Calculating Frequency With Earned Exposure

Why Exposure Belongs in a Frequency Calculation

A claim count by itself does not show how often claims occurred relative to the amount of insurance in force. One period may have more claims simply because it had more insured exposure. Observed claim frequency puts the count in context by dividing it by the corresponding earned exposure.

The denominator is not an arbitrary scaling choice: it must represent the insured units that generated the claims being counted. For example, if a hypothetical analysis counts claims under vehicle coverage, earned vehicle-years can be an appropriate denominator when the count and exposure refer to the same coverage and experience period. A frequency stated per vehicle-year is a rate with units; it is not automatically the probability that a vehicle has a claim.

Definition: For a specified claim definition and experience period, observed claim frequency is the number of claims $N$ divided by the matching earned exposure $E$: $$ \widehat{f}=\frac{N}{E}. $$ The units are claims per exposure unit. A reporting base such as 1,000 exposure units changes the way the answer is displayed, not the underlying rate.

The hat indicates a value calculated from observed experience. No claim-count probability distribution is required for this descriptive ratio. Interpreting the result as a model parameter or a probability would require additional assumptions and analysis.

Make the Numerator and Denominator Match

Before dividing, specify what counts as a claim and what exposure unit is being used. Then ensure that the claim count and exposure refer to the same coverage, population, and experience period. If, for example, the numerator counts claims from one coverage but the denominator includes exposure for several coverages, the ratio does not describe the intended frequency.

Earned exposure is the measure of insured risk that corresponds to the time or units actually covered during the experience period. The appropriate unit depends on the analysis: it might be earned vehicle-years, policy-years, or another defined exposure unit. State the unit alongside the result. A value of $0.04$ claims per vehicle-year and a value of 40 claims per 1,000 vehicle-years express the same frequency, but the latter is often easier to interpret.

For periods $i=1,\ldots,k$, let $N_i$ be the claim count and $E_i$ the earned exposure for period $i$. Calculate each period’s observed frequency as $N_i/E_i$. To obtain the observed frequency across all periods, use the total count and total exposure:

\widehat{f}_{\mathrm{combined}} = \frac{\sum_{i=1}^{k}N_i}{\sum_{i=1}^{k}E_i}.

This calculation describes the combined experience represented by those counts and exposures. It does not assign equal importance to each period’s rate: a period with more exposure contributes more to the combined result.

Why the Combined Frequency Is Exposure-Weighted

The pooled frequency can also be written as an exposure-weighted average of the individual period frequencies. For positive exposures, define $\widehat{f}_i=N_i/E_i$. Since $N_i=E_i\widehat{f}_i$, substituting into the combined ratio gives:

\widehat{f}_{\mathrm{combined}} = \frac{\sum_i E_i\widehat{f}_i}{\sum_i E_i} = \sum_i \left(\frac{E_i}{\sum_j E_j}\right)\widehat{f}_i.

Each period’s weight is its share of the total exposure, $E_i/\sum_j E_j$. The weights sum to 1, so the combined frequency is an average of the period frequencies weighted by exposure. An unweighted arithmetic mean instead gives a small period the same influence as a large one, even when the periods represent very different amounts of insured risk.

Worked Example: Converting a Count to Frequency per 1,000 Vehicle-Years

A hypothetical insurer reviews one complete experience year of a specified vehicle coverage. The matching data contain 72 claims and 2,400 earned vehicle-years. Calculate the observed frequency per vehicle-year and per 1,000 vehicle-years. Assume the claims and exposure use the same coverage definition and period; no claim-count distribution is assumed.

Method. Divide the claim count by earned exposure, then multiply by 1,000 to express the result on the requested base.

Step 1: Calculate claims per vehicle-year. $$ \widehat{f}=\frac{72}{2{,}400}=0.03 \quad\text{claims per vehicle-year}. $$

Step 2: Convert the reporting base. $$ 0.03\times1{,}000=30 \quad\text{claims per 1,000 vehicle-years}. $$

Check. At 0.03 claims per vehicle-year, 2,400 vehicle-years correspond to $0.03\times2{,}400=72$ claims, matching the observed count. The answers are 0.03 claims per vehicle-year and 30 claims per 1,000 vehicle-years. The common mistake avoided is treating 0.03 as 0.03 claims per 1,000 vehicle-years; the multiplication by 1,000 is necessary to change the base.

Worked Example: Combining Two Experience Periods

For a hypothetical coverage, period A has 84 claims and 2,800 earned policy-years. Period B has 126 claims and 5,600 earned policy-years. Calculate each period’s frequency and the combined frequency per 1,000 earned policy-years. The claim definition and coverage are consistent across both periods, and no claim-count distribution is assumed.

Step 1: Calculate period A’s frequency. $$ \widehat{f}_A=\frac{84}{2{,}800}=0.03 \quad\text{claims per policy-year}. $$

Step 2: Calculate period B’s frequency. $$ \widehat{f}_B=\frac{126}{5{,}600}=0.0225 \quad\text{claims per policy-year}. $$

Step 3: Add counts and exposures separately. $$ N_{\mathrm{total}}=84+126=210, \qquad E_{\mathrm{total}}=2{,}800+5{,}600=8{,}400. $$

Step 4: Divide the totals and convert the base. $$ \widehat{f}_{\mathrm{combined}} =\frac{210}{8{,}400}=0.025 \quad\text{claims per policy-year}, $$ $$ 0.025\times1{,}000=25 \quad\text{claims per 1,000 policy-years}. $$

Check. Period A contributes one-third of total exposure and period B contributes two-thirds. The weighted calculation is $(1/3)(0.03)+(2/3)(0.0225)=0.025$, agreeing with the ratio of totals. The unweighted average, $(0.03+0.0225)/2=0.02625$, is not the combined frequency because it gives the smaller period the same weight as the larger one. The combined answer is 25 claims per 1,000 policy-years.

Read Counts and Frequencies Together

A claim count can rise while the observed frequency falls if earned exposure rises by a greater proportion. Conversely, a lower count does not necessarily indicate a lower frequency if exposure has fallen substantially. Compare ratios on the same exposure base before drawing conclusions about changes in frequency.

When comparing two periods, a relative change in frequency can be calculated from the two observed rates:

\text{Relative change} = \frac{\widehat{f}_{\mathrm{new}}-\widehat{f}_{\mathrm{old}}} {\widehat{f}_{\mathrm{old}}}.

This is a comparison of the measured frequencies. It does not, by itself, explain why they differ or establish that the difference is statistically meaningful.

Worked Example: More Claims but a Lower Observed Frequency

A hypothetical property coverage records 96 claims over 3,200 earned exposure units in an older period, and 117 claims over 4,500 earned exposure units in a newer period. Calculate both frequencies per 1,000 exposure units and the relative change from the older period to the newer period. The numerator and denominator use the same claim definition in both periods; no claim-count distribution is assumed.

Step 1: Calculate the older-period frequency. $$ \widehat{f}_{\mathrm{old}}=\frac{96}{3{,}200}=0.03, \qquad 0.03\times1{,}000=30 \quad\text{claims per 1,000 exposure units}. $$

Step 2: Calculate the newer-period frequency. $$ \widehat{f}_{\mathrm{new}}=\frac{117}{4{,}500}=0.026, \qquad 0.026\times1{,}000=26 \quad\text{claims per 1,000 exposure units}. $$

Step 3: Calculate the relative change. $$ \frac{0.026-0.03}{0.03} = \frac{-0.004}{0.03} = -0.1333\ldots $$ The relative change is approximately $-13.33\%$, or a 13.33% decrease.

Check. The count increased by $117-96=21$, but exposure increased from 3,200 to 4,500 units. The calculated rates show that frequency fell from 30 to 26 claims per 1,000 units. The common mistake avoided is interpreting the increase in raw claim count as an increase in frequency.

Common Calculation Traps

Using the wrong denominator. Dividing claims by a count of policies, insured objects, or another measure that does not match the stated exposure unit produces a different quantity. Name the unit and use it consistently.
Combining rates without their exposure weights. For example, averaging period frequencies of 0.03 and 0.0225 gives 0.02625. If the periods have exposures of 2,800 and 5,600, respectively, the combined rate is instead $210/8{,}400=0.025$.
Confusing a reporting base with the underlying rate. Multiplying a per-exposure-unit rate by 1,000 reports it per 1,000 units; it does not change the observed experience.
Comparing mismatched data. A count and exposure must correspond to the same period, coverage, and claim definition for their ratio to have the intended interpretation.
Calling a frequency a probability. The ratio is claims per exposure unit. It is not automatically the probability of at least one claim; such an interpretation requires a model and a clearly defined event.

Key takeaway: Calculate observed frequency as matching claims divided by matching earned exposure. For multiple periods, add the counts, add the exposures, and divide the totals; this is equivalent to an exposure-weighted average of the period frequencies.

Check Your Understanding

For each problem, use the stated exposure unit and show the ratio before converting to any requested reporting base.

A hypothetical fleet coverage has 63 claims over 2,100 earned vehicle-years. What is the observed frequency per 1,000 vehicle-years? (Answer: 30 claims per 1,000 vehicle-years)
Period A has 48 claims and 1,600 earned policy-years; period B has 70 claims and 2,800 earned policy-years. Find the combined frequency per 1,000 policy-years. (Answer: $118/4{,}400=0.026818\ldots$, or 26.8 claims per 1,000 policy-years)
A hypothetical coverage records 55 claims over 2,500 earned exposure units in one period and 105 claims over 3,500 units in another. Calculate the combined frequency per 100 exposure units. (Answer: $160/6{,}000=0.026666\ldots$, or 2.67 claims per 100 exposure units)
A portfolio’s count rises from 90 claims over 3,000 earned policy-years to 120 claims over 5,000 earned policy-years. Calculate each frequency per 1,000 policy-years and the relative change from the first period to the second. (Answer: 30 and 24 claims per 1,000 policy-years; a 20% decrease)
Two periods have frequencies of 0.04 and 0.025 claims per exposure unit, with 1,000 and 3,000 exposure units, respectively. Find the combined frequency and explain why the simple average is not the answer. (Answer: $(40+75)/(1{,}000+3{,}000)=0.02875$ claims per exposure unit; the simple average 0.0325 gives equal weight to periods with unequal exposure)

What You'll Learn