Fixed Claim Counts Versus Random Claim Counts

Why the Claim-Count Assumption Matters

Aggregate loss varies because individual claim amounts vary. It can also vary because the number of claims varies. These are distinct sources of uncertainty, and a variance calculation that treats a random claim count as fixed omits an important component.

To isolate the effect of claim-count uncertainty, compare two models with the same expected claim count and the same claim severity distribution. In one model, the count is fixed. In the other, the count is random. The expected aggregate losses agree, but the random-count model generally has greater variance whenever claim amounts have a nonzero mean and the count itself has positive variance.

Definition: Let $N$ be a nonnegative integer-valued claim count. Let $X_1,X_2,\ldots$ be claim severities, and define aggregate loss by $S_N=\sum_{j=1}^{N}X_j$, with $S_N=0$ when $N=0$. In the standard random-sum model, the severities are independent and identically distributed, are independent of $N$, and have finite mean $E[X]$ and variance $\operatorname{Var}(X)$.

Under a fixed count of $n$, aggregate loss is $S_n=X_1+\cdots+X_n$. Under a random count, the number of terms in the sum is uncertain. Throughout the comparison, the severity distribution is held constant: the claim amount is not assumed to change just because the count changes.

Variance With a Fixed Claim Count

If exactly $n$ claims occur, and their severities are independent with common variance $\operatorname{Var}(X)$, then the variance of their sum is the sum of the individual variances:

E[S_n]=nE[X], \qquad \operatorname{Var}(S_n)=n\operatorname{Var}(X).

This is the fixed-count benchmark. It includes severity variation but has no frequency variation: the claim count cannot move above or below $n$.

Worked Example: A Two-Point Random Claim Count

A hypothetical coverage has a claim count $N$ equal to 1 or 3, each with probability $0.5$. Claim severities are independent of the count and of one another, with $E[X]=\$1{,}000$ and $\operatorname{Var}(X)=\$400{,}000$. Find the random-count aggregate mean and variance, then compare the variance with a fixed count equal to the expected count.

Step 1: Calculate the moments of the count.

$$ E[N]=1(0.5)+3(0.5)=2. $$

The second moment is $E[N^2]=1^2(0.5)+3^2(0.5)=0.5+4.5=5$. Therefore,

\operatorname{Var}(N)=E[N^2]-(E[N])^2=5-2^2=1.

Step 2: Find the aggregate mean. The expected count is 2, so the expected aggregate loss is $2(\$1{,}000)=\$2{,}000$.

Step 3: Find the random-count variance. The severity contribution is $E[N]\operatorname{Var}(X)=2(\$400{,}000)=\$800{,}000$. The frequency contribution is $\operatorname{Var}(N)(E[X])^2=1(\$1{,}000)^2=\$1{,}000{,}000$. Thus the random-count aggregate variance is $\$800{,}000+\$1{,}000{,}000=\$1{,}800{,}000$.

Step 4: Compare with a fixed count of 2. The fixed-count variance is $2(\$400{,}000)=\$800{,}000$. The random-count variance exceeds it by $\$1{,}800{,}000-\$800{,}000=\$1{,}000{,}000$, exactly the frequency contribution. The expected aggregate loss is $\$2{,}000$ under either model.

Method and check. The count moments were calculated from the stated two-point distribution before applying the random-sum formula. The result is checked by subtracting the fixed-count variance: the difference equals $\operatorname{Var}(N)(E[X])^2=\$1{,}000{,}000$. A common mistake would be to use only the fixed-count formula and report $\$800{,}000$ for both models.

Deriving the Random-Sum Variance

Condition first on the observed claim count. Given $N=n$, the sum contains $n$ independent severities, so its conditional mean is $nE[X]$ and its conditional variance is $n\operatorname{Var}(X)$. The law of total variance says that total variance is the expected conditional variance plus the variance of the conditional mean:

\begin{aligned} \operatorname{Var}(S_N) &=E[\operatorname{Var}(S_N\mid N)] +\operatorname{Var}(E[S_N\mid N])\\ &=E[N\operatorname{Var}(X)] +\operatorname{Var}(NE[X])\\ &=E[N]\operatorname{Var}(X) +\operatorname{Var}(N)(E[X])^2. \end{aligned}

The first term is the average severity variance across the claims that occur. The second is the variance in the conditional aggregate mean as the count changes. Because $E[X]$ is the expected loss per claim, a fluctuation of one claim changes the conditional mean by $E[X]$; squaring that amount gives the $(E[X])^2$ factor.

Theorem: Under the independent, identically distributed severity assumptions above, with finite $E[N]$, $\operatorname{Var}(N)$, $E[X]$, and $\operatorname{Var}(X)$, $$ \operatorname{Var}(S_N) =E[N]\operatorname{Var}(X)+\operatorname{Var}(N)(E[X])^2. $$ If $E[N]=n$ and $n$ is an integer, the fixed-count benchmark has variance $n\operatorname{Var}(X)$. The random-count model's variance exceeds that benchmark by $\operatorname{Var}(N)(E[X])^2$.

The comparison holds the expected count, not the realized count, constant across models. When $E[N]=n$, the expected aggregate loss is $nE[X]$ in both models. The variances differ because a random count adds uncertainty about how many expected claim amounts are included.

Poisson Counts as a Special Case

If $N$ has a Poisson distribution with parameter $\lambda$, then $E[N]=\operatorname{Var}(N)=\lambda$. Substitution gives a particularly useful form:

\operatorname{Var}(S_N) =\lambda\operatorname{Var}(X)+\lambda(E[X])^2 =\lambda E[X^2].

The last equality follows from $E[X^2]=\operatorname{Var}(X)+(E[X])^2$. This result applies to a Poisson count independent of iid severities; it is not the fixed-count variance.

Worked Example: Poisson Frequency Compared With a Fixed Count

A hypothetical portfolio has a Poisson claim count with parameter $\lambda=5$. Claim severities are independent of the count and of one another, with $E[X]=\$800$ and $\operatorname{Var}(X)=\$360{,}000$. Calculate the aggregate mean and variance, then compare with exactly five claims.

Step 1: Identify the count moments. For a Poisson count, $E[N]=\operatorname{Var}(N)=\lambda=5$.

Step 2: Calculate the expected aggregate loss.

E[S_N]=E[N]E[X]=5(\$800)=\$4{,}000.

Step 3: Calculate the random-count variance. The severity contribution is $5(\$360{,}000)=\$1{,}800{,}000$. The frequency contribution is $5(\$800)^2=5(\$640{,}000)=\$3{,}200{,}000$. Therefore,

\operatorname{Var}(S_N) =\$1{,}800{,}000+\$3{,}200{,}000 =\$5{,}000{,}000.

Step 4: Compare with a fixed count of five. Its aggregate variance is $5(\$360{,}000)=\$1{,}800{,}000$. The random-count variance is greater by $\$3{,}200{,}000$, the frequency contribution. Both models have expected aggregate loss $\$4{,}000$.

Check and common mistake. The second severity moment is $\$360{,}000+(\$800)^2=\$1{,}000{,}000$, so the Poisson form gives $5(\$1{,}000{,}000)=\$5{,}000{,}000$, agreeing with the two-term calculation. Using only $5\operatorname{Var}(X)$ would incorrectly treat the Poisson count as fixed and omit $\$3{,}200{,}000$.

Checking Count Distributions Before Using Them

For a count distribution given as a probability table, calculate its moments from that table rather than relying on an asserted variance that may not match the probabilities. In particular, compute $E[N]$, then $E[N^2]$, and use $\operatorname{Var}(N)=E[N^2]-(E[N])^2$. This check matters because the frequency contribution depends directly on $\operatorname{Var}(N)$.

Worked Example: Verify the Count Variance Before Comparing

A hypothetical risk has $N=0,2,4$ with probabilities $0.25,0.50,0.25$, respectively. Severities are iid and independent of $N$, with $E[X]=\$1{,}200$ and $\operatorname{Var}(X)=\$160{,}000$. Find the random-count aggregate variance and compare it with a fixed count equal to the expected count.

Step 1: Verify the probabilities and mean. The probabilities sum to $0.25+0.50+0.25=1$, and

$$ E[N]=0(0.25)+2(0.50)+4(0.25)=0+1+1=2. $$

Step 2: Calculate the variance from the distribution.

E[N^2]=0^2(0.25)+2^2(0.50)+4^2(0.25)=0+2+4=6, \qquad \operatorname{Var}(N)=6-2^2=2.

Step 3: Apply the random-sum formula. The severity contribution is $2(\$160{,}000)=\$320{,}000$. The frequency contribution is $2(\$1{,}200)^2=2(\$1{,}440{,}000)=\$2{,}880{,}000$. Thus,

\operatorname{Var}(S_N)=\$320{,}000+\$2{,}880{,}000=\$3{,}200{,}000.

Step 4: Compare and check. A fixed count of 2 has variance $\$320{,}000$, so the difference is $\$3{,}200{,}000-\$320{,}000=\$2{,}880{,}000$. This equals $\operatorname{Var}(N)(E[X])^2$, as required. The expected aggregate loss is $2(\$1{,}200)=\$2{,}400$ under either model.

Method and common mistake. The variance was derived from the actual probability table. A common error is to use a stated or guessed count variance without checking it against the distribution; that changes the frequency contribution and therefore the aggregate variance.

Assumptions, Interpretation, and Exam Traps

The extra variance term is nonnegative. It is zero if the count is fixed, or if $E[X]=0$. For ordinary nonnegative insurance claim amounts with a positive mean, any genuinely random count with positive variance adds aggregate-loss variance relative to the fixed-count benchmark with the same expected count.

The standard formula depends on the model assumptions. If claim severities are dependent on the count, or if severities are not identically distributed, then the conditional mean and variance given $N$ may not equal $NE[X]$ and $N\operatorname{Var}(X)$. The general law of total variance still holds, but its conditional terms must be calculated from the actual model. Do not apply the iid independent-severity formula automatically in those settings.

Omitting frequency variance: $E[N]\operatorname{Var}(X)$ alone is the fixed-count-style contribution, not the full random-sum variance.
Using the wrong count moment: the additional term uses $\operatorname{Var}(N)$, not $E[N]$, except when the count is Poisson, for which those moments happen to be equal.
Confusing severity variance with severity second moment: the random-count formula uses $\operatorname{Var}(X)$ in its first term and $(E[X])^2$ in its second. For Poisson frequency, these combine to $E[X^2]$.
Comparing different expected counts: to isolate the effect of random frequency, compare models with the same $E[N]$ and the same severity distribution.
Skipping distribution checks: for a discrete $N$, verify that probabilities sum to 1 and calculate its moments consistently before finding aggregate variance.

The distinction is useful in credibility calculations because aggregate-loss fluctuation depends on both claim size and claim count. A fixed-count calculation can understate variability when the actual process has random frequency. The additional term quantifies precisely how much variance is attributable to that frequency randomness under the stated model, a distinction aligned with the CAS Exam 8 (Advanced Ratemaking) syllabus.

Check Your Understanding

For each problem, identify the count moments and separate the severity and frequency contributions before comparing models.

A hypothetical program has exactly 4 claims, each with independent severity mean $\$500$ and variance $\$90{,}000$. Find the aggregate mean and variance. (Answer: mean $\$2{,}000$; variance $\$360{,}000$.)
A random claim count takes values 0, 2, and 4 with probabilities 0.25, 0.50, and 0.25. Independent severities have mean $\$1{,}200$ and variance $\$160{,}000$. Find the count variance, random-count aggregate variance, and the difference from a fixed count of 2. (Answer: $\operatorname{Var}(N)=2$; random-count variance $\$3{,}200{,}000$; difference $\$2{,}880{,}000$.)
A hypothetical portfolio has a Poisson claim count with parameter 3. Independent severities have mean $\$900$ and variance $\$190{,}000$. Calculate the random-count aggregate variance and compare it with exactly 3 claims. (Answer: random-count variance $\$3{,}000{,}000$; fixed-count variance $\$570{,}000$; difference $\$2{,}430{,}000$.)
A claim count is 1 or 5, each with probability 0.5. Independent severities have mean $\$700$ and variance $\$250{,}000$. Calculate $E[N]$, $\operatorname{Var}(N)$, the random-count aggregate variance, and its excess over the fixed-count variance at $E[N]$. (Answer: $E[N]=3$; $\operatorname{Var}(N)=4$; random-count variance $\$2{,}710{,}000$; fixed-count variance $\$750{,}000$; excess $\$1{,}960{,}000$.)

What You'll Learn