Avoiding Ratio-Averaging Errors In Development Credibility

Why the Choice of Average Matters

A development ratio compares cumulative loss amounts at two successive development ages. When periods have different amounts of loss already reported, their ratios do not carry equal weight in a projection of the aggregate book. A simple average may describe the average period ratio, but it generally does not describe how the total current amount develops. The distinction matters whenever the selected factor is used to project an aggregate ultimate or reserve.

This tutorial focuses on selecting and applying ratios when experience is available by accident period, policy year, or another group. It is not enough to calculate a factor correctly: the weights must correspond to the quantity being estimated. A factor for an average period, an average exposure, and an aggregate dollar projection can be different answers to different questions.

Definition: Let $C_{i,j}$ be the cumulative loss amount for period $i$ at development age $j$, and let $C_{i,j+1}$ be the amount at the next age. If $C_{i,j}>0$, the period-level age-to-age development ratio is $r_i=C_{i,j+1}/C_{i,j}$. The simple average is $\sum_{i=1}^{n}r_i/n$. The volume-weighted age-to-age factor is $\sum_{i=1}^{n}C_{i,j+1}/\sum_{i=1}^{n}C_{i,j}$, provided the total denominator is positive.

The Aggregate Factor Is a Volume-Weighted Average

If a factor is intended to project the total cumulative amount at age $j$ to the total at age $j+1$, the matching calculation is to divide the sum of the later-age amounts by the sum of the current-age amounts. This ratio can also be written as a weighted average of period ratios, with each period’s current cumulative amount as its weight.

f_{\mathrm{vol}} =\frac{\sum_{i=1}^{n}C_{i,j+1}}{\sum_{i=1}^{n}C_{i,j}} =\frac{\sum_{i=1}^{n}C_{i,j}r_i}{\sum_{i=1}^{n}C_{i,j}}

The identity follows by substituting $C_{i,j+1}=C_{i,j}r_i$ into the numerator. Thus, a period with a large current cumulative amount has more influence on the factor for the aggregate dollar volume. If every period has the same current amount, the volume-weighted factor equals the simple average. Otherwise, the two generally differ.

This calculation assumes that the amounts being summed are measured on a consistent basis and are appropriate to combine—for example, all are cumulative paid losses at the same two development ages. It also assumes a positive aggregate denominator. The factor describes observed aggregate development across the included periods; it does not by itself establish that future periods will develop the same way.

Worked Example: Simple Average Versus Aggregate Development

A hypothetical insurer reviews cumulative paid losses for two accident periods at a particular development age. Period A has $C_{A,j}=\$100{,}000$ and $C_{A,j+1}=\$120{,}000$. Period B has $C_{B,j}=\$900{,}000$ and $C_{B,j+1}=\$990{,}000$. Calculate both the simple average ratio and the factor that projects the aggregate current amount.

Step 1: Calculate the period ratios. For Period A, $r_A=120{,}000/100{,}000=1.20$. For Period B, $r_B=990{,}000/900{,}000=1.10$.

Step 2: Calculate the simple average. The arithmetic mean is $(1.20+1.10)/2=1.15$. It gives each accident period equal weight.

Step 3: Calculate the volume-weighted factor. The aggregate current amount is $\$100{,}000+\$900{,}000=\$1{,}000{,}000$. The aggregate next-age amount is $\$120{,}000+\$990{,}000=\$1{,}110{,}000$. Therefore, $f_{\mathrm{vol}}=\$1{,}110{,}000/\$1{,}000{,}000=1.11$.

Step 4: Check the aggregate projection. Applying 1.11 to $\$1{,}000{,}000$ gives $\$1{,}110{,}000$, exactly the combined next-age amount. Applying 1.15 would give $\$1{,}150{,}000$, which does not reproduce the aggregate development observed in these data.

Method and trap avoided. Use the volume-weighted factor when the target is the development of aggregate dollars. The simple average is not an arithmetic error; it answers a different question. The trap is to assume that every period ratio should receive equal weight merely because each period contributes one ratio.

Volume Weights and Exposure Weights Answer Different Questions

Exposure weights may be useful when the target is an average development experience per unit of exposure. But they are not interchangeable with current-loss-volume weights. If the goal is to multiply a total current cumulative loss by a factor to project total losses, the factor must reflect the relative contributions of those current losses. Weighting period ratios by exposure instead will generally produce a different factor.

An exposure-based approach can instead focus on incremental development per exposure. Let $E_i$ be the exposure associated with period $i$, measured consistently for all periods, and suppose $E_i>0$. Define the observed incremental amount per exposure as $q_i=(C_{i,j+1}-C_{i,j})/E_i$. The exposure-weighted average incremental amount per exposure is $\sum_i E_iq_i/\sum_iE_i$. If this rate is judged suitable for a target book with total exposure $E_{\mathrm{target}}$, the projected incremental amount is the selected rate multiplied by $E_{\mathrm{target}}$.

This method projects an increment, not a ratio. It is useful when exposure is the intended basis for the projection and a per-exposure incremental rate is meaningful. The exposure measure must match the loss measure and the projection period. A per-policy rate, for example, should not be applied to exposure measured in earned vehicle-years unless the units are converted appropriately.

Worked Example: An Exposure-Based Increment and an Aggregate Factor

Two hypothetical groups have exposures of 20 and 80 units. Their cumulative amounts increase from $\$400{,}000$ to $\$480{,}000$ and from $\$600{,}000$ to $\$690{,}000$, respectively. Calculate the exposure-weighted ratio, the aggregate factor, and an exposure-weighted incremental projection based on these observed amounts.

Step 1: Find the group ratios. The first ratio is $480{,}000/400{,}000=1.20$. The second is $690{,}000/600{,}000=1.15$.

Step 2: Calculate the exposure-weighted average ratio. It is $(20(1.20)+80(1.15))/(20+80)=(24+92)/100=1.16$. This describes an exposure-weighted average of the two ratios; it is not the factor that projects the combined current amount.

Step 3: Calculate the aggregate factor. The combined current amount is $\$400{,}000+\$600{,}000=\$1{,}000{,}000$, and the combined next-age amount is $\$480{,}000+\$690{,}000=\$1{,}170{,}000$. Thus, the aggregate factor is $1{,}170{,}000/1{,}000{,}000=1.17$. Applying the exposure-weighted ratio of 1.16 to the aggregate current amount would project $\$1{,}160{,}000$, not the observed $\$1{,}170{,}000$.

Step 4: Calculate the incremental rates per exposure. The first group’s increment is $\$480{,}000-\$400{,}000=\$80{,}000$, so its rate is $\$80{,}000/20=\$4{,}000$ per exposure unit. The second group’s increment is $\$90{,}000$, so its rate is $\$90{,}000/80=\$1{,}125$ per exposure unit.

Step 5: Average the increments by exposure. The weighted rate is $(20(\$4{,}000)+80(\$1{,}125))/100=(\$80{,}000+\$90{,}000)/100=\$1{,}700$ per exposure unit. Applied to the observed total exposure of 100 units, it gives an increment of $\$170{,}000$. Adding this to the current $\$1{,}000{,}000$ gives $\$1{,}170{,}000$, consistent with the combined next-age amount.

Method and trap avoided. Use exposure-weighted incremental rates when the projection is explicitly on an exposure basis. Do not treat an exposure-weighted average ratio as a substitute for the volume factor in an aggregate-dollar projection.

Applying Credibility After Selecting the Relevant Factor

A development factor may be adjusted toward a benchmark using a credibility weight when a model or selection procedure specifies that approach. Let $f_{\mathrm{obs}}$ be the selected factor from the relevant experience basis, $f_0$ the benchmark factor, and $Z$, with $0\leq Z\leq1$, the credibility assigned to the observed factor. Under the stated linear credibility rule, the selected factor is $f_{\mathrm{sel}}=Zf_{\mathrm{obs}}+(1-Z)f_0$.

The key is to identify $f_{\mathrm{obs}}$ before applying credibility. If the target is aggregate development, that observed factor is ordinarily based on the aggregate amounts, equivalently the volume-weighted period ratios. Credibility weighting does not correct a mismatch between the quantity being averaged and the target projection.

Worked Example: Credibility Weighting the Aggregate Factor

Three hypothetical periods have current cumulative amounts of $\$100{,}000$, $\$300{,}000$, and $\$600{,}000$. Their respective development ratios are 1.30, 1.10, and 1.05. A specified credibility procedure assigns $Z=0.60$ to the observed aggregate factor and uses a benchmark factor of 1.04. Apply the selected factor to the combined current amount.

Step 1: Calculate the aggregate factor. The next-age amounts are $\$100{,}000(1.30)=\$130{,}000$, $\$300{,}000(1.10)=\$330{,}000$, and $\$600{,}000(1.05)=\$630{,}000$. Their sum is $\$1{,}090{,}000$, and the current amounts sum to $\$1{,}000{,}000$. Therefore, $f_{\mathrm{obs}}=1.09$.

Step 2: Apply the stated credibility rule. The selected factor is $0.60(1.09)+0.40(1.04)=0.654+0.416=1.070$.

Step 3: Project the aggregate amount. The selected next-age amount is $\$1{,}000{,}000(1.070)=\$1{,}070{,}000$. The selected increment above the current amount is $\$70{,}000$.

Step 4: Check the averaging order. The simple mean of the three ratios is $(1.30+1.10+1.05)/3=1.15$. Credibility-weighting that mean would give $0.60(1.15)+0.40(1.04)=1.106$, and a projection of $\$1{,}106{,}000$. That is not the stated aggregate-based selection.

Method and trap avoided. First select the observed factor that matches the projection target; then apply the specified credibility formula. Credibility does not justify using an unweighted mean when the target is aggregate loss volume.

A Practical Selection Sequence

Before averaging any development ratios, write down the quantity the factor is supposed to estimate. That decision determines the relevant weights. The following sequence helps prevent a mechanically correct calculation from answering the wrong question.

State the projection target.
Decide whether the target is aggregate loss dollars, an average period ratio, or an incremental amount per unit of exposure.

Check the data basis.
Confirm that cumulative amounts refer to consistent loss measures and development ages, and that exposures use compatible units.

Match the weights to the target.
For an aggregate link factor, use current cumulative amounts as weights. For an explicitly exposure-based incremental rate, use exposure weights on increments per exposure.

Apply credibility only as specified.
Blend the correctly selected observed quantity with its benchmark using the stated credibility rule.

Check the projected amount.
Multiply the selected factor by the aggregate current amount, or add the projected increment to that amount. Compare the result with the intended basis of the projection.

A volume-weighted factor is not automatically the best prediction of future development. Changes in claim handling, coverage, case reserves, or the mix of business may make historical relationships less appropriate. The calculation establishes which factor matches the observed aggregate relationship; actuarial selection still requires assessing whether the experience is relevant to the projection.

Common Exam Trap

Common mistake: Calculate one ratio per period, take their arithmetic mean, and multiply the total current amount by that mean without checking the target. The arithmetic mean gives every period equal weight, even when one period represents far more current loss volume. For an aggregate projection, calculate $\sum_i C_{i,j+1}/\sum_i C_{i,j}$, or equivalently weight each ratio by $C_{i,j}$. Use exposure weights only when the target and method are explicitly exposure-based, and do not confuse an exposure-weighted ratio with an exposure-weighted incremental rate.

Another frequent error is to apply credibility to the ratios individually and then average, or to credibility-weight a simple mean when the specified observed quantity is the aggregate factor. The order and basis of the calculations matter: determine the target, construct the matching observed estimate, and only then apply the stated credibility adjustment.

Key takeaway: The weight in a development average encodes what is being estimated. Current cumulative amounts weight an aggregate link factor; exposure weights can support an exposure-based incremental projection. An average ratio should never be selected solely because it is easy to calculate.

Check Your Understanding

For each problem, identify the quantity being projected before choosing or applying a factor.

A hypothetical portfolio has current cumulative amounts of $\$200{,}000$ and $\$800{,}000$, with development ratios 1.25 and 1.05. Find the simple average, the aggregate factor, and the projected aggregate next-age amount. (Answer: simple average $1.15$; aggregate factor $1.09$; projected amount $\$1{,}090{,}000$.)
Two periods have current cumulative amounts of $\$300{,}000$ and $\$700{,}000$, and development ratios of 1.10 and 1.04. What volume-weighted factor projects the aggregate current amount? (Answer: $1.058$.)
Two groups have exposures of 25 and 75 units, current cumulative amounts of $\$500{,}000$ each, and ratios of 1.20 and 1.08. Calculate the exposure-weighted average ratio and the aggregate factor. (Answer: exposure-weighted ratio $1.11$; aggregate factor $1.14$.)
A specified credibility rule uses $Z=0.25$, a volume-weighted observed factor of 1.12, and a benchmark of 1.06. If current aggregate losses are $\$2{,}000{,}000$, calculate the selected factor and projected amount. (Answer: selected factor $1.075$; projected amount $\$2{,}150{,}000$.)
Two groups have exposures of 40 and 60 units and observed next-age increments of $\$120{,}000$ and $\$90{,}000$. If total current cumulative losses are $\$800{,}000$, calculate the exposure-weighted increment per unit and project the next-age total using the observed total exposure. (Answer: $\$2{,}100$ per exposure unit; projected total $\$1{,}010{,}000$.)

What You'll Learn