From Analysis to a Defensible Rationale
A fairness review does more than report a disparity metric. It explains why a rating factor is being considered, what evidence connects it to expected insurance costs, what disparities appear under the selected comparisons, and what the evidence cannot establish. The resulting rationale should make the reasoning traceable without presenting a statistical finding as a legal conclusion.
The earlier material on fairness metrics and calibration tradeoffs provides tools for describing group outcomes. A defensible rationale uses those tools in context: it states the measurement choices, distinguishes observed results from interpretation, and identifies the limitations that matter to the decision. It should be possible for a reviewer to follow the chain from factor definition to evidence to proposed action.
A useful rationale keeps four questions distinct. First, what cost component is the factor intended to predict or reflect? Second, what evidence supports that relationship, and how reliable is it? Third, what disparity evidence is relevant, and how does it depend on the chosen metric or comparison? Fourth, what constraints or competing objectives limit the available choices?
Connect the Factor to Expected Costs
Begin by stating the factor precisely: how it is defined, how it is measured, and how its categories are assigned. Then identify the cost outcome being assessed, such as claim frequency, average severity, or pure premium. Explain the observed relationship using the relevant experience or model results, including the period, population, and material adjustments used. A factor’s predictive association with cost is evidence for its actuarial purpose, but association alone does not establish causation or resolve whether its use is permissible.
Where rates and expected costs can be compared on a consistent basis, the comparison can expose calibration differences. Define the rate-to-cost ratio for group \(g\) as \(R_g/C_g\), where \(R_g\) is the rate or premium measure and \(C_g\) is expected cost on the same exposure and coverage basis. Ratios are meaningful only when the compared quantities use compatible units, coverage, time periods, and expense treatment.
This identity follows by dividing the rate ratio by the cost ratio and rearranging: \((R_A/R_B)/(C_A/C_B)=(R_A/R_B)(C_B/C_A)=(R_A/C_A)/(R_B/C_B)\). Thus, the relative comparison is the ratio of the groups’ rate-to-cost ratios. A value below one means Group A’s rate is lower relative to its expected cost than Group B’s rate is relative to its expected cost. It does not mean the absolute rate is below cost, nor does it identify the cause of the difference.
Worked Examples: Building and Checking the Evidence
Worked Example: Comparing Rate Ratios With Cost Ratios
A hypothetical insurer compares two rating groups on the same coverage and annual exposure basis. Group A has a rate of $840 and estimated expected cost of $700. Group B has a rate of $700 and expected cost of $560. Calculate each ratio and state what the comparison supports. Assume the cost estimates and rates are expressed on a consistent basis.
Step 1: Calculate the rate ratio. Group A’s rate divided by Group B’s rate is \(840/700=1.20\).
Step 2: Calculate the expected-cost ratio. Group A’s expected cost divided by Group B’s expected cost is \(700/560=1.25\).
Step 3: Calculate each group’s rate-to-cost ratio. For Group A, \(840/700=1.20\). For Group B, \(700/560=1.25\).
Step 4: Compare the ratios. The relative comparison is \(1.20/1.25=0.96\). Equivalently, \((840/700)/(700/560)=1.20/1.25=0.96\). Group A’s rate-to-cost ratio is 96% of Group B’s, or 4% lower on this relative measure.
Conclusion and check. The rate ratio of 1.20 is below the cost ratio of 1.25, consistent with the relative comparison of 0.96. Group A’s rate is still 20% above its estimated cost, while Group B’s is 25% above its estimated cost. The result is a calibration comparison, not a finding about why the difference arose or whether it is legally acceptable. The method avoids the common mistake of reporting 1.20 as the relative rate-to-cost comparison; that is the rate ratio, whereas the requested comparison is 0.96.
Worked Example: Connecting a Rating Category to Cost Evidence
A hypothetical auto insurer is considering annual mileage bands as a rating factor. In a defined experience period, the lower-mileage band has 0.10 claims per exposure and average severity of $5,000. The higher-mileage band has 0.13 claims per exposure and the same average severity. A proposed rate indication is $540 for the lower-mileage band and $690 for the higher-mileage band. Treat the frequency and severity estimates as applicable to the same coverage and exposure definition, and assume the indicated rates are on the same basis as the expected costs.
Step 1: Estimate expected cost per exposure. Under the stated frequency-severity approach, expected cost per exposure is claim frequency multiplied by average severity. For the lower-mileage band, \(0.10\times \$5{,}000=\$500\). For the higher-mileage band, \(0.13\times \$5{,}000=\$650\).
Step 2: Compare the rate and cost ratios. The rate ratio is \(690/540=1.2778\) to four decimals. The cost ratio is \(650/500=1.30\). The relative comparison is \(1.2778/1.30=0.9829\) to four decimals. Using the unrounded rate ratio, \((690/540)/(650/500)=1.277777\ldots/1.30=0.982906\ldots\), which also rounds to 0.9829.
Step 3: Interpret the group-level calibration. The lower-mileage rate-to-cost ratio is \(540/500=1.08\). The higher-mileage ratio is \(690/650=1.061538\ldots\), or 1.0615 to four decimals. The higher-mileage group’s rate-to-cost ratio is about 98.29% of the lower-mileage group’s ratio.
Conclusion and check. The experience supports a relationship between mileage band and estimated expected cost in this hypothetical data: estimated cost rises from $500 to $650 per exposure. The proposed rates rise from $540 to $690, a somewhat smaller relative increase than the cost increase. That comparison does not establish that mileage causes losses or that the proposed factor is permissible in every jurisdiction. The common trap is to compare only the $150 absolute differences; the relative ratios show that costs rise 30% while rates rise about 27.78%.
Summarize Disparity Without Overclaiming
A rationale should identify which groups are compared, what outcome is measured, and why that measure is relevant to the decision. Where several reasonable measures give different views, report the material results rather than selecting only the most favorable one. Describe whether a result is a point estimate, an uncertainty range, a model-based estimate, or a descriptive statistic. State the comparison population and any exclusions that could affect interpretation.
Disparity evidence has boundaries. A difference in average rates does not, by itself, show a difference in expected costs. A difference in observed costs does not by itself establish that a rating factor is a suitable or allowable basis for charging different rates. Results can also depend on the chosen groups, period, exposure definition, adjustments, and uncertainty. When group volumes are small or estimates are unstable, say so directly and avoid treating a point estimate as exact.
Evidence of a factor’s predictive association should be separated from evidence about its possible proxy effects and from conclusions about fairness. Those are related questions, but they are not interchangeable. A defensible account names what was assessed, summarizes the result, and makes clear what remains outside the analysis.
Worked Example: Documenting an Uncertain Group Comparison
A hypothetical review estimates expected costs of $800 for Group A and $700 for Group B. The estimated standard errors are $80 and $25, respectively. Assume the estimates are independent and approximately normally distributed, so a 95% interval can be calculated as estimate plus or minus 1.96 standard errors. Proposed rates are $760 for Group A and $720 for Group B. Calculate the uncertainty in the cost difference and compare the point-estimate ratios.
Step 1: Calculate the estimated cost difference. Group A minus Group B is \(\$800-\$700=\$100\).
Step 2: Calculate the standard error of the difference. Independence means the variance of the difference is the sum of the variances. Therefore, the standard error is \(\sqrt{80^2+25^2}=\sqrt{6{,}400+625}=\sqrt{7{,}025}=83.815\) dollars, to three decimals.
Step 3: Calculate a 95% interval for the difference. The margin is \(1.96\times83.815=164.277\) dollars, to three decimals. The interval is \(\$100\pm\$164.277\), or approximately \([-\$64.28,\ \$264.28]\) to the nearest cent. It includes zero, so these estimates do not establish a clear difference at this approximate interval level.
Step 4: Compare the point-estimate ratios. The rate ratio is \(720/760=0.9474\) to four decimals; the expected-cost ratio is \(700/800=0.8750\). Their relative comparison is \(0.947368\ldots/0.875=1.0827\) to four decimals. The proposed rate ratio is about 8.27% higher than the point-estimate cost ratio.
Conclusion and check. The point estimates suggest that the rate relationship is less favorable to Group B than the cost relationship, but the wide uncertainty in the estimated cost difference calls for caution. The interval describes uncertainty under the stated approximation; it does not settle a legal or ethical question. A rationale could note that the comparison is inconclusive, explain the small-group uncertainty, and identify additional monitoring or evidence needed before relying heavily on the point estimate. The common mistake is to treat the $100 point difference as certain; its standard error is about $83.82, and the stated interval includes zero.
State Constraints and Unresolved Tradeoffs
A sound rationale separates actuarial evidence from constraints on how that evidence may be used. Applicable laws and regulations vary by jurisdiction, line of business, and rating practice. They may restrict particular variables, classifications, data sources, or forms of discrimination. The rationale should identify the relevant review and its scope, record unresolved questions for appropriate legal or compliance review, and avoid presenting an actuarial opinion as a legal determination.
Practical limitations also matter. A factor may be supported by available data but measured with error, unavailable for some applicants, difficult to explain, or burdensome to administer. A change may affect rate adequacy, stability, affordability, competitiveness, or operational systems. Explain which considerations materially affected the recommendation and which were not evaluated. If a constraint prevents the most direct predictive approach, state the resulting limitation rather than implying that the adopted alternative is equally predictive.
State the factor, intended use, affected coverage, population, and decision being considered.
Identify the cost outcome, data period, comparison basis, and evidence linking the factor to expected costs, including important uncertainty.
Name the groups and measures, summarize the relevant rate and cost comparisons, and describe sensitivity to assumptions or alternative measures.
Identify legal, data, model, implementation, and consumer considerations, then state what remains uncertain and how it will be addressed.
The final rationale need not force a single conclusion where the evidence does not support one. It can state that cost indications support a factor while group comparisons remain uncertain, or that the evidence is predictive but a legal or operational constraint limits its use. A clear account distinguishes the observed facts, the actuarial interpretation, the decision, and any follow-up plan. That discipline makes the reasoning reviewable and aligned with the CAS Exam 8 (Advanced Ratemaking) syllabus.
Common Rationale Errors
- Confusing prediction with justification. A predictive relationship is relevant evidence, but it does not automatically resolve legal, fairness, or implementation questions.
- Mixing incompatible quantities. Rate and cost comparisons require the same coverage, exposure, time basis, and treatment of expenses or other included components.
- Reporting a ratio without its meaning. State whether the result compares rates, costs, or rate-to-cost ratios, and identify the reference group.
- Ignoring uncertainty. Sparse data, unstable estimates, and wide intervals should temper the strength of the conclusion.
- Hiding tradeoffs. If a constraint or competing objective affects the recommendation, describe it rather than implying that all objectives were simultaneously achieved.
Check Your Understanding
For each hypothetical case, calculate or identify the requested result and consider what a rationale should say.
- A rating group has a rate of $990 and expected cost of $900; a reference group has a rate of $900 and expected cost of $750. Calculate the rate ratio, cost ratio, and relative rate-to-cost comparison. (Answer: Rate ratio = 1.10; cost ratio = 1.20; relative comparison = 1.10/1.20 = 0.9167.)
- A proposed vehicle-use category has estimated frequency of 0.08 claims per exposure and average severity of $6,000. Its proposed rate is $504 per exposure on the same coverage basis. Calculate expected cost and the rate-to-cost ratio, then state what the comparison indicates. (Answer: Expected cost = $480; rate-to-cost ratio = $504/$480 = 1.05, or 5% above estimated cost.)
- Two groups have proposed rates of $600 and $720 and estimated costs of $500 and $660, respectively. Calculate rate ratio, cost ratio, and their relative comparison. (Answer: Rate ratio = 1.20; cost ratio = 1.32; relative comparison = 1.20/1.32 = 0.9091.)
- A cost estimate for a small group has a standard error of $50. Using the stated approximate 95% multiplier of 1.96, calculate the margin of error and give one appropriate limitation to record. (Answer: Margin = 1.96 × $50 = $98; record that the estimate is uncertain, so its point value should not be treated as exact.)
- A factor has a stable association with expected costs in the reviewed data, but its use may be restricted under rules that vary by jurisdiction. What should the rationale conclude, and what should it avoid claiming? (Answer: State the actuarial evidence and its scope, identify the jurisdiction-specific issue for legal or compliance review, and avoid claiming that predictive support alone establishes legal permissibility.)