Aggregating Counts Across Unequal Exposure

Why Unequal Exposure Changes the Average

Claim records rarely contribute identical amounts of exposure. One group may contain many policies in force for only part of a year, while another contains fewer policies with longer coverage periods. To combine their experience fairly, the aggregate frequency must reflect both the number of claims and the amount of exposure behind those claims.

The calculation is simple, but a common shortcut is not: averaging the groups’ individual frequencies without weighting them by exposure. That gives a small group the same influence as a large group. The resulting number can differ substantially from the frequency calculated from all claims and all exposure together.

Definition: For group $j$, let $N_j$ be its observed claim count and $e_j$ its exposure, measured on a consistent basis. Its observed claim frequency is $f_j=N_j/e_j$, provided $e_j>0$. For $J$ groups with positive exposure, the aggregate frequency is $$ f_{\mathrm{agg}}=\frac{\sum_{j=1}^{J}N_j}{\sum_{j=1}^{J}e_j}. $$ It is the total observed count divided by total exposure, and is measured in claims per unit of exposure.

The unit of exposure might be earned policy-years, vehicle-years, or another appropriate measure defined for the risk and analysis. The exact choice is established before aggregating; all component exposures must use the same basis. Likewise, the claim counts must refer to the same kind of claims and align with the exposure periods being analyzed.

This is a rate, not necessarily a probability. In particular, claims per policy-year should not automatically be read as the probability that a policy has at least one claim. A group can have more than one claim per exposure unit, so its observed frequency can exceed one. The distinction between frequency and probability of any claim is a separate question.

Why the Aggregate Is an Exposure-Weighted Average

The pooled calculation has an equivalent weighted-average form. Since $N_j=e_j f_j$, substitute this identity into the definition of aggregate frequency:

f_{\mathrm{agg}} =\frac{\sum_{j=1}^{J}N_j}{\sum_{j=1}^{J}e_j} =\frac{\sum_{j=1}^{J}e_j f_j}{\sum_{j=1}^{J}e_j} =\sum_{j=1}^{J}w_j f_j, \qquad w_j=\frac{e_j}{\sum_{k=1}^{J}e_k}.

Each weight $w_j$ is the group’s share of total exposure. The weights are nonnegative and sum to one, so the aggregate is a weighted average of the group frequencies. A group with twice as much exposure receives twice as much weight. This derivation also shows why an unweighted mean is generally different: it uses equal weights $1/J$, regardless of the groups’ exposure sizes.

There is a useful model interpretation, but no claim-count distribution is needed for the calculation itself. If groups are being treated as having a common underlying frequency $\lambda$, so that expected count is proportional to exposure, the expected count in group $j$ is $E[N_j]=\lambda e_j$. By additivity of expectation,

E\!\left[\sum_{j=1}^{J}N_j\right] =\sum_{j=1}^{J}E[N_j] =\lambda\sum_{j=1}^{J}e_j.

Thus total exposure is the denominator that matches the expected total count under a common rate. This justifies pooling when a common rate is the quantity of interest. It does not establish that all groups truly have the same frequency. If their underlying risks differ, the pooled value describes the combined experience and can conceal those differences.

Calculating the Pooled Rate

Confirm comparability.
Check that the counts refer to the same claim definition and that each exposure is measured on a consistent basis for the relevant period.

Add the counts.
Calculate $N_{\mathrm{tot}}=\sum_j N_j$. Do not average the counts across groups.

Add the exposures.
Calculate $e_{\mathrm{tot}}=\sum_j e_j$, including fractional exposures where appropriate.

Divide and interpret.
Calculate $N_{\mathrm{tot}}/e_{\mathrm{tot}}$, state the units, and retain precision until the final rounding.

Worked Example: Three Groups With Unequal Exposure

A hypothetical insurer reviews claim experience for three groups of commercial vehicles. Group A has 120 vehicle-years and 6 claims; Group B has 30 vehicle-years and 3 claims; Group C has 50 vehicle-years and 1 claim. Calculate the aggregate frequency and compare it with the simple average of the group frequencies. Use claims per vehicle-year and round to four decimal places. No probability distribution is assumed.

Method: pool counts and exposures. First calculate the individual frequencies:

f_A=\frac{6}{120}=0.05,\qquad f_B=\frac{3}{30}=0.10,\qquad f_C=\frac{1}{50}=0.02.

The total claim count is $6+3+1=10$, and the total exposure is $120+30+50=200$ vehicle-years. Therefore,

f_{\mathrm{agg}}=\frac{10}{200}=0.05 \text{ claims per vehicle-year}.

For comparison, the unweighted average is $(0.05+0.10+0.02)/3=0.17/3=0.056666\ldots$, or $0.0567$ to four decimal places. It is higher than the pooled rate because it gives the 30-vehicle-year group the same weight as the group with 120 vehicle-years.

Check using exposure weights. The group weights are $120/200=0.60$, $30/200=0.15$, and $50/200=0.25$. Then $0.60(0.05)+0.15(0.10)+0.25(0.02)=0.030+0.015+0.005=0.050$. This agrees with the pooled count-over-exposure calculation. The common mistake avoided is treating the three groups as equally informative despite their unequal exposure.

Fractional Exposure and Combined Periods

Exposure can represent a partial policy period. For example, a policy in force for one quarter of a year contributes $0.25$ policy-years if the chosen exposure basis is policy-years. Aggregation works the same way: add the fractional exposure amounts, then divide the total count by that total. Do not replace the actual exposure by the number of policies unless each policy contributes exactly one unit of exposure.

For records from different periods, pooled frequency is an exposure-weighted summary of those observed periods. It is not automatically the frequency for a particular future period. If rates changed over time or the groups have different risk profiles, a single pooled number can hide that variation; the calculation alone does not adjust for it.

Worked Example: Unequal Experience Across Periods

In a hypothetical fleet program, Period 1 contributes 40 vehicle-years and 2 claims; Period 2 contributes 80 vehicle-years and 7 claims; Period 3 contributes 20 vehicle-years and 1 claim. Calculate the pooled frequency and the unweighted average of the three period frequencies. Report claims per vehicle-year to four decimal places. Treat the calculation as descriptive; no claim-count distribution is assumed.

Method: combine the observed counts and exposures. The period frequencies are

f_1=\frac{2}{40}=0.05,\qquad f_2=\frac{7}{80}=0.0875,\qquad f_3=\frac{1}{20}=0.05.

The total count is $2+7+1=10$ claims. Total exposure is $40+80+20=140$ vehicle-years. Hence,

f_{\mathrm{agg}}=\frac{10}{140}=0.071428\ldots =0.0714\text{ claims per vehicle-year}.

The unweighted average is $(0.05+0.0875+0.05)/3=0.1875/3=0.0625$. The two answers differ because Period 2 has the largest exposure and also has the highest observed frequency.

Check using weights. The exposure shares are $40/140=2/7$, $80/140=4/7$, and $20/140=1/7$. Therefore, $(2/7)(0.05)+(4/7)(0.0875)+(1/7)(0.05)=0.0142857+0.05+0.0071429=0.0714286$, which rounds to $0.0714$. The common mistake avoided is giving each period one-third of the influence, even though the periods contribute different amounts of exposure.

Worked Example: Combining Partial and Extended Policy Terms

A hypothetical equipment portfolio contains 12 policies that each contributed three months of exposure and generated 2 claims in total. Four other policies each contributed 18 months of exposure and generated 3 claims in total. Calculate the aggregate frequency in claims per policy-year. Also calculate the unweighted average of the two group frequencies. Round both rates to four decimal places. No probability distribution is assumed.

Method: convert every duration to policy-years before adding. Each three-month policy contributes $3/12=0.25$ policy-years. The first group therefore contributes $12(0.25)=3$ policy-years. Each 18-month policy contributes $18/12=1.5$ policy-years, so the second group contributes $4(1.5)=6$ policy-years.

The group frequencies are $2/3=0.666666\ldots$ and $3/6=0.5$ claims per policy-year. The pooled totals are $2+3=5$ claims and $3+6=9$ policy-years, giving

f_{\mathrm{agg}}=\frac{5}{9}=0.555555\ldots =0.5556\text{ claims per policy-year}.

The unweighted average is $(0.666666\ldots+0.5)/2=0.583333\ldots=0.5833$. As a check, the exposure weights are $3/9=1/3$ and $6/9=2/3$, and $(1/3)(0.666666\ldots)+(2/3)(0.5)=0.222222\ldots+0.333333\ldots=0.555555\ldots$.

The common mistake avoided is dividing the counts by policy counts while ignoring duration. The total of 16 policies is not 16 policy-years: the actual denominator is 9 policy-years.

Common Traps and Practical Checks

An aggregate calculation is easy to perform and easy to mis-specify. Before accepting a result, verify both sides of the ratio. A count numerator that includes a broader or different set of claims than the exposure denominator will not describe the intended frequency. Nor can differently scaled exposure units be added without conversion.

Taking a simple average of rates. This is generally wrong for an overall frequency when exposures differ. Use exposure weights, or return to total counts divided by total exposure.
Using policy count instead of exposure. Policy count is a suitable denominator only when each policy contributes the same amount of exposure on the chosen basis.
Mixing exposure units. Convert, for example, policy-months to policy-years before combining them with policy-years.
Rounding too early. Keep exact counts and exposure values through the calculation; round the final frequency to the requested precision.
Assuming the pooled rate explains every group. The aggregate is a weighted summary. It does not show whether group frequencies are alike or why they differ.

A quick reasonableness check is to compare the aggregate with the group frequencies: with positive exposure, an exposure-weighted average must lie between the smallest and largest group frequencies. Also verify that the weighted-average result matches the direct ratio. If those checks fail, revisit the units, totals, and arithmetic.

Key takeaway: For comparable claim counts and exposures, calculate aggregate frequency as total claims divided by total exposure. This is exactly the exposure-weighted average of group frequencies; an unweighted average answers a different question and can misrepresent the combined experience.

Check Your Understanding

For each problem, calculate the aggregate frequency from the combined counts and exposures. Use the stated exposure unit, and do not assume a claim-count probability distribution.

A hypothetical service-vehicle portfolio has 75 vehicle-years and 3 claims in one division, and 25 vehicle-years and 4 claims in another. Calculate the pooled frequency and the unweighted average of the two division frequencies. (Answer: pooled frequency $=0.0700$ claims per vehicle-year; unweighted average $=0.1000$.)
A hypothetical boat insurer observes 1 claim over 50 boat-years in one season and 9 claims over 150 boat-years in another. What is the aggregate frequency? (Answer: $0.0500$ claims per boat-year.)
A hypothetical property program has 20 policies contributing 0.5 policy-years each and 4 claims, plus 10 policies contributing 1.2 policy-years each and 3 claims. Calculate the pooled frequency to four decimal places. (Answer: $7/22=0.3182$ claims per policy-year.)
Three hypothetical repair shops contribute 60, 30, and 10 equipment-years, with 6, 0, and 4 claims, respectively. Calculate the pooled frequency and the unweighted average of the shop frequencies. (Answer: pooled frequency $=0.1000$ claims per equipment-year; unweighted average $=(0.1000+0+0.4000)/3=0.1667$.)
Two hypothetical customer groups have frequencies of 0.04 and 0.12 claims per exposure unit. The first contributes 300 exposure units and the second contributes 100. Calculate the combined rate using exposure weights. (Answer: $(300(0.04)+100(0.12))/400=0.0600$ claims per exposure unit.)

What You'll Learn