Combining Historical Experience Across Trend Periods

Why Historical Periods Need a Common Cost Level

Experience used in a prospective premium indication may come from several periods. If claim costs have changed over time, a dollar of loss from an older period is not directly comparable with a dollar of loss from a more recent period. Combining the unadjusted results can give too much weight to experience measured at an outdated cost level. The solution is to trend each period’s loss cost to a common prospective level, then combine the adjusted costs using exposure weights.

This adjustment must occur before credibility weighting when the experience is blended with a prospective complement. Both components of the blend should represent costs at the same level. The methods here are aligned with the CAS Exam 8 (Advanced Ratemaking) syllabus.

Definition: A trended loss cost is a historical loss cost adjusted from its period midpoint to a specified prospective cost level. If the annual compound trend rate is $r$, the historical period midpoint is $n$ years before the prospective midpoint, and the same trend applies throughout, the trend factor is $(1+r)^n$. The trended loss cost is the historical loss cost multiplied by that factor.

The period midpoint matters: trend time is measured between the midpoint of the experience period and the midpoint of the prospective period, not necessarily between the dates when the data are collected and the premium is issued. If a trend index is supplied directly, use the ratio of the prospective index to the historical index. For changing annual rates, multiply the applicable annual factors rather than applying one constant rate for the entire interval.

The examples assume that the stated trend applies to the loss-cost measure being used, that coverage and the exposure base are comparable across periods, and that the supplied period loss costs are suitable for the calculation. There is no interest discounting or mortality assumption in these examples; they are deterministic calculations, and no probability distribution for losses is needed.

Trend First, Then Exposure-Weight

Let $L_j$ be aggregate losses in historical period $j$, $w_j$ be its exposure units, and $C_j=L_j/w_j$ be its loss cost per exposure unit. If its trend factor to the prospective midpoint is $T_j$, its adjusted loss cost is $C_j^*=C_jT_j$. The combined experience indication is the exposure-weighted average of those adjusted costs.

C_{\mathrm{exp}}^* = \frac{\sum_j w_j C_j^*}{\sum_j w_j} = \frac{\sum_j w_j C_jT_j}{\sum_j w_j}.

This is also the total of the trended period losses divided by total exposure: each period contributes $w_jC_jT_j=L_jT_j$ to the adjusted loss total. Exposure weights matter because a period representing many units of risk should contribute more to the combined cost than a period representing few units. Equal weighting is appropriate only when the periods have equal exposure.

When a credibility weight $Z$ and a prospective complement $M$ are given, the credibility indication is

P_{\mathrm{cred}}=ZC_{\mathrm{exp}}^*+(1-Z)M.

Here $M$ must also be expressed at the prospective cost level. In a comparison with an untrended calculation, use the same $Z$ and $M$ unless the problem specifically says otherwise. In practical work, the credibility parameters may need review if trend adjustments alter the assumptions used to determine credibility.

Why the Trend Factor Has This Form

Under a constant annual compound trend rate $r$, a cost level $C$ becomes $C(1+r)$ after one year. Applying the same annual change repeatedly gives $C(1+r)^2$ after two years, and so on. Thus, for $n$ years, the prospective-level cost is $C(1+r)^n$. Dividing by $C$ gives the factor $(1+r)^n$.

This derivation assumes the same compound rate applies over the full interval and that the cost measure remains comparable. If annual rates differ, for example $r_1$ in the first year and $r_2$ in the second, the factor is $(1+r_1)(1+r_2)$, not $(1+r_1+r_2)^2$ or a simple sum of rates. If trend is already supplied as an index ratio, do not apply an additional trend factor to it.

Worked Applications

Worked Example: Unequal Exposures at Three Trend Ages

A hypothetical portfolio has 50 exposure units with a loss cost of $800 two years before the prospective midpoint, 75 units with a loss cost of $900 one year before it, and 100 units at the prospective midpoint with a loss cost of $1,200. The annual compound trend rate is 8%. Find the combined trended loss cost per exposure unit. No interest or mortality assumption is used; the supplied costs are treated as deterministic, and no loss distribution is required.

Method. Trend each period cost to the prospective midpoint, multiply each adjusted cost by its exposure, and divide the total by total exposure.

The three trend factors are $(1.08)^2=1.1664$, $1.08$, and $1.00$. The corresponding trended costs are:

Two years before: $800(1.1664)=\$933.12$. One year before: $900(1.08)=\$972.00$. At the prospective midpoint: $1{,}200(1.00)=\$1{,}200.00$.

The exposure-weighted adjusted loss total is $50(933.12)+75(972.00)+100(1{,}200.00)=46{,}656+72{,}900+120{,}000=\$239{,}556$. Total exposure is $50+75+100=225$ units. Therefore,

$\$239{,}556/225=\$1{,}064.6933\ldots$, or $1,064.69 per exposure unit, rounded to the nearest cent.

Check. The result lies between the lowest and highest trended costs, as a positive-weighted average should. The common mistake is to average the three trended costs equally: $(933.12+972+1{,}200)/3=\$1{,}035.04$. That calculation incorrectly gives each period the same weight despite the different exposures.

Worked Example: Comparing Trended and Untrended Credibility Indications

A hypothetical risk has 100 exposure units in each of two historical periods. Its loss costs are $700 two years before the prospective midpoint and $900 one year before it. Annual trend is 10%, the prospective complement is $850 per exposure unit, and $Z=0.50$. Calculate the trended credibility indication, the untrended indication, and their difference. No interest or mortality assumption is used; the costs are deterministic inputs, with no loss distribution specified.

Method. First trend the experience to the prospective midpoint. Then apply the stated credibility formula, keeping the complement and credibility weight the same in the comparison.

The trended costs are $700(1.10)^2=700(1.21)=\$847$ and $900(1.10)=\$990$. Since both periods have 100 units, the combined trended experience cost is $[100(847)+100(990)]/200=(84{,}700+99{,}000)/200=\$918.50$.

The trended credibility indication is $0.50(918.50)+0.50(850)=459.25+425=\mathbf{\$884.25}$ per exposure unit.

Without trend, the exposure-weighted experience cost is $[100(700)+100(900)]/200=\$800$. The untrended credibility indication is $0.50(800)+0.50(850)=400+425=\mathbf{\$825.00}$. Thus, the trended indication is $\$884.25-\$825.00=\mathbf{\$59.25}$ per exposure unit higher.

Check. The trended indication is between the trended experience cost of $918.50 and the complement of $850; the untrended indication is between $800 and $850. The trap is to trend the experience but compare it with an untrended complement, or to omit trend and call the result a prospective-level indication. Each blend must use components on a consistent cost basis.

Worked Example: Three Periods with Unequal Exposure and Credibility

A hypothetical program has 80 exposure units with a loss cost of $640 three years before the prospective midpoint, 120 units with a loss cost of $720 two years before it, and 100 units with a loss cost of $900 at the midpoint. Annual trend is 5%. The prospective complement is $780 per exposure unit and $Z=0.50$. Calculate the trended experience cost and the credibility indication, then compare it with the indication obtained without trend. No interest or mortality assumption is used; the stated costs are deterministic and no loss distribution is required.

Method. Adjust each period cost separately, use its exposure as the weight, and then blend the resulting experience cost with the prospective-level complement.

The factors are $(1.05)^3=1.157625$, $(1.05)^2=1.1025$, and $1$. The trended loss costs are $640(1.157625)=\$740.88$, $720(1.1025)=\$793.80$, and $900(1)=\$900.00$.

The adjusted loss total is $80(740.88)+120(793.80)+100(900)=59{,}270.40+95{,}256+90{,}000=\$244{,}526.40$. Total exposure is $80+120+100=300$. Thus, trended experience cost is $244{,}526.40/300=\$815.088$, or $815.09 per exposure unit, rounded to the nearest cent.

The trended credibility indication, using the unrounded experience cost, is $0.50(815.088)+0.50(780)=407.544+390=\$797.544$, or $797.54 per exposure unit.

For the untrended calculation, the loss total is $80(640)+120(720)+100(900)=51{,}200+86{,}400+90{,}000=\$227{,}600$. The untrended experience cost is $227{,}600/300=\$758.6667$. The resulting indication is $0.50(758.6667)+0.50(780)=\$769.3333$, or $769.33 per exposure unit. The trended indication is therefore $\$797.544-\$769.3333=\$28.2107$, or $28.21 higher.

Check. The trended experience cost of $815.09 exceeds the untrended cost of $758.67, consistent with positive trend. Both credibility indications lie between their respective experience costs and the $780 complement. The calculation also shows why the exposure totals must be used: the 120-unit period contributes more than either 80-unit period.

Common Traps and Practical Checks

The adjustment is straightforward, but several errors can produce plausible-looking indications. Use the following checks before finalizing a calculation.

Trend in the wrong direction. To move older experience forward under positive trend, multiply by a factor greater than one. A period already at the prospective midpoint has factor 1.
Use the wrong time interval. Count years between the historical and prospective midpoints. A two-year interval at 8% uses $(1.08)^2$, not $1+2(0.08)$.
Average periods rather than exposures. Use $\sum w_jC_j^*/\sum w_j$. A simple average is valid only for equal exposure weights.
Trend exposure instead of cost. The stated trend adjusts the loss-cost level. Exposure is used as a weight; it is not multiplied by the trend factor.
Blend inconsistent levels. A credibility complement for the prospective period should be on the same prospective cost basis as trended experience.
Round too early. Retain precision through the weighted average and credibility blend, then round the final monetary result as requested.

For a quick reasonableness check, a positive trend should make each older period’s cost no lower than its historical cost. The combined trended cost must lie between the smallest and largest adjusted costs when all exposures are positive. A credibility indication with $0\leq Z\leq1$ must lie between the adjusted experience cost and its complement. These checks do not replace the calculation, but they often reveal a misplaced trend factor or an incorrect weight.

Key takeaway: Bring every historical period to the same prospective cost level first. Then combine period costs using exposure weights, and only then blend the result with a prospective-level complement. This keeps both the experience and the complement on a consistent basis.

Check Your Understanding

For each problem, retain full precision until the final monetary answer and use exposure weights where periods differ in size.

A hypothetical period has a loss cost of $520 per exposure unit, measured two years before the prospective midpoint. Annual compound trend is 6%. Find its trended cost. (Answer: $584.27 per exposure unit)
A hypothetical book has 60 exposure units at a loss cost of $500 two years before the prospective midpoint and 90 units at $700 one year before it. Annual trend is 4%. Find the combined trended experience cost. (Answer: $653.12 per exposure unit)
A hypothetical risk has 40 units at $600 two years before the prospective midpoint, 80 units at $750 one year before it, and 80 units at $820 at the midpoint. Annual trend is 5%, the complement is $760, and $Z=0.40$. Find the trended credibility indication. (Answer: $766.12 per exposure unit)
A hypothetical risk has 100 units at $800 two years before the prospective midpoint and 100 units at $1,000 one year before it. Annual trend is 5%, the complement is $900, and $Z=0.25$. Calculate the trended and untrended credibility indications and the difference. (Answer: $916.50 trended; $900.00 untrended; trended is $16.50 higher)

What You'll Learn