Scaling Poisson Frequency By Exposure

Why Exposure Scaling Matters

A claim frequency rate is meaningful only when its exposure basis is clear. A rate stated per policy-year cannot be applied directly to a group observed for six months without accounting for the shorter period. Likewise, a mean claim count for one portfolio cannot be used unchanged for a portfolio twice its size. Exposure scaling converts time and portfolio size to a common basis, then adjusts the Poisson mean proportionally.

The method is simple, but it depends on a specific model: the claim rate per exposure unit is constant, and the claim count follows a Poisson distribution. Scaling exposure does not mean that a larger group must have exactly a proportional number of claims in a particular realization. It means that its expected count, and the parameter of its Poisson model, scale proportionally.

Definition: Let $\lambda$ be a Poisson claim rate per exposure unit, and let $e$ be the amount of exposure measured in those units. Under a constant-rate Poisson model, the claim count $N$ has distribution $N\sim\operatorname{Poisson}(\lambda e)$. Its mean and variance are both $\lambda e$, and for an integer $k\geq 0$, $$ P(N=k)=e^{-\lambda e}\frac{(\lambda e)^k}{k!}. $$ An exposure unit might be one policy-year, one vehicle-month, or another defined combination of insured units and time. The rate and exposure must use matching units.

Calculating Exposure Before the Poisson Mean

For identical risks each observed for the same duration, total exposure is the number of risks multiplied by the duration per risk. If the duration is a fraction of a year, convert it to years when the rate is per policy-year. For example, 80 policies each observed for six months provide $80(6/12)=40$ policy-years.

If risks have different observation periods, add their individual exposures. For risk $j$, let $t_j$ denote its duration measured in the chosen time unit. With a common rate $\lambda$ per unit of exposure, total exposure is $e=\sum_j t_j$, and the Poisson mean is $\lambda e$. This calculation assumes that the risks share the same claim rate on that basis. If rates differ, their expected counts must instead be calculated using their respective rates and added when their counts are independent.

Identify the rate basis.
Write down the unit in the rate, such as claims per policy-year. Keep the same time unit throughout the calculation.

Convert the portfolio to exposure.
Multiply the number of insured units by their exposure duration, or add the individual exposures if durations differ.

Scale the mean.
Multiply the rate by total exposure to obtain $\mu=\lambda e$, the Poisson mean.

Calculate the requested probability.
Use the Poisson probability formula or an appropriate sum of its probabilities, with mean $\mu$.

Why the Mean Scales Proportionally

The proportional scaling rule follows from the additivity of independent Poisson counts. Suppose two nonoverlapping exposure blocks have the same rate $\lambda$ and exposures $e_1$ and $e_2$. Their counts are independent Poisson random variables with means $\lambda e_1$ and $\lambda e_2$. The sum of independent Poisson random variables is Poisson, with parameter equal to the sum of their parameters. Therefore, the total count has mean $\lambda e_1+\lambda e_2=\lambda(e_1+e_2)$.

N_1+N_2\sim\operatorname{Poisson}\bigl(\lambda e_1+\lambda e_2\bigr) =\operatorname{Poisson}\bigl(\lambda(e_1+e_2)\bigr).

Dividing an exposure period into smaller pieces therefore does not change the model for the total, provided the rate remains constant and the counts in disjoint pieces are independent. This is the basis for scaling from a rate per year to a fraction of a year, or from one portfolio size to another. The result does not say that a realized count in a particular block is proportional to exposure; only the Poisson parameter is proportional.

Worked Example: Six Months of Policy-Year Exposure

A hypothetical insurer models claim counts at a constant rate of $0.05$ claims per policy-year. It observes 80 policies for six months each. Find the expected claim count and the probability of exactly three claims. Also calculate the probability of at least one claim.

Method and exposure. The rate is per policy-year, so six months is $6/12=0.5$ years per policy. Total exposure is $$ e=80(0.5)=40\text{ policy-years}. $$ The Poisson mean is $$ \mu=\lambda e=(0.05)(40)=2. $$ Thus $N\sim\operatorname{Poisson}(2)$, and the expected claim count is 2.

Exactly three claims. Substitute $k=3$ and $\mu=2$: $$ P(N=3)=e^{-2}\frac{2^3}{3!} =e^{-2}\frac{8}{6} =0.180447\ldots. $$ To four decimal places, $P(N=3)=0.1804$.

At least one claim. Use the complement of zero claims: $$ P(N\geq 1)=1-P(N=0) =1-e^{-2}\frac{2^0}{0!} =1-e^{-2} =0.864664\ldots. $$ To four decimal places, $P(N\geq 1)=0.8647$.

Check and common mistake. The expected count of 2 is consistent with 40 policy-years at a rate of 0.05 per policy-year. Treating the 80 policies as 80 full policy-years would incorrectly give a mean of 4 and the wrong probabilities. The six-month duration must be included before applying the Poisson formula.

Scaling Across Portfolio Sizes

Portfolio size changes are handled in the same way as time changes: both change total exposure. If a group has $m$ identical risks observed for duration $t$, its exposure is $mt$. Doubling the number of risks while holding the duration fixed doubles exposure and therefore doubles the Poisson mean. Halving the observation period halves the mean, all else equal.

This proportionality requires a common rate per exposure unit. It is not appropriate to multiply the mean by a portfolio-size ratio if the new group has a different rate or if the rate changes over the observation period. For instance, if risk characteristics or coverage terms differ, the relevant rate may not be the same. The model must describe the risks being counted, not just their number.

Worked Example: Resizing a Portfolio and Finding an Exact Count

A hypothetical travel-protection portfolio has a constant claim rate of $0.06$ claims per insured-year. For a new block of business, 100 insureds are covered for six months each. Find the expected claim count and the probability of exactly two claims.

Method and exposure. Each insured contributes $6/12=0.5$ insured-year, so $$ e=100(0.5)=50\text{ insured-years}. $$ The mean claim count is $$ \mu=(0.06)(50)=3. $$ Thus $N\sim\operatorname{Poisson}(3)$, and $E[N]=3$.

Exactly two claims. Apply the probability function with $k=2$: $$ P(N=2)=e^{-3}\frac{3^2}{2!} =e^{-3}\frac{9}{2} =0.224041\ldots. $$ To four decimal places, the probability is $0.2240$.

Check and common mistake. The same 100 insureds would contribute 100 insured-years if covered for a full year, giving mean $0.06(100)=6$. The six-month block has half that exposure and mean, namely 3. Using the full-year mean of 6 for this block would fail to adjust for duration.

Combining Different Durations and Checking Assumptions

Exposure need not be a whole number. A group of 150 risks covered for nine months contributes $150(9/12)=112.5$ risk-years. A Poisson mean such as $0.04(112.5)=4.5$ is valid: the count $N$ remains an integer-valued random variable, but its mean parameter need not be an integer.

When combining separate blocks, add their expected counts if the blocks have independent Poisson counts. If the blocks share a common rate, adding their exposures first gives the same result. When rates differ, calculate each block's mean separately. Independence is needed for the total count to have the Poisson distribution with the sum of the means; without it, the summed mean may still be calculated from expectations, but the Poisson distribution for the total does not follow from the scaling argument alone.

Worked Example: A Fractional Exposure and a Cumulative Probability

A hypothetical equipment program uses a constant claim rate of $0.04$ claims per covered equipment-year. It covers 150 units for nine months each. Calculate the probability that the total claim count is at most two.

Method and exposure. Nine months is $9/12=0.75$ years. Total exposure and the Poisson mean are $$ e=150(0.75)=112.5\text{ equipment-years}, \qquad \mu=(0.04)(112.5)=4.5. $$ Therefore $N\sim\operatorname{Poisson}(4.5)$.

“At most two” includes zero, one, and two claims, so add those three probabilities: $$ \begin{aligned} P(N\leq 2) &=P(N=0)+P(N=1)+P(N=2)\\ &=e^{-4.5}\left(1+4.5+\frac{4.5^2}{2!}\right)\\ &=e^{-4.5}(1+4.5+10.125)\\ &=e^{-4.5}(15.625)\\ &=0.173578\ldots. \end{aligned} $$ To four decimal places, $P(N\leq 2)=0.1736$.

Check and common mistake. The exposure is fractional because it counts total equipment-years, not whole calendar years. The probability sum includes exactly the three integer counts $0,1,2$; using only $P(N=2)$ would answer a different question.

Common Exam Traps

Mixing rate and exposure units. A rate per policy-year must be multiplied by policy-years, not by the number of policies alone when coverage is shorter or longer than one year.
Scaling the rate instead of the mean. If the rate is quoted per exposure unit, retain that rate and scale total exposure. Equivalently, the mean changes in proportion to exposure; the underlying per-unit rate does not.
Treating a mean as a guaranteed count. A mean of 3 does not mean that exactly three claims will occur. It is the parameter used to calculate the full count distribution.
Rounding exposure too early. Keep exact fractions such as $9/12$ through the calculation when possible, and round a requested probability only at the end.
Assuming proportional scaling despite a changing rate. A constant-rate model is essential. If the rate varies by time or risk group, use the appropriate rate for each exposure block rather than applying one rate indiscriminately.

Key takeaway: Under a constant-rate Poisson model, first express the portfolio in units matching the rate. Then calculate $\mu=\lambda e$ and use that mean for the desired claim-count probability. Proportional exposure scaling changes the Poisson mean, not the fact that realized counts remain random.

Check Your Understanding

For each question, identify the matching exposure units before calculating a mean or probability.

A hypothetical device portfolio has a rate of $0.025$ claims per device-month. Eighty devices are covered for three months. Find the expected count and the probability of zero claims. (Answer: mean $=6$; $P(N=0)=0.0025$, to four decimal places.)
A hypothetical portfolio has a constant rate of $0.08$ claims per policy-year. There are 120 policies, each covered for nine months. What is the expected claim count? (Answer: $7.2$ claims.)
A hypothetical program has a rate of $0.03$ claims per unit-year and covers 250 units for four months each. Find the probability of exactly one claim. (Answer: $0.2052$, to four decimal places.)
A hypothetical block has an expected 2 claims over 40 exposure units. Assume a constant Poisson rate. Find the rate per exposure unit, then find the probability of at least two claims over 40 exposure units. (Answer: rate $=0.05$ per exposure unit; probability $=0.5940$, to four decimal places.)
Two independent hypothetical groups have the same rate of $0.04$ claims per exposure unit. The first has 25 units and the second has 50 units. Find the probability of zero claims in the first group and exactly two in the second group. (Answer: $2e^{-3}=0.0996$, to four decimal places.)

What You'll Learn