Why the Experience Period Matters
An indication uses historical experience to inform a prospective rate level. The experience period is the span of historical data selected for that analysis. A very short period may give too much weight to random fluctuations or unusual events. A very long period may include business, costs, or claim practices that no longer represent the risks expected to be insured. The choice is therefore a balance, not simply a rule to use the latest available years or the largest possible data set.
The selection should be defensible in terms of four questions: Is the data recent enough to be relevant? Is there enough exposure volume to limit random variation? Does the historical business resemble the future business? Are the records sufficiently complete and reliable? The best choice is the period that gives an appropriate answer to all four, subject to the data available.
Four Considerations in Selecting a Period
Recency matters because the objective is to inform future costs, not merely describe the distant past. Historical experience may become less representative as the cost environment, insured population, coverage, underwriting, or claims practices change. More recent observations often have greater relevance, but recency alone cannot rescue a period with too little volume or unreliable records.
Volume matters because observed experience varies randomly. More exposure generally gives a more stable average when the observations are reasonably comparable. However, additional exposure is useful only to the extent that it measures business relevant to the indication. A large amount of experience from a substantially different book may be less informative than a smaller amount from the current target business.
Relevance asks whether the historical risks and conditions resemble those expected in the business for which rates are being indicated. For example, a major change in coverage, underwriting rules, distribution, or the mix of insureds may weaken the relevance of older data. When only part of the historical portfolio is comparable, the analyst should assess that comparable portion rather than assume the entire historical book represents the future.
Completeness and reliability concern whether the exposure and loss records are present, consistently defined, and sufficiently reconciled for the intended analysis. A recent year with missing claim records, incomplete exposure files, or unresolved data errors may be less useful than an older, complete period. Missing observations must not be silently treated as zero losses or zero exposure.
| Consideration | Question to ask | Warning sign |
|---|---|---|
| Recency | How closely does this experience reflect the conditions relevant to future costs? | The period includes years before a material change in the business. |
| Volume | Is there enough comparable exposure to make the experience reasonably stable? | A short period is dominated by a small number of observations. |
| Relevance | Does the historical population resemble the future target business? | Older business has different coverage, selection, or claim handling. |
| Completeness | Are exposure and loss records present and reliable for the whole period? | Recent records are missing, unmatched, or not reconciled. |
Why More Data Can Help—and Still Mislead
A simple model explains the benefit of volume. Suppose each observation is an independent, comparable exposure unit with a loss-cost observation having mean \(\mu\) and variance \(\sigma^2\). For \(n\) such observations, the average has expected value \(\mu\). Independence gives the variance of the average:
The model assumes a common mean and variance and independent observations. Under those assumptions, increasing \(n\) reduces the variance of the average. This is the statistical reason that additional comparable experience can improve stability. It does not show that every older observation should be included: if the underlying mean has changed, the common-mean assumption fails.
The trade-off can also be stated in terms of mean squared error. If the average has variance \(V\) and its expected value differs from the future target mean \(\mu_F\) by a bias \(b\), then:
A longer period may reduce \(V\) by adding observations while increasing \(b^2\) if those observations come from a less relevant business or cost environment. There is no universal number of years that resolves this trade-off. The analyst must use the available evidence about volume, changes, and data quality.
When exposures differ by year, a pooled average should ordinarily reflect the exposure volume in each year rather than give each year's average equal weight. If \(L_t\) is the loss amount and \(E_t\) is the exposure for year \(t\), the pooled loss cost over a set of years is:
This ratio is exposure-weighted: years with more exposure contribute more observations to the pooled result. It is a descriptive comparison for the examples below, not by itself a complete rate indication.
A Practical Selection Process
State which future risks the indication is intended to represent. Identify material differences between that target and historical business.
Determine which candidate years have complete, reconciled exposure and loss records. Exclude or separately investigate periods that cannot support a dependable comparison.
For the usable periods, calculate exposure totals and descriptive experience measures. Check whether a shorter recent period has enough volume for the intended use.
Weigh the stability gained by longer experience against the risk of including less comparable years. Record why the selected period is preferable to the main alternatives.
This process does not require a mechanical score. A score can conceal important differences—for example, a severe completeness problem should not necessarily be offset by a high recency score. The decision should explain the material trade-offs in the context of the data.
Worked Example: Balancing Recency Against Exposure Volume
A hypothetical insurer is comparing complete annual experience for a relatively stable product. For this comparison, assume the records are consistently defined and the losses and exposures are directly comparable. The insurer considers a three-year period, a four-year period, and the latest two years. A planning guideline is to have at least 15,000 exposures. Which period is most defensible?
Step 1: Calculate each year's experience. The annual data are:
| Year | Exposure | Losses | Loss cost |
|---|---|---|---|
| 2021 | 4,000 | $1,000,000 | $250 |
| 2022 | 5,000 | $1,300,000 | $260 |
| 2023 | 6,000 | $1,620,000 | $270 |
| 2024 | 8,000 | $2,320,000 | $290 |
For example, the 2024 loss cost is $2,320,000 divided by 8,000, or $290 per exposure.
Step 2: Calculate the latest three-year pooled loss cost. Exposures are \(5{,}000+6{,}000+8{,}000=19{,}000\). Losses are \($1{,}300{,}000+$1{,}620{,}000+$2{,}320{,}000=$5{,}240{,}000\). The pooled loss cost is \($5{,}240{,}000/19{,}000=$275.79\), rounded to the nearest cent.
Step 3: Compare the alternatives. The latest two years have \(6{,}000+8{,}000=14{,}000\) exposures and \($1{,}620{,}000+$2{,}320{,}000=$3{,}940{,}000\) losses, giving \($3{,}940{,}000/14{,}000=$281.43\). That period is more recent, but it does not meet the stated 15,000-exposure guideline. The four-year period has \(23{,}000\) exposures and \($6{,}240{,}000\) losses, giving \($6{,}240{,}000/23{,}000=$271.30\).
Conclusion and check. The latest three years meet the volume guideline and omit the oldest year, so they are a reasonable choice if there is no evidence that the remaining years are materially unlike the target business. The four-year period adds 4,000 exposures but includes older experience; the two-year period is short of the guideline. The selected calculation checks because \(19{,}000 \times $275.79\) is approximately \($5{,}240{,}000\), with the small difference due to rounding. Method: compare candidate periods using pooled loss cost and explicit selection criteria. Common mistake avoided: averaging annual loss costs equally despite different exposure volumes.
Worked Example: Rejecting an Incomplete Recent Year
A hypothetical insurer has complete records for 2022 through 2024. For 2025, the exposure file contains 10,000 exposures, but the loss file is known to include only 75% of the records after a system conversion. The recorded 2025 losses are $2,400,000. The complete-year figures are: 2022, 7,000 exposures and $1,820,000 in losses; 2023, 8,000 exposures and $2,160,000 in losses; 2024, 9,000 exposures and $2,610,000 in losses. Compare the complete 2023–2024 experience with a naive 2024–2025 calculation.
Step 1: Calculate the complete two-year period. Exposures are \(8{,}000+9{,}000=17{,}000\). Losses are \($2{,}160{,}000+$2{,}610{,}000=$4{,}770{,}000\). The pooled loss cost is \($4{,}770{,}000/17{,}000=$280.59\), rounded to the nearest cent.
Step 2: Calculate the tempting but invalid comparison. Dividing the reported 2024–2025 losses by all reported exposures gives \($5{,}010{,}000/(9{,}000+10{,}000)=$5{,}010{,}000/19{,}000=$263.68\). This calculation treats the unrecorded 2025 losses as if they were zero, so it is not a valid experience measure.
Step 3: Decide what to use. The 2025 records are incomplete, and the supplied data do not establish the amount of missing losses. Therefore, the analyst cannot repair the comparison by simply using the known losses with the full exposure denominator. The complete 2023–2024 period is the defensible choice among these options until the 2025 records can be reconciled or otherwise supported.
Check: The complete-period totals are internally consistent: \(17{,}000 \times $280.59\) is approximately $4,770,000 after rounding. Method: screen candidate periods for completeness before comparing their averages. Common mistake avoided: interpreting missing losses as zero losses.
Worked Example: Giving Greater Weight to the Future Target Business
A hypothetical insurer has changed its product design and underwriting approach. Its planned future business is the new product. The older product produced 30,000 exposures and $7,500,000 in losses during 2021–2023. The new product has complete experience for 2024–2025: 12,000 exposures and $3,600,000 in losses. Assume the future book is expected to consist predominantly of the new product, and that the figures are comparable within each product. Should the older, higher-volume period automatically be preferred?
Step 1: Calculate the older product's average. The loss cost is \($7{,}500{,}000/30{,}000=$250\) per exposure.
Step 2: Calculate the new product's average. The loss cost is \($3{,}600{,}000/12{,}000=$300\) per exposure.
Step 3: Relate the calculations to the target. The older product has 18,000 more exposures, but it is not the product expected to dominate the future book. The recent new-product experience is more relevant to that target, and its records are complete. The new-product period is therefore the more defensible starting period for this hypothetical decision, while its lower volume should be disclosed as a limitation.
Check: The calculated loss costs recover the given totals: \(30{,}000 \times $250=$7{,}500{,}000\), and \(12{,}000 \times $300=$3{,}600{,}000\). Method: prioritize comparability to the future target while explicitly recognizing the volume trade-off. Common mistake avoided: selecting the period with the most exposure without asking whether it represents the future business.
Common Exam Traps and Documentation
A common trap is to treat the experience period as a purely numerical choice. For example, one might select the longest period because it has the most exposure, or the latest period because it is most recent, without checking whether the data are complete or representative. Both approaches can be wrong. Volume reduces random variation under appropriate comparability assumptions; it does not eliminate bias from stale or mismatched experience.
Another trap is to compare candidate periods using the unweighted average of annual loss costs when annual exposures differ. In Worked Example 1, the latest three annual loss costs are $260, $270, and $290, whose unweighted average is $273.33. The pooled loss cost is $275.79 because the years have different exposure volumes. If the goal is the average per exposure across the combined period, the pooled ratio is the appropriate calculation.
A selection memo should identify the candidate periods considered, their exposure volumes and record status, and the main reasons for including or excluding years. It should also explain material changes in the book and note limitations, such as a shorter period or incomplete records. This makes the decision reviewable and prevents the selected dates from appearing arbitrary.
Check Your Understanding
For each hypothetical case, choose or compare the candidate periods using the information provided.
- A product has three complete years: Year A has 4,000 exposures and $1,080,000 in losses; Year B has 6,000 exposures and $1,680,000; Year C has 10,000 exposures and $3,000,000. A guideline calls for at least 15,000 exposures. Compare Years B–C with Years A–C and state which meets the guideline with less historical data. (Answer: Years B–C have 16,000 exposures and a pooled loss cost of $292.50; they meet the guideline and use fewer years than A–C, which have 20,000 exposures and a pooled loss cost of $288.00.)
- A complete 2023–2024 period has 7,000 exposures and $1,890,000 in losses in 2023, and 8,000 exposures and $2,240,000 in 2024. In 2025, 10,000 exposures are recorded, but only 80% of loss records are confirmed complete; reported losses are $2,400,000. Calculate the complete-period loss cost and explain whether the 2025 data should be included as supplied. (Answer: 2023–2024 has 15,000 exposures and $4,130,000 in losses, or $275.33 per exposure; do not include 2025 as supplied because the loss records are incomplete.)
- A future book will mainly consist of a new coverage. The older coverage has 24,000 exposures and $6,000,000 in losses; the new coverage has 9,000 exposures and $2,880,000 in complete recent losses. Calculate each loss cost and identify the more relevant period for the stated future book. (Answer: Older coverage, $250; new coverage, $320; the recent new-coverage period is more relevant to the future book, although it has lower volume.)
- A three-year period has annual exposures of 2,000, 5,000, and 9,000, with annual loss costs of $200, $240, and $280. Calculate the pooled loss cost and the unweighted average of annual loss costs. Which is the average per exposure across the combined period? (Answer: Total losses are $400,000 + $1,200,000 + $2,520,000 = $4,120,000; total exposure is 16,000; pooled loss cost is $257.50, while the unweighted annual average is $240.00; $257.50 is the average per exposure across the combined period.)
- A complete, comparable period has 20,000 exposures and a pooled loss cost of $260. A newer, complete period has 8,000 exposures and a pooled loss cost of $300, and a material change means the newer period better represents the expected future business. State the principal trade-off and why neither volume nor recency alone settles the choice. (Answer: The newer period is more relevant but has lower volume and may be less stable; the older period has greater volume but is less representative, so the selection depends on balancing those limitations.)