Converting Model Predictions Into Rate Indications

From Fitted Losses to an Indication

A fitted model can estimate expected claim losses for a risk or rating cell, but that prediction is not automatically a premium. A rate indication must put the prediction on the correct exposure basis and, when the objective is a total-premium indication, account for selected expenses and other provisions. The distinction matters: a loss-cost indication describes expected losses per exposure unit, while a total-premium indication also provides for the specified expenses and premium-based charges.

The model prediction and the indication must refer to the same coverage, time period, and exposure definition. For example, an annual expected loss per policy-year should be multiplied by policy-years, not by the number of claims or the number of insured items unless those are the defined exposure units. Earlier material on generalized linear models supplies the framework for obtaining fitted predictions; here the task is to translate those predictions into indications.

Definition: Let $e_i$ be the exposure for risk or cell $i$, and let $\hat m_i$ be the model-predicted expected claim loss per unit of that exposure. The indicated aggregate claim loss is $\sum_i e_i\hat m_i$, and the indicated claim loss cost per exposure unit is $\sum_i e_i\hat m_i/\sum_i e_i$. If the model prediction is already an aggregate expected loss, do not multiply it by exposure again.

In this tutorial, a claim loss prediction excludes loss adjustment expense (LAE) unless the problem explicitly says otherwise. LAE that varies as a proportion of claim loss can be added as a separate provision. A fixed expense per exposure unit is added on that exposure basis. An expense or provision stated as a percentage of premium is different: because its dollar amount depends on the premium being calculated, it must be included in a gross-up equation.

Definition: A loss-cost indication is expected claim loss per exposure unit under a stated prediction basis. A total-premium indication is the premium required to fund the selected loss provision, fixed expenses, and provisions stated as percentages of premium. Whether LAE is included in the loss provision must be stated explicitly.

Building a Total-Premium Indication

Suppose the aggregate expected claim loss is $L$. Let $q$ be a selected LAE percentage of claim loss, $F$ be aggregate fixed expenses, and $v$ be the sum of all selected provisions stated as percentages of premium. Examples of such provisions may include commission, premium taxes, or other explicitly identified premium-based expenses. Under this setup, the total premium $P$ must fund $L$, LAE of $qL$, fixed expenses $F$, and premium-based provisions of $vP$.

$$ P=L+qL+F+vP $$

Collecting the premium-based term gives $(1-v)P=(1+q)L+F$. Thus, provided $v<1$, the indicated total premium is:

P=\frac{(1+q)L+F}{1-v}

This is a gross-up, not an extra percentage added to the pre-expense amount. The premium-based provision is $vP$, so it must be calculated from the final premium. If the indicated premium is needed per exposure unit, divide the aggregate premium by total exposure after calculating the aggregate components. The formula assumes the selected provisions are proportional to premium at the stated rates, the selected fixed expenses are not themselves premium-based, and no provision is counted in two places. It does not apply unchanged if, for example, an expense is a percentage of claim loss rather than premium; that expense belongs with the loss-based terms.

Worked Example: Aggregating Cell Predictions and Expenses

A hypothetical portfolio has three rating cells. A fitted model predicts annual claim loss per policy-year of $240, $360, and $500 for cells A, B, and C, respectively. Their exposures are 100, 50, and 20 policy-years. The selected LAE is 10% of claim loss, fixed expense is $30 per policy-year, and premium-based provisions total 22%. Calculate the portfolio loss cost and total-premium indication.

Method. First aggregate the predicted claim losses using their corresponding exposures. Then add loss-based LAE and fixed expenses, and gross up for the premium-based provisions.

Cell A has predicted claim losses of $100(240)=\$24{,}000$. Cell B has $50(360)=\$18{,}000$, and cell C has $20(500)=\$10{,}000$. Therefore, aggregate claim loss is $24{,}000+18{,}000+10{,}000=\$52{,}000$. Total exposure is $100+50+20=170$ policy-years.

The indicated loss cost is $\$52{,}000/170=\$305.8824$ per policy-year before rounding. LAE is $0.10(\$52{,}000)=\$5{,}200$, and fixed expense is $170(\$30)=\$5{,}100$. The amount to be funded before premium-based provisions is therefore $\$52{,}000+\$5{,}200+\$5{,}100=\$62{,}300$.

With $v=0.22$, total premium is $\$62{,}300/(1-0.22)=\$62{,}300/0.78=\$79{,}871.7949$, or $79,871.79 to the nearest cent. Per policy-year, that is $\$79{,}871.7949/170=\$469.8341$, or $469.83 to the nearest cent.

Check. Premium-based provisions are $0.22(\$79{,}871.7949)=\$17{,}571.7949$. Adding the components gives $\$52{,}000+\$5{,}200+\$5{,}100+\$17{,}571.7949=\$79{,}871.7949$, which agrees with the indicated premium. The common mistake avoided is applying 22% to the pre-premium amount and treating that dollar figure as the final provision.

Using Frequency-Severity Predictions

A model may predict expected claim loss directly, or it may provide predictions that are combined through a frequency-severity calculation. If frequency is the expected number of claims per exposure unit and severity is the expected claim amount, expected claim loss per exposure unit is their product, provided the predictions refer to the same risk, coverage, and period. This product supplies the loss prediction; it does not by itself supply LAE, fixed expenses, or premium-based provisions.

Result: When the expected claim count per exposure unit is $\hat f_i$ and expected severity per claim is $\hat s_i$, the predicted claim loss per exposure unit is $\hat m_i=\hat f_i\hat s_i$. Aggregate expected claim loss is then $\sum_i e_i\hat f_i\hat s_i$. This calculation requires frequency and severity to be defined consistently and the severity prediction to represent the expected payment per claim under the modeled coverage.

Worked Example: Frequency-Severity Predictions by Class

A hypothetical insurer models annual claim frequency and mean claim severity for two classes. Class A has predicted frequency 0.04 claims per policy-year and mean severity $3,000; its exposure is 80 policy-years. Class B has predicted frequency 0.07 and mean severity $5,000; its exposure is 120 policy-years. LAE is 7% of claim loss, fixed expense is $24 per policy-year, and premium-based provisions total 18%. Find the portfolio loss cost and total-premium indication.

Method. Multiply predicted frequency by mean severity to obtain expected claim loss per policy-year in each class, then exposure-weight the results.

For A, expected claim loss per policy-year is $0.04(\$3{,}000)=\$120$. Aggregate expected claim loss for A is $80(\$120)=\$9{,}600$. For B, it is $0.07(\$5{,}000)=\$350$ per policy-year and $120(\$350)=\$42{,}000$ in aggregate. Total expected claim loss is $\$9{,}600+\$42{,}000=\$51{,}600$, across $80+120=200$ policy-years.

The portfolio loss cost is $\$51{,}600/200=\$258.00$ per policy-year. LAE is $0.07(\$51{,}600)=\$3{,}612$, and fixed expense is $200(\$24)=\$4{,}800$. The pre-premium-provision amount is $\$51{,}600+\$3{,}612+\$4{,}800=\$60{,}012$.

The total premium is $\$60{,}012/(1-0.18)=\$60{,}012/0.82=\$73{,}185.3659$, or $73,185.37 to the nearest cent. The portfolio indication per policy-year is $\$73{,}185.3659/200=\$365.9268$, or $365.93 to the nearest cent.

Check. Premium-based provisions are $0.18(\$73{,}185.3659)=\$13{,}173.3659$. The components sum to $\$51{,}600+\$3{,}612+\$4{,}800+\$13{,}173.3659=\$73{,}185.3659$. The common mistake avoided is multiplying frequency by exposure and adding severity, rather than multiplying frequency by severity to get expected loss per exposure unit.

Model Output and Provision Bases

Before applying expenses, identify exactly what the model predicts. A fitted loss cost might represent claim payments only, claim payments plus allocated LAE, or another explicitly defined response. The expense calculation must follow that definition. If predicted losses already include LAE, adding a separate LAE percentage to those predictions counts the same provision twice. Conversely, if predictions exclude LAE and the selected indication requires it, omitting LAE understates the amount to be funded.

The same discipline applies to the denominator used for a percentage. A percentage of claim loss is calculated from claim loss; a percentage of premium is calculated from final premium. Fixed expenses are added as dollar amounts on the stated exposure basis. These bases cannot be interchanged just because all the items ultimately contribute to the premium.

Worked Example: Avoiding Double Counting of LAE

A hypothetical model predicts annual loss, including allocated LAE, of $350 per policy-year for a portfolio of 120 policy-years. Fixed expense is $20 per policy-year. Commission, taxes, and other selected provisions together equal 16% of premium. Calculate total premium and explain why no additional LAE percentage is applied.

Method. Since the model prediction already includes LAE, use it directly as the loss provision. Add only the separately selected fixed expense, then gross up for the premium-based provisions.

Aggregate predicted loss including LAE is $120(\$350)=\$42{,}000$. Aggregate fixed expense is $120(\$20)=\$2{,}400$. The amount before premium-based provisions is $\$42{,}000+\$2{,}400=\$44{,}400$. With $v=0.16$, total premium is $\$44{,}400/(1-0.16)=\$44{,}400/0.84=\$52{,}857.1429$, or $52,857.14 to the nearest cent. Per policy-year, the indication is $\$52{,}857.1429/120=\$440.4762$, or $440.48 to the nearest cent.

Check. Premium-based provisions are $0.16(\$52{,}857.1429)=\$8{,}457.1429$. Then $\$42{,}000+\$2{,}400+\$8{,}457.1429=\$52{,}857.1429$. The common mistake avoided is multiplying the $42,000 prediction by an additional LAE factor even though LAE is already included.

Checks Before Reporting an Indication

A calculated indication is only meaningful if its units and components are internally consistent. A practical sequence is to define the prediction basis, aggregate it on matching exposures, identify each provision's basis, and reconcile the resulting premium. This does not replace model validation or the selection of assumptions; it makes the translation from selected predictions and provisions arithmetically transparent.

Confirm the prediction.
Establish whether each model output is expected claim loss per exposure, expected loss per policy, or an aggregate amount, and whether it includes LAE.

Match the exposure.
Multiply per-unit predictions by exposures measured on the same basis and covering the same time period.

Classify provisions.
Apply loss-based percentages to their defined loss base, add fixed expenses on their stated basis, and reserve premium-based percentages for the gross-up.

Reconcile.
Verify that the final premium equals the sum of the loss provision, fixed expenses, and premium-based provisions.

Common exam traps include using the wrong exposure denominator, treating a frequency prediction as a loss-cost prediction, applying a premium percentage to losses rather than premium, and adding an expense twice because it is already embedded in the model response. Always state whether a reported amount is a loss cost or a total premium and name its unit, such as dollars per policy-year.

Key takeaway: A model supplies an expected-loss estimate on a defined basis; an indication translates that estimate into a rate or premium using explicitly selected provisions. Aggregate with matching exposures, apply each provision to its correct base, and check that the components reconcile to the final premium. This treatment is aligned with the CAS Exam 8 (Advanced Ratemaking) syllabus.

Check Your Understanding

For each problem, identify the requested basis and keep loss-based, fixed, and premium-based amounts separate.

A fitted model predicts claim loss of $180 per policy-year for 40 policy-years and $420 per policy-year for 60 policy-years. Calculate the portfolio claim loss cost before expenses. (Answer: $324.00 per policy-year.)
Expected claim loss is $275 per policy. LAE is 6% of claim loss, fixed expense is $18 per policy, and premium-based provisions are 20%. Find the total-premium indication per policy, rounded to cents. (Answer: $386.88.)
A frequency-severity model predicts frequency 0.03 and mean severity $4,000 for 50 policy-years in one class, and frequency 0.05 and mean severity $6,000 for 30 policy-years in another. LAE is 10% of claim loss, fixed expense is $12 per policy-year, and premium-based provisions are 15%. Find the aggregate total premium, rounded to cents. (Answer: $20,541.18.)
A portfolio has $40,000 of expected claim loss, LAE equal to 5% of claim loss, fixed expenses of $2,000, and premium-based provisions of 15%. Calculate the total-premium indication and the dollar amount of premium-based provisions, each rounded to cents. (Answer: total premium $51,764.71; premium-based provisions $7,764.71.)
A model predicts $52,000 of loss per portfolio, explicitly including LAE. Fixed expense is $2,000 and premium-based provisions are 10%. What is the total-premium indication, and should a separate LAE load be added? (Answer: $60,000.00; no separate LAE load should be added.)

What You'll Learn