Aggregate Tail Probabilities By Conditioning

Why Condition on the Claim Count?

An aggregate loss is a random sum: the number of claims is uncertain, and the amounts of those claims are uncertain as well. To find the probability that the total exceeds a threshold, it is often more efficient to first fix the claim count. For each possible count, we can ask a simpler question about a sum of severities, then weight that conditional probability by the chance of observing that count.

This approach is particularly useful for discrete models. A threshold may make many conditional probabilities immediately zero or one, while a complement can reduce the remaining calculations to just a few severity combinations. The method is aligned with the SOA Exam FAM (Fundamentals of Actuarial Mathematics) syllabus.

Definition: Let $N$ be a nonnegative integer-valued claim count. Let $X_1,X_2,\ldots$ be claim severities, independent and identically distributed, and independent of $N$. The aggregate loss is $S=\sum_{j=1}^{N}X_j$, with $S=0$ when $N=0$. For a threshold $s$, the aggregate tail probability is $P(S>s)$. Conditional on $N=n$, write $S_n=\sum_{j=1}^{n}X_j$, where $S_0=0$.

The Conditioning Formula

The possible values of $N$ divide the event $S>s$ into mutually exclusive cases. In the case $N=n$, the relevant probability is the tail of the sum of $n$ severities. The law of total probability therefore gives the main calculation:

P(S>s)=\sum_{n=0}^{\infty}P(N=n)P(S_n>s).

This formula does not require a particular frequency distribution. For a binomial or other bounded claim count, the sum has only finitely many terms. For an unbounded count such as a Poisson count, the sum is infinite in principle, but the threshold and severity support may make some terms zero or one, or allow a convenient remainder probability.

Important boundary rule: The event is $S>s$, not $S\ge s$. If severities are discrete, a sum equal to $s$ does not count as an exceedance. When using a complement, use $P(S_n>s)=1-P(S_n\le s)$.

For example, if every severity is at least $m>0$, then $S_n\ge nm$. If $nm>s$, the conditional tail is one. If the largest possible severity is $M$ and $nM\le s$, the conditional tail is zero. These bounds can eliminate counts before any detailed probability calculation.

Why the Formula Holds

For each nonnegative integer $n$, the events $\{N=n\}$ are disjoint and together cover all possible values of $N$. Splitting the event $\{S>s\}$ according to these cases gives:

\begin{aligned} P(S>s) &=\sum_{n=0}^{\infty}P(S>s,\ N=n)\\ &=\sum_{n=0}^{\infty}P(N=n)P(S>s\mid N=n)\\ &=\sum_{n=0}^{\infty}P(N=n)P(S_n>s). \end{aligned}

The last equality follows because, given $N=n$, the aggregate is the sum of $n$ severities. The independence assumption ensures that conditioning on the count does not change the severity distribution. If severities depend on the count, the same partition is valid, but $P(S>s\mid N=n)$ must use the appropriate count-conditional severity model; it cannot automatically be calculated from one common severity distribution.

A Practical Calculation Sequence

The main task is usually not applying the conditioning formula, but calculating only the conditional tails that matter. Use the following sequence to keep the work focused and to avoid threshold errors.

Write the frequency and severity assumptions.
State the distribution and parameters of $N$, the severity distribution, and the independence assumptions.

Fix the event boundary.
Translate the question exactly. For $P(S>s)$, outcomes equal to $s$ are excluded.

Find the relevant counts.
Use the minimum and maximum possible severity to identify counts whose conditional tails are zero or one.

Calculate the remaining conditional tails.
Enumerate severity combinations when manageable, or calculate the complement $P(S_n\le s)$ when that has fewer cases.

Weight and check.
Multiply each conditional tail by $P(N=n)$, add the contributions, and verify that the result lies between zero and one.

Worked Examples

Worked Example: A Binomial Count and a Strict Threshold

A hypothetical portfolio has $N\sim\operatorname{Binomial}(2,0.5)$. Given the count, each claim severity is independent, with $P(X_j=\$2)=0.5$ and $P(X_j=\$5)=0.5$. The count and severities are independent. Calculate $P(S>\$5)$. State the method and identify a common mistake it avoids.

Method. Condition on $N$, and calculate the severity-sum tail for each possible count. The main trap is to treat one claim of $\$5$ as exceeding $\$5$; it equals the threshold and is not an exceedance.

The count probabilities are $P(N=0)=0.25$, $P(N=1)=2(0.5)(0.5)=0.50$, and $P(N=2)=(0.5)^2=0.25$. With zero claims, $S=0$, so $P(S>\$5\mid N=0)=0$. With one claim, the severity is either $\$2$ or $\$5$, neither of which exceeds $\$5$, so $P(S>\$5\mid N=1)=0$.

With two claims, the four equally likely ordered pairs of severities are $(2,2)$, $(2,5)$, $(5,2)$, and $(5,5)$, in dollars. Their aggregate amounts are $\$4,\$7,\$7,\$10$. Three of the four totals exceed $\$5$, giving $P(S>\$5\mid N=2)=3/4$. Thus:

P(S>\$5)=0.25(0)+0.50(0)+0.25\left(\frac{3}{4}\right)=0.1875.

The answer is $0.1875$, or $18.75\%$. As a check, only the two-claim case can exceed the threshold, so the answer cannot be greater than $P(N=2)=0.25$; $0.1875$ satisfies that bound.

Worked Example: Poisson Frequency and a Complement

A hypothetical small portfolio has $N\sim\operatorname{Poisson}(1)$. Each independent claim severity is $\$1$ with probability $0.75$ and $\$3$ with probability $0.25$, independently of $N$. Find $P(S>\$3)$, giving the probability to four decimal places.

Method. Condition on $N$. For the two- and three-claim cases, use the complement of the event that the sum is at most $\$3$. This avoids listing every ordered combination. A common mistake is to include totals equal to $\$3$ as exceedances.

For $N=0$ or $N=1$, the total cannot exceed $\$3$. For $N=2$, the total exceeds $\$3$ unless both severities are $\$1$. Therefore:

P(S>\$3\mid N=2)=1-(0.75)^2=\frac{7}{16}.

For $N=3$, the total is $\$3$ only if all three severities are $\$1$; any $\$3$ claim makes the total greater than $\$3$. Hence:

P(S>\$3\mid N=3)=1-(0.75)^3=\frac{37}{64}.

For $N\ge4$, even the minimum possible total is $4(\$1)=\$4$, so the conditional tail is one. Since $N$ is Poisson with parameter 1, $P(N=n)=e^{-1}/n!$. The count probabilities needed are $P(N=2)=e^{-1}/2$, $P(N=3)=e^{-1}/6$, and $P(N\ge4)=1-P(N\le3)=1-\frac{8}{3e}$. Combining the contributions:

\begin{aligned} P(S>\$3) &=\frac{e^{-1}}{2}\frac{7}{16} +\frac{e^{-1}}{6}\frac{37}{64} +\left(1-\frac{8}{3e}\right)\\ &=1-\frac{301}{128e}\\ &\approx 0.1349. \end{aligned}

The result is $0.1349$. It is below $P(N\ge2)$, as it must be because the aggregate can exceed $\$3$ only when there are at least two claims.

Worked Example: Identify Counts Before Enumerating Outcomes

A hypothetical portfolio has $N\sim\operatorname{Binomial}(4,0.5)$. Each claim independently has severity $\$1$ or $\$3$, each with probability $0.5$, independently of the count. Calculate $P(S>\$6)$.

Method. Condition on the number of claims and count how many $\$3$ severities are needed to exceed $\$6$. This avoids enumerating all possible severity sequences for every value of $N$. The common trap is to assume that three claims always exceed $\$6$: three $\$1$ claims total only $\$3$.

With at most two claims, the maximum total is $\$6$, so the conditional tail is zero for $N=0,1,2$. For $N=3$, let $K$ be the number of $\$3$ claims. The total is $3+2K$ dollars, which exceeds $\$6$ exactly when $K\ge2$. Since $K\sim\operatorname{Binomial}(3,0.5)$ conditional on $N=3$:

P(S>\$6\mid N=3)=P(K\ge2)=\frac{\binom{3}{2}+\binom{3}{3}}{2^3}=\frac{4}{8}=\frac{1}{2}.

For $N=4$, the total is $4+2K$ dollars. It exceeds $\$6$ when $K\ge2$, so:

P(S>\$6\mid N=4)=P(K\ge2) =\frac{\binom{4}{2}+\binom{4}{3}+\binom{4}{4}}{2^4} =\frac{6+4+1}{16} =\frac{11}{16}.

The relevant count probabilities are $P(N=3)=\binom{4}{3}(0.5)^4=4/16$ and $P(N=4)=(0.5)^4=1/16$. Therefore:

P(S>\$6)=\frac{4}{16}\frac{1}{2}+\frac{1}{16}\frac{11}{16} =\frac{32}{256}+\frac{11}{256} =\frac{43}{256} \approx 0.1680.

The answer is $0.1680$. The other count cases contribute zero, and the final result is no greater than $P(N\ge3)=5/16$, which provides a useful check.

Common Errors and Final Checks

Conditioning keeps frequency probabilities separate from severity-sum probabilities. A conditional tail such as $P(S_n>s)$ is not yet the aggregate tail: it must be multiplied by $P(N=n)$. Conversely, the unconditional count probability must not be multiplied into the conditional severity calculation more than once.

Changing the threshold event. For $P(S>s)$, a total equal to $s$ is excluded. Check equality cases before using a complement.
Stopping at a conditional probability. A value such as $P(S_2>s)$ must be weighted by $P(N=2)$ before it contributes to $P(S>s)$.
Ignoring the zero-claim case. When $N=0$, $S=0$. This case contributes to an exceedance only if $0>s$, and for a nonnegative threshold it contributes zero.
Assuming severities are independent of the count without stating it. If that assumption does not hold, the conditional severity-sum distribution may change with $N$.
Calculating unnecessary cases. Use the minimum and maximum possible totals first; many count values may have conditional tail zero or one.

Key takeaway: Calculate an aggregate tail by partitioning on the claim count, finding the conditional severity-sum tail for each relevant count, and weighting by the count probabilities. Strict inequalities, independence assumptions, and the zero-claim case should be checked explicitly.

Check Your Understanding

For each problem, condition on the claim count and pay close attention to whether the requested event is a strict exceedance.

A hypothetical portfolio has $N\sim\operatorname{Binomial}(4,0.25)$, with every claim severity equal to $\$3$. Find $P(S>\$6)$. (Answer: $0.0508$)
A hypothetical portfolio has $N\sim\operatorname{Binomial}(2,0.5)$. Each severity is independently $\$1$ or $\$4$, each with probability $0.5$. Find $P(S>\$4)$. (Answer: $0.1875$)
A hypothetical portfolio has $N\sim\operatorname{Poisson}(1)$, and every claim severity is $\$2$. Find $P(S>\$4)$, to four decimal places. (Answer: $0.0803$)
A hypothetical portfolio has $N\sim\operatorname{Binomial}(3,0.5)$. Each severity is independently $\$1$ or $\$2$, each with probability $0.5$, independently of $N$. Find $P(S>\$4)$. (Answer: $0.0625$)
A hypothetical portfolio has $N\sim\operatorname{Binomial}(3,0.4)$. Each severity is independently $\$1$ or $\$4$, each with probability $0.5$, independently of $N$. Find $P(S>\$4)$. (Answer: $0.2720$)

What You'll Learn