaid is $1,706,400 − $1,580,000 = $126,400.

The incorrect split understates unpaid loss by $364,000 − $126,400 = $237,600. Check: Paid loss is $1,080,000 + $500,000 = $1,580,000 under either treatment; only the classification and associated factors change. The example shows why event evidence and rules must be applied before selecting category-specific assumptions.

Common Mistakes and Practical Controls

Classifying by claim size alone. A threshold defined on event aggregate must be applied to the event total, not separately to each claim.
Assuming every large loss is catastrophic. Severity by itself does not establish a shared event or meet a stated catastrophe rule.
Combining claims without evidence. Similar dates, causes, or locations may prompt investigation but do not necessarily prove a common event.
Using an unweighted average factor. When combining category factors to reproduce a total estimate, weight each factor by its corresponding paid amount.
Mixing measures or valuation dates. A paid-based factor and paid amount must be on compatible definitions and dates; otherwise the multiplication is not meaningful.
Treating analytical classification as contract interpretation. An internal catastrophe group does not by itself establish which losses meet a reinsurance contract’s definition of an occurrence.

A useful control is to preserve the claim-to-event mapping, record the rule and valuation date used, and reconcile the classified totals to the full population. Where event assignment is uncertain, retain an explicit uncertainty flag and assess whether plausible alternative classifications materially change the estimate. These practices support review and make later changes in classification traceable.

Key takeaway: Define the event rule before aggregating, classify at the unit the rule specifies, and reconcile all claims. If category-specific factors are used, calculate each category consistently; a pooled factor reproduces their total only when it is weighted by the corresponding paid amounts.

Check Your Understanding

Use the stated classification rules and factors; treat each problem as a hypothetical analysis.

An internal rule classifies an event as catastrophe at $750,000 or more. Event A has claims of $410,000 and $260,000; Event B has claims of $500,000 and $300,000; and a separate, unlinked claim is $720,000. Find the catastrophe and attritional totals. (Answer: Catastrophe $800,000; attritional $1,390,000.)
Catastrophe paid loss is $150,000 with factor 1.40; attritional paid loss is $250,000 with factor 1.08. Find total estimated ultimate and unpaid loss using separate factors. (Answer: Ultimate $480,000; unpaid $80,000.)
A $1,000,000 event threshold applies to event aggregate. Three claims of $540,000, $280,000, and $210,000 share a confirmed event identifier. An unrelated claim is $460,000. Classify the losses and state the totals. (Answer: Catastrophe $1,030,000; attritional $460,000.)
Catastrophe paid loss is $300,000 with factor 1.25, and attritional paid loss is $700,000 with factor 1.05. A separately selected pooled factor is 1.10. Compare unpaid estimates. (Answer: Separate-factor unpaid $110,000; pooled-factor unpaid $100,000; pooled is $10,000 lower.)
An event threshold is $900,000. One confirmed event has claims of $470,000, $290,000, and $180,000; a separate claim is $880,000. Use factors 1.30 for catastrophe and 1.10 for attritional. Compare correct grouping with the result if the event is split into separate claims before applying the threshold. (Answer: Correctly grouped unpaid $370,000; incorrectly split unpaid $182,000; the split understates unpaid by $188,000.)

What You'll Learn