What Is Being Counted?
A frequency model begins with a precisely defined count. Before assigning a probability distribution, identify what event is counted, which risks are observed, and the period or exposure basis. Without these details, a numerical answer such as “12 claims” may be impossible to interpret.
For example, one insured vehicle could have no claims, one claim, or several claims during a year. The number of vehicles is not the number of claims, and the total dollars paid on claims are not either count. These distinctions are foundational for frequency modeling aligned with the SOA Exam FAM (Fundamentals of Actuarial Mathematics) syllabus.
The rule matters. A single incident involving several damaged items might be recorded as one claim under one administrative definition and as several claim records under another. Similarly, the date used to assign a claim to a period might be the loss date or the report date. A count is meaningful only when the counting rule and time window are clear.
A claim count does not require that different claims come from different risks. If one policy has three qualifying claim events during the observation period, those events contribute three to the total count. A risk with no qualifying claim contributes zero.
Claims, Risks, Amounts, and Exposure
It is useful to separate four quantities that are often mixed together:
- Number of insured risks: how many vehicles, homes, people, or other units are covered.
- Claim count: how many qualifying claim events occur under the stated rule.
- Claim amount: the dollar amount associated with an individual claim.
- Exposure: the amount of time or coverage represented by the observed risks, measured in a stated unit such as policy-years.
Exposure allows observations with different coverage durations to be put on a consistent basis. One policy covered for one full year contributes one policy-year. A policy covered for six months contributes one-half policy-year. The exposure measure should reflect the coverage relevant to the question; a lapse or a change in covered status may mean that a policy contributes less than a full year.
The rate can exceed 1. For instance, 140 claims over 100 policy-years gives 1.4 claims per policy-year. That is possible because a policy-year can contain more than one claim. By contrast, a probability that a risk has at least one claim cannot exceed 1. The two quantities answer different questions.
Worked Example: Claims Versus Insured Vehicles
A hypothetical insurer observes 160 delivery vehicles for a full year. The claim register contains 28 qualifying claim events. Calculate the number of insured risks, the claim count, and the observed frequency per vehicle-year. Explain whether the frequency is a probability.
Method. Count risks and claim events separately, then divide the claim count by the total exposure in vehicle-years.
Insured risks. There are 160 insured vehicles. Because each is covered for one full year, the total exposure is \(160\times 1=160\) vehicle-years.
Claim count. The number of qualifying claim events is \(N=28\). This total may include multiple claims from the same vehicle.
Observed frequency.
The result is not a probability. It is a count per unit of exposure. In particular, it does not say that 17.5% of vehicles had a claim; that percentage would require knowing how many distinct vehicles had one or more claims.
Check. Multiplying the rate by the exposure recovers the observed count: \(0.175\times160=28\) claims. The common mistake avoided is treating “claims per vehicle-year” as “the proportion of vehicles with a claim.”
Combining Different Coverage Durations
When risks have different coverage durations, add their exposure amounts before calculating a pooled frequency. Do not divide by the number of policies if the desired measure is claims per policy-year and the policies were not all covered for a full year.
Worked Example: Partial-Year Policy Exposure
A hypothetical book of business contains 75 policies covered for one year, 30 policies covered for six months, and 20 policies covered for three months. During those coverage periods, 19 qualifying claims are recorded. Find the number of distinct policies, total policy-year exposure, and observed claims per policy-year.
Method. Convert each coverage duration to years, multiply by the number of policies in that group, and add the resulting exposures.
Distinct policies. The number of policies is \(75+30+20=125\).
Exposure from each group. The first group contributes \(75\times1=75\) policy-years. The second contributes \(30\times0.5=15\) policy-years. The third contributes \(20\times0.25=5\) policy-years. Thus the total is \(75+15+5=95\) policy-years.
Observed frequency. Divide the 19 claims by 95 policy-years:
Check. \(0.20\times95=19\), so the rate and exposure reproduce the recorded claim count. The common denominator error would be \(19/125=0.152\) claims per policy. That ratio describes claims per distinct policy over this particular collection of unequal coverage periods; it is not the requested annualized rate per policy-year.
Why the Pooled Rate Uses Total Claims Over Total Exposure
The pooled rate can also be understood as an exposure-weighted average of group rates. Suppose group \(j\) has \(n_j\) claims over \(e_j\) exposure units, with \(e_j>0\). Its observed rate is \(n_j/e_j\). Weight that rate by its exposure, add across groups, and divide by total exposure:
The equality holds because \(e_j(n_j/e_j)=n_j\) for each group. Thus the pooled rate is total claims divided by total exposure. A simple, unweighted average of group rates generally does not give this result: it gives equal weight to groups even when their exposures differ greatly.
This is an arithmetic justification for the descriptive rate, not an assumption that claims follow any particular probability distribution. The distributional model, when introduced, will add assumptions about how claim counts behave. For now, the ratio summarizes observed experience on a chosen exposure basis.
Worked Example: Why Group Rates Must Be Exposure-Weighted
A hypothetical insurer observes 40 small commercial policies for one year and records 8 claims. It also observes 20 seasonal policies for half a year and records 6 claims. Calculate each group’s rate and the pooled rate. Compare the pooled rate with the unweighted average of the two group rates.
Method. Calculate exposure and rate for each group, then pool by totaling claims and exposure.
First group. Exposure is \(40\times1=40\) policy-years, so its rate is \(8/40=0.20\) claims per policy-year.
Second group. Exposure is \(20\times0.5=10\) policy-years, so its rate is \(6/10=0.60\) claims per policy-year.
Pooled rate. Total claims are \(8+6=14\), and total exposure is \(40+10=50\) policy-years. Therefore,
Comparison. The unweighted average of the group rates is \((0.20+0.60)/2=0.40\), not 0.28. It gives the 10-exposure-unit seasonal group the same weight as the 40-exposure-unit group. The pooled rate instead weights the rates by exposure: \((40(0.20)+10(0.60))/50=(8+6)/50=0.28\).
Check. The pooled rate times total exposure is \(0.28\times50=14\) claims. The common mistake avoided is averaging group rates without accounting for their different exposure sizes.
Claim Count Is Not Claim Amount
Frequency describes how many claims occur. Severity describes the amount associated with a claim. To make the distinction explicit, let \(N\) be the number of claims and let \(X_j\) be the amount for claim \(j\). The total amount for the period is the sum of the amounts for the claims that occurred. If \(N=0\), this sum is defined to be zero:
This identity follows by adding the amount attached to every counted claim event. It does not say that the total amount can be found from the count alone: the individual amounts are also needed. Nor does a large total amount necessarily mean there were many claims; a small number of large claims could produce it.
Worked Example: Separating Count and Dollars
Three hypothetical insured warehouses are observed for one year. Warehouse A has two qualifying claims with amounts of $1,200 and $800. Warehouse B has no claims. Warehouse C has two qualifying claims with amounts of $2,400 and $1,600. Find the number of insured risks, the claim count, the total claim amount, and the average amount per claim.
Method. Count covered risks and claim events separately. Add the claim amounts only after identifying the four events.
Insured risks. There are three warehouses.
Claim count. Warehouse A contributes 2 claims, Warehouse B contributes 0, and Warehouse C contributes 2. Therefore \(N=2+0+2=4\) claims.
Total claim amount. Add the four amounts: \(1{,}200+800=2{,}000\), \(2{,}400+1{,}600=4{,}000\), and \(2{,}000+4{,}000=\$6{,}000\).
Average amount per claim. Divide the total amount by the number of claims: \(\$6{,}000/4=\$1{,}500\) per claim.
Check. The four amounts average $1,500 because \(4\times\$1{,}500=\$6{,}000\). The number of risks is 3, the claim count is 4, and the aggregate amount is $6,000; none can be substituted for another. The common mistake avoided is assuming one claim per insured risk.
Common Interpretation Traps
- Using the number of risks as the claim count. A risk can have zero, one, or multiple claims in the period.
- Using policy count instead of exposure. This is inappropriate for a rate per policy-year when coverage durations differ.
- Calling a count rate a probability. Claims per exposure unit can exceed one; a probability cannot.
- Confusing claim count with claim dollars. Counts measure events; amounts measure the dollars attached to events.
- Leaving the counting rule unstated. Occurrence date, reporting date, and claim-record conventions can affect which period receives a claim.
A reliable setup states the claim event definition, the covered risks, the observation period, and the exposure unit before doing arithmetic. Only after those choices are clear should a frequency distribution or a model for claim amounts be applied.
Check Your Understanding
For each situation, distinguish the relevant counts and calculate the requested frequency where exposure is given.
- A hypothetical fleet has 90 scooters covered for one full year and records 12 claims. State the number of insured risks and claims, then calculate claims per scooter-year. (Answer: 90 insured risks, 12 claims, 0.1333 claims per scooter-year.)
- A hypothetical insurer covers 50 homes for one year and 24 homes for six months. It records 16 claims. Find the total exposure and observed claims per home-year. (Answer: 62 home-years and \(16/62\approx0.2581\) claims per home-year.)
- One insured research vessel has three qualifying claims during a year, with amounts of $700, $1,100, and $2,200. Find the number of risks, claim count, total amount, and average amount per claim. (Answer: 1 risk, 3 claims, $4,000 total, and $1,333.33 per claim.)
- A book has 40 policies covered for one year with 8 claims, and 20 policies covered for half a year with 6 claims. Find the pooled frequency and explain why the simple average of the group rates is not the pooled rate. (Answer: 0.28 claims per policy-year; the simple average is 0.40 because it gives unequal-exposure groups equal weight.)
- A hypothetical portfolio contains 100 policies, each covered for two years. There are 31 claims, arising from 17 distinct policies. Find the number of risks, claim count, total policy-year exposure, and observed frequency. (Answer: 100 risks, 31 claims, 200 policy-years, and 0.155 claims per policy-year.)