Blending Exposure And Burning-Cost Indications

Why Blend Two Layer-Loss Indications?

An excess layer can be priced using information about the risk portfolio or using its adjusted loss experience. Exposure rating estimates expected layer loss from exposure data and an expected loss distribution or exposure curve. Burning cost estimates it from historical losses falling within the layer, after the specified development, trend, and exposure adjustments. The two sources answer different questions: exposure rating supplies a model-based estimate, while burning cost reflects observed experience.

Neither source is automatically preferable in every situation. A limited amount of historical experience may be volatile, especially for a high layer with few losses. An exposure estimate can be less responsive to the particular portfolio’s experience, and can depend on assumptions in the exposure model. When both indications are available, a selected estimate can combine them using a stated credibility weight. The weight is an instruction about how much of each indication to use; it does not change either indication’s underlying calculation.

The calculations below concern hypothetical annual expected losses for the same per-occurrence layer and portfolio. All amounts are undiscounted dollars at the same stated cost level. No interest rate, mortality assumption, or parametric loss distribution is needed for these blending calculations; the exposure indication is assumed to have been calculated already using the applicable exposure-rating method.

Definition: Let $E$ be an exposure-rated expected layer loss and $B$ a burning-cost indication for the same layer, period, and portfolio basis. If $Z$ is the weight assigned to burning cost, the selected expected layer loss is $$ S=ZB+(1-Z)E,\qquad 0\leq Z\leq 1. $$ Here $Z=0$ selects exposure rating, $Z=1$ selects burning cost, and intermediate values blend the two. The weight $Z$ is sometimes called the credibility assigned to the burning-cost indication.

Put Both Indications on a Common Basis

Before blending, verify that the estimates refer to the same insurance layer and are expressed for the same portfolio, period, and cost level. For example, a burning cost may need development to ultimate, trend to the rating period, and adjustment for changes in exposure or coverage. Those adjustments produce a comparable historical estimate; the blending formula does not perform them automatically.

A common way to express the exposure estimate is as an expected layer loss cost per exposure unit multiplied by the current number of exposure units. A burning-cost indication can be expressed in the same way: divide adjusted historical layer losses by the comparable historical exposure units to obtain a cost per unit, then apply that cost to the current exposure units. If the indications are already total expected losses for the same current portfolio, blend the totals directly.

The following sequence keeps the calculation organized.

Align the basis.
Confirm that both estimates use the same layer attachment and limit, coverage, loss definition, rating period, and cost level.

Calculate each indication.
Determine the exposure-rated expected loss and the adjusted burning-cost estimate separately. Preserve their units until each is on a common basis.

Apply the weight to the correct source.
Multiply burning cost by $Z$ and exposure rating by $1-Z$, then add the weighted amounts.

Check the selection.
For a weight between zero and one, the selected amount must lie between the two indications. Check the total-loss units and the arithmetic.

Worked Example: Blend Per-Unit Indications

A hypothetical excess insurer has 80,000 current-equivalent exposure units. The exposure-rated expected layer loss cost is $6.25 per unit. Adjusted historical losses for the same layer total $2,160,000 over 360,000 comparable exposure units. The burning-cost weight is $Z=0.35$. Calculate the two total-loss indications and the selected expected layer loss.

Method and basis. Treat both indications as annual, undiscounted expected losses for the same layer and current portfolio. No parametric distribution is required because the exposure-rated cost per unit is given. The adjusted historical loss total is assumed already to reflect the required development, trend, and portfolio adjustments.

Exposure indication. Multiply the stated exposure-rated cost per unit by current-equivalent units: $$ E=80{,}000\times \$6.25=\$500{,}000. $$

Burning-cost indication. First calculate historical layer loss cost per comparable exposure unit: $$ \frac{\$2{,}160{,}000}{360{,}000}=\$6.00\text{ per unit}. $$ Apply it to the 80,000 current-equivalent units: $$ B=80{,}000\times \$6.00=\$480{,}000. $$

Selection. Assign 35% to burning cost and the remaining 65% to exposure rating: $$ S=0.35(\$480{,}000)+0.65(\$500{,}000) =\$168{,}000+\$325{,}000 =\$493{,}000. $$

The selection lies between $480,000 and $500,000, as it must for a weighted average with a weight between zero and one. The method avoids a common mistake: multiplying the $500,000 exposure indication by 35% and the $480,000 burning-cost indication by 65%, which would reverse the stated weight.

What the Weight Means

A stated credibility weight represents the chosen relative influence of the two estimates. If a pricing instruction assigns $Z=0.35$ to burning cost, that indication supplies 35% of the selected estimate and exposure rating supplies 65%. The weight is not a percentage adjustment to the premium, nor does it mean that 35% of the portfolio is “historical” exposure. It is a weight on the estimates themselves.

The weight may be selected under a prescribed procedure or judgment framework. This tutorial’s blending formula does not determine a credibility weight from exposure volume, claim count, or layer characteristics. If a problem gives the weight, use it as stated. If a problem asks for a weight that minimizes variance under specified assumptions, the statistical result below provides one particular method.

Worked Example: Apply a Weight to Total-Loss Indications

For a hypothetical liability layer, the exposure-rated annual expected loss is $1,050,000. Adjusted historical losses are $2,400,000 over 300,000 comparable exposure units. The current portfolio has 120,000 units, and burning cost receives weight $Z=0.60$. Calculate the selected expected loss.

Method. Calculate the burning-cost rate per unit, apply it to current exposure, and then blend the two total-loss indications. The data are stated on a common annual, undiscounted basis, and the adjusted historical amount is treated as suitable for comparison.

The historical burning-cost rate is $$ \frac{\$2{,}400{,}000}{300{,}000}=\$8.00\text{ per unit}. $$ Therefore the current-portfolio burning-cost indication is $$ B=120{,}000\times \$8.00=\$960{,}000. $$ The exposure indication is already a total for the current portfolio, so $E=\$1{,}050{,}000$. Apply the stated weight: $$ S=0.60(\$960{,}000)+0.40(\$1{,}050{,}000) =\$576{,}000+\$420{,}000 =\$996{,}000. $$

The answer is between $960,000 and $1,050,000. Using $2,400,000 directly as $B$ would be a units error: that is the historical loss total, not the burning-cost indication for the current portfolio.

Variance-Minimizing Weight: Derivation and Conditions

A variance-minimizing weight is available under a specific statistical model. Suppose $E$ and $B$ estimate the same fixed expected loss $\theta$, and their estimation errors are independent with variances $\sigma_E^2$ and $\sigma_B^2$. The blend is $S=(1-Z)E+ZB$. Independence is important here because it makes the covariance between the estimates zero.

The derivation minimizes the variance of the blend. Independence gives

\operatorname{Var}(S)=(1-Z)^2\sigma_E^2+Z^2\sigma_B^2.

Differentiate this quadratic with respect to $Z$ and set the derivative to zero:

\frac{d}{dZ}\operatorname{Var}(S) =-2(1-Z)\sigma_E^2+2Z\sigma_B^2=0.

Rearranging,

Z(\sigma_E^2+\sigma_B^2)=\sigma_E^2, \qquad Z=\frac{\sigma_E^2}{\sigma_E^2+\sigma_B^2}.

When the sum of the variances is positive, this weight minimizes variance. The second derivative is $2(\sigma_E^2+\sigma_B^2)>0$, so the stationary point is a minimum. It assigns more weight to the estimate with the smaller variance: if burning cost is more variable, $Z$ decreases. If both variances are zero, every blend has zero variance and there is no unique variance-minimizing weight.

Unbiasedness is not needed for this variance-minimization calculation. It is needed if the goal is for the blend to be unbiased for $\theta$: if both $E$ and $B$ are unbiased, their weighted average is unbiased. If either estimate is biased, the variance-minimizing weight still has the formula above under independence, but the blend need not be unbiased. Minimizing mean squared error in that case requires accounting for bias as well. If the errors are dependent, the covariance must also be included, so the independent-error formula does not apply as written.

Worked Example: Calculate a Variance-Based Burning-Cost Weight

Two independent estimates of a hypothetical layer’s annual expected loss are available. The exposure indication is $1,200,000 with standard error $120,000; the burning-cost indication is $1,000,000 with standard error $80,000. Assume both estimates are unbiased for the same expected loss and their estimation errors are independent. Find the variance-minimizing burning-cost weight and selected loss.

Method and assumptions. Use the independent-error variance-minimizing formula. The standard errors are expressed in dollars, so square them to obtain variances in squared dollars: $$ \sigma_E^2=(\$120{,}000)^2=14{,}400{,}000{,}000, \qquad \sigma_B^2=(\$80{,}000)^2=6{,}400{,}000{,}000. $$ The burning-cost weight is $$ Z=\frac{14{,}400{,}000{,}000} {14{,}400{,}000{,}000+6{,}400{,}000{,}000} =\frac{14.4}{20.8} =0.692307\ldots. $$

Thus the exposure weight is $1-Z=0.307692\ldots$. Using the unrounded ratio for the calculation, $$ S=(0.692307\ldots)(\$1{,}000{,}000) +(0.307692\ldots)(\$1{,}200{,}000) =\$692{,}307.69+\$369{,}230.77 =\$1{,}061{,}538.46. $$ Rounded to the nearest dollar, the selected loss is $1,061,538. The selection is closer to the burning-cost estimate, which has the smaller standard error. A common error is to use $\sigma_B^2$ in the numerator for the burning-cost weight; under the derived formula, the numerator for $Z$ is the variance of the other estimate, $E$.

Common Checks and Pricing Interpretation

A blended expected layer loss is still a loss indication, not automatically a final premium. Expenses, commissions, taxes, risk loads, and other pricing provisions are separate considerations. The blend answers the narrower question of how to combine the two stated expected-loss estimates. Applying later pricing provisions to the selected loss is a separate calculation.

Check comparability. Do not blend a historical layer total on an old exposure base directly with a current exposure-rated total unless the problem explicitly establishes that they are comparable.
Check which weight belongs to which source. In $S=ZB+(1-Z)E$, $Z$ multiplies burning cost.
Check units. A loss cost per unit must be multiplied by the relevant number of units before it is blended with a total loss.
Check the range. For $0\leq Z\leq1$, the selected estimate cannot be below both inputs or above both inputs.
Do not overinterpret the variance formula. It requires independent estimation errors for the stated result. A prescribed credibility weight need not be the variance-minimizing weight.

Key takeaway: First make the exposure and burning-cost indications comparable; then apply the burning-cost weight to $B$ and the remaining weight to $E$. A variance-based weight is justified only under its stated error assumptions, and the selected amount should pass both a units check and a range check.

Check Your Understanding

For each hypothetical problem, calculate or identify what is requested. Treat amounts as annual and undiscounted; assume all explicitly described adjustments have already been made.

A portfolio has 45,000 current-equivalent units and an exposure-rated layer cost of $11.20 per unit. Adjusted historical losses are $1,890,000 over 210,000 comparable units. Burning cost receives weight 0.40. Find both current-portfolio indications and the selected loss. (Answer: $E=\$504{,}000$, $B=\$405{,}000$, $S=\$464{,}400$.)
An exposure indication for a hypothetical layer is $1,020,000. Adjusted historical losses of $2,520,000 arose over 180,000 comparable units, and the current portfolio has 75,000 units. Burning cost receives weight 0.25. Find the selected loss. (Answer: $B=\$1{,}050{,}000$; $S=\$1{,}027{,}500$.)
A layer has an exposure indication of $640,000. Adjusted historical losses total $1,320,000 over 120,000 comparable units; current exposure is 50,000 units. Burning cost receives weight 0.70. Calculate the burning-cost indication and selected loss. (Answer: $B=\$550{,}000$; $S=\$577{,}000$.)
Independent unbiased exposure and burning-cost estimates have standard errors of $100,000 and $200,000, respectively. Their indications are $900,000 and $1,100,000. Find the variance-minimizing burning-cost weight and the selected loss. (Answer: $Z=0.20$; $S=\$940{,}000$.)
An analyst proposes blending a current exposure-rated total with an historical loss total that has not been adjusted for development, trend, or changes in exposure. What must be addressed before the analyst can interpret the result as a selected current expected layer loss? (Answer: Adjust the historical experience as needed and put both indications on the same layer, period, cost level, and portfolio basis before blending.)

What You'll Learn