Estimating Frequency from Pooled Experience

Why Pool Claim Experience?

When claim experience is reported across several periods or groups, each period may have a different number of claims and a different amount of exposure. A pooled frequency combines the counts and exposures before calculating a rate. This gives more influence to periods that represent more exposure, rather than treating a small period as equally informative as a large one.

For example, suppose one period has 20 claims over 500 exposure units and another has 40 claims over 2,000 exposure units. Averaging the two period frequencies equally does not give the same answer as dividing all 60 claims by all 2,500 exposure units. Which calculation is appropriate depends on the quantity being estimated. For the frequency across the combined exposure, use the pooled calculation.

Definition: For period $j$, let $N_j$ be the number of claims counted under a consistent claim definition, and let $E_j$ be the corresponding exposure on a consistent basis. The period frequency is $f_j=N_j/E_j$. The pooled frequency across $k$ periods is $$ f_{\mathrm{pool}}=\frac{\sum_{j=1}^{k}N_j}{\sum_{j=1}^{k}E_j}. $$ It is a rate per exposure unit. To report it per 1,000 exposure units, multiply by 1,000.

Exposure must represent the opportunity for the claims being counted to occur. It might be earned vehicle-years, insured units, or another defined exposure base. Counts and exposures must refer to compatible periods, populations, and coverage definitions. If one period counts claims from a different coverage or uses a different exposure basis, simply combining the figures does not produce a meaningful pooled frequency.

Why the Pooled Result Is Exposure-Weighted

The pooled frequency can be written as a weighted average of the period frequencies. Since $N_j=E_j f_j$, substituting this into the pooled calculation gives

f_{\mathrm{pool}} =\frac{\sum_{j=1}^{k}E_jf_j}{\sum_{j=1}^{k}E_j} =\sum_{j=1}^{k}\left(\frac{E_j}{\sum_{r=1}^{k}E_r}\right)f_j.

The weight for period $j$ is its share of total exposure, $E_j/\sum_r E_r$. These weights are nonnegative and sum to 1. A period with twice as much exposure as another period receives twice as much weight in the pooled frequency. This is the key distinction from an unweighted average, which assigns each period weight $1/k$, regardless of its size.

Key result: The pooled frequency equals the exposure-weighted average of the period frequencies. It equals their unweighted average when all periods have equal exposure, or in special cases where the different weights happen to produce the same result. Otherwise, the two calculations generally differ.

A simple model also explains why this estimator is standard when a common underlying frequency is assumed. Suppose exposures $E_j$ are known and fixed, and the counts are independent with $N_j\sim\operatorname{Poisson}(\lambda E_j)$, where $\lambda$ is the common expected number of claims per exposure unit. Under this model, the likelihood, apart from factors that do not depend on $\lambda$, is proportional to

L(\lambda)\propto \lambda^{\sum_jN_j}\exp\left(-\lambda\sum_jE_j\right).

The log-likelihood is $\ell(\lambda)=(\sum_jN_j)\log(\lambda)-\lambda\sum_jE_j+\text{constant}$. Its derivative is $(\sum_jN_j)/\lambda-\sum_jE_j$. Setting that derivative to zero gives $\widehat{\lambda}=\sum_jN_j/\sum_jE_j$. The second derivative is $-(\sum_jN_j)/\lambda^2$, which is negative when the total count is positive, so this stationary point maximizes the likelihood. If the total count is zero, the likelihood is maximized at the boundary $\widehat{\lambda}=0$.

This likelihood argument relies on the independent Poisson model and a common frequency across the pooled experience. The ratio of total claims to total exposure can still be calculated descriptively without those assumptions, but interpreting it as an estimate of one common underlying $\lambda$ requires them. If underlying frequencies differ across groups, the pooled ratio describes the combined experience and is an exposure-weighted average of their rates, not necessarily the rate of any one group.

Worked Example: Pooling Two Periods With Different Exposure

A hypothetical insurer reviews two completed periods for the same coverage and claim definition. Period A has 24 claims and 800 earned vehicle-years; Period B has 30 claims and 1,200 earned vehicle-years. Calculate the pooled frequency per vehicle-year and per 100 vehicle-years. Compare the result with the unweighted average of the period frequencies.

Method. Add the claim counts and exposure amounts separately, then divide total claims by total exposure.

Period A frequency is $24/800=0.030$ claims per vehicle-year. Period B frequency is $30/1{,}200=0.025$ claims per vehicle-year.

The combined claim count is $24+30=54$, and the combined exposure is $800+1{,}200=2{,}000$ vehicle-years. Therefore,

f_{\mathrm{pool}}=\frac{54}{2{,}000}=0.027

The pooled frequency is 0.027 claims per vehicle-year. Per 100 vehicle-years, it is $0.027\times100=2.7$ claims.

The unweighted average of the two period frequencies is $(0.030+0.025)/2=0.0275$ claims per vehicle-year, or 2.75 per 100 vehicle-years. The exposure-weighted calculation gives the same answer directly: Period A has weight $800/2{,}000=0.4$, and Period B has weight $1{,}200/2{,}000=0.6$. Thus, $0.4(0.030)+0.6(0.025)=0.012+0.015=0.027$.

Check. Multiplying the pooled rate by total exposure gives $0.027(2{,}000)=54$ claims, matching the observed total. The unweighted average is larger because it gives the smaller, higher-frequency period more weight than its share of exposure warrants. The common mistake avoided here is treating each period as equally large.

Applying the Calculation and Checking Units

A reliable calculation keeps the numerator, denominator, and rate unit visible throughout. Counts are not themselves rates, and a rate per exposure unit should not be compared with a rate per 1,000 units unless the scale is converted. A final multiplication by 1,000 changes the reporting scale; it does not change the underlying frequency.

Check comparability.
Confirm that the claim definition, coverage, population, and exposure basis are consistent across the periods being combined.

Combine counts and exposures separately.
Calculate $N_{\mathrm{total}}=\sum_j N_j$ and $E_{\mathrm{total}}=\sum_j E_j$.

Divide and state the unit.
Calculate $N_{\mathrm{total}}/E_{\mathrm{total}}$, then convert to the requested number of claims per exposure unit, such as per 100 or per 1,000.

Check the arithmetic.
Multiply the pooled frequency by total exposure. The result should equal the combined claim count, subject only to any rounding already applied.

Worked Example: Converting a Pooled Rate to Claims per 1,000 Units

A hypothetical equipment insurer observes 9 claims over 300 insured equipment-years in one period and 51 claims over 2,550 insured equipment-years in another. Calculate the pooled frequency per equipment-year and per 1,000 equipment-years. Also calculate the unweighted average of the period frequencies and explain the difference.

Method. Pool the raw counts and exposures, then convert the resulting rate to the requested scale.

The period frequencies are $9/300=0.030$ and $51/2{,}550=0.020$ claims per equipment-year. Total claims are $9+51=60$, and total exposure is $300+2{,}550=2{,}850$ equipment-years. Therefore,

f_{\mathrm{pool}}=\frac{60}{2{,}850}=0.0210526\ldots

Per 1,000 equipment-years, this is $0.0210526\ldots\times1{,}000=21.0526\ldots$, or 21.05 claims per 1,000 when rounded to two decimal places. The unweighted average is $(0.030+0.020)/2=0.025$, or 25 claims per 1,000.

The exposure shares are $300/2{,}850=0.105263\ldots$ and $2{,}550/2{,}850=0.894737\ldots$. Weighting the period frequencies gives $0.105263\ldots(0.030)+0.894737\ldots(0.020)=0.0031579\ldots+0.0178947\ldots=0.0210526\ldots$, as required.

Check. The pooled rate times total exposure is $0.0210526\ldots(2{,}850)=60$ claims. The unweighted calculation overstates the combined rate because the period with the higher rate accounts for only a small share of exposure. The common mistake avoided is averaging rounded or unrounded period rates equally instead of using the total exposure denominator.

Interpreting the Pooled Frequency

Pooling does not make all periods equally representative. It answers: “How many claims occurred per unit of exposure across all of this experience?” It does not answer: “What is the average of the period-specific rates when each period is given equal importance?” Those are different summaries when exposure varies.

The pooled rate is appropriate when the target is the combined experience and the exposure units are comparable. It is also the maximum-likelihood estimate of a common frequency under the stated Poisson model. If a particular analysis instead calls for an average period rate, each period may intentionally receive equal weight, but that quantity should be labeled as an unweighted average rather than the pooled frequency.

Pooling alone does not establish that the underlying frequency is constant across periods or risk groups. It summarizes the observed combined experience. A common-frequency interpretation depends on an appropriate common-rate assumption; if rates vary, the pooled result remains exposure-weighted and may hide that variation. Assessing changes between periods or deciding how to project future frequency requires additional analysis beyond this calculation.

Common Mistakes

Taking a simple average by default. The mean of period rates gives every period equal weight. For pooled frequency, divide total claims by total exposure.
Adding rates instead of counts and exposures. Add the underlying numerators and denominators first; rates alone do not preserve the amount of experience in each period.
Mixing exposure units. Do not combine, for example, policy-years and vehicle-years as though they were interchangeable. First use a common, appropriate exposure basis.
Reporting a rate without its scale. State whether the answer is per exposure unit, per 100 units, or per 1,000 units.
Assuming pooling proves a common rate. The ratio is a valid combined descriptive rate, but interpreting it as one common underlying frequency requires a suitable model and assumptions.

Key takeaway: Estimate frequency for combined experience by dividing total claims by total exposure. Equivalently, weight each period frequency by that period’s share of total exposure. Use an unweighted average only when equal weighting of periods is the intended target.

Check Your Understanding

For each problem, calculate the requested pooled frequency and compare it with an unweighted average when asked.

A hypothetical portfolio has 14 claims over 400 policy-years in one period and 36 claims over 1,600 policy-years in another. Find the pooled frequency per 1,000 policy-years and the unweighted average of the two period frequencies. (Answer: pooled frequency $=25.00$ claims per 1,000 policy-years; unweighted average $=28.75$ claims per 1,000 policy-years.)
A hypothetical insurer records 18 claims over 900 insured-unit-years and 12 claims over 300 insured-unit-years. Find the pooled frequency per insured-unit-year and compare it with the unweighted average. (Answer: pooled frequency $=30/1{,}200=0.025$; unweighted average $=(0.020+0.040)/2=0.030$ claims per insured-unit-year.)
A hypothetical book has 6 claims over 250 vehicle-years and 27 claims over 1,500 vehicle-years. Calculate the pooled frequency per 1,000 vehicle-years, and verify it by weighting the period frequencies by exposure share. (Answer: $33/1{,}750\times1{,}000=18.857$ claims per 1,000 vehicle-years; the exposure-weighted rate is $0.142857(0.024)+0.857143(0.018)=0.018857$ per vehicle-year.)
A hypothetical insurer observes 40 claims over 2,000 exposure units. Under a common-frequency model, use the pooled estimate to calculate the expected number of claims over 1,800 comparable future exposure units. (Answer: estimated frequency $=40/2{,}000=0.020$ per unit; expected claims $=0.020(1{,}800)=36$.)
Three hypothetical periods have, respectively, 16 claims over 400 exposure units, 12 claims over 600 units, and 18 claims over 1,500 units. Calculate the pooled frequency and the unweighted average of the three period frequencies, both per 1,000 exposure units. (Answer: pooled $=46/2{,}500\times1{,}000=18.4$; unweighted average $=(40+20+12)/3=24$ claims per 1,000 exposure units.)

What You'll Learn