Secondary Perils and Multi-Event Accumulation

Why Secondary Perils Require an Annual View

A catastrophe portfolio can be affected by a severe, infrequent event and by less severe events that occur repeatedly. Flood, wildfire, severe convective storm, and other secondary perils may each produce losses that are moderate in isolation, yet their annual accumulation can be material. Portfolio decisions therefore need to account for both how large an event can be and how often losses can recur within the year.

The annual view also matters because a recovery arrangement may respond differently to one large event and to several smaller events. An occurrence recovery applies to a defined event, while an annual aggregate recovery depends on the total eligible loss during the year. The attachment and limit can leave losses from repeated events partly or wholly retained after the available recovery is exhausted.

Definition: A secondary peril is a catastrophe peril that is often characterized by relatively frequent events and potentially material aggregate losses, rather than only by rare, extreme events. A multi-event annual loss is the sum of eligible losses from all events in a specified annual period. For events with losses $Z_1,\ldots,Z_N$, the annual gross loss is $L=\sum_{j=1}^{N} Z_j$, with $L=0$ when $N=0$.

The term “secondary” does not mean “unimportant.” A peril’s contribution depends on its frequency, severity, exposure distribution, and interaction with the portfolio’s recovery structure. A sound comparison keeps these elements distinct: frequency describes event counts, severity describes loss conditional on an event, and accumulation describes the annual total.

Frequency-Severity Models for Annual Loss

A standard starting point is a compound Poisson model. Assume that the annual number of events $N$ follows a Poisson distribution with parameter $\lambda$, that event severities $Z_1,Z_2,\ldots$ are independent and identically distributed, and that the severities are independent of $N$. Define $L=\sum_{j=1}^{N}Z_j$. These assumptions are useful for calculation, but they may not capture clustered events, changing conditions, or dependence between event frequency and severity.

Theorem: Under the compound Poisson assumptions, if $E[Z]$ and $E[Z^2]$ are finite, then $E[L]=\lambda E[Z]$ and $\operatorname{Var}(L)=\lambda E[Z^2]$. For independent peril losses $L_1,\ldots,L_m$, the annual total $L_{\mathrm{total}}=\sum_{r=1}^{m}L_r$ has mean $\sum_r E[L_r]$ and variance $\sum_r\operatorname{Var}(L_r)$. Independence is required for the variance sum; without it, covariance terms must be included.

The variance uses the severity second moment, not just the severity variance. In particular, $E[Z^2]=\operatorname{Var}(Z)+(E[Z])^2$. This distinction matters when translating event-level variability into annual portfolio variability.

Why the Compound Poisson Formulas Hold

Condition on the event count. Given $N=n$, the annual loss is a sum of $n$ independent severities, so $E[L\mid N=n]=nE[Z]$ and $\operatorname{Var}(L\mid N=n)=n\operatorname{Var}(Z)$. Taking expectations gives $E[L]=E[N]E[Z]=\lambda E[Z]$. The law of total variance then gives:

\begin{aligned} \operatorname{Var}(L) &=E[\operatorname{Var}(L\mid N)]+\operatorname{Var}(E[L\mid N])\\ &=E[N]\operatorname{Var}(Z)+\operatorname{Var}(N)(E[Z])^2\\ &=\lambda\operatorname{Var}(Z)+\lambda(E[Z])^2\\ &=\lambda E[Z^2]. \end{aligned}

The last equality follows from $E[Z^2]=\operatorname{Var}(Z)+(E[Z])^2$, and a Poisson variable has both mean and variance $\lambda$. This derivation also shows why treating the event count as fixed would generally understate or misstate annual variability.

Worked Example: Combining Primary and Secondary Peril Losses

Consider a hypothetical portfolio with independent annual peril losses, measured in millions of dollars. Primary-peril events follow a Poisson distribution with $\lambda_P=0.2$ events per year and a fixed loss of $10$ per event. Secondary-peril events follow a Poisson distribution with $\lambda_S=2$ events per year and a fixed loss of $1$ per event. Find the total expected annual loss, its variance and standard deviation, and the probability of at least one secondary-peril event.

Method and assumptions. Apply the compound Poisson formulas separately to each peril. Event counts are Poisson; event losses are fixed at the stated amounts; and the two peril totals are independent. All loss figures in the calculations below are in millions of dollars.

For the primary peril, $E[Z_P]=10$ and $E[Z_P^2]=10^2=100$. Therefore:

E[L_P]=0.2(10)=2,\qquad \operatorname{Var}(L_P)=0.2(100)=20.

For the secondary peril, $E[Z_S]=1$ and $E[Z_S^2]=1^2=1$. Therefore:

E[L_S]=2(1)=2,\qquad \operatorname{Var}(L_S)=2(1)=2.

Independence allows the means and variances to be added. Thus $E[L_P+L_S]=2+2=4$ million, and $\operatorname{Var}(L_P+L_S)=20+2=22$ million-squared. The standard deviation is $\sqrt{22}=4.690415\ldots$ million, or $4,690,415.76 to the nearest cent.

For a Poisson count with mean $2$, the probability of at least one secondary-peril event is $1-P(N_S=0)=1-e^{-2}=0.8646647\ldots$, or 0.8647 to four decimals. This probability describes whether an event occurs, not whether the annual secondary-peril loss exceeds a specified threshold.

Check. The expected total of $4 million equals the sum of the two peril means, and the secondary-peril event probability is below 1. A common mistake is to calculate the secondary variance as $2\operatorname{Var}(Z_S)=0$ because severity is fixed. That omits variation in the event count; the correct variance is $\lambda E[Z_S^2]=2$ million-squared.

Repeated Events and Annual Aggregate Recovery

For an annual aggregate layer with attachment $A$ and limit $M$, the recovery on annual eligible loss $L$ is $\min((L-A)_+,M)$, where $(x)_+=\max(x,0)$. The attachment is applied to the annual total, and the limit caps the annual recovery. This formula assumes the stated losses are eligible under the contract and ignores other terms such as exclusions or reinstatements.

Repeated events can make the order of calculation important. Calculate each event’s eligible loss, sum those losses for the year, and then apply the annual attachment and limit. Applying the attachment separately to every event would instead model an occurrence structure, not the annual aggregate layer described here.

Worked Example: Applying an Annual Aggregate Layer After Repeated Events

A hypothetical insurer faces three independent periods in a year. In each period, an event occurs with probability $0.2$, independently of the other periods. Each event produces an eligible loss of $3 million. An annual aggregate recovery attaches at $4 million and has a $5 million limit. Find the expected annual recovery and expected annual net loss.

Method and assumptions. Let $K$ be the number of events. Then $K\sim\operatorname{Binomial}(3,0.2)$, and the gross annual loss is $3K$ million. Apply the aggregate attachment and limit once to that annual total.

The probabilities are $P(K=0)=0.8^3=0.512$, $P(K=1)=3(0.2)(0.8^2)=0.384$, $P(K=2)=3(0.2^2)(0.8)=0.096$, and $P(K=3)=0.2^3=0.008$. They sum to $0.512+0.384+0.096+0.008=1.000$.

If $K=0$, gross loss is $0 million and recovery is $0 million. If $K=1$, gross loss is $3 million, below the $4 million attachment, so recovery is $0 million. If $K=2$, gross loss is $6 million, so recovery is $\min(6-4,5)=2$ million. If $K=3$, gross loss is $9 million, so recovery is $\min(9-4,5)=5$ million.

Weighting those recoveries by their probabilities gives:

E[\text{recovery}] =(0.512)(0)+(0.384)(0)+(0.096)(2)+(0.008)(5) =0.192+0.040=0.232\text{ million}.

Expected gross loss is $E[3K]=3E[K]=3(3)(0.2)=1.8$ million. Hence expected net loss is $1.8-0.232=1.568$ million. As a direct check, the net losses for $K=0,1,2,3$ are respectively $0,3,4,4$ million, giving $(0.512)(0)+(0.384)(3)+(0.096)(4)+(0.008)(4)=1.568$ million.

Common mistake avoided. Applying the $4 million attachment separately to each $3 million event would produce no recovery at all. That is not the contract in this problem: the annual aggregate attachment is applied after adding the eligible losses. The computed recovery is below the $5 million annual limit, as required.

Concentration, Diversification, and Recovery Capacity

Exposure concentration changes the way an event translates into portfolio loss. A concentrated portfolio can have a large loss from one regional event, while a dispersed portfolio may require several independent regional events to reach a comparable annual total. Their expected losses may be equal even though their chances of crossing a recovery attachment, exhausting a limit, or exceeding an internal risk tolerance differ.

A recovery limit is a finite source of loss transfer, not a guarantee that every event will be fully reimbursed. Portfolio decisions should therefore consider gross annual loss, net annual loss, and the conditions under which recovery becomes available. The following comparison isolates the effects of concentration under a simple hypothetical model; it is not a claim that geographic diversification always reduces risk, since dependence between regions can materially change the result.

Worked Example: Equal Expected Loss, Different Aggregate Recoveries

Compare two hypothetical portfolios. In Portfolio A, one concentrated region has a 0.25 annual probability of an event causing an $8 million loss, and otherwise has no loss. In Portfolio B, two dispersed regions each independently have a 0.25 annual probability of an event causing a $4 million loss. An annual aggregate recovery attaches at $5 million and is limited to $4 million. Compare expected gross loss, the probability gross loss exceeds $5 million, and expected recovery.

Method and assumptions. The regional events in Portfolio B are independent. In both portfolios, apply the aggregate layer to the total annual loss; recovery is $\min((L-5)_+,4)$, with all amounts in millions of dollars.

For Portfolio A, expected gross loss is $(0.25)(8)+(0.75)(0)=2$ million. Gross loss exceeds $5 million exactly when the event occurs, so that probability is $0.25$. When it occurs, recovery is $\min(8-5,4)=3$ million. Expected recovery is $(0.25)(3)=0.75$ million, and expected net loss is $2-0.75=1.25$ million.

For Portfolio B, the number of regional events $K$ is binomial with $n=2$ and $p=0.25$. Thus $P(K=0)=0.75^2=0.5625$, $P(K=1)=2(0.25)(0.75)=0.3750$, and $P(K=2)=0.25^2=0.0625$. Gross loss is $0,4,8$ million for $K=0,1,2$, respectively. Its expected value is $(0.5625)(0)+(0.3750)(4)+(0.0625)(8)=1.5+0.5=2$ million. Gross loss exceeds $5 million only when both regions have an event, so the probability is $0.0625$.

With one event, Portfolio B’s $4 million gross loss is below the attachment. With two events, gross loss is $8 million and recovery is $\min(8-5,4)=3$ million. Therefore, expected recovery is $(0.0625)(3)=0.1875$ million, and expected net loss is $2-0.1875=1.8125$ million.

Check. Both portfolios have $2 million expected gross loss, but their probabilities of gross loss exceeding $5 million are 0.25 and 0.0625. Portfolio A has higher expected recovery under this particular attachment because its single event crosses the attachment more often. The comparison is specific to the assumptions and layer: a different attachment, event dependence, or limit could change the result.

Using the Results in Portfolio Decisions

Annual catastrophe analysis supports practical decisions about underwriting, portfolio steering, reinsurance, and capital planning. The key is to connect modeled event loss to the amount and location of exposure, then carry event losses through the annual recovery structure. A peril with modest average severity can still be important if it generates repeated losses, especially when those losses accumulate in the same region or consume a limited annual recovery.

Describe frequency and severity.
Specify the annual event-count model and the conditional loss distribution or scenario severity. State whether count and severity are assumed independent.

Map events to exposed business.
Estimate how each event footprint affects insured values, policy terms, and geographic concentrations. Do not assume equal event severity across portfolios unless the scenario supports it.

Accumulate losses for the year.
Sum eligible event losses before applying an annual aggregate attachment or limit. Include dependence or clustering where the model supports it.

Evaluate gross and net outcomes.
Compare annual loss distributions, recovery utilization, and retained losses. Check whether a proposed portfolio change improves the measure that actually governs the decision, not only expected gross loss.

The compound Poisson calculations are not appropriate without qualification when events are strongly clustered, event rates vary materially over the year, or severity depends on the count or prior losses. In those cases, a more suitable frequency model, a scenario approach, or explicit dependence assumptions may be needed. Likewise, an expected recovery alone does not describe the chance of no recovery, recovery-limit exhaustion, or a large retained annual loss.

Common Exam Traps

Confusing event severity with annual loss. A $1 million event loss is not the same as a $1 million annual loss when multiple events can occur.
Using severity variance in place of the second moment. For a compound Poisson total, $\operatorname{Var}(L)=\lambda E[Z^2]$, which includes variation from event frequency.
Applying an aggregate attachment event by event. First sum eligible losses over the annual period; then apply the annual attachment and limit.
Assuming equal expected loss means equal tail behavior. Concentration, event dependence, and the recovery attachment can produce different exceedance probabilities and net losses.
Ignoring the recovery limit. Once the annual layer is exhausted, further eligible loss is not recovered under that layer.

Key takeaway: Secondary-peril risk is an annual accumulation problem as well as an event-severity problem. Model frequency and severity explicitly, map losses to concentrated exposures, and apply recovery terms to the correct event or annual total before drawing portfolio conclusions.

Check Your Understanding

For each hypothetical problem, show the event-frequency calculation and apply any stated recovery terms to the specified loss total.

A secondary peril has a Poisson annual event count with mean $1.2$, and every event causes a fixed loss of $0.5 million. Find expected annual loss, annual loss variance, and the probability of at least one event. (Answer: $0.60 million; $0.30 million-squared; 0.6988)
Two independent periods each have event probability $0.4$; an event causes an eligible loss of $2 million. An annual aggregate recovery attaches at $1.5 million and is limited to $2.5 million. Find expected recovery. (Answer: $0.64 million)
A peril has a Poisson annual event count with mean $0.5$, and every event causes a fixed loss of $6 million. Find expected annual loss, annual loss variance, and the probability of at least one event. (Answer: $3 million; $18 million-squared; 0.3935)
Portfolio A has one region with a 0.2 probability of a $10 million annual loss and otherwise no loss. Portfolio B has two independent regions, each with a 0.2 probability of a $5 million loss. An annual aggregate recovery attaches at $6 million and is limited to $3 million. For each portfolio, find expected gross loss, the probability gross loss exceeds $6 million, and expected recovery. (Answer: Both expected gross losses are $2 million; exceedance probabilities are 0.2000 and 0.0400; expected recoveries are $0.60 million and $0.12 million, respectively.)

What You'll Learn