Why Counts Need a Time Period and an Exposure Base
A claim count is meaningful only when its scope is clear. “There were 12 claims” leaves important questions unanswered: over what period, for which policies, and under what definition of a claim? A count for a month cannot be compared directly with a count for a year unless the observation periods and the amount of risk exposed are also considered.
In frequency analysis, the claim count describes how many claim events occurred in a specified portfolio during a stated observation period. An exposure base measures how much risk was observed during that period. Dividing the count by exposure gives an observed frequency rate. The rate is a ratio with units, such as claims per policy-year; it is not itself a claim count.
A claim count is a nonnegative integer. A rate need not be an integer: 3 claims over 10 policy-years is 0.3 claims per policy-year. Nor is a frequency rate generally a probability. It is possible for a rate to exceed one, because a policy can generate more than one claim during an exposure period.
The claim-counting rule matters. A study might count each distinct covered claim event, each reported claim, or each claim arising from a particular coverage. Those rules can produce different counts. The examples below use the stated hypothetical claim events as given; they do not infer how a real insurer records or groups claims.
Exposure Units and Policy-Years
The exposure base should reflect both the number of risks observed and how long each was observed. A common base is the policy-year: one policy in force for one year contributes one policy-year of exposure. A policy in force for half a year contributes 0.5 policy-years. Thus, exposure is not necessarily the number of policies at a single date.
For a portfolio of policies with durations \(t_1,\ldots,t_m\), measured in years under the chosen time convention, total policy-year exposure is the sum of their durations:
This calculation assumes the listed durations represent the periods during which the policies are exposed under the study's definition. If a policy terminates, starts partway through the observation window, or has a gap in coverage, only the eligible time in force should contribute. A study should use a consistent convention for converting dates into fractions of a year, such as actual days divided by 365, when exact dates matter.
Other exposure units may be appropriate for other risks, such as vehicle-years or insured-item-years. The unit must match the risks being counted. The numerator and denominator must also refer to the same portfolio, coverage, and time window. Dividing a count of claims from one group by exposure from another does not produce a meaningful rate for either group.
Worked Example: Counting Claims Across Partial Policy Periods
A hypothetical insurer reviews four policies under one stated claim-counting rule. During the observation window, the policies were in force for 12, 12, 6, and 3 months, respectively. They generated 1, 0, 2, and 1 claims. Find the total claim count, total policy-year exposure, and observed rate per policy-year.
Method. Sum the claim events to obtain the count, convert each duration into years, then divide the count by the total exposure.
Claim count. The total is \(1+0+2+1=4\) claims.
Exposure. The four durations in years are \(12/12=1\), \(12/12=1\), \(6/12=0.5\), and \(3/12=0.25\). Therefore,
Rate. The observed rate is \(4/2.75=1.4545\ldots\) claims per policy-year, or 1.4545 claims per policy-year to four decimal places.
Check. The exposure is greater than two but less than three policy-years, and four claims divided by that exposure gives a rate between \(4/3\) and \(4/2\), as expected. The mistake to avoid is dividing by four policies and reporting \(4/4=1\) as the policy-year rate. That denominator ignores the unequal durations.
Changing the Observation Period
The count depends on the period being observed: extending a window can include additional claim events. A frequency rate also depends on its denominator, so comparisons across periods should use exposure measured over the corresponding periods. The rate describes the observed claims per exposure unit; by itself, it does not establish that a later period will have the same experience.
For example, a six-month count should not be compared with a twelve-month count as though they covered equal amounts of time. One useful comparison is to calculate claims per policy-year in each period. This puts the ratios on the same time-based unit when the underlying exposure is policy-years. It does not make the periods identical in every other respect, nor does it prove that their risk conditions were the same.
Worked Example: Comparing Two Half-Year Periods
A hypothetical group of 10 policies is continuously in force throughout a year. There are 3 claims in the first six months and 2 claims in the second six months. Calculate the count and policy-year rate for each half-year, then for the full year.
First half-year. Each of 10 policies contributes \(6/12=0.5\) years, so exposure is \(10(0.5)=5\) policy-years. The count is 3 and the rate is \(3/5=0.6\) claims per policy-year.
Second half-year. The exposure is also \(10(0.5)=5\) policy-years. The count is 2 and the rate is \(2/5=0.4\) claims per policy-year.
Full year. The count is \(3+2=5\) claims. The exposure is \(5+5=10\) policy-years, so the full-year rate is \(5/10=0.5\) claims per policy-year.
Check. The total-period rate equals the exposure-weighted combination of the half-year rates: \(0.6(5/10)+0.4(5/10)=0.5\). The raw counts, 3 and 2, are not rates; their difference alone does not account for the amount of exposure. The annualized figures are descriptive ratios, not guaranteed forecasts for a future year.
Pooling Experience Across Groups
When several groups have different exposures, the pooled rate is found by adding their counts and exposures separately and then dividing. This is equivalent to an exposure-weighted average of group rates. It is generally not the simple arithmetic average, which gives every group the same weight regardless of how much exposure it represents.
The weights in the last expression are the groups' shares of total exposure, \(E_j/\sum_k E_k\). This identity follows by substituting \(C_j=E_jr_j\) into the pooled count and dividing by total exposure. Consequently, a group with more exposure contributes more weight to the pooled rate.
Worked Example: Pooling Unequal Segment Exposures
Two hypothetical portfolio segments have the following experience: Segment A has 2 claims over 20 policy-years; Segment B has 6 claims over 180 policy-years. Find each segment rate, the pooled rate, and the unweighted arithmetic average of the segment rates.
Segment rates. For A, \(r_A=2/20=0.10\) claims per policy-year. For B, \(r_B=6/180=0.033333\ldots\) claims per policy-year.
Pooled rate by totals. The combined count is \(2+6=8\), and the combined exposure is \(20+180=200\) policy-years. Thus,
Unweighted average. Giving the two segment rates equal weight instead yields
Check and interpretation. The pooled rate is 0.04, not 0.0667. Segment B represents \(180/200=0.9\), or 90%, of total exposure, whereas Segment A represents only 10%. The pooled calculation therefore gives B more influence. The common mistake is to average the two segment rates equally and call that the pooled rate; this incorrectly gives the small segment the same influence as the segment with nine times as much exposure.
Choosing and Reporting a Rate
A rate is only interpretable when its unit is explicit. For instance, “0.25” could mean 0.25 claims per policy-year, 0.25 claims per vehicle-year, or another ratio entirely. State the numerator, denominator, period, and unit rather than reporting a bare number.
It can be useful to express a rate per 100 exposure units. If 12 claims occur over 48 vehicle-years, then the rate is \(12/48=0.25\) claims per vehicle-year, equivalently \(100(0.25)=25\) claims per 100 vehicle-years. The count remains 12; changing the scale of the rate does not change the count.
A rate per policy can also be calculated for a defined observation period, but that denominator answers a different question from policy-year exposure when policies have different durations. For example, claims divided by the number of policies gives claims per policy in the stated study, while claims divided by policy-years gives claims per unit of time in force. These quantities should not be given the same label.
The following reporting sequence helps prevent denominator errors:
State what is counted and which coverage or population is included.
Use the same start and end boundaries for the count and the matching exposure.
Add the eligible policy durations or other exposure amounts using a consistent convention.
Give the count as a number of claims and the ratio with its unit, such as claims per policy-year.
Common Exam Traps
- Using the number of policies as exposure automatically. This is appropriate only when the intended denominator is policies for the specified period or when each policy contributes the same relevant duration. With unequal policy durations, calculate the time-based exposure if the rate is requested per policy-year.
- Comparing counts over unequal periods. A longer observation period may produce a larger count simply because more exposure was observed. Keep the period explicit and compare rates with matched units when appropriate.
- Averaging group rates equally. The pooled rate is total claims divided by total exposure. An unweighted mean is not the pooled rate when segment exposures differ.
- Calling a rate a probability. Claims per policy-year is a frequency ratio, and it can exceed one. Do not interpret it as the probability that a policy has a claim.
- Mixing exposure units. Claims counted for policies should not be divided by an unrelated exposure base without a clear reason and matching definition.
Check Your Understanding
For each problem, identify the count and calculate the rate using the requested exposure unit.
- Five hypothetical policies are each observed for one full year. Their claim counts are 0, 2, 1, 0, and 1. Find the total count and rate per policy-year. (Answer: 4 claims; 0.8 claims per policy-year.)
- Eight hypothetical contracts are each in force for nine months, and the group has 3 claims during that period. Find total policy-year exposure and the rate per policy-year. (Answer: 6 policy-years; 0.5 claims per policy-year.)
- A hypothetical portfolio has 9 claims over 30 policy-years in one quarter and 6 claims over 20 policy-years in the next quarter. Find the pooled rate for the two-quarter period and each quarter's rate. (Answer: 15 claims over 50 policy-years; each rate and the pooled rate are 0.3 claims per policy-year.)
- Segment A has 3 claims over 12 vehicle-years, and Segment B has 5 claims over 100 vehicle-years. Find the pooled rate and the unweighted average of the two segment rates. (Answer: Pooled rate \(8/112\approx0.0714\) claims per vehicle-year; unweighted average \((0.25+0.05)/2=0.15\) claims per vehicle-year.)
- Twelve hypothetical policies are each observed for three months, during which 2 claims occur. Find the count, the rate per policy for the three-month observation period, and the rate per policy-year. (Answer: 2 claims; \(2/12=0.1667\) claims per policy over three months; exposure is 3 policy-years, so \(2/3=0.6667\) claims per policy-year.)