Catastrophe Exposure and Geographic Accumulation

Why Catastrophe Exposure Accumulation Matters

A portfolio can contain many policies yet still be highly exposed to a single catastrophe. If a large share of insured property is located in one hazard-prone area, one event may affect many policies at once. Counting policies alone will not reveal this concentration: a few high-value buildings can represent more exposure than many small risks, and nearby locations may share the same hazard.

Catastrophe exposure analysis therefore begins with a structured view of the insured portfolio. Location, insured value, construction, and occupancy describe different dimensions of accumulation. Together, they help an actuary assess how much property could be affected by a common hazard and why the resulting claim liabilities may not behave like independent claims.

Organizing Exposure by Location and Risk Attributes

A useful schedule should identify each insured location at a level fine enough to distinguish meaningful hazard areas. Depending on the purpose of the analysis, location data might be summarized by address, postal area, county, or a defined geographic zone. The finer the geographic detail, the more useful it may be for identifying local concentrations—but only if the location data are accurate.

For each location, record the insured values relevant to the coverage being assessed. A building value, contents value, and business-interruption value may each contribute to the exposure, but they should not be combined blindly if coverage scope, limits, or policy terms differ. Construction and occupancy add context: construction can affect vulnerability, while occupancy can be associated with different property contents or business-interruption exposure. These categories do not, by themselves, determine the amount of loss.

A first summary is the total insured value (TIV) within each group. For geographic zone \(j\), let \(T_j\) be the sum of the relevant insured values in that zone, and let \(T=\sum_j T_j\) be the portfolio total. The zone's exposure share is \(s_j=T_j/T\). These shares describe how the portfolio's insured value is distributed; they do not describe the probability or severity of a catastrophe.

Concentration Measures and Their Limits

The Herfindahl index summarizes how unevenly exposure is distributed across the groups being measured. If there are \(n\) zones with positive exposure, the index is \(H=\sum_{j=1}^{n}s_j^2\). A portfolio spread evenly across \(n\) zones has \(H=1/n\); a portfolio concentrated entirely in one zone has \(H=1\). The reciprocal, \(1/H\), expresses the concentration as the number of equally sized groups that would produce the same index.

The index is useful for comparing distributions only when the grouping is reasonably consistent. Splitting one zone into many smaller zones can change the index even if the physical exposure has not changed. It also treats the defined groups as separate categories; it does not measure how close two zones are or whether they share a hazard footprint. A zone-level index is a portfolio summary, not a substitute for examining individual locations and their geography.

Construction and occupancy can be summarized in the same way. For example, the share of TIV in wood-frame buildings is the total insured value in that class divided by portfolio TIV. A cross-tabulation can show the joint distribution—such as wood-frame residential property—rather than reporting each attribute separately. Separate totals can conceal an important joint concentration if a large share of one construction type is also in one occupancy or geographic area.

Why Accumulated Exposures Create Correlated Liabilities

If risks are exposed to the same event, their losses may rise or fall together. This is a different situation from assuming that each location's loss is independent of every other location's loss. A common event can affect multiple locations, so uncertainty in whether the event occurs creates dependence in their liabilities.

A simple model makes the mechanism explicit. Let \(Z\) be an indicator of a hypothetical catastrophe: \(Z=1\) if it occurs and \(Z=0\) otherwise, with \(P(Z=1)=p\). Suppose two locations have losses \(L_1=aZ\) and \(L_2=bZ\), where \(a\) and \(b\) are fixed loss amounts if the event occurs. This deliberately simplified model assumes no loss in the absence of the event and fixed losses conditional on its occurrence.

To see why the covariance formula holds, use the definition \(\operatorname{Cov}(L_1,L_2)=E[L_1L_2]-E[L_1]E[L_2]\). Since \(Z^2=Z\), \(L_1L_2=abZ\), so \(E[L_1L_2]=abp\). Also \(E[L_1]E[L_2]=(ap)(bp)=abp^2\). Subtracting gives \(abp(1-p)\). For positive \(a\), \(b\), and \(0<p<1\), this covariance is positive.

For aggregate liability \(L=L_1+L_2\), the variance identity is \(\operatorname{Var}(L)=\operatorname{Var}(L_1)+\operatorname{Var}(L_2)+2\operatorname{Cov}(L_1,L_2)\). Omitting the covariance term would understate aggregate variance in this shared-event model. The model is not a claim that all catastrophe losses are fixed or perfectly proportional: actual losses can vary by location, hazard intensity, vulnerability, and policy terms.

Practical Interpretation and Common Traps

Exposure concentration measures are most useful when they are interpreted alongside the data definitions and assumptions used to produce them. A high TIV share in one area signals a possible accumulation, but not the amount of loss from any particular event. That amount depends on the hazard, which locations are affected, vulnerability, coverage terms, and other factors. Conversely, a low share in a broad region does not guarantee low local concentration if the region contains a dense cluster of high-value properties.

A common exam trap is to use a policy count where the question requires insured-value weighting. Suppose one zone has one policy worth $10 million and another has nine policies worth $1 million each. The first zone has 10% of the policies but \(10/19=52.63\%\) of the insured value. A count-based comparison would therefore give a very different impression of accumulation.

Another frequent error is to omit the covariance term when aggregating losses exposed to one event. In the worked example, adding only the individual variances gives \(0.6664\), not the common-event variance of \(1.2544\) million-dollars-squared. The omitted amount is \(2\operatorname{Cov}(L_1,L_2)=0.5880\). Independent-loss aggregation is appropriate only when independence is a justified assumption for the quantities being combined.

A further trap is treating a concentration index as a universal risk score. Its value depends on the grouping chosen, and it does not account for proximity between groups, hazard intensity, or differences in vulnerability. Use it to describe a distribution, then inspect the underlying location and attribute data. Missing or stale addresses, inconsistent construction codes, and unclear insured-value definitions can undermine even correctly executed calculations.