Why a Changing Risk Mix Matters
Historical experience is useful for ratemaking only to the extent that it represents the risks to which the indicated rate will apply. A portfolio can have a higher aggregate loss cost in a later period even when risk-specific loss costs have not changed, simply because the portfolio contains a larger share of higher-risk exposures. Conversely, an improving aggregate result may reflect a shift toward lower-risk business rather than an improvement in the underlying risks.
Credibility does not automatically correct this problem. A credibility formula combines observed experience with a collective estimate under its model assumptions. If the observed experience is not comparable to the target population because its risk mix changed, the credibility-weighted result can retain that distortion. The first task is therefore to identify the population represented by each observation and, where appropriate, restate experience to a common mix.
Let \(g=1,\ldots,G\) index mutually exclusive risk groups. In period \(t\), let \(E_{tg}\) be the exposure in group \(g\), \(L_{tg}\) the corresponding loss, and \(r_{tg}=L_{tg}/E_{tg}\) the group loss cost. The period's exposure share in group \(g\) is \(p_{tg}=E_{tg}/\sum_h E_{th}\). Provided all groups use the same exposure basis and loss definition, the aggregate rate is the exposure-weighted average of the group rates:
If \(p_{tg}\) changes over time, then \(r_t\) can change even if every \(r_{tg}\) stays constant. For a chosen target mix \(q_g\), where \(q_g\geq 0\) and \(\sum_g q_g=1\), the standardized rate is:
This calculation answers a different question from the aggregate rate. The aggregate rate describes the actual mix in period \(t\); the standardized rate estimates what that period's group experience would imply under the fixed target mix. Standardization does not, by itself, adjust for trend, benefit changes, claim handling changes, or random fluctuation within groups. Those issues require separate consideration.
Why Standardization Separates Mix From Rate Changes
The basic justification follows directly from the weighted-average identity. Suppose each group's rate is unchanged between two periods, so \(r_{1g}=r_{2g}=r_g\), but the shares differ. Then:
The aggregate difference is entirely due to changes in shares; there is no within-group rate change in this case. Under a fixed target mix, both standardized rates instead equal \(\sum_g q_g r_g\), so their difference is zero. More generally, when rates as well as shares change, standardization removes the effect of using different shares in the two periods, while leaving differences in the group rates visible.
Worked Example: Aggregate Change With No Within-Group Change
Worked Example: Aggregate Change With No Within-Group Change
A hypothetical equipment insurer has two risk groups. Group A has a loss cost of $500 per exposure and Group B has a loss cost of $1,500 per exposure in both years. Each year has 1,000 exposures. In Year 1, 800 exposures are in A and 200 in B. In Year 2, 200 are in A and 800 in B. Calculate both aggregate rates and standardize both years to a 50% A, 50% B mix.
Method. First calculate each year's actual aggregate rate from its exposure shares. Then apply the same target shares to the unchanged group rates.
Year 1. The shares are \(p_{1A}=800/1{,}000=0.8\) and \(p_{1B}=200/1{,}000=0.2\). The aggregate rate is:
The implied total loss is \(800(500)+200(1{,}500)=400{,}000+300{,}000=\$700{,}000\), and \(\$700{,}000/1{,}000=\$700\) per exposure, which checks the rate.
Year 2. The shares are \(p_{2A}=0.2\) and \(p_{2B}=0.8\). Therefore:
The implied total loss is \(200(500)+800(1{,}500)=100{,}000+1{,}200{,}000=\$1{,}300{,}000\), which divided by 1,000 exposures is $1,300.
Standardize both years. With \(q_A=q_B=0.5\), each standardized rate is:
The aggregate rate rose by $600, but the standardized rate did not change. That increase is explained entirely by the shift toward Group B. Common mistake avoided: treating the raw aggregate increase as evidence that the underlying group rates worsened.
Using Standardized Experience in Credibility
In a Bühlmann-style credibility calculation, the usual form of an estimate is \(Z\) times the observed experience plus \(1-Z\) times a collective mean. For this tutorial, the important issue is the meaning of the observed experience: it must estimate the quantity being rated. If the target is a rate for a specified mix, use experience standardized to that mix, not an aggregate that represents a different mix.
For a single standardized observation \(X^{\mathrm{std}}\), a credibility estimate can be written:
where \(\mu^{\mathrm{std}}\) is the collective expected rate on the same target-mix basis. This expression is appropriate only if the observation and collective mean refer to the same target, measurement basis, and relevant period adjustments, and if the credibility model's assumptions are suitable. For multiple periods, the experience average must use the weights justified by the data and model; equal weights should not be replaced by unequal weights without a reason.
If credible group-specific rates are available, another approach is to estimate each group's rate and then combine the estimates using fixed target shares. For group \(g\), let \(\widehat r_g=Z_g r_g+(1-Z_g)\mu_g\), where \(r_g\) is that group's observed rate and \(\mu_g\) its collective rate. The target-mix result is \(\sum_g q_g\widehat r_g\). This approach requires group-specific credibility estimates and should not be confused with applying a single credibility factor to an aggregate that has a changing mix.
Worked Example: Credibility After Standardizing Each Period
A hypothetical class has two risk groups, A and B. In Period 1, their rates are $600 and $1,200, and the observed mix is 80% A and 20% B. In Period 2, their rates are $600 and $1,600, and the observed mix is 8% A and 92% B. Standardize each period to a 50%/50% mix. Give the two-period averages using equal period weights, then calculate credibility estimates with \(Z=2/3\) and a collective standardized mean of $1,000.
Method. Calculate raw and standardized rates separately in each period. Use equal weights for the two period rates in both averages, then apply credibility. The stated equal-weight instruction determines the averaging; it is not an exposure-weighted average across periods.
Period 1 raw rate. The aggregate is:
Period 1 standardized rate. At a 50%/50% target mix, it is:
Period 2 raw rate. The aggregate is:
Period 2 standardized rate. The target-mix rate is:
Equal-weight averages. The raw average is \((720+1{,}520)/2=\$1{,}120\). The standardized average is \((900+1{,}100)/2=\$1{,}000\).
Credibility estimates. Applying the given formula to the raw average would give:
Applying it to the standardized average gives:
The result based on raw data is $80 higher because the raw average includes the changing mix. The standardized estimate equals the collective mean here because the standardized average itself equals that mean. Check: the two-period raw average is $1,120, not an unequal-weight average; the standardized average is $1,000. Common mistake avoided: inadvertently assigning unequal weights to the periods despite an equal-weight instruction.
Worked Example: Standardize Groupwise Credibility Estimates
Worked Example: Standardize Groupwise Credibility Estimates
A hypothetical specialty insurer targets a portfolio that is 70% Group A and 30% Group B. For a class under review, the observed losses and exposures are $126,000 on 180 exposures in A and $96,000 on 60 exposures in B. The group collective rates are $600 and $1,400, respectively. Given \(Z_A=0.50\) and \(Z_B=0.25\), calculate the target-mix credibility estimate and compare it with the raw class rate.
Method. Calculate each group's observed rate, credibility-weight it with its own group collective rate, and then apply the target shares. Finally, calculate the unstandardized aggregate using the class's actual exposures.
Group A observed rate: \(\$126{,}000/180=\$700\) per exposure. Its credibility estimate is:
Group B observed rate: \(\$96{,}000/60=\$1{,}600\) per exposure. Its credibility estimate is:
Target-mix estimate:
Raw class rate. Total losses are \(\$126{,}000+\$96{,}000=\$222{,}000\), and total exposures are \(180+60=240\). Thus the raw class rate is \(\$222{,}000/240=\$925\). Equivalently, the actual mix is 75% A and 25% B, giving \((0.75)(700)+(0.25)(1{,}600)=525+400=\$925\).
The $890 estimate is on the specified 70%/30% target basis and includes the stated group credibility weights; the $925 raw rate reflects the observed 75%/25% mix and no credibility adjustment. These figures answer different questions. Check: the target weights sum to 1, and \(455+435=890\). Common mistake avoided: using the observed exposure shares instead of the requested target shares when combining group estimates.
Practical Checks and Limitations
State exactly which population the final rate should represent and give its group weights \(q_g\). Confirm that the weights are nonnegative and sum to 1.
Use consistent exposure units, loss definitions, coverage terms, and period adjustments. A mix adjustment does not correct different limits, deductibles, inflation, or claim practices.
Calculate \(r_{tg}\) and \(p_{tg}\) before interpreting aggregate changes. This reveals whether the issue is composition, group experience, or both.
Form the observation on the target basis and ensure the collective mean is on that same basis. Use only weights supported by the stated credibility model and data.
Standardization can be unstable when a group has little experience, and it cannot create evidence for a group absent from the data. If group-specific rates are noisy, credible estimates may be needed before standardizing; the resulting estimate still depends on the group definitions, collective rates, and credibility assumptions. If material risk characteristics are omitted, a standardized result may remain noncomparable because the selected groups do not capture the true mix shift.
A further exam trap is to read a changed aggregate rate as a changed expected cost without checking the exposure distribution. Another is to standardize the observed rates but leave the collective mean on the original portfolio mix. A credibility blend is interpretable only when both terms estimate the same target quantity. These checks are aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.
Check Your Understanding
For each problem, distinguish the observed mix from the target mix and show the calculation before checking the final answer.
- A portfolio has two groups with rates of $400 and $1,000. In one period, 70% of exposures are in the first group and 30% in the second. Find the aggregate rate and the rate standardized to a 50%/50% mix. (Answer: Aggregate $580; standardized $700.)
- Two periods have group rates \((500,1{,}100)\) and \((700,1{,}300)\), for Groups A and B. The target mix is 60% A and 40% B. Calculate each standardized rate and their equal-weight average. (Answer: Period rates $740 and $940; equal-weight average $840.)
- A class has two equal-weight period rates of $880 and $1,120. Their standardized counterparts are $800 and $1,000. With \(Z=0.50\) and a collective standardized mean of $900, calculate the credibility estimate from the raw average and from the standardized average. (Answer: Raw-based estimate $950; standardized estimate $900.)
- For Groups A and B, observed rates are $600 and $1,400; collective rates are $500 and $1,200; and credibility factors are 0.40 and 0.60. Standardize the two group credibility estimates to a target mix of 25% A and 75% B. (Answer: Group estimates $540 and $1,320; target-mix estimate $1,125.)
- A period has a raw aggregate rate of $1,050, while standardization to the stated target mix gives $900. The collective mean is also $900 and \(Z=0.75\). Calculate the raw-based and standardized credibility estimates, and state what the difference indicates. (Answer: Raw-based estimate $1,012.50; standardized estimate $900; the raw result retains a mix effect relative to the target.)