Why Blend Exposure Curves?
An internal exposure curve reflects the portfolio being rated, but it may be volatile when the portfolio has limited experience, especially at high limits where claims are sparse. An external benchmark can provide a more stable reference, but may not represent the portfolio’s coverage, mix, or severity pattern exactly. Credibility blending provides a structured way to combine these sources while retaining the internal indication to the extent supported by its experience.
The blending operation should be applied to the curve values before calculating increased limits factors (ILFs). A factor is a ratio of two curve values, so an internal factor and benchmark factor need not have the same base denominator. Directly averaging those ratios generally gives a different answer from blending the curves and then taking the ratio.
For the usual cumulative exposure-curve interpretation, \(C(L)\) is the expected loss limited at \(L\), divided by the ground-up expected loss. It is therefore a cumulative proportion, generally nondecreasing in \(L\). This tutorial assumes the internal and benchmark values have already been put on a consistent loss basis and have the same definition of limit and curve. A credibility blend cannot repair inconsistencies in loss development, trend, coverage, or censoring adjustments; those need to be addressed before blending.
Blending the Curve, Then Taking the Ratio
At each limit, the blended value is a weighted average of the two curve values. Apply the same selected \(Z\) across the limits being compared unless the method explicitly justifies limit-specific weights. A common weight keeps the blended curve tied to one stated credibility judgment and avoids a curve whose shape changes merely because the weight was changed from one limit to another.
The resulting ILF is not generally the weighted average of the component ILFs. The reason follows by writing each component curve value at the target limit as its base value multiplied by its own ILF.
For internal and benchmark factors \(F_I=C_I(L)/C_I(b)\) and \(F_B=C_B(L)/C_B(b)\), the target values are \(C_I(b)F_I\) and \(C_B(b)F_B\).
The blended ILF is \(\{ZC_I(b)F_I+(1-Z)C_B(b)F_B\}/\{ZC_I(b)+(1-Z)C_B(b)\}\).
This ratio equals \(\alpha F_I+(1-\alpha)F_B\), where \(\alpha=ZC_I(b)/\{ZC_I(b)+(1-Z)C_B(b)\}\). The factor weights depend on the base curve values, so they are not generally \(Z\) and \(1-Z\).
This derivation also explains the special cases. If \(Z=1\), the result is the internal factor; if \(Z=0\), it is the benchmark factor. If the internal and benchmark curve values at the base limit are equal, then \(\alpha=Z\), and direct factor blending happens to agree. Otherwise, curve blending and direct factor blending generally differ.
Worked Example: Calculate the ILF from Blended Curve Values
A hypothetical portfolio has an internal curve value of 0.52 at the base limit and 0.78 at a target limit. A benchmark curve has corresponding values of 0.43 and 0.72. Use \(Z=0.70\) for the internal curve. Calculate the blended-curve ILF and compare it with the direct weighted average of the component ILFs. Round results to three decimals.
Method. First blend the curve values separately at the base and target limits. Then divide the blended target value by the blended base value.
At the base limit, $$ C_Z(b)=0.70(0.52)+0.30(0.43)=0.364+0.129=0.493. $$ At the target limit, $$ C_Z(L)=0.70(0.78)+0.30(0.72)=0.546+0.216=0.762. $$ Therefore, the ILF from the blended curve is $$ \operatorname{ILF}_Z(L;b)=\frac{0.762}{0.493}=1.5456389\ldots=1.546. $$
For comparison, the internal factor is \(0.78/0.52=1.500\), and the benchmark factor is \(0.72/0.43=1.6744186\ldots\). Their direct weighted average is $$ 0.70(1.500)+0.30(1.6744186\ldots)=1.050+0.5023256\ldots=1.5523256\ldots=1.552. $$ The two results differ because the component factors use different base curve values, 0.52 and 0.43. The correct result under the stated curve-blending method is 1.546. Common mistake avoided: treating the credibility weight on curve values as though it were automatically the same weight on the component factors.
Check. The blended target value exceeds the blended base value, so the resulting factor is greater than 1. The result also lies between the two component factors.
How the Credibility Weight Affects Factors
The credibility weight controls the relative influence of the internal curve at every limit. When \(Z\) increases toward 1, the blended curve values move toward the internal values; when \(Z\) decreases toward 0, they move toward the benchmark. But the resulting ILF need not move linearly with \(Z\), because both its numerator and denominator change.
The following example shows why it is useful to calculate factors at more than one target limit. The same weight is applied at the base and both target limits, but each ILF is calculated with the blended base value as its denominator.
Worked Example: Compare Indications at Two Target Limits
For a hypothetical liability portfolio, the internal curve values at the base limit, a 250,000 target limit, and a 500,000 target limit are 0.40, 0.66, and 0.82. The benchmark values at those limits are 0.44, 0.62, and 0.78. Use \(Z=0.60\). Calculate both target ILFs, rounding to three decimals.
At the base limit, $$ C_Z(b)=0.60(0.40)+0.40(0.44)=0.240+0.176=0.416. $$ At the 250,000 limit, $$ C_Z(250{,}000)=0.60(0.66)+0.40(0.62)=0.396+0.248=0.644, $$ so $$ \operatorname{ILF}_Z(250{,}000;b)=\frac{0.644}{0.416}=1.5480769\ldots=1.548. $$ At the 500,000 limit, $$ C_Z(500{,}000)=0.60(0.82)+0.40(0.78)=0.492+0.312=0.804, $$ and therefore $$ \operatorname{ILF}_Z(500{,}000;b)=\frac{0.804}{0.416}=1.9326923\ldots=1.933. $$
The factors increase with the target limit, consistent with the increasing blended curve values. As a check on the role of \(Z\), the internal factors are \(0.66/0.40=1.650\) and \(0.82/0.40=2.050\), while the benchmark factors are \(0.62/0.44=1.409\) and \(0.78/0.44=1.773\). Each blended result falls between its corresponding component factors.
Common mistake avoided: using the internal base value of 0.40 or the benchmark base value of 0.44 as the denominator for a blended target value. Both target ILFs must use the blended base value, 0.416, because the indicated factors are derived from the blended curve.
Selecting and Stating a Credibility Weight
A credibility weight should have an explicit rationale. It may be prescribed for an analysis, selected by a documented judgmental process, or obtained from a credibility framework appropriate to the data. Do not present a convenient weight as though it were uniquely determined by the curve values alone.
One illustrative volume-based rule is the Bühlmann form \(Z=n/(n+K)\), where \(n\) is the volume of relevant internal experience and \(K\) is a positive credibility parameter. In the classical Bühlmann model, \(K\) is the ratio of expected process variance to the variance of hypothetical means. Applying this form to a curve blend requires an appropriate definition of experience volume and a rationale for the parameter; it is not a universal rule for every exposure-curve estimation problem. A common \(Z\) across limits is a modeling choice that should be stated.
Worked Example: Apply a Volume-Based Credibility Rule
Suppose a hypothetical internal curve is based on \(n=80\) effective exposure units. Use the stipulated Bühlmann-style rule \(Z=n/(n+K)\) with \(K=120\). The internal curve values at the base limit, a 150,000 limit, and a 400,000 limit are 0.36, 0.58, and 0.76. The benchmark values are 0.40, 0.54, and 0.70. Find the credibility weight and the two blended ILFs. Round the weight and factors to three decimals.
First calculate the weight: $$ Z=\frac{80}{80+120}=\frac{80}{200}=0.400. $$ The benchmark weight is \(1-Z=0.600\). The blended base value is $$ C_Z(b)=0.400(0.36)+0.600(0.40)=0.144+0.240=0.384. $$ At 150,000, $$ C_Z(150{,}000)=0.400(0.58)+0.600(0.54)=0.232+0.324=0.556, $$ giving $$ \operatorname{ILF}_Z(150{,}000;b)=\frac{0.556}{0.384}=1.4479167\ldots=1.448. $$ At 400,000, $$ C_Z(400{,}000)=0.400(0.76)+0.600(0.70)=0.304+0.420=0.724, $$ so $$ \operatorname{ILF}_Z(400{,}000;b)=\frac{0.724}{0.384}=1.8854167\ldots=1.885. $$
The internal factors at the two target limits are \(0.58/0.36=1.6111\ldots\) and \(0.76/0.36=2.1111\ldots\); the benchmark factors are \(0.54/0.40=1.350\) and \(0.70/0.40=1.750\). The blended factors lie between their respective component factors. Common mistake avoided: calculating \(Z\) correctly but then blending component ILFs with that same weight without checking whether the base curve values match.
The weight is meaningful only to the extent that \(n\) and \(K\) are defensible for this application. It is not enough to label a weight “credibility” without identifying the experience measure, rule, and parameter used.
Practical Checks Before Using a Blend
A pointwise blend is easy to compute, but its usefulness depends on the inputs and on how the result is applied. Before relying on an indicated factor, check the following:
- Consistent basis. Confirm that internal and benchmark curves use compatible definitions of loss, limit, coverage, and adjustment basis. Differences in basis can masquerade as differences in curve shape.
- Valid curve values. Check that values are within the intended range and are nondecreasing with limit. The blended values inherit monotonicity if both input curves are nondecreasing and the same \(Z\) is used at each limit.
- One denominator. Calculate every blended ILF using the blended curve value at the selected base limit, not a component-specific base value.
- Weight interpretation. Record why the chosen weight is appropriate and whether it is judgmental, prescribed, or derived from a stated credibility rule.
- Limit range. Do not assume blending resolves the limitations of either input curve outside its supported range. Extrapolation and the associated uncertainty require separate consideration.
A useful diagnostic is to calculate the internal, benchmark, and blended factors together at each target limit. With a common \(Z\) and positive base values, the blended factor must lie between the two component factors, as the effective-weight derivation shows. If it does not, check the arithmetic, the base denominators, and whether a different weight was used at different limits.
These methods are aligned with the CAS Exam 8 (Advanced Ratemaking) syllabus. The central exam skill is to distinguish credibility weighting of curve values from credibility weighting of factors and to keep the base limit consistent throughout the calculation.
Check Your Understanding
For each problem, calculate from the stated curve values and use the stated credibility rule or weight. Round factors to three decimals unless the question asks for a weight only.
- An internal curve has values 0.45 at the base and 0.72 at a target limit; the benchmark has values 0.50 and 0.68. Use \(Z=0.30\). Calculate the blended-curve ILF. (Answer: blended base \(=0.485\), blended target \(=0.692\), ILF \(=1.427\))
- At a base limit, the internal and benchmark values are 0.38 and 0.42. At a higher limit, they are 0.63 and 0.59. Use \(Z=0.50\). Calculate the blended ILF. (Answer: blended base \(=0.400\), blended target \(=0.610\), ILF \(=1.525\))
- Use \(Z=n/(n+K)\) with \(n=60\) and \(K=90\). At the base limit, the internal and benchmark curve values are 0.30 and 0.36; at the target limit, they are 0.57 and 0.60. Find \(Z\) and the blended ILF. (Answer: \(Z=0.400\), blended base \(=0.336\), blended target \(=0.588\), ILF \(=1.750\))
- The internal curve has base and target values 0.40 and 0.68; the benchmark has values 0.50 and 0.70. With \(Z=0.60\), calculate the ILF from the blended curve and compare it with the direct weighted average of component factors. (Answer: blended-curve ILF \(=0.688/0.440=1.564\); direct factor blend \(=0.60(1.700)+0.40(1.400)=1.580\))
- An internal curve has base and target values 0.35 and 0.63; a benchmark has values 0.45 and 0.69. Find the weight \(Z\) that makes the blended-curve ILF exactly 1.600. (Answer: \(Z=0.300\); blended base \(=0.420\), blended target \(=0.672\), and \(0.672/0.420=1.600\))