Accident, Policy, Calendar, and Trend Periods

Why Period Definitions Matter

A loss amount has more than one relevant date. A claim may arise from an accident during one calendar year, be covered by a policy that began in an earlier year, and be paid in a later year. Those dates place the claim in different periods for different analyses. Before trending experience, an actuary must know which date defines the experience period and where that period lies relative to the future period being priced.

This tutorial focuses on period definitions and timelines, not on estimating a trend rate. The key discipline is to label each endpoint and state what it represents. That prevents a common error: counting the same time both as historical adjustment and as prospective projection.

Three Ways to Assign Experience

An accident period follows the occurrence date. For example, losses from events occurring between January 1 and December 31 of a year belong to that accident year, even if some claims are reported or settled later. Accident-period data can be immature: a claim may develop after the year in which the accident occurred. The assignment date does not change merely because later payments or estimates change the amount ultimately recorded.

A policy period follows the selected policy date or coverage interval. When policies are grouped by effective year, a policy effective in 2024 is ordinarily assigned to policy year 2024, even if it covers events occurring in 2025. A policy period is not automatically the same as the calendar period in which its premium is earned or its claims are paid.

A calendar period follows the date used to record the experience. Calendar-year paid losses include payments made during that calendar year, which can relate to accidents from several earlier years. Calendar-year incurred losses also require a stated accounting or valuation convention: the amounts are those recorded for that calendar period under that convention. Do not treat “calendar-year loss” as a synonym for “loss from an accident in that year.”

Historical and Prospective Trend Periods

A trend analysis compares experience at one point in time with losses or costs expected at another. The experience period is the span from which the historical observations are drawn. Its relevant date for trending is often represented by an average or midpoint date, selected consistently with the experience being measured. The prospective period is the future period to which the analysis is intended to apply; its relevant date is likewise represented by an average or midpoint date when appropriate.

A historical trend interval moves from the experience date to a stated reference date, such as the date at which the current rate level is expressed. A prospective trend interval moves from that reference date to the relevant date of the future period. When the analysis defines the intervals this way, they meet at the reference date: the end of the historical interval is the start of the prospective interval.

Some analyses instead describe one full experience-to-future interval. That is the interval from the experience date directly to the future date. It is not an additional interval to be added to a historical adjustment that already covers part of the same dates. The analyst must say whether a duration is a historical interval, a prospective interval, or the full interval.

Finding Midpoints and Elapsed Time

For a uniform interval, the midpoint follows from averaging its endpoints. If time is measured on a continuous scale from \(a\) to \(b\), the average date is:

The first equality defines the average date under a uniform distribution of experience through time. Evaluating the integral gives \((b^2-a^2)/(2(b-a))\), which factors to \((a+b)/2\). Thus the midpoint is justified when experience is uniform across the interval. If exposures or measured experience are concentrated at particular dates, a simple midpoint can misrepresent their average timing.

For a set of dated observations \(t_j\) with nonnegative weights \(w_j\), the weighted average date is:

The weights must match the purpose of the analysis. For instance, exposure weights may be suitable when the target is an exposure-weighted average date; they should not be substituted automatically for claim or loss weights when a different measure is being analyzed.

Once the relevant dates are identified, elapsed time is measured from the start date to the end date. Use one consistent time unit, and do not count both a historical and prospective interval unless they are intended to be separate, consecutive segments. With consecutive intervals, the durations add to the duration of the full interval because the shared boundary is counted only as an endpoint, not as extra elapsed time.

Keeping Historical and Prospective Intervals Separate

The following example uses three dates: an experience midpoint, a reference date, and a future midpoint. They divide the full timeline into two consecutive intervals. This is the safest way to avoid overlapping periods when both historical adjustment and prospective projection are described.

A Practical Timeline Procedure

Before calculating any trend duration, write the dates in chronological order and label their roles. The process below separates the assignment of observations from the definition of the trend intervals.

When the Midpoint Needs Care

The midpoint of a date range represents experience well only when experience is reasonably uniform over that range. A policy-effective period may have uneven writings, and the experience may be concentrated in particular months. In that situation, the simple midpoint of the period boundaries is not necessarily the average date of the relevant exposures or losses. The weighted average date can provide a better summary, but the selected weights must correspond to the measure being analyzed.

For example, suppose a hypothetical portfolio has 100 exposure units effective January 1 and 300 units effective May 1. Measure time in months from January 1, so those dates have offsets 0 and 4. The exposure-weighted date is:

The result is April 1, not the midpoint between January 1 and May 1, which is March 1. The later effective date has three times as much exposure weight. This calculation illustrates why a period’s boundaries and the average date of its experience are related but not interchangeable.

Why These Distinctions Matter

Period definitions determine what a trend comparison means. Comparing accident-year losses with calendar-year paid losses without recognizing their different assignment rules can create an apparent timing shift that is not a change in underlying claim cost. Likewise, using a policy-period boundary as though it were the average date of earned exposure can misstate the distance to a future target date.

For an analysis aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus, make the period basis and endpoint convention explicit. A clear timeline often resolves the main difficulty before any trend calculation begins. The essential checks are: which date assigns the experience, what date represents each period, and whether the historical and prospective intervals are consecutive or overlapping.