Why Frequency and Any-Claim Probability Differ
Claim frequency and the probability that a policy has at least one claim answer different questions. Frequency measures the expected number of claims per unit of exposure. The any-claim probability measures the chance that a policy or exposure unit has one or more claims during a specified period. The distinction matters whenever one policy can have multiple claims: expected claim counts include every claim, while an any-claim indicator records only whether the count is zero or positive.
A Poisson model gives a direct relationship between the two measures when its assumptions are appropriate. In particular, it converts an expected count over a specified exposure period into the probability of at least one claim during that period. The exposure period must be explicit: a rate quoted per policy-year cannot be used as if it were already a one-year probability.
For example, a frequency of \(0.20\) claims per policy-year means the expected claim count is \(0.20\) for one policy-year of exposure under the model. It does not mean that exactly 20% of policies have claims. Some policies can have more than one claim, and the frequency is an average count rather than a proportion of policies.
Deriving the Any-Claim Probability
The Poisson model assumes a constant claim rate per unit of exposure and independent claim occurrences over disjoint exposure periods. With these assumptions, the count over exposure \(t\) has Poisson mean \(\lambda t\). The any-claim event is the complement of having zero claims, so its probability follows from the probability of zero under that count model.
This derivation shows why the exposure amount appears in the exponent: it scales the expected count from a per-unit rate \(\lambda\) to the period mean \(\lambda t\). It also shows why frequency and probability are generally unequal. For any positive \(\lambda t\), \(1-e^{-\lambda t}\) is less than \(\lambda t\); the gap reflects the possibility of multiple claims being counted in the expected number but not adding extra units to the any-claim indicator.
The result is useful for translating a count-based rate into a probability for a specified exposure period. If a policy has a fraction of a year of exposure, use that fraction as \(t\) when \(\lambda\) is per policy-year. If the period is longer than a year, use the full amount of exposure. Do not substitute the annual rate alone unless the period is exactly one policy-year.
Worked Example: Converting Annual Frequency to a Claim Probability
A hypothetical small-business policy segment is modeled with a constant frequency of \(0.24\) claims per policy-year. Calculate the probability of at least one claim for one policy-year and for two policy-years of exposure. Also compare the two-year probability with the expected number of claims.
Method and assumptions. Use \(N(t)\sim\operatorname{Poisson}(\lambda t)\), with \(\lambda=0.24\) claims per policy-year. This assumes a constant rate over the period and a suitable Poisson claim-count process.
One policy-year. Here \(t=1\), so the mean claim count is \(\lambda t=0.24(1)=0.24\). Therefore,
The modeled probability of at least one claim is \(0.2134\), or about \(21.34\%\). The expected number of claims is \(0.24\), not \(0.2134\).
Two policy-years. Set \(t=2\), giving mean count \(0.24(2)=0.48\). Then
The expected count is \(0.48\) claims over two policy-years, while the probability of one or more claims is \(0.3812\). The probability is lower because a policy-year with two or more claims still contributes only one occurrence of the event “at least one claim.” Probabilities are rounded to four decimal places. Check: both probabilities are below 1, and the two-year probability exceeds the one-year probability, as expected for a longer exposure period.
Common mistake avoided: Treating \(0.24\) claims per policy-year as a \(24\%\) chance of a claim skips the Poisson conversion. The model gives \(1-e^{-0.24}=0.2134\), not \(0.24\).
Applying Frequency to a Portfolio
For a portfolio, the observed claim frequency is commonly calculated as total claims divided by total earned exposure. That ratio is measured in claims per exposure unit. If it is used as an estimate of a Poisson rate, the resulting any-claim probability is calculated with the rate and the relevant exposure period. The ratio itself is not the share of policies with at least one claim.
A related distinction is between the expected number of claims and the expected number of policies with claims. For \(m\) policies with the same exposure \(t\) and rate \(\lambda\), the expected total claim count is \(m\lambda t\). The expected number of policies with at least one claim is \(m(1-e^{-\lambda t})\). These expectations are different because a policy with multiple claims adds more than one to the total count, but only one to the count of policies with any claim.
Worked Example: From Portfolio Frequency to Expected Claimants
A hypothetical portfolio has 72 claims over 300 earned policy-years. Assume this frequency is used as the rate for a homogeneous Poisson model. For a new group of 500 policies, calculate the probability of at least one claim for a full policy-year and for six months. For the 500 full-year policies, calculate expected claims and expected policies with at least one claim.
Method. First estimate the rate as claims divided by exposure; then multiply by the applicable exposure duration to get the Poisson mean. Use the any-claim formula for the probability of a positive count.
Step 1: Estimate frequency.
Step 2: Full-year any-claim probability. For \(t=1\),
Step 3: Six-month any-claim probability. Six months is \(t=0.5\) policy-years, so the mean count is \(0.24(0.5)=0.12\). Thus,
Step 4: Expected values for 500 full-year policies. The expected total claims are \(500(0.24)=120\). The expected number of policies with at least one claim is \(500(0.2134)=106.7\), using the unrounded probability before rounding the result. The first expectation counts multiple claims on a policy separately; the second counts each policy at most once.
The probabilities are rounded to four decimal places, and the expected number of policies is rounded to one decimal. Check: 120 expected claims exceed 106.7 expected policies with claims, which is consistent with a model that permits multiple claims on one policy.
Common mistake avoided: Multiplying 500 by \(0.24\) gives expected claims, not expected claimants. For the expected number of policies with any claim, multiply by \(1-e^{-0.24}\).
Unequal Exposure Periods
When policies have different exposure durations, a single probability based on the average duration may not equal the average policy-level probability. Each policy’s any-claim probability depends on its own exposure amount through \(1-e^{-\lambda t}\). The total expected claim count can still be calculated by adding the expected counts across policies, but the expected number of policies with claims is found by adding their individual any-claim probabilities.
This distinction follows from the nonlinearity of the exponential function: in general, the average of \(1-e^{-\lambda t}\) over policies is not \(1-e^{-\lambda\bar{t}}\), where \(\bar{t}\) is their average exposure. The expression using \(\bar{t}\) gives the probability for a hypothetical policy with that exposure under the model; it need not give the portfolio’s average probability.
Worked Example: Different Policy Terms in One Portfolio
A hypothetical book has 100 policies with six months of exposure each and 50 policies with two years of exposure each. Assume a common Poisson frequency of \(0.20\) claims per policy-year. Find the expected total claims and the expected number of policies with at least one claim. Compare the portfolio’s average any-claim probability with the probability for a one-year policy.
Method and assumptions. Apply the same constant rate to each policy, using each policy group’s own exposure duration. The count for an individual policy with exposure \(t\) has mean \(0.20t\).
Step 1: Total exposure and expected claims. The first group contributes \(100(0.5)=50\) policy-years. The second contributes \(50(2)=100\) policy-years. Total exposure is \(50+100=150\) policy-years, so
Step 2: Expected policies with a claim. For a six-month policy, the probability is \(1-e^{-0.20(0.5)}=1-e^{-0.10}=0.0952\). The expected number in that group is \(100(0.0952)=9.52\). For a two-year policy, the probability is \(1-e^{-0.20(2)}=1-e^{-0.40}=0.3297\). The expected number in that group is \(50(0.3297)=16.49\). Adding the groups gives about \(9.52+16.49=26.01\) policies with claims, using probabilities rounded to four decimals. Using unrounded probabilities gives about \(26.00\).
Step 3: Compare average probabilities. The average across the 150 policies is approximately \(26.00/150=0.1733\). A one-year policy has probability \(1-e^{-0.20}=0.1813\). They differ because the book contains both shorter and longer terms, and averaging policy-level probabilities is not equivalent to applying the formula to average exposure.
The expected claim count is 30, while the expected number of policies with at least one claim is approximately 26.00. The slight difference between 26.01 from rounded intermediate probabilities and 26.00 from unrounded probabilities is a rounding effect; final expected values are reported to two decimals. Check: the longer-term group has the higher any-claim probability, and total expected claims exceed the expected number of policies with claims.
Common mistake avoided: Applying \(1-e^{-0.20}\) to all 150 policies treats every policy as having one year of exposure. The actual policy terms differ, so each term must be used in the exponent.
Assumptions, Interpretation, and Exam Traps
The conversion is only as appropriate as the Poisson model. Its standard interpretation assumes a constant claim rate over the exposure period and independent occurrence increments. A homogeneous rate also treats the policies being modeled as having the same underlying frequency after relevant differences have been addressed. If risk levels vary materially, a single portfolio-average rate may not describe each policy’s probability accurately. In that setting, the aggregate frequency alone is not enough to recover every policy-level probability.
The formula also describes the probability of at least one event in the modeled claim count. It does not by itself address claim severity, whether a claim is covered after applying a deductible or limit, or whether a reporting or settlement definition changes which claims are counted. The claim definition, exposure basis, and time period must be consistent between the frequency estimate and the probability being calculated.
- Do not label a count rate as a percentage. A frequency of \(0.24\) claims per policy-year is a count per exposure unit, not automatically a \(24\%\) any-claim probability.
- Keep time units consistent. If the rate is annual, convert months to fractions of a year before using the formula.
- Do not confuse expected claims with expected claimants. The former counts all claims; the latter counts policies with one or more claims.
- Do not assume average exposure gives the average probability. With unequal policy terms, calculate probabilities at the policy or group level and then aggregate.
The practical value of the distinction is that a ratemaking or portfolio question may ask for either a count-based measure or a policy-based measure. The former is naturally expressed as expected claims per earned exposure. The latter concerns the probability, or expected number, of policies experiencing any claim. Converting between them requires an explicit count model and a stated exposure period, not merely a change in units.
Check Your Understanding
For each problem, keep the frequency unit and exposure period explicit. Probabilities are rounded to four decimal places.
- A hypothetical equipment policy has Poisson frequency \(0.15\) claims per policy-year. What is the probability of at least one claim in one year? (Answer: \(0.1393\))
- A modeled frequency is \(0.40\) claims per policy-year. What is the probability of at least one claim over a three-month policy period? (Answer: \(0.0952\))
- A portfolio has 90 claims over 450 earned policy-years. Under a homogeneous Poisson model, find the expected claims and expected policies with at least one claim among 100 one-year policies. (Answer: frequency \(0.2000\) per policy-year; expected claims \(20\); expected policies with claims \(18.13\))
- A hypothetical policy has Poisson frequency \(0.30\) claims per policy-year. Over two years, what are the expected claim count and probability of at least one claim? (Answer: expected count \(0.60\); probability \(0.4512\))
- A group has 200 policies, each with six months of exposure, and a modeled frequency of \(0.12\) claims per policy-year. Find the expected total claims and the expected number of policies with at least one claim. (Answer: expected claims \(12\); expected policies with claims \(11.65\))