Selecting Consistent Frequency and Severity Trend Periods

Why the Trend Periods Must Be Consistent

A projected loss cost combines two quantities: expected claim frequency and expected severity per claim. Those quantities may come from data measured over different periods or represented by different central dates. Each must be projected from its own measurement date to the same prospective endpoint before they are multiplied. Otherwise, the combined result mixes cost levels from different dates.

For example, suppose frequency is measured two years before the prospective endpoint while severity is measured one year before it. The appropriate frequency trend period is two years, but the appropriate severity trend period is one year. Applying two years of trend to both components would add an extra year of severity trend; applying one year to both would omit a year of frequency trend.

Definition: A trend interval for a component is the elapsed time from that component’s measurement date (or stated central date) to the selected prospective endpoint. Frequency and severity intervals are consistent when each begins at its own correct measurement date and both end at the same prospective date. Their lengths need not be equal.

A common date convention makes the alignment visible. Set the frequency measurement date at time $t_F$, the severity measurement date at time $t_S$, and the prospective endpoint at time $T$. Then the trend periods are $T-t_F$ for frequency and $T-t_S$ for severity. A date’s position on this timeline—not whether the component is called “frequency” or “severity”—determines its trend period.

Model and Calculation

Assume frequency is measured as claims per exposure, severity as dollars per claim, and both refer to compatible claim and exposure definitions. Assume, for the calculation below, that each component follows a constant multiplicative annual trend: a rate $r$ changes the level by a factor of $1+r$ for each year. A fractional-year period $n$ uses the factor $(1+r)^n$, which assumes geometric interpolation of that annual trend. If trend rates vary by year or subperiod, use the applicable successive factors for the actual interval instead.

Result: If frequency $f$ is measured $n_F$ years before the endpoint with annual trend $r_F$, and severity $s$ is measured $n_S$ years before it with annual trend $r_S$, the projected loss cost per exposure is

\text{Projected loss cost} = \left[f(1+r_F)^{n_F}\right] \left[s(1+r_S)^{n_S}\right].

Here $n_F$ and $n_S$ are each measured from the corresponding component’s data date to the same endpoint. The expression assumes the component trend rates are suitable for the coverage and exposure basis being projected. It does not correct changes in coverage, claim definitions, exposure mix, or other data-basis differences; those need to be addressed separately.

The formula follows from the definition of loss cost. At the endpoint, expected claims per exposure are the projected frequency, while expected dollars per claim are the projected severity. Multiplying claims per exposure by dollars per claim cancels “claims,” leaving dollars per exposure. The two trend factors multiply because each changes a different component of that product.

\begin{aligned} \text{Loss cost at endpoint} &= \text{frequency at endpoint} \times \text{severity at endpoint} \\ &= f(1+r_F)^{n_F}\,s(1+r_S)^{n_S} \\ &= fs(1+r_F)^{n_F}(1+r_S)^{n_S}. \end{aligned}

This derivation also shows why trend factors must be applied to the correct components. The combined factor is not generally a single trend rate applied over a single interval: it is the product of the factor for frequency over its interval and the factor for severity over its interval.

Worked Applications

Worked Example: Matching Both Components to One Endpoint

A hypothetical small-equipment policy has frequency of $0.080$ claims per exposure and severity of $\$1{,}500$ per claim, both measured two years before the prospective endpoint. Use annual frequency trend of $4\%$ and annual severity trend of $2\%$. Find projected loss cost per exposure.

Method. Both components have the same measurement date, so both trend periods are two years. Project each component separately before multiplying.

Step 1: Project frequency.

0.080(1.04)^2 =0.080(1.0816) =0.086528 \text{ claims per exposure}.

Step 2: Project severity.

\$1{,}500(1.02)^2 =\$1{,}500(1.0404) =\$1{,}560.60 \text{ per claim}.

Step 3: Multiply the projected components.

0.086528(\$1{,}560.60) =\$135.0355968 \approx \$135.04 \text{ per exposure}.

Check. The untrended loss cost is $0.080(\$1{,}500)=\$120$. The combined factor is $(1.04)^2(1.02)^2=1.12529664$, and $\$120(1.12529664)=\$135.0355968$, matching the component calculation. The result is rounded to the nearest cent. This method avoids treating the frequency trend rate as if it were the severity rate.

Worked Example: Different Measurement Dates

A hypothetical delivery-van program has frequency of $0.10$ claims per exposure measured two years before the prospective endpoint. Severity is $\$2{,}000$ per claim measured 1.5 years before that endpoint. Annual frequency trend is $5\%$, and annual severity trend is $3\%$. Find projected loss cost per exposure.

Method. Use two years for frequency and 1.5 years for severity. The components do not have the same trend interval because they do not have the same measurement date.

Step 1: Project frequency over two years.

0.10(1.05)^2 =0.10(1.1025) =0.11025 \text{ claims per exposure}.

Step 2: Project severity over 1.5 years. The fractional-year factor is $1.03^{1.5}=1.04533596\ldots$, so

\$2{,}000(1.03)^{1.5} =\$2{,}000(1.04533596\ldots) =\$2{,}090.67192\ldots \text{ per claim}.

Step 3: Multiply and round.

0.11025(\$2{,}090.67192\ldots) =\$230.49658\ldots \approx \$230.50 \text{ per exposure}.

Check. The calculation projects each component to the stated endpoint, and the units multiply to dollars per exposure. A common error is to use two years for severity as well: that would project severity beyond the endpoint and overstate the loss cost. The correct answer, rounded to the nearest cent, is $\$230.50$.

Worked Example: Measuring the Effect of a Mismatched Interval

In a hypothetical specialty property portfolio, frequency is $0.060$ claims per exposure at time 0, while severity is $\$2{,}500$ per claim at time 1. The prospective endpoint is time 3. Assume annual frequency trend of $2\%$ and annual severity trend of $6\%$. Calculate the properly aligned loss cost, then show the result if severity is mistakenly trended from time 0.

Method. Frequency is measured three years before the endpoint; severity is measured two years before it. The incorrect calculation uses three years for both components, effectively moving the severity measurement date back by one year.

Step 1: Project frequency for three years.

0.060(1.02)^3 =0.060(1.061208) =0.06367248 \text{ claims per exposure}.

Step 2: Project severity for two years.

\$2{,}500(1.06)^2 =\$2{,}500(1.1236) =\$2{,}809.00 \text{ per claim}.

Step 3: Calculate the correctly aligned result.

0.06367248(\$2{,}809) =\$178.85599632 \approx \$178.86 \text{ per exposure}.

Step 4: Calculate the mistaken result. If severity is incorrectly trended for three years, it becomes

\$2{,}500(1.06)^3 =\$2{,}500(1.191016) =\$2{,}977.54.

Using the same projected frequency then gives

0.06367248(\$2{,}977.54) =\$189.587356\ldots \approx \$189.59 \text{ per exposure}.

Check and interpretation. The erroneous result is higher because it includes one extra year of positive severity trend. It does not represent the specified endpoint. The properly aligned answer is $\$178.86$ per exposure, rounded to the nearest cent. This comparison isolates the period error; it does not imply that the two components were observed on a common historical date.

A Practical Alignment Check

Before combining frequency and severity trends, write down the two measurement dates and the single endpoint. Then calculate each elapsed interval independently. This simple timeline check is especially useful when the data are described using different labels, such as “accident-year average severity” and “calendar-year frequency.” The labels alone do not establish the correct interval; the stated or assumed central dates do.

Identify what each component measures.
Confirm the frequency denominator and the severity claim basis so the product represents loss dollars per exposure.

Mark the measurement dates.
Record the date or central date associated with frequency and the date or central date associated with severity.

Choose the prospective endpoint.
Use one endpoint for both components, consistent with the intended projection.

Calculate the two intervals and factors.
For each component, use its own elapsed time and its applicable trend rate.

Project, multiply, and check.
Project frequency and severity separately, multiply them, and verify that the resulting units and endpoint are correct.

A useful exam check is to compare any stated trend period with the timeline. A period that starts before the component’s measurement date or ends after the common endpoint signals a mismatch. Also check the direction: positive trend over a positive elapsed period should increase the component; negative trend should decrease it.

Common Mistakes and Why They Matter

Forcing both components to use the same number of years. Equal intervals are appropriate only when the measurement dates are equal. Different measurement dates usually imply different interval lengths.
Using the endpoint as the start date. The interval runs from the component’s measurement date to the endpoint. If a component is already measured at the endpoint, its trend period is zero.
Applying trend to a combined loss cost without checking the component dates. A single combined trend factor can hide which interval each component requires. Projecting each component separately makes the assumptions visible.
Confusing a rate with a trend factor. An annual trend rate of $3\%$ over two years gives a factor of $(1.03)^2$, not $1+2(0.03)$ under the multiplicative model.
Ignoring fractional periods or changing units. If dates are in months, convert both intervals consistently to years when annual trend rates are used. Do not round a period prematurely if the calculation uses a fractional-year factor.

Period alignment matters because trend changes the cost level attributed to each component. An extra interval can create an apparent increase that comes only from the calculation, while an omitted interval can understate projected cost. A sound projection therefore requires both a defensible trend assumption and a correctly matched interval for each input.

Key takeaway: Frequency and severity do not have to share a measurement date or a trend-period length. They do have to be projected from their respective correct dates to the same prospective endpoint before they are combined.

Check Your Understanding

For each problem, identify the component-specific trend periods before calculating.

A hypothetical coverage has frequency $0.12$ claims per exposure measured two years before the endpoint and severity of $\$900$ measured one year before it. Annual trends are $3\%$ for frequency and $4\%$ for severity. Find projected loss cost per exposure. (Answer: $\$119.16$ per exposure)
A hypothetical repair plan has frequency $0.20$ measured at time 0 and severity $\$1{,}200$ measured at time 1. The endpoint is time 4. Annual trends are $3\%$ for frequency and $4\%$ for severity. Find projected loss cost per exposure. (Answer: $\$303.85$ per exposure)
For a hypothetical product, frequency is $0.050$ at time 0, severity is $\$3{,}000$ at time 1, and the endpoint is time 2. Annual trends are $10\%$ and $5\%$, respectively. Find the aligned projected loss cost. (Answer: $\$190.58$ per exposure)
A hypothetical book has frequency $0.080$ at January 1, 2023, and severity of $\$1{,}000$ at July 1, 2023. The endpoint is January 1, 2025. Annual trends are $2\%$ for frequency and $4\%$ for severity. Use 2 years for frequency and 1.5 years for severity. (Answer: $\$88.28$ per exposure)
A hypothetical product has frequency $0.10$ already measured at the endpoint and severity of $\$500$ measured two years earlier. Annual trends are $3\%$ for frequency and $4\%$ for severity. Find projected loss cost, and state why frequency receives no trend. (Answer: $\$54.08$ per exposure; frequency is already at the endpoint)

What You'll Learn