Why Portfolio Aggregation Requires Event-Level Thinking
A catastrophe portfolio may contain policies at many locations, but a portfolio loss is not obtained by treating every location as if it experienced a separate catastrophe. A single storm, wildfire, or other event can affect multiple locations at once. To aggregate correctly, retain the identity of each event and the losses it produces at each location. Those shared event outcomes are the basis for both event-level and annual portfolio losses.
The calculation also depends on what each location loss represents. Losses should be on a consistent basis: for example, all gross of reinsurance or all net of specified policy terms and reinsurance. If policy terms are applied location by location, first calculate each location’s covered payment under those terms, then aggregate those payments for the event. A portfolio-level deductible, limit, or other nonlinear term may instead need to be applied to the appropriate portfolio total.
The event total and annual total answer different questions. An event loss combines locations affected by one event. An annual aggregate loss combines the event totals for the entire year. Consequently, if a simulated year contains several events, its annual aggregate can exceed each individual event loss. Conversely, an OEP calculation selects the largest event in that year rather than adding every event.
From Location Losses to Event and Annual Totals
A reliable aggregation keeps a record of at least three things for each modeled loss: the year, the event identifier, and the location. The year identifies which losses belong in an annual total; the event identifier identifies which location losses belong in the same event total. Summing all losses in a year without retaining event identities gives an annual aggregate, but it cannot recover the OEP loss for that year.
The event-total formula assumes that the location amounts can be added. This is appropriate when they are compatible payments or losses and there is no additional interaction among locations. It does not mean that every contract term is applied independently. For instance, if a contract has a single portfolio aggregate limit, applying that limit separately to every location can produce an incorrect total. The calculation must follow the contract’s defined order of operations.
Worked Example: Aggregating a Shared Storm Event
A hypothetical wind portfolio has three locations affected by one modeled storm. The gross covered damage at the locations is $1,200,000, $800,000, and $500,000. Each location has a $100,000 deductible and a $1,000,000 per-location limit. Assume the payment at each location is \(\min(\max(X-d,0),u)\), where \(X\) is that location’s gross damage, \(d\) is its deductible, and \(u\) is its limit. Find the event payment for the portfolio.
Method. Apply the specified location-level deductible and limit to each location’s damage, then add the resulting location payments because all three losses belong to the same event.
At location 1, damage after the deductible is \(1{,}200{,}000-100{,}000=1{,}100{,}000\). The per-location limit reduces this payment to \(\min(1{,}100{,}000,1{,}000{,}000)=1{,}000{,}000\).
At location 2, the amount after the deductible is \(800{,}000-100{,}000=700{,}000\), which is below the limit. Its payment is $700,000.
At location 3, the amount after the deductible is \(500{,}000-100{,}000=400{,}000\), also below the limit. Its payment is $400,000.
Check. The event total includes three location payments for one storm, and each payment is no greater than its $1,000,000 limit. The common mistake avoided is applying one $100,000 deductible to the portfolio total: the stated terms impose a separate deductible at each location.
Building Annual Portfolio Losses
For annual simulation, keep event totals within each simulated year and then add those event totals to obtain the annual aggregate. A year with no modeled events has annual aggregate loss zero under the simulation. For a collection of simulated years with equal weight, the empirical average annual loss is the arithmetic mean of their annual aggregate losses. A different simulation weighting scheme requires using its specified weights instead.
Worked Example: Aggregating Events Across Simulated Years
A hypothetical portfolio has two locations. The following event losses are already on the same net basis and are stated in thousands of dollars. Find each year’s annual aggregate loss and largest event loss, then calculate the equally weighted average annual aggregate loss across the four simulated years.
| Year | Event | Location 1 loss ($000) | Location 2 loss ($000) |
|---|---|---|---|
| 1 | A | 400 | 200 |
| 1 | B | 0 | 300 |
| 2 | C | 100 | 800 |
| 2 | D | 200 | 0 |
| 3 | None | 0 | 0 |
| 4 | E | 700 | 100 |
| 4 | F | 200 | 200 |
Step 1: Calculate each event total. In year 1, event A totals \(400+200=600\) thousand dollars and event B totals \(0+300=300\) thousand dollars. In year 2, event C totals \(100+800=900\) thousand dollars and event D totals \(200+0=200\) thousand dollars. In year 4, event E totals \(700+100=800\) thousand dollars and event F totals \(200+200=400\) thousand dollars.
Step 2: Calculate the annual aggregate and largest event. Year 1 has annual aggregate \(600+300=900\) and largest event 600, both in thousands. Year 2 has annual aggregate \(900+200=1{,}100\) and largest event 900. Year 3 has both amounts equal to zero. Year 4 has annual aggregate \(800+400=1{,}200\) and largest event 800.
Step 3: Average the annual aggregate losses. The four annual totals are 900, 1,100, 0, and 1,200 thousand dollars. Their sum is 3,200 thousand dollars, so the equally weighted average is \(3{,}200/4=800\) thousand dollars, or $800,000.
Check. The largest event in each year is no greater than that year’s annual aggregate. The common mistake avoided is reporting the largest event as the annual aggregate: in year 4, for example, those amounts are $800,000 and $1,200,000, respectively.
Geographic Dependence and the Aggregate Distribution
Location losses are often dependent because the same event can affect multiple locations. A storm’s track, intensity, and footprint can make losses occur together; locations far apart or exposed to different perils may have less shared event exposure. Dependence is about how outcomes occur jointly, not just the separate distribution of loss at each location.
For two location losses \(X_1\) and \(X_2\), the portfolio total is \(S=X_1+X_2\). Its expectation is the sum of the location expectations whether or not the losses are dependent. Its variance, however, includes a covariance term. This distinction explains why knowing each location’s marginal loss distribution does not, by itself, determine the portfolio loss distribution.
To justify the formula, write each loss as its mean plus a deviation: \(X_j=E[X_j]+(X_j-E[X_j])\). The portfolio deviation from its mean is the sum of the two location deviations. Squaring this sum produces the square of each deviation and twice their product. Taking expectations gives the two variances and twice the covariance, because covariance is the expected product of deviations from the means. For more than two locations, the variance has one variance term for each location and twice the covariance for each distinct pair.
Worked Example: Shared Event Versus Independent Local Impacts
Consider two hypothetical locations. Each has a loss of $100,000 if hit during the year and zero otherwise. Each location has probability 0.2000 of being hit. Compare the aggregate loss distribution when a single shared event hits both locations with probability 0.2000, with the distribution when the locations are hit independently with probability 0.2000 each.
Shared event model. With probability 0.2000, both locations are hit and the aggregate is $200,000. With probability \(1-0.2000=0.8000\), neither is hit and the aggregate is zero. Thus the aggregate distribution has probability 0.8000 at zero and 0.2000 at $200,000.
Independent impact model. The probability both are hit is \(0.2000\times0.2000=0.0400\), giving an aggregate of $200,000. The probability only the first is hit is \(0.2000\times0.8000=0.1600\), giving $100,000. The probability only the second is hit is also \(0.8000\times0.2000=0.1600\), also giving $100,000. The probability neither is hit is \(0.8000\times0.8000=0.6400\), giving zero. Combining the two one-location outcomes gives probability 0.3200 at $100,000.
Compare the results. In both models, the expected aggregate is \(0.2000(200{,}000)=\$40{,}000\) for the shared event model and \(0.0400(200{,}000)+0.3200(100{,}000)=\$40{,}000\) for the independent model. But the probability of a $200,000 aggregate is 0.2000 under the shared event model and 0.0400 under the independent model. The marginal hit probabilities and amounts are identical; their joint behavior differs.
Check. Each model has the stated 0.2000 hit probability at each location. The common mistake avoided is assuming that identical location-level probabilities imply an identical portfolio distribution.
Aggregation Checks and a Common Exam Trap
A common mistake is to add location-level tail measures as if they were portfolio tail measures. A location percentile describes that location’s loss distribution; it does not specify whether high losses at different locations happen in the same simulated years. The dependence structure and event footprint determine how the location results combine.
For example, in the preceding example, each location’s 99th percentile loss is $100,000, using the discrete quantile convention \(q_{0.99}=\inf\{x:F(x)\geq0.99\}\). Adding these marginal percentiles gives $200,000. Yet under independent impacts, the aggregate distribution has probability 0.6400 at zero, 0.3200 at $100,000, and 0.0400 at $200,000. Its 99th percentile is $100,000, because its cumulative probability at $100,000 is 0.9600, while it reaches 1.0000 at $200,000—so, applying the stated convention, the 99th percentile is actually $200,000. Under the shared-event model it is also $200,000. This example shows why checking the cumulative probabilities matters: a percentile is not obtained by mechanically adding marginal percentiles, and the correct result must be determined from the aggregate distribution.
For practical validation, confirm that every location contribution is assigned to the intended event and year, that all amounts have compatible units and financial treatment, and that each event total is the sum of the intended location amounts. Then confirm that annual aggregates sum the event totals, while annual largest-event values select a maximum. These checks catch common errors such as double-counting a location, combining losses from distinct events into one event, or confusing annual aggregate loss with OEP loss.
Specify whether amounts are gross or net and identify which policy or reinsurance terms have already been applied.
Use event identifiers to combine affected locations for each event, and preserve the simulated year for annual accumulation.
Sum location amounts for event loss; sum event totals for annual aggregate loss; select the largest event for the annual OEP amount.
Check whether locations share event outcomes and verify the resulting aggregate distribution rather than adding marginal tail measures.
Check Your Understanding
Work each problem by identifying the correct aggregation level and retaining the stated loss basis.
- Three locations have net payments of $180,000, $90,000, and zero from the same event. Assuming no further portfolio-level term, what is the event loss? (Answer: $270,000)
- In one simulated year, event A produces losses of $300,000 and $500,000 at two locations; event B produces $200,000 and zero. Find the annual aggregate and the largest event loss. (Answer: annual aggregate $1,000,000; largest event loss $800,000)
- Three locations have gross losses of $80,000, $40,000, and $10,000. Each has a separate $25,000 deductible and no limit. Find the total payment for one shared event. (Answer: $70,000)
- Two locations each lose $50,000 when hit. Each has hit probability 0.1000. Compare the probability of a $100,000 aggregate if hits always occur together versus if they occur independently. (Answer: 0.1000 together; 0.0100 independently)
- Two location losses have expected values of $20,000 and $30,000, variances of \(9\) and \(16\) million dollars squared, and covariance of \(3\) million dollars squared. Find the expected portfolio loss and its variance. (Answer: expected loss $50,000; variance \(31\) million dollars squared)