Deriving the Mean of Compound Aggregate Loss

Why the Aggregate-Loss Mean Separates into Two Parts

A portfolio’s aggregate loss depends on two distinct quantities: how many claims occur and how large each claim is. Separating these components makes the expected loss easier to calculate and helps prevent a common error: using the right severity mean with a claim count or time period that does not match the question.

Let $N$ be the number of claims in a specified period. Let $X_j$ be the severity, or loss amount, of claim $j$. The aggregate loss $S$ is the sum of the claim severities; if no claims occur, the sum is zero. The standard compound frequency-severity model assumes the severities are identically distributed and independent of the claim count.

Definition: In a compound frequency-severity model, $N$ is a nonnegative integer-valued claim count, $X_1,X_2,\ldots$ are claim severities, and aggregate loss is $$ S=\sum_{j=1}^{N}X_j, $$ where $S=0$ when $N=0$. Under the standard model, the severities are identically distributed, have finite mean $E[X]$, and are independent of $N$. The expected aggregate loss is $$ E[S]=E[N]E[X]. $$

The product has a direct interpretation: expected number of claims multiplied by expected loss per claim. It is not generally the product of a claim probability and a severity mean; the frequency component must be the expected count for the period being modeled. The formula also requires that the expected severity per claim does not change with the claim count.

Deriving the Formula by Conditioning on the Claim Count

Conditioning on $N$ means first considering the aggregate loss when the claim count is fixed, then averaging over the possible values of that count. If $N=n$, the aggregate loss is $X_1+\cdots+X_n$. By linearity of expectation, its conditional mean is $nE[X]$, because each severity has the same mean and is independent of the count.

E[S\mid N=n] =E\left[\sum_{j=1}^{n}X_j\ \middle|\ N=n\right] =\sum_{j=1}^{n}E[X_j\mid N=n] =nE[X].

For $n=0$, the sum is empty and the conditional mean is zero, which also agrees with $nE[X]$. The law of total expectation now averages these conditional means over all possible claim counts:

\begin{aligned} E[S] &=E\big[E[S\mid N]\big]\\ &=E[NE[X]]\\ &=E[X]E[N]. \end{aligned}

In the last step, $E[X]$ is a constant: the common severity mean does not depend on the realized value of $N$. The derivation needs finite means so these expectations are defined. It does not require the severities to be continuous; the result applies to discrete, continuous, or mixed severity distributions.

Key condition: The product $E[N]E[X]$ applies when the expected severity per claim is the same for each possible claim count. Independence of the severity sequence from $N$, together with identically distributed severities, is a standard sufficient assumption. If severity depends on $N$, use $E[S]=E[E[S\mid N]]$ and calculate the conditional means directly.

Worked Applications

Worked Example: A Binomial Claim Count and Mixed Severity

A hypothetical insurer covers 10 independent pieces of equipment for one year. Each piece has probability $0.12$ of producing one claim during the year, and no piece can produce more than one claim. Claim occurrence is independent of severity. Each claim is for $ \$2{,}000$ with probability $0.70$ or $ \$7{,}000$ with probability $0.30$. Find the expected annual aggregate loss.

Method and assumptions. The claim count is $N\sim\operatorname{Binomial}(10,0.12)$. The severities are identically distributed and independent of $N$, so the compound mean formula applies.

Step 1: Find the expected claim count. For a binomial count with 10 trials and claim probability $0.12$, $$ E[N]=10(0.12)=1.2\text{ claims}. $$

Step 2: Find the mean severity. Weight each possible claim amount by its probability: $$ E[X]=(0.70)(\$2{,}000)+(0.30)(\$7{,}000) =\$1{,}400+\$2{,}100 =\$3{,}500\text{ per claim}. $$

Step 3: Calculate the expected aggregate loss. $$ E[S]=(1.2\text{ claims})(\$3{,}500\text{ per claim})=\$4{,}200. $$

Check. The units cancel from claims times dollars per claim to dollars per year. The expected annual aggregate loss is $ \$4{,}200$. A common mistake is to multiply $0.12$ by the severity mean and stop at $ \$420$; that is the expected loss per covered piece, not for all 10 pieces.

Worked Example: Converting a Quarterly Poisson Rate to an Annual Mean

A hypothetical commercial portfolio has a Poisson claim count with mean $0.75$ claims per quarter. Quarterly claim counts are independent, and each claim’s severity is independent of frequency and exponentially distributed with mean $ \$8{,}000$. Find the expected aggregate loss over one year.

Method and assumptions. The annual count is the sum of four quarterly Poisson counts. The sum of independent Poisson counts is Poisson, with mean equal to the sum of the quarterly means. Severity is independent of the count, and the stated exponential mean is already the expected loss per claim.

Step 1: Put the frequency on an annual basis. $$ E[N_{\text{year}}]=4(0.75)=3\text{ claims per year}. $$

Step 2: Record the severity mean. The exponential severity distribution has mean $E[X]=\$8{,}000$ per claim, as given.

Step 3: Apply the compound mean formula. $$ E[S_{\text{year}}]=(3\text{ claims per year})(\$8{,}000\text{ per claim}) =\$24{,}000\text{ per year}. $$

Check. The result is an annual expected loss. A common mistake is to multiply the quarterly count mean $0.75$ by the annual severity mean and report $ \$6{,}000$ as the annual result. That calculation uses a quarterly frequency with a one-year target period; the frequency must first be converted to the year.

Worked Example: Two Claim Types with Different Severity Means

A hypothetical insurer has two types of claims during a year. Type A claims have a Poisson count with mean $1.2$ and mean severity $ \$9{,}000$. Type B claims have a binomial count with 6 trials and probability $0.10$ per trial, and mean severity $ \$3{,}000$. Within each type, severities are identically distributed and independent of that type’s count. Find the expected total annual loss.

Method and assumptions. The two types have different severity means, so there is no single common severity mean to multiply by the combined count without first weighting by the expected numbers of claims. By linearity of expectation, calculate each type’s expected aggregate loss and add them. Dependence between the two type counts would not change this expectation calculation.

Step 1: Find the expected counts. $$ E[N_A]=1.2,\qquad E[N_B]=6(0.10)=0.6. $$

Step 2: Find each type’s expected aggregate loss. $$ E[S_A]=(1.2)(\$9{,}000)=\$10{,}800, \qquad E[S_B]=(0.6)(\$3{,}000)=\$1{,}800. $$

Step 3: Add the type-specific means. $$ E[S_A+S_B]=\$10{,}800+\$1{,}800=\$12{,}600. $$

Check. The result is $ \$12{,}600$ per year. A tempting but incorrect shortcut is to average $ \$9{,}000$ and $ \$3{,}000$ equally, then multiply by the total expected count $1.8$, giving $ \$10{,}800$. That treats the two types as equally represented; their expected counts are $1.2$ and $0.6$, not equal.

When the Product Formula Needs Care

A practical calculation should identify the modeled period, the frequency measure for that period, and the severity mean on the same loss basis. For example, if frequency is stated per quarter but the requested aggregate loss is annual, convert the expected count to an annual amount before multiplying. Likewise, do not mix a severity mean for one claim type with a total claim count that includes other types unless the severity mean is appropriately weighted.

If the severity distribution changes with the claim count, then $E[X_j\mid N]$ need not equal a constant. In that case the conditional-expectation step remains valid, but the simplification to $E[N]E[X]$ may not. For example, if larger claim counts are associated with larger mean severities, calculate $E[S\mid N=n]$ for each count value and average those conditional means using the probabilities of the counts.

Match the period.
Write the period for the expected claim count and for the requested aggregate loss. Convert the count mean if the periods differ.

Identify the severity basis.
Confirm that the severity mean applies to the claims included in the count. For multiple claim types, keep type-specific means and counts aligned.

Check the model condition.
Use $E[S]=E[N]E[X]$ when the severity mean per claim is constant across claim counts. Otherwise, use conditional means and total expectation.

Check the units.
Expected claims multiplied by dollars per claim gives dollars of expected aggregate loss for the stated period.

The aggregate-loss mean is useful as a baseline for pricing, budgeting, and comparing modeled expected losses across portfolios. It describes the center of the aggregate-loss distribution, not the likelihood of a particular loss or the size of an unusually adverse outcome. A mean alone does not describe the distribution’s spread or tail; those require additional measures.

Common exam trap: Multiplying the wrong frequency by the right severity mean can produce a plausible-looking but incorrectly scaled answer. State the frequency period, severity units, and independence or constant-mean condition before applying the product.

Check Your Understanding

For each problem, identify the expected claim count and severity mean before calculating the expected aggregate loss. Assume all stated severities are independent of frequency unless the problem says otherwise.

A hypothetical warranty portfolio has 15 covered devices, each with probability $0.04$ of one claim during the year. A claim is $ \$2{,}000$ with probability $0.70$ and $ \$7{,}000$ otherwise. Find expected annual aggregate loss. (Answer: $ \$2{,}100$)
A hypothetical property portfolio has a Poisson claim rate of $0.25$ per month and mean severity $ \$8{,}000$. Find expected aggregate loss over 12 months. (Answer: $ \$24{,}000$)
A portfolio has two claim types. Type A has a Poisson count with mean $0.4$ and mean severity $ \$30{,}000$; Type B has a binomial count with 20 trials and probability $0.03$, and mean severity $ \$4{,}000$. Find the combined expected aggregate loss. (Answer: $ \$14{,}400$)
A claim count $N$ equals 0, 1, or 3 with probabilities $0.20$, $0.50$, and $0.30$, respectively. The common severity mean is $ \$5{,}000$, independent of $N$. Find expected aggregate loss. (Answer: $ \$7{,}000$)
A hypothetical model has $P(N=0)=0.25$, $P(N=1)=0.50$, and $P(N=2)=0.25$. Given $N=1$, the mean severity is $ \$4{,}000$; given $N=2$, each claim has mean severity $ \$7{,}000$. Find expected aggregate loss, using conditional means rather than assuming a constant severity mean. (Answer: $ \$5{,}500$)

What You'll Learn