Aligning Policy-Year Experience With Trend

Why Policy-Year Timing Matters

A policy-year loss total combines losses from policies that began at different times. Even when all policies have one-year terms, losses attached to an early effective date and losses attached to a late effective date generally occur at different times. To project the experience to a prospective cost level, the trend interval must reflect when the losses occurred—not simply the beginning or end of the policy year.

The relevant date may be the actual loss occurrence date, if the data are available and appropriate for the projection. When losses are assumed to be spread uniformly through each policy term, the term midpoint is a conventional estimate of their average occurrence date. That assumption is useful, but it should be stated: different claim timing, seasonal effects, or a concentration of large losses can make a midpoint unrepresentative.

Definition: The policy effective date is the date coverage begins. A policy-year experience period groups policies by their effective dates, but the losses in that period can occur during the policies’ terms. The trend interval for a loss is the time from its occurrence date (or an explicitly selected representative occurrence date) to the prospective trend endpoint.

For an annual effective trend rate $r$, a loss amount $L$ at date $t$ projected to date $T$ is multiplied by $(1+r)^{T-t}$, where $T-t$ is measured in years. Fractional-year intervals use fractional exponents under this convention. Use the same date and year-fraction convention throughout a calculation.

L_T = L(1+r)^{T-t}

Finding a Representative Loss Date

Suppose an annual policy begins on date $e$, and losses are assumed to occur uniformly over its one-year term. Its representative loss date is six months after $e$, not at $e$ itself. More generally, for a policy term of length $q$ years with uniform loss timing, the representative date is $e+q/2$. The interval to prospective date $T$ is therefore $T-e-q/2$.

\text{Representative loss date}=e+\frac{q}{2}, \qquad \text{trend interval}=T-e-\frac{q}{2}

The midpoint follows from the assumed timing distribution. If loss occurrence time $U$ is uniform from 0 to $q$ years after inception, then its mean is

E[U]=\int_0^q u\frac{1}{q}\,du =\frac{1}{q}\left[\frac{u^2}{2}\right]_0^q =\frac{q}{2}.

Thus the midpoint is the average occurrence time under uniform timing. It is not necessarily an exact way to trend a total of many losses. Since the trend factor is nonlinear in time, in general $E[(1+r)^{T-U}]$ is not equal to $(1+r)^{T-E[U]}$. The midpoint method is a timing approximation; date-level losses, when available and suitable, can instead be trended using their own dates.

Worked Example: Two Annual Policy Cohorts

A hypothetical insurer groups annual policies by effective date. One group takes effect April 1, 2024, and has $25,000 of losses; a second takes effect October 1, 2024, and has $35,000 of losses. Assume losses occur uniformly during each policy term. Project the losses to October 1, 2026, at an 8% annual effective trend rate.

Method. Use each group’s policy-term midpoint as its representative loss date, then compound each loss from that date to the target date.

April group. Its representative date is October 1, 2024. The interval to October 1, 2026, is 2 years. The factor is $1.08^2=1.1664$, so the projected amount is $25{,}000(1.1664)=\$29{,}160.00$.

October group. Its representative date is April 1, 2025. The interval to October 1, 2026, is 1.5 years. The factor is $1.08^{1.5}=1.122368923$ to the shown precision, so the projected amount is $35{,}000(1.122368923)=\$39{,}282.91$, rounded to cents.

Adding the unrounded projected amounts gives $\$29{,}160.00+\$39{,}282.9123=\$68{,}442.9123$, or $68,442.91 to the nearest cent. The calculation uses a 1.5-year interval for the second group because its losses are centered on April 1, 2025, not on its October 1, 2024 effective date.

Policies Effective Throughout the Year

When policies take effect throughout a year, one policy-year total can represent losses from many different exposure periods. A useful calculation separates the experience into cohorts with known effective dates and loss amounts. For cohort $j$, let $L_j$ be its losses and $m_j$ its selected representative loss date. Project and sum the cohorts:

L_T=\sum_j L_j(1+r)^{T-m_j}

If each cohort consists of annual policies with uniform loss timing, $m_j$ is six months after that cohort’s effective date. Do not assign every cohort the same midpoint merely because all the policies belong to the same policy year.

Worked Example: Quarterly Effective-Date Cohorts

A hypothetical book has four annual-policy cohorts, effective January 1, April 1, July 1, and October 1, 2024. Each cohort has $20,000 of losses, assumed uniform over its policy terms. Project to January 1, 2027, at an 8% annual effective trend rate.

Method. Determine each cohort’s policy-term midpoint and its interval to January 1, 2027. Apply the fractional-year trend factor to each cohort separately.

Effective date	Representative loss date	Interval to January 1, 2027	Projected losses
January 1, 2024	July 1, 2024	2.5 years	$20,000(1.08)^2.5 = $24,243.17
April 1, 2024	October 1, 2024	2.25 years	$20,000(1.08)^2.25 = $23,781.18
July 1, 2024	January 1, 2025	2 years	$20,000(1.08)^2 = $23,328.00
October 1, 2024	April 1, 2025	1.75 years	$20,000(1.08)^1.75 = $22,883.45

For example, the first factor is $1.08^{2.5}=1.212158437$, and the second is $1.08^{2.25}=1.189059123$. Summing the four projected amounts before rounding gives $\$94{,}235.8049$, or $94,235.80 to the nearest cent. The check is that earlier representative loss dates receive longer trend intervals and therefore larger factors.

Use Actual Loss Dates When They Are the Appropriate Basis

The policy effective date is a way to locate coverage, not automatically the date on which a loss occurred. If the experience provides actual loss occurrence dates, and those dates are appropriate for the selected trend, use them rather than substituting a policy-term midpoint. This is particularly important when losses are concentrated in a part of the term or when a large claim materially affects a cohort total.

Worked Example: Known Occurrence Dates Within Policy-Year Experience

A hypothetical policy-year portfolio includes three losses: $10,000 occurring January 1, 2022; $20,000 occurring January 1, 2023; and $30,000 occurring January 1, 2024. Project all three to January 1, 2026, using a 5% annual effective trend rate.

Method. Use the supplied occurrence date for each loss, rather than inferring dates from policy effective dates.

The first loss has a 4-year interval: $10{,}000(1.05)^4=10{,}000(1.21550625)=\$12{,}155.0625$. The second has a 3-year interval: $20{,}000(1.05)^3=20{,}000(1.157625)=\$23{,}152.50$. The third has a 2-year interval: $30{,}000(1.05)^2=30{,}000(1.1025)=\$33{,}075.00$.

The unrounded total is $\$12{,}155.0625+\$23{,}152.50+\$33{,}075.00=\$68{,}382.5625$, or $68,382.56 to the nearest cent. This date-based answer depends on the stated occurrence dates; a uniform-within-term assumption would be a different timing model.

Keep Loss Timing Separate From Premium Treatment

Trending policy-year losses does not mean that premium should automatically receive the same trend factor. Loss trend projects the cost of losses from their experience-period timing to a prospective cost level. Premium is handled according to the rate level and exposure basis appropriate to the calculation. If premium is being used as a denominator, its amount and period must be identified consistently, but it is not made into a loss by multiplying it by the loss trend factor.

For a cohort analysis, document at least the policy effective dates, term lengths, exposure or loss weights, loss-timing assumption, trend rate, and prospective endpoint. If the midpoint assumption is used, say so. If a policy is written on one date but coverage becomes effective later, use the effective date to locate its covered exposure.

Common Mistake: Trending From Policy Inception

A frequent error is to treat the policy effective date as the occurrence date for all losses in the policy term. In the first worked example, that mistake would use intervals of 2.5 years for the April 1, 2024 cohort and 2 years for the October 1, 2024 cohort. The resulting incorrect calculation is

25{,}000(1.08)^{2.5}+35{,}000(1.08)^2 =\$30{,}303.96+\$40{,}824.00 =\$71{,}127.96.

That calculation applies too much trend under the uniform-loss-timing assumption. The correct representative dates are October 1, 2024, and April 1, 2025, giving intervals of 2 and 1.5 years and a total of $68,442.91. The policy effective date anchors the term; it does not by itself determine the loss occurrence date.

Key takeaway: Align each policy-year loss amount to the prospective trend endpoint using its actual occurrence date when appropriate, or an explicitly stated representative date when it is not known. For annual policies with uniform loss timing, the representative date is six months after the effective date; staggered cohorts therefore have different trend intervals.

Check Your Understanding

For each problem, identify the loss-timing basis before calculating the trend interval. Round dollar answers to cents.

An annual policy takes effect October 1, 2024, has $8,000 of losses assumed uniform over its term, and is projected to October 1, 2026, at a 6% annual effective trend rate. What is its projected loss amount? (Answer: $8,730.69)
Two hypothetical six-month policies take effect January 1 and July 1, 2025. Each has $10,000 of losses, uniform over its six-month term. Project their combined losses to January 1, 2027, at a 5% annual effective trend rate. (Answer: $21,520.20)
A $12,000 loss occurs July 1, 2023, and an $18,000 loss occurs January 1, 2024. Project both to January 1, 2026, at a 4% annual effective trend rate. (Answer: $32,705.04)
A $40,000 policy-year loss total comes from annual policies effective January 1, 2022, and is assumed uniform over their terms. Project it to January 1, 2025, at an 8% annual effective trend rate. (Answer: $48,486.34)
A policy-year loss total is $50,000, with a representative loss date 1.5 years before the prospective date. The corresponding premium has already been adjusted to the intended rate level and is $70,000. At a 4% annual effective loss trend rate, calculate the projected loss and the loss-to-premium ratio; do not trend the premium. (Answer: projected loss $53,029.80; loss-to-premium ratio 75.76%)

What You'll Learn