Tutorials › Exam FAM - Fundamentals of Actuarial Mathematics › Estimating Claim Frequency per Exposure Unit

Home  /  Tutorials  /  Exam FAM / Empirical Estimation from Data

Exam FAM · Empirical Estimation from Data · Tutorial 825 of 988

Estimating Claim Frequency per Exposure Unit

Learn to calculate an exposure-based claim rate, express it in a convenient unit, and apply it consistently to estimate expected counts.

Intermediate 10 min read

What You'll Learn

  • Calculate an empirical claim rate as observed claims divided by total exposure.
  • Convert a rate between exposure units without changing its meaning.
  • Convert a portfolio size and coverage period into matching exposure units.
  • Use an estimated rate to project an expected claim count.
  • Identify denominator and time-unit errors that distort a frequency estimate.

Why Claim Frequency Uses Exposure

A claim count alone is hard to interpret. For example, 30 claims might represent a high frequency in a small group observed briefly, or a low frequency in a large group observed for many years. To compare experience or estimate claims for a target portfolio, relate the count to the amount of time the insured units were exposed to risk.

An exposure unit is the chosen denominator for that comparison. Common choices include a policy-year, vehicle-year, machine-year, or policy-month. A policy-year represents one policy exposed for one year; 500 policies covered for a full year contribute 500 policy-years. The unit must fit the data and the question, and the numerator and denominator must refer to the same claim definition and covered population.

Definition: If \(N\) claims are observed over \(e\) exposure units, the empirical claim frequency per exposure unit is \(\hat{\lambda}=N/e\). If the requested rate is per \(k\) exposure units, it is \(k\hat{\lambda}=kN/e\). The hat indicates an estimate from observed data, not a known underlying rate.

The result is a rate, not a claim count. For example, \(0.02\) claims per machine-year means that the observed count was two claims per 100 machine-years. It does not mean that each machine will make exactly \(0.02\) claims, nor does it specify the probability that a machine has at least one claim. Those are different questions.

From Observed Rate to Expected Count

To project a claim count, first express the target portfolio and its coverage period in the same exposure unit used for the estimated rate. Then multiply the target exposure by the rate per exposure unit. The calculation gives an expected count, which can be fractional; it is an average-level projection, not a promise that the realized count will equal it.

$$ \widehat{\text{expected claims in target}} = \hat{\lambda}\times e_{\text{target}} = \frac{N_{\text{observed}}}{e_{\text{observed}}}\,e_{\text{target}}. $$

This plug-in estimate is meaningful when the observed experience is relevant to the target: the claim definition, insured population, coverage, and conditions affecting frequency should be sufficiently comparable. If the underlying rate changes materially between the observation and projection periods, simple multiplication by the historical rate does not account for that change. The calculation itself does not require a probability distribution for claim counts; it estimates a mean count from an observed rate and a target exposure.

The conversion of exposure units is often the most important step. If a rate is quoted per 100 policy-years, target exposure in policy-years must be divided by 100 before multiplying by the quoted rate. Alternatively, convert the rate to claims per policy-year and multiply directly. Both routes must give the same result.

Why the Exposure-Based Calculation Works

The logic follows from the meaning of a stable rate. Suppose each part of a portfolio contributes exposure \(e_j\), and the same underlying claim rate \(\lambda\) applies to each part. The expected claim count for part \(j\) is \(\lambda e_j\). Adding the parts adds both their expected counts and their exposures.

$$ E[N_{\text{total}}] = \sum_j E[N_j] = \sum_j \lambda e_j = \lambda\sum_j e_j. $$

Thus, if the total observed count is used to estimate the common rate, dividing by total exposure gives the empirical rate \(N_{\text{total}}/\sum_j e_j\). Applying that rate to target exposure estimates the target count. This reasoning relies on using a comparable rate across the included exposure. It does not justify combining unlike groups when their claim frequencies differ substantially; the resulting overall rate describes the combined experience, not necessarily either group on its own.

Worked Calculations

Worked Example: Claims per 100 Vehicle-Years

Method and assumptions. A hypothetical motor insurer observes 27 claims over 1,800 vehicle-years. Assume the claims and exposure are defined consistently and use the observed rate as a suitable estimate for a target portfolio of 6,400 vehicle-years. Find the observed rate per 100 vehicle-years and the estimated target claim count.

Step 1: Find the rate per vehicle-year. Divide the 27 observed claims by the 1,800 vehicle-years:

$$ \hat{\lambda}=\frac{27}{1800}=0.015 \quad\text{claims per vehicle-year}. $$

Step 2: Express it per 100 vehicle-years. Multiply by 100:

$$ 100(0.015)=1.5 \quad\text{claims per 100 vehicle-years}. $$

Step 3: Project the target count. Using the per-vehicle-year rate gives \(6{,}400(0.015)=96\) expected claims. As a check, the target has \(6{,}400/100=64\) groups of 100 vehicle-years, so \(64(1.5)=96\) expected claims.

Answer and check. The estimate is 1.5 claims per 100 vehicle-years, and the projected count is 96. Both calculation routes agree. The common mistake is to multiply 6,400 by 1.5, even though 1.5 is a rate per 100 vehicle-years rather than per single vehicle-year.

Worked Example: Converting Machine-Months to Machine-Years

Method and assumptions. A hypothetical equipment coverage portfolio records 46 claims over 2,300 machine-years. Use this rate to estimate claims for 375 machines covered continuously for eight months. Assume each machine is exposed for the stated period and the historical rate is appropriate for this target.

Step 1: Calculate the historical rate.

$$ \hat{\lambda}=\frac{46}{2300}=0.02 \quad\text{claims per machine-year}. $$

Step 2: Convert the target coverage period to years. Eight months is \(8/12=2/3\) of a year. Therefore, 375 machines contribute

$$ e_{\text{target}}=375\left(\frac{8}{12}\right)=375\left(\frac{2}{3}\right)=250 \quad\text{machine-years}. $$

Step 3: Estimate the count.

$$ \widehat{\text{expected claims}}=0.02(250)=5. $$

Answer and check. The projected expected count is 5 claims. Equivalently, the observed rate is 2 claims per 100 machine-years, and the target is 2.5 blocks of 100 machine-years, giving \(2(2.5)=5\). A common mistake is to treat eight months as eight years, which would produce \(0.02(375)(8)=60\) and overstate the exposure by a factor of 12.

Worked Example: Combining Compatible Data Recorded in Different Units

Method and assumptions. A hypothetical property program has one observation summarized as 9 claims over 1,800 policy-months and another as 36 claims over 1,200 policy-years. Assume these observations refer to comparable policies, claim definitions, and risk conditions, so a common rate is appropriate. Estimate the combined rate per 100 policy-years, then project claims for 420 policies covered for nine months.

Step 1: Convert all observed exposure to policy-years. Since 12 policy-months equal one policy-year, the first exposure is

$$ 1800\div 12=150 \quad\text{policy-years}. $$

The second exposure is already 1,200 policy-years, so combined exposure is \(150+1{,}200=1{,}350\) policy-years. The combined observed count is \(9+36=45\) claims.

Step 2: Calculate the combined rate.

$$ \hat{\lambda}=\frac{45}{1350}=\frac{1}{30} =0.033333\ldots \quad\text{claims per policy-year}. $$

Per 100 policy-years, this is \(100/30=3.333333\ldots\) claims.

Step 3: Find target exposure. Nine months is \(9/12=3/4\) of a year, so the target exposure is

$$ 420\left(\frac{9}{12}\right)=420\left(\frac{3}{4}\right)=315 \quad\text{policy-years}. $$

Step 4: Project the count.

$$ \widehat{\text{expected claims}}=\frac{1}{30}(315)=10.5. $$

Answer and check. The estimated rate is 3.333 claims per 100 policy-years, and the projected expected count is 10.5 claims. The check using the displayed rate is \(315/100\times 3.333333\ldots=10.5\). The fractional answer is valid as an expected count. The main trap is adding 1,800 policy-months directly to 1,200 policy-years; the denominators must be converted to one common unit before combining exposures.

Common Errors to Avoid

A frequency estimate is only as interpretable as its numerator, denominator, and unit. Check each one before multiplying by target exposure.

  • Using the number of insured units instead of exposure. If coverage lasts for only part of a year, the exposure is less than one policy-year per policy. Multiply the number of units by the fraction of the year covered.
  • Mixing units. Claims per 100 policy-years cannot be multiplied directly by a target stated in policy-years. Convert either the rate or the target exposure first.
  • Adding incompatible exposure measures. Convert policy-months, vehicle-years, or other units into a common unit before combining them. Do not add a count of months to a count of years.
  • Confusing an estimate with a guaranteed count. Multiplying a rate by exposure gives an estimated expected count. Actual observed claims may be greater or smaller.
  • Ignoring comparability. A historical rate may not be suitable for a target with different coverage or materially different risk characteristics. The arithmetic does not correct for such differences.
Key takeaway: Divide observed claims by observed exposure to obtain a rate, state the exposure unit explicitly, convert target exposure to that same unit, and multiply. Keep the rate, exposure, and claim definition consistent throughout.

Check Your Understanding

For each problem, keep the claim count and exposure in matching units before estimating the target count.

  1. A hypothetical boat portfolio has 18 claims over 900 boat-years. Find the rate per 100 boat-years and the expected count over 1,250 boat-years at that rate. (Answer: 2 claims per 100 boat-years; 25 expected claims)
  2. A device program records 32 claims over 4,000 device-months. Find the rate per 100 device-years, then estimate claims for 160 devices covered for 15 months. (Answer: 9.6 claims per 100 device-years; 19.2 expected claims)
  3. A hypothetical service plan has 63 claims over 2,100 unit-years. Estimate the rate per unit-year and the expected count for 700 units covered for two years. (Answer: 0.03 claims per unit-year; 42 expected claims)
  4. A portfolio records 12 claims over 6,000 policy-months. Find the rate per 100 policy-years and estimate the count for 250 policies covered for two years. (Answer: 2.4 claims per 100 policy-years; 12 expected claims)
  5. Two comparable hypothetical groups have 14 claims over 700 policy-years and 22 claims over 1,100 policy-years. Find their combined rate per 100 policy-years and estimate claims for 850 policies covered for 18 months. (Answer: 2 claims per 100 policy-years; 25.5 expected claims)

Previous

Next

Back to Tutorials
Clinical Biostats