Common Errors In Random-Sum Calculations

Why Random-Sum Calculations Need an Assumption Check

Aggregate loss combines two sources of uncertainty: how many claims occur and how large each claim is. Many calculation errors arise from treating these sources as if one could be ignored, or from applying a formula without checking its assumptions. A useful first step is to write down the model explicitly before calculating.

Let $N$ be a nonnegative integer-valued claim count, let $X_1,X_2,\ldots$ denote claim severities, and define aggregate loss by $S=\sum_{j=1}^{N}X_j$, with $S=0$ when $N=0$. The standard formulas below require that $N$ is independent of the severities and that the severities are independent and identically distributed, with finite mean $\mu=E[X_1]$ and variance $\sigma^2=\operatorname{Var}(X_1)$.

Definition: For a random sum $S=\sum_{j=1}^{N}X_j$, if $N$ is independent of an iid severity sequence with mean $\mu$ and variance $\sigma^2$, then $$ E[S]=E[N]\mu,\qquad \operatorname{Var}(S)=E[N]\sigma^2+\operatorname{Var}(N)\mu^2. $$ These formulas require the stated independence and identical-distribution assumptions, as well as finite moments. If severities differ by claim position or depend on $N$, use conditioning on $N$ with the actual conditional severity model instead.

Deriving the Mean and Variance Formulas

The formulas follow from conditioning on the claim count. Given $N=n$, the aggregate is a sum of $n$ iid severities. Therefore its conditional mean is $n\mu$, and its conditional variance is $n\sigma^2$. Applying the laws of total expectation and total variance gives:

E[S]=E\big[E[S\mid N]\big]=E[N\mu]=E[N]\mu,

For the variance, the law of total variance separates the average within-count severity variation from the variation in the conditional mean:

\begin{aligned} \operatorname{Var}(S) &=E[\operatorname{Var}(S\mid N)] +\operatorname{Var}(E[S\mid N])\\ &=E[N\sigma^2]+\operatorname{Var}(N\mu)\\ &=E[N]\sigma^2+\operatorname{Var}(N)\mu^2. \end{aligned}

The first term is the severity contribution: even at a fixed claim count, severities can vary. The second is the frequency contribution: the count varies, changing the conditional aggregate mean. Notice that the frequency term uses $\operatorname{Var}(N)$, not $E[N^2]$.

Useful check: Since $E[N^2]=\operatorname{Var}(N)+(E[N])^2$, the second moment of the count is generally larger than its variance. Substituting $E[N^2]$ for $\operatorname{Var}(N)$ adds an erroneous extra term $(E[N])^2\mu^2$ to the aggregate variance.

Worked Example: Using the Count Variance, Not Its Second Moment

A hypothetical portfolio has $N\sim\operatorname{Poisson}(2)$, independent of iid severities that equal \$10 or \$30 with equal probabilities. Find $E[X_1]$, $\operatorname{Var}(X_1)$, $E[S]$, and $\operatorname{Var}(S)$. Then find the variance produced by the specific error of substituting $E[N^2]$ for $\operatorname{Var}(N)$ in the correct random-sum variance formula.

Method and assumptions. The count is Poisson with parameter $2$, so $E[N]=\operatorname{Var}(N)=2$. The severities are iid and independent of $N$, so the standard random-sum formulas apply.

Severity mean. The two outcomes have probabilities $1/2$ each: $$ E[X_1]=\tfrac12(10)+\tfrac12(30)=20. $$ Thus the mean severity is \$20.

Severity variance. First calculate the second moment: $$ E[X_1^2]=\tfrac12(10^2)+\tfrac12(30^2) =\tfrac12(100)+\tfrac12(900)=500. $$ Therefore $\operatorname{Var}(X_1)=500-20^2=500-400=100$, or \$100 squared.

Aggregate mean and variance. $$ E[S]=2(20)=40, $$ so the aggregate mean is \$40. Also, $$ \operatorname{Var}(S)=2(100)+2(20^2)=200+800=1{,}000. $$ The aggregate variance is $1{,}000$ dollars squared.

Incorrect substitution and correction. For a Poisson count, $E[N^2]=\operatorname{Var}(N)+(E[N])^2=2+2^2=6$. If 6 is incorrectly substituted for $\operatorname{Var}(N)$, the resulting calculation is $$ 2(100)+6(20^2)=200+6(400)=200+2{,}400=2{,}600. $$ Thus the erroneous result is \$2,600 in squared-dollar units, not the correct \$1,000. The mistake is using the count's second moment where the variance is required.

Check. The correct aggregate variance is $1{,}000$, and the incorrect calculation exceeds it by $2{,}600-1{,}000=1{,}600=2^2(20^2)$, exactly the spurious $(E[N])^2\mu^2$ term.

The Zero-Claim Case Is Part of the Distribution

When $N=0$, the sum is empty and $S=0$. Consequently, the aggregate distribution has at least the probability $P(N=0)$ at zero; if zero severities are possible, there may be additional ways to have $S=0$. A distribution calculated only for positive claim counts is conditional on $N>0$, not the unconditional aggregate distribution. The distinction affects probabilities and expectations.

Worked Example: Keeping the Zero-Count Probability

A hypothetical policy group has $N\sim\operatorname{Binomial}(2,0.5)$. Given a claim, its severity is \$10 or \$20 with equal probability, independently across claims and independently of $N$. Find the unconditional probabilities of $S=0,10,20,30,40$, and check that they sum to one. Also calculate $E[S]$.

Method and assumptions. Use the claim-count probabilities first, then condition on the count. The binomial probabilities are $P(N=0)=0.25$, $P(N=1)=0.50$, and $P(N=2)=0.25$.

If $N=0$, then $S=0$, contributing probability $0.25$. If $N=1$, each severity is equally likely, so this case contributes $0.50(0.5)=0.25$ to each of $S=10$ and $S=20$. If $N=2$, the two severities are independent: sums of \$20, \$30, and \$40 have conditional probabilities $0.25,0.50,0.25$, respectively. Multiplying by $P(N=2)=0.25$, this case contributes $0.0625,0.125,0.0625$.

Combining contributions for each total gives $$ \begin{aligned} P(S=0)&=0.25,\\ P(S=10)&=0.25,\\ P(S=20)&=0.0625+0.25=0.3125,\\ P(S=30)&=0.125,\\ P(S=40)&=0.0625. \end{aligned} $$

Probability check. The probabilities sum to $0.25+0.25+0.3125+0.125+0.0625=1$.

Mean check. The mean severity is $\tfrac12(10)+\tfrac12(20)=15$, and $E[N]=2(0.5)=1$. Hence $E[S]=1(15)=15$. Directly from the distribution, $E[S]=10(0.25)+20(0.3125)+30(0.125)+40(0.0625)=2.5+6.25+3.75+2.5=15$.

Common mistake avoided. Omitting $N=0$ would discard $P(S=0)=0.25$; the remaining probabilities would sum to $0.75$, not one. They describe the positive-count portion without its unconditional probability mass.

A Mean Severity Is Not a Realized Claim

The identity $E[S\mid N=n]=n\mu$ is an expectation, not a statement that the realized aggregate equals $n\mu$. Replacing every random severity by its mean may preserve the aggregate mean, but it removes the severity variation. In the iid model, the lost variance contribution is $E[N]\sigma^2$.

Worked Example: What the Mean-Severity Shortcut Leaves Out

A hypothetical portfolio has $N\sim\operatorname{Binomial}(2,0.5)$, independent of iid severities that are \$10 or \$30 with equal probabilities. Calculate the aggregate mean and variance. Then compare the variance with the shortcut that replaces every severity by its mean.

Method and assumptions. Here $E[N]=2(0.5)=1$ and $\operatorname{Var}(N)=2(0.5)(0.5)=0.5$. The severity mean is $\mu=20$. Since $E[X_1^2]=500$, the severity variance is $\sigma^2=500-400=100$.

The correct aggregate mean is $E[S]=1(20)=20$. Its variance is $$ \operatorname{Var}(S)=1(100)+0.5(20^2)=100+200=300. $$

Under the shortcut, each claim is treated as if it were exactly \$20, giving $S^*=20N$. This shortcut has mean $E[S^*]=20(1)=20$, but its variance is only $$ \operatorname{Var}(S^*)=20^2\operatorname{Var}(N)=400(0.5)=200. $$ It misses the severity contribution $E[N]\sigma^2=100$, so its variance is $300-200=100$ too small.

Check. The shortcut preserves the mean in this iid independent model because $E[X_1]=20$, but it does not preserve the variance because the actual severities are random. The shortcut is valid for the aggregate itself only if each severity is deterministic at \$20, not merely because \$20 is the mean.

Unequal Severities Require a Different Calculation

The iid formula treats claims as exchangeable in severity: each claim has the same distribution. If a claim's severity depends on its position, that assumption fails. For a sequence of independent, potentially nonidentically distributed severities $X_j$, independent of $N$, write $\mu_j=E[X_j]$ and $\sigma_j^2=\operatorname{Var}(X_j)$. Conditional on $N=n$, the aggregate mean is $\sum_{j=1}^n\mu_j$ and the aggregate variance is $\sum_{j=1}^n\sigma_j^2$. Unconditionally,

E[S]=\sum_{j\geq 1}P(N\geq j)\mu_j,\qquad \operatorname{Var}(S)=E\!\left[\sum_{j=1}^N\sigma_j^2\right] +\operatorname{Var}\!\left(\sum_{j=1}^N\mu_j\right).

These expressions account for which severities are included at each possible count. They cannot generally be replaced by $E[N]$ times one average severity and the iid variance formula.

Worked Example: A Random Count with Two Different Claim Amounts

A hypothetical system can produce at most two claims. The first claim, if present, is always \$10; the second, if present, is always \$30. The count is $N\sim\operatorname{Binomial}(2,0.5)$, independent of the claim amounts, and the aggregate includes the first $N$ claims. Find $E[S]$ and $\operatorname{Var}(S)$, and assess the iid shortcut that uses the average of \$10 and \$30 as a common severity.

Method and assumptions. The severities are fixed by position, so they are not identically distributed. The count probabilities are $P(N=0)=0.25$, $P(N=1)=0.50$, and $P(N=2)=0.25$. Thus $S$ equals \$0, \$10, or \$40 with those respective probabilities.

The mean is $$ E[S]=0.25(0)+0.50(10)+0.25(40)=0+5+10=15. $$ The second moment is $$ E[S^2]=0.25(0^2)+0.50(10^2)+0.25(40^2) =0+50+400=450. $$ Therefore $\operatorname{Var}(S)=450-15^2=450-225=225$.

Why the iid shortcut fails. Averaging the two claim amounts gives \$20, and multiplying by $E[N]=1$ predicts an aggregate mean of \$20, not the correct \$15. The first claim is included whenever $N\geq1$, with probability $0.75$, while the second is included only when $N=2$, with probability $0.25$. Their expected contributions are therefore $0.75(10)=7.5$ and $0.25(30)=7.5$, totaling \$15. A single average severity does not capture these different inclusion probabilities.

Check. The distribution gives $E[S]=15$ and variance $225$. Both calculations use the actual claim positions rather than an unsupported iid assumption.

Common Error Checks

Before accepting an aggregate calculation, verify the model and the role of each quantity. These checks are particularly useful when a problem supplies several moments or asks for an unconditional probability.

Write the aggregate definition.
State $S=\sum_{j=1}^{N}X_j$, including $S=0$ when $N=0$.

Check independence and identical distribution.
Confirm whether $N$ is independent of severity and whether all claim severities share one distribution. If not, condition on the count and use the actual severity structure.

Separate count moments.
Use $\operatorname{Var}(N)$ in the standard aggregate variance formula; do not substitute $E[N^2]$.

Check probability mass and moments.
Include the zero-count case, ensure probabilities sum to one, and compare a computed mean or variance with a conditioning calculation when possible.

These distinctions matter because aggregate results feed into pricing, reserving, and risk analysis. A small-looking substitution can materially distort a variance, while omitting the zero-count case can make an otherwise plausible probability table invalid. This error-checking approach is aligned with the SOA Exam FAM (Fundamentals of Actuarial Mathematics) syllabus.

Key takeaway: The iid random-sum formulas are consequences of specific independence and identical-distribution assumptions. Use the count variance, retain the zero-count case, and treat $N\mu$ as a conditional expected aggregate—not as the realized aggregate.

Check Your Understanding

For each problem, identify the model before calculating; answers include the requested final values for checking.

A hypothetical portfolio has $N\sim\operatorname{Poisson}(1.5)$, independent of iid severities that equal \$4 or \$12 with equal probabilities. Find the severity mean and variance, then find the aggregate mean and variance. (Answer: $E[X_1]=\$8$, $\operatorname{Var}(X_1)=16$ dollars squared, $E[S]=\$12$, $\operatorname{Var}(S)=120$ dollars squared.)
A hypothetical group has $N\sim\operatorname{Binomial}(3,0.4)$, and every claim pays exactly \$5. Find $P(S=0)$, $E[S]$, and $\operatorname{Var}(S)$. (Answer: $P(S=0)=0.2160$, $E[S]=\$6$, $\operatorname{Var}(S)=18$ dollars squared.)
Let $N\sim\operatorname{Poisson}(2)$, independent of iid severities equal to \$2 or \$6 with equal probabilities. Find the correct aggregate mean and variance, and the variance from incorrectly using $E[N^2]$ in place of $\operatorname{Var}(N)$. (Answer: $E[S]=\$8$, correct $\operatorname{Var}(S)=40$ dollars squared, incorrect variance $=104$ dollars squared.)
A system has at most two claims: the first pays \$8 and the second pays \$20, and $N\sim\operatorname{Binomial}(2,0.25)$ counts the first $N$ claims. Find $E[S]$ and $\operatorname{Var}(S)$. (Answer: $E[S]=\$4.75$, $\operatorname{Var}(S)=50.4375$ dollars squared.)
Two independent hypothetical claim types have $N_A\sim\operatorname{Poisson}(1)$ claims paying \$3 each and $N_B\sim\operatorname{Poisson}(2)$ claims paying \$4 each; the counts are independent. Find the aggregate mean, variance, and probability of no payment. (Answer: $E[S]=\$11$, $\operatorname{Var}(S)=41$ dollars squared, $P(S=0)=e^{-3}=0.0498$.)

What You'll Learn