Why Document a Trend Selection?
A trend indication is not self-explanatory. A reviewer needs to know what measure is being trended, which observations and periods support the estimate, what methods were considered, and why the selected rate is appropriate for the projection. A documented selection makes those judgments visible and helps distinguish evidence from assumptions.
This tutorial focuses on presenting and testing a trend selection. The calculations use deterministic annual compounding: a selected annual rate is held constant over the projection horizon. No interest-rate, mortality, or probability-distribution assumption is involved. The relevant modeling assumption is that the chosen annual trend rate applies throughout the stated period; if the trend is expected to vary over time, a single constant rate may not be suitable.
What a Clear Selection Should State
Begin by naming the measure. A trend in average claim severity is not necessarily a trend in claim frequency or aggregate loss cost. State the units and population represented, and note any adjustments that make the observations comparable. The earlier material on distinguishing loss trend from loss development is important here: the selected trend should represent changes in the cost level, not changes caused by claims maturing or developing.
Next, specify the data periods and how the observations are summarized. Annual observations may be compared through year-to-year percentage changes, a fitted trend, or a compound annual rate over multiple years. State the method rather than reporting a rate without its basis. Then describe relevant external evidence, such as medical, wage, or repair-cost indicators, and explain why it is informative for the particular measure. An external index is context, not an automatic substitute for insurance claim data.
Finally, state the selected rate and explain any judgment between indications. A concise rationale can mention whether the data are stable, whether there is evidence of a structural change or mix shift, how well an external indicator matches the covered costs, and whether the selection reflects the conditions expected during the projection period. If evidence is mixed, report that fact instead of implying false precision.
Name the quantity being trended, its units, and the population or coverage to which it applies.
Give the observation periods, data adjustments, and the methods used to summarize historical changes.
Compare historical indications with relevant external evidence and diagnostics; explain material differences.
State the selected rate and projection horizon, then show how reasonable alternative rates change the result.
Compounding and Sensitivity
If a measure starts at \(C_0\), a constant effective annual trend rate \(r\) is applied for \(t\) years, and fractional years are represented by fractional exponents, the projected value is:
The rate \(r\) is written as a decimal in the formula: for example, 5.5% is \(0.055\). The horizon \(t\) is measured in years. The formula assumes the same annual rate applies throughout the full period and that the stated fractional-year convention is appropriate. It does not describe a situation in which annual rates change from year to year.
The formula follows from repeated annual application of the selected rate. After one year the value is \(C_0(1+r)\); after two years it is \(C_0(1+r)(1+r)=C_0(1+r)^2\). Repeating this process gives \(C_0(1+r)^n\) for \(n\) whole years. For a fractional horizon, the same model uses the corresponding fractional power, so the projection factor remains \((1+r)^t\).
Sensitivity analysis calculates the projection under plausible alternative rates while holding the starting measure and horizon fixed. This isolates the effect of the rate assumption. It does not establish which rate is correct; that judgment still depends on the evidence and rationale.
Worked Example: Summarizing Historical Evidence and Selecting a Rate
A hypothetical insurer reviews annual average repair claim costs for a stable group of vehicle coverages. The recorded amounts are $1,000 in 2021, $1,060 in 2022, $1,113 in 2023, and $1,179 in 2024. A hypothetical repair-cost indicator rose 4.8% over the most recent year, while an indicator for a labor-intensive subset rose 6.1%. The group’s coverage mix appears stable, and the review finds no clear break in the annual observations. Prepare a short selection rationale and calculate a compound annual indication.
Method: calculate annual changes and the compound annual rate. The three year-to-year rates are:
For the final year, the increase is \(1{,}179-1{,}113=66\), so:
The arithmetic mean of these annual rates is \((6.00\%+5.00\%+5.9299\%)/3=5.6433\%\). The compound annual rate uses the start and end values across three years:
The two summaries are close, but they answer slightly different questions: the arithmetic mean averages the observed annual percentages, whereas the compound rate is the constant annual rate that takes the initial value to the final value. The compound indication is approximately 5.64%.
Selection rationale. For this hypothetical coverage, a 5.5% annual selection is a reasonable judgmental choice: it is near the historical compound indication, below the 6.1% labor-intensive indicator, and above the broader 4.8% repair-cost indicator. The mixed external evidence supports a range rather than an exact answer. Stable mix and the absence of a clear break make the historical series relevant, but the indicators’ differences should be retained in the documentation as uncertainty. The common mistake avoided is treating one external index as the selected insurance trend without checking how well it represents the covered claims.
Check. Applying the compound indication for three years gives \(1{,}000(1.056423)^3\), approximately $1,179, consistent with the observed endpoint after rounding.
Worked Sensitivity Comparisons
Worked Example: Comparing Alternative Rates Over Three Years
A hypothetical portfolio has a current loss-cost index of 1,000. The documented central selection is 5.78% per year, with 4.8% and 6.4% used as lower and upper sensitivity scenarios. Calculate each projected index after three years, the changes from the central result, and the full range. Assume each annual rate is constant for all three years.
Method: apply annual compounding separately to each scenario. At 4.8%, the factor is:
Thus the lower projection is \(1{,}000(1.151022592)=1{,}151.022592\), or 1,151.02 to two decimal places.
At the central 5.78% rate:
The central projection is \(1{,}000(1.183615621)=1{,}183.615621\), or 1,183.62 to two decimal places.
At 6.4%:
The upper projection is \(1{,}000(1.204550144)=1{,}204.550144\), or 1,204.55 to two decimal places. Relative to the central projection, the lower scenario is \(1{,}151.02-1{,}183.62=-32.60\), and the upper scenario is \(1{,}204.55-1{,}183.62=20.93\). The full range is \(1{,}204.55-1{,}151.02=53.53\).
Check. The lower rate produces the lowest result and the higher rate the highest; the three-year projections are all above the starting index of 1,000. The common mistake avoided is substituting a rounded rate for the stated 5.78% central scenario: 5.8% would be a different assumption and would produce a different projection.
Worked Example: Sensitivity With a Fractional-Year Horizon
A hypothetical liability coverage has a current loss cost of $800. The projection horizon is 1.5 years. Calculate projections at 4%, 5%, and 6% annual trend, rounding money to the nearest cent. Assume a constant effective annual rate in each scenario and use fractional-year compounding.
Method: use the stated 1.5-year exponent for every scenario. At 4%:
At 5%:
At 6%:
The 5% scenario is $12.26 above the 4% scenario, and the 6% scenario is $12.33 above the 5% scenario. Across the full sensitivity range, the difference is \($873.07-$848.48=\)$24.59.
Check. Each projection is greater than $800 because each selected trend rate is positive. The common mistake avoided is treating 1.5 years as one year, or rounding the horizon to two years; either changes the modeled period. Also, \(800(1.04)^{1.5}\) rounds to $848.48, not $848.30.
How to Present the Selection and Its Limits
A useful record separates the selected central rate from the alternatives used to describe uncertainty. It should also make clear whether the projection is for a single annual measure or for a combined quantity. If frequency and severity are examined separately, document those measures and their respective selections before explaining how they support the overall loss-cost view. Do not imply that a sensitivity range is a statistical confidence interval unless it was constructed as one.
The choice of sensitivity scenarios should be interpretable. Alternatives can reflect nearby rates supported by historical or external evidence, or a wider range when the data are especially uncertain. Report the starting value, horizon, rates, factors, and projected values together. That lets a reviewer distinguish the size of the rate assumption from the effect of the projection period.
| Documentation item | What to record | Why it matters |
|---|---|---|
| Measure | Quantity, units, coverage, and population | Defines what the selected rate applies to |
| Historical basis | Observation periods, adjustments, and calculation method | Makes the historical indication reproducible |
| External evidence | Relevant indicators and their limitations | Provides context without treating an index as a direct substitute |
| Selection | Chosen rate and concise rationale | Shows how evidence and judgment were combined |
| Sensitivity | Alternative rates, horizon, and projected results | Quantifies how much the result depends on the selection |
Common Mistakes
- Leaving the measure vague. “Trend is 5%” does not say whether the rate applies to severity, frequency, or another measure.
- Reporting a rate without its data periods or method. A reader cannot tell whether it is a one-year change, an average of annual changes, or a compound rate.
- Equating an external indicator with insurance loss trend. Explain how the indicator relates to the covered costs and why it may differ.
- Changing the rate or horizon without labeling the change. Sensitivity scenarios should vary the intended assumption while keeping the other inputs fixed.
- Rounding too early. Keep adequate precision in the factor and projection, then round the final reported amount as specified.
- Calling a range a probability statement. A set of alternative projections is a scenario analysis unless a probability model justifies a probabilistic interpretation.
Check Your Understanding
For each original problem, use constant annual compounding unless the problem states otherwise. Round projected money to the nearest cent.
- A hypothetical equipment loss-cost index is 500. Project it for three years at 3% annually. Give the factor to six decimal places and the projected index to two decimal places. (Answer: factor \(1.03^3=1.092727\); projected index \(500(1.092727)=546.36\).)
- A hypothetical claim-cost series has annual changes of 4%, 6%, and 5%. Calculate the arithmetic mean annual change, then project a current cost of $1,200 for two years at that mean rate. (Answer: mean \(=(4\%+6\%+5\%)/3=5\%\); projection \(1{,}200(1.05)^2=\$1{,}323.00\).)
- A hypothetical starting index is 900, and the projection horizon is two years. Calculate projections at 4.5% and 5.5%, then give the difference between them. (Answer: \(900(1.045)^2=900(1.092025)=982.82\); \(900(1.055)^2=900(1.113025)=1{,}001.72\); difference \(=\) 1,001.72 − 982.82 \(=18.90\).)
- A hypothetical severity index rises from 100 to 121 over two years. What constant compound annual rate links the two observations? If the selected rate is rounded to 10.0%, what should be stated about that rounding in a selection note? (Answer: \((121/100)^{1/2}-1=10.0\%\); identify 10.0% as the rounded compound indication, rather than implying greater precision.)