Why Period Labels Matter
A ratemaking analysis brings together historical observations, measures of insured risk, and a future period for which rates are intended. These are different roles on a timeline. A loss may be recorded in one experience window, an exposure measure may represent a different span, and the resulting rate may be intended for policies in force later still. Calling all three spans “the experience period” hides those distinctions.
The dates are important because a trend adjustment depends on the interval between the cost level represented by historical data and the future cost level being estimated. The interval must be specified, not guessed from a label such as “2023 data” or “next year’s rates.” The period definitions below provide a consistent way to classify dates. In practice, an organization’s data and exposure conventions should also be stated.
These definitions distinguish purpose, not necessarily length. Two periods can each cover twelve months and still have different start and end dates. Conversely, two periods can share dates while representing different things: one may describe historical losses, the other the exposure associated with those losses. A period label alone does not establish what the underlying records measure.
State Boundaries Before Comparing Periods
For clarity, this tutorial writes date intervals in the half-open form \([a,b)\): the start date \(a\) is included, and the end date \(b\) is excluded. Thus \([2023\text{-}01\text{-}01,2024\text{-}01\text{-}01)\) covers the twelve months of calendar year 2023. This convention avoids counting a boundary date in both adjacent intervals. Other conventions can be used, but they must be explicit and applied consistently.
A valuation date is a separate item: it is the date as of which the available or valued data are measured. It is not automatically the experience period’s end date, nor does it say which dates the observations represent. For example, losses valued on March 31, 2024, might be assigned to accidents occurring during calendar year 2022. The valuation date describes the data’s status; the experience period describes their assignment.
A useful first step is to write a short period statement for each input and output. Include the type of data, the date basis, the start and end dates, and, for the projection period, the intended application. This prevents ambiguous statements such as “use 2023 experience for 2025 rates” from concealing different exposure and target dates.
Ask whether the dates describe historical experience, represented exposure, or the future application of rates.
Say how the observations are assigned to time, for example by accident date or by the organization’s stated exposure convention.
Give the start and end dates and indicate whether endpoints are included or excluded.
Do not infer the projection dates from the experience dates or assume the exposure window is identical to either one.
Worked Example: Classifying Three Different Windows
A hypothetical insurer selects losses assigned by accident date during calendar year 2023. Its exposure file contains earned exposure assigned to the twelve months beginning July 1, 2022. The proposed rates are intended for a future twelve-month period beginning July 1, 2024. Classify each window and state its boundaries using the half-open convention. The method is role-based period classification; it avoids treating every twelve-month window as the same “year.”
Experience period. The losses are assigned by accident date to calendar year 2023, so the experience period is \([2023\text{-}01\text{-}01,2024\text{-}01\text{-}01)\). It is a historical loss-data window.
Exposure period. The earned exposure is assigned to the twelve months beginning July 1, 2022, so its represented period is \([2022\text{-}07\text{-}01,2023\text{-}07\text{-}01)\). It is an exposure window, not the loss experience window in this description.
Projection period. The rates are intended for the twelve months beginning July 1, 2024, so the target is \([2024\text{-}07\text{-}01,2025\text{-}07\text{-}01)\). It is a future application window.
Check. Each interval is twelve months long, but the starts are July 1, January 1, and July 1 in different years. Equal duration does not make the periods interchangeable. The classification does not, by itself, determine whether the loss and exposure data are suitably matched for an indication.
Why Experience and Exposure Boundaries May Differ
The experience window is determined by how the selected historical observations are assigned. The exposure window is determined by what dates the exposure measure represents and by its stated convention. Those assignments need not produce matching boundaries. A loss dataset could be grouped by accident date, while earned exposure is summarized over a fiscal year. A portfolio could also have an exposure window defined for a different analysis purpose than the chosen loss window.
A mismatch is not automatically an error. It is a fact to identify and evaluate. The analyst should not silently change one period’s label to make the dates look aligned. Nor should an exam candidate infer that a stated accident-year loss period means the exposure is also accident-year exposure. The data description must support that conclusion.
The projection period is different again: it describes the future period whose cost level is relevant to the intended rates. It is not the period in which the historical data were collected, and it is not necessarily the calendar year in which the analysis is performed. Stating “rates effective in 2025” may be insufficient if the rates apply to policies with different effective dates or coverages; the intended future window should be spelled out.
Worked Example: Separating the Valuation Date From the Experience Period
A hypothetical commercial auto analysis uses claims with accident dates from January 1 through December 31, 2022. The claims are valued as of March 31, 2024. Exposure is reported for the policy year beginning October 1, 2021, and ending October 1, 2022. Rates are intended for the period October 1, 2024, through October 1, 2025. Classify each date or interval. The method is to distinguish a data status date from the period represented; it avoids mistaking the valuation date for the end of experience.
Loss experience period. Since losses are assigned by accident date in 2022, the experience window is \([2022\text{-}01\text{-}01,2023\text{-}01\text{-}01)\).
Valuation date. March 31, 2024 is a single date as of which the claims are valued. It is neither a twelve-month experience period nor the projection period.
Exposure period. The policy-year exposure is stated as \([2021\text{-}10\text{-}01,2022\text{-}10\text{-}01)\). This is a twelve-month exposure window with boundaries different from the accident-year loss window.
Projection period. The intended future application is \([2024\text{-}10\text{-}01,2025\text{-}10\text{-}01)\).
Check. The valuation date is later than the experience period’s end, as expected for data that have been valued after the accidents occurred. The exposure and experience windows overlap, but neither is identical to the other. The dates alone do not establish whether the claims are fully developed or whether the inputs are appropriate for a rate indication.
Reference Dates and Trend Intervals
Period boundaries describe windows; a trend calculation also requires a specified starting cost-level date and a target cost-level date. If those reference dates are provided, the elapsed interval can be calculated. Do not substitute the beginning or end of a period for a reference date unless the problem or analysis explicitly directs that choice. Selecting which dates best represent a period is a separate analytical decision.
Under a constant effective annual trend rate \(t\), a cost level grows by the factor \(1+t\) over one year. Over \(n\) years of compounding, it grows by \((1+t)^n\). This follows because each year’s level is the prior level multiplied by \(1+t\): after two years the factor is \((1+t)(1+t)=(1+t)^2\), and repeating the multiplication \(n\) times gives \((1+t)^n\). For a fractional year, this formula assumes the same annual compound trend is applicable over that fraction, so the factor is \((1+t)^n\) with fractional \(n\). It is not a suitable shortcut when the trend varies over time or the analysis requires a different trend model.
Worked Example: Calculating a Factor Between Stated Reference Dates
A hypothetical property analysis has a historical experience period of \([2023\text{-}01\text{-}01,2024\text{-}01\text{-}01)\) and a projection period of \([2025\text{-}01\text{-}01,2026\text{-}01\text{-}01)\). For this calculation only, the problem specifies that the historical cost level is referenced to April 1, 2023, and the target cost level to January 1, 2025. Assume a constant effective annual trend rate of 4%. Find the factor. The method is compound trending between the given reference dates; it avoids assuming the period starts or ends are automatically the reference dates.
Step 1: Find the elapsed time. From April 1, 2023, to April 1, 2024 is 1 year. From April 1, 2024, to January 1, 2025 is 9 months, or \(9/12=0.75\) year. Therefore \(n=1+0.75=1.75\) years.
Step 2: Apply the compound factor. With \(t=0.04\),
The factor is approximately 1.0710, rounded to four decimal places. Thus, under the stated constant-trend assumption, the target cost level is about 1.0710 times the level at the specified historical reference date.
Check. The target date is later than the historical reference date, so a positive trend produces a factor greater than 1. The dates span less than two years, so the factor should be greater than one year’s factor of 1.04 but less than two years’ factor, \(1.04^2=1.0816\). The computed 1.0710 satisfies both checks.
Common Period-Labeling Traps
- Equating the valuation date with the experience period. A valuation date is a point in time; an experience period is a window to which observations are assigned.
- Assuming matching lengths mean matching dates. Two twelve-month periods can have different boundaries and represent different quantities.
- Calling the future rate application “experience.” Historical data belong to an experience period; the future application belongs to a projection period.
- Inferring a trend interval from a year label. “2023 to 2025” does not establish the exact start and end reference dates or elapsed time.
- Leaving the exposure convention implicit. State whether the exposure measure is written, earned, or otherwise assigned, and give its represented dates.
These distinctions matter because an adjustment built from the wrong dates may shift the estimated cost level too little or too much. Before calculating, write the experience, exposure, and projection periods separately, mark any valuation date separately, and identify the reference dates actually specified for the calculation. This timeline discipline is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.
Check Your Understanding
For each problem, classify the relevant periods and show how the stated dates determine the answer.
- A hypothetical liability analysis assigns claims by accident date in 2021, values them on June 30, 2023, and uses exposure reported for calendar year 2021. Rates are intended for July 1, 2024, through June 30, 2025. Identify the experience period, valuation date, exposure period, and projection period. (Answer: experience \([2021\text{-}01\text{-}01,2022\text{-}01\text{-}01)\); valuation date June 30, 2023; exposure \([2021\text{-}01\text{-}01,2022\text{-}01\text{-}01)\); projection \([2024\text{-}07\text{-}01,2025\text{-}07\text{-}01)\).)
- A hypothetical insurer has losses assigned to \([2022\text{-}07\text{-}01,2023\text{-}07\text{-}01)\), exposure assigned to \([2022\text{-}01\text{-}01,2023\text{-}01\text{-}01)\), and a target rate period of \([2024\text{-}01\text{-}01,2025\text{-}01\text{-}01)\). Which is the experience period, which is the exposure period, and which is the projection period? (Answer: respectively, July 1, 2022–July 1, 2023; January 1, 2022–January 1, 2023; and January 1, 2024–January 1, 2025, with the end dates excluded.)
- A cost level is specified at October 1, 2022, and projected to April 1, 2024. Under a constant effective annual trend of 3%, find the elapsed time and trend factor, rounded to four decimals. (Answer: \(1.5\) years; \((1.03)^{1.5}\approx1.0453\).)
- Historical losses are assigned to calendar year 2023 and valued as of December 31, 2024. The intended rate period is calendar year 2025. Is December 31, 2024 the experience-period end, the valuation date, or the projection-period end? (Answer: it is the valuation date; the experience period ends January 1, 2024, and the projection period is \([2025\text{-}01\text{-}01,2026\text{-}01\text{-}01)\).)
- A hypothetical analysis uses a stated experience interval of \([2023\text{-}04\text{-}01,2024\text{-}04\text{-}01)\), exposure interval of \([2023\text{-}01\text{-}01,2024\text{-}01\text{-}01)\), and projection interval of \([2025\text{-}07\text{-}01,2026\text{-}07\text{-}01)\). A candidate says all are “2023 data” because each begins or overlaps 2023. Identify the error and give the correct labels. (Answer: the label ignores distinct roles and boundaries; the first interval is experience, the second exposure, and the third projection.)