Selecting Historical and Prospective Trend Periods

Why Trend Periods Are Measured Between Midpoints

A ratemaking trend connects the time represented by historical experience with the time represented by the rates being developed. Those are not generally the start or end dates of either period. When experience is reasonably spread across its interval, the historical period is represented by its midpoint. A prospective rate period is represented in the same way: by the midpoint of the period during which the rates will apply.

The duration used for a trend calculation is therefore the elapsed time from the historical midpoint to the prospective midpoint. It is not simply the time between the end of experience and the start of the prospective period. That gap omits part of each interval: experience is occurring before its midpoint, and the prospective rate period continues after its midpoint.

A reliable way to calculate the duration is to draw or write a timeline, identify the boundary dates for each interval, find each midpoint, and subtract the historical midpoint from the prospective midpoint. For annual trend calculations, convert the resulting months or days into years using a consistent convention.

Representing Periods on a Timeline

Treat an interval as running from its start boundary to its end boundary. For example, calendar years 2022 and 2023 together run from January 1, 2022, to January 1, 2024. The final date of 2023 is December 31, but the next boundary is January 1, 2024. Using boundaries avoids ambiguity about whether an endpoint is included.

For an interval with evenly distributed experience, the midpoint is halfway between its start and end boundaries. A one-year calendar period has a midpoint six months after its start; a two-year period has a midpoint one year after its start. A partial-year prospective period is handled the same way. Do not assume that every period midpoint falls on January 1 or July 1.

The subtraction is elapsed calendar time, not subtraction of year labels alone. For example, January 1, 2023, to October 1, 2024, is 21 months, or 1.75 years. If dates do not fall on month boundaries, use exact elapsed days or a stated day-count convention; do not silently treat a partial month as a full year.

Partial-Year Periods Require the Same Method

A rate period does not have to begin on January 1, and the historical experience period does not have to be a calendar year. Find each midpoint from its own endpoints before comparing dates. This matters because using a period’s start date or end date in place of its midpoint can shift the calculated duration by several months.

Why the Midpoint Is a Useful Representative Date

The midpoint convention gives a single representative date for an interval whose experience is spread through time. One justification is that, under uniform weighting, the average date in the interval is its midpoint. Put dates on a numerical time scale, let the interval start at \(a\) and end at \(b\), and let the relative weight per unit of time be constant. The average time is:

Thus the midpoint is the average date when the time weighting is uniform. Using the historical and prospective midpoints identifies the interval between the central dates represented by the two periods. This is the standard period-selection convention for a reasonably uniform experience measure.

The uniformity assumption matters. If experience is concentrated in a particular part of the interval, its representative date may not be the calendar midpoint. For instance, if most exposures or losses occur late in a period, the simple midpoint can misrepresent the timing. A weighted average date may be more appropriate when the data support one. The midpoint convention is a practical representation of the interval, not a claim that all observations occurred on that date.

Also distinguish the midpoint convention from a claim that averaging a nonlinear trend factor over every date in an interval always gives exactly the factor at the midpoint. The convention identifies the representative time for the period. It does not remove the need to consider the assumptions behind a trend model or the distribution of experience within the interval.

Common Errors and Practical Checks

Before using a selected duration in a trend calculation, check that the timeline matches the question’s experience and prospective periods. These are common sources of error:

• Using the gap between intervals. The gap from the end of experience to the start of the rate period is generally shorter than the midpoint-to-midpoint duration.
• Using the start or end of one period as its representative date. Find both midpoints independently, including for partial-year intervals.
• Counting by year labels only. A date in October is not equivalent to a date in January; calculate elapsed months or days.
• Mixing units. If the result is 21 months and the trend period is in years, convert it to \(21/12=1.75\) years.
• Assuming uniformity without checking the setup. A simple calendar midpoint is appropriate when the period’s experience is reasonably uniform or the problem directs that convention. A concentrated or changing distribution may call for a weighted representative date.