Why Rating Variable Restrictions Require a Separate Review
A variable can have a clear definition and appear predictive in a dataset without being permitted for every rating use in a particular jurisdiction. A filing review therefore asks two related but distinct questions: what does the variable measure, and does the applicable rule allow this variable to be used in this way? Statistical support alone does not answer the second question.
The answer depends on the applicable jurisdiction, line of business, coverage, and filing requirements. There is no universal list of permitted rating variables that can replace checking the controlling sources. Those sources may include statutes, regulations, jurisdictional bulletins, approved rating plans, filing instructions, and required filing forms. A rule may limit a variable’s definition, its data source, the coverage to which it applies, or the conditions under which it may be used.
The practical task is to review the proposed variable and its use against the requirements that actually apply, then preserve enough information for another reviewer to understand and reproduce the conclusion. This is not a substitute for jurisdiction-specific legal interpretation. If the controlling language is unclear, document the ambiguity and seek the appropriate legal or compliance review rather than treating uncertainty as permission.
A Structured Screen for Variables and Their Uses
Start with the proposed use, not just the variable’s name. “Driving score,” for example, is not a sufficiently precise description to compare against a rule. Determine how the score is constructed, what information it uses, who or what it describes, and which coverage and premium components it would affect.
Confirm the jurisdiction, line, coverage, filing type, and effective period. Check the applicable legal sources, approved plan, and filing instructions. Record the source and the specific provision or instruction relied upon.
Specify the unit measured, data source, measurement period, transformations, categories, thresholds, and treatment of missing or invalid values. A label alone is not a reproducible definition.
Check whether the rule permits the variable for the relevant coverage and whether it limits the measurement method, source, or conditions. Review each use separately; permission for one application does not automatically extend to another.
Describe the data, period, exposure basis, calculations, and limitations relevant to the proposed use. Keep the empirical support distinct from the conclusion about what the rule permits.
Record whether the proposed use is permitted, not permitted, or requires clarification under the sources reviewed. State the allowed scope precisely and identify any use that is excluded.
A useful workpaper or filing exhibit can organize the review in a compact matrix. The matrix does not replace the governing source or analysis; it makes the comparison and the scope of the conclusion visible.
| Review item | What to record |
|---|---|
| Rule source | Jurisdiction, document, provision or instruction, and effective date |
| Variable definition | Measurement, source, period, transformations, categories, and missing-value handling |
| Proposed use | Line, coverage, premium component, and how the variable changes the rating |
| Conditions or limits | Any permitted-use restriction, required data source, or stated condition |
| Supporting analysis | Data period, exposure basis, calculations, checks, and material limitations |
| Conclusion | Permitted scope, excluded uses, and issues requiring clarification |
Separate Empirical Support From Permission
A supporting analysis may help explain why a proposed variable is relevant to expected costs. It cannot expand a restriction. The converse also matters: a rule allowing a variable does not establish that a proposed relativity or implementation is adequately supported. The filing should address both questions, in their proper order.
For a simple comparison of two groups, an observed loss cost can be calculated as losses divided by exposures, provided the loss and exposure measures are defined consistently for the groups. If group \(A\) is the reference, a descriptive relativity for group \(B\) is the loss cost for \(B\) divided by the loss cost for \(A\). This calculation describes the observed comparison; it does not establish legal permission, statistical significance, or that the result should be adopted as a final factor.
The exposure denominator matters because raw loss totals reflect both the cost experience and the amount of business observed. The ratio is interpretable only when the groups use comparable definitions, loss measures, and exposure bases. This is a descriptive calculation, not a substitute for the full ratemaking analysis needed to support a filed factor.
Worked Example: A Geographic Variable With a Defined Boundary
Hypothetical scenario. Assume a fictional jurisdiction’s applicable rule permits three filed territories based on the insured vehicle’s garaging address, but does not permit ZIP code to be used as a separate rating factor. A proposed filing uses ZIP-level relativities. For a preliminary review, the hypothetical experience is 6,000 exposures and 120 claims in Territory A, and 6,000 exposures and 180 claims in Territory B. Assume the same exposure definition and claim period in both territories. No interest, mortality, or probability-distribution assumption is used; the calculation is a direct summary of the stated counts.
Method. First compare observed claim frequency per exposure within the permitted territory definitions. Then compare the proposed ZIP-level use with the stated rule. The frequency calculation is supporting context, not a determination of permission.
For Territory A, frequency is \(120 \div 6{,}000 = 0.0200\) claims per exposure. For Territory B, it is \(180 \div 6{,}000 = 0.0300\) claims per exposure. The descriptive ratio is \(0.0300 \div 0.0200 = 1.500\).
Review and documentation. The stated rule permits territory classification based on garaging address but prohibits ZIP code as a separate factor. Therefore, the proposed ZIP-level rating use does not satisfy the assumed restriction, even though the two territory frequencies differ. The filing could document the permitted territorial definition, its address-based assignment method, and the supporting data at that level. It should not present the frequency ratio as authorization for ZIP-level factors.
Check. Both frequencies use the same exposure denominator, and \(0.0300\) is 1.500 times \(0.0200\). The common mistake is to treat a measurable difference as permission to use a more granular variable; the correction is to apply the rule’s stated boundary first.
Document the Variable as It Will Actually Be Used
A filing should make it possible to determine how a policy receives a value for the variable. State the data source and the relevant date or measurement period. Explain any conversion, score, band, or threshold. Define the treatment of missing, conflicting, or invalid information. Where the rule permits a variable only under specific conditions, show how the proposed process meets those conditions and keep the use within the stated scope.
Avoid descriptions such as “based on vehicle data” or “uses a customer score.” They do not identify the underlying measure or explain the rating operation. If a proposed composite contains multiple inputs, describe those inputs separately and check the proposed composite against the applicable requirements. A permitted input does not automatically make every combination or application of that input permissible.
Worked Example: A Permitted Mileage Measure and an Unpermitted Composite
Hypothetical scenario. Assume a fictional jurisdiction permits verified annual mileage as a liability rating variable but does not permit a telematics composite score. The hypothetical experience is 300 low-mileage exposures with $54,000 in incurred losses and 600 high-mileage exposures with $144,000 in incurred losses. Assume the bands, loss measure, and exposure definition are consistently applied. The proposed filing describes a “driving score” combining mileage, harsh braking, and night-driving frequency. No interest, mortality, or probability-distribution assumption is used.
Method. Calculate observed loss costs for the two mileage groups, then compare the proposed definition with the assumed rule. The calculation supports a mileage comparison only; it does not validate the other score inputs or override the restriction on a composite.
Low-mileage observed loss cost is \(\$54{,}000 \div 300 = \$180\) per exposure. High-mileage observed loss cost is \(\$144{,}000 \div 600 = \$240\) per exposure. The descriptive high-to-low ratio is \(\$240 \div \$180 = 1.333\), to three decimals.
Review and documentation. Under the assumed rule, the filing may propose the verified annual-mileage measure within its permitted liability use, subject to the rule’s conditions. It should define how annual mileage is measured and verified, how each band is assigned, and how missing or invalid readings are handled. The proposed composite should not be treated as permitted merely because mileage is one of its inputs. Its other components and combined form fall outside the use allowed by the assumed rule.
Check. The two loss costs are $180 and $240, and their ratio is 1.333. The common mistake is to document only the score’s name and predictive result; the correction is to specify each input and keep any proposed use within the rule’s defined scope.
Check Coverage and Other Use Boundaries
Restrictions may distinguish among coverages within the same line. A filing that uses a variable for every coverage should therefore be reviewed coverage by coverage, even if the variable is defined consistently. Supporting experience should also be shown on the basis relevant to the proposed use rather than pooled in a way that obscures a coverage-specific result.
Worked Example: A Vehicle-Age Factor Limited to Certain Coverages
Hypothetical scenario. Assume a fictional jurisdiction permits vehicle age as a rating variable for collision coverage but not for liability coverage. A proposed filing applies one vehicle-age factor to both. For newer vehicles, hypothetical experience is 800 exposures, $160,000 in collision losses, and $240,000 in liability losses. For older vehicles, it is 400 exposures, $120,000 in collision losses, and $108,000 in liability losses. Assume exposures and loss measures are consistently defined within each coverage. No interest, mortality, or probability-distribution assumption is used.
Method. Calculate the observed loss cost separately by coverage and age group. Then compare each proposed use with the assumed coverage restriction.
For newer vehicles, collision loss cost is \(\$160{,}000 \div 800 = \$200\) per exposure. For older vehicles, it is \(\$120{,}000 \div 400 = \$300\) per exposure. The older-to-newer collision ratio is \(\$300 \div \$200 = 1.500\).
For newer vehicles, liability loss cost is \(\$240{,}000 \div 800 = \$300\) per exposure. For older vehicles, it is \(\$108{,}000 \div 400 = \$270\) per exposure. The older-to-newer liability ratio is \(\$270 \div \$300 = 0.900\).
Review and documentation. The assumed rule permits the vehicle-age variable for collision, not liability. The filing should therefore limit the proposed use to collision coverage and describe the variable and supporting collision analysis there. It should not carry the factor over to liability, even though liability experience is available. The different observed ratios reinforce why the coverages should not be combined for this comparison; they do not change the jurisdictional restriction.
Check. Collision produces a 1.500 ratio and liability a 0.900 ratio. The common mistake is to apply one factor across all coverages because the variable itself is allowed somewhere; the correction is to document and test each coverage use against the rule.
Common Filing Traps
- Assuming a general permission applies everywhere. A rule may allow a variable only for specified coverages, measurement methods, or conditions. State the precise scope rather than summarizing it as “allowed.”
- Using a predictive result as the permission test. A favorable loss-cost comparison does not establish that the variable or its proposed use is permitted. Record the rule-based conclusion separately.
- Leaving the variable undefined. A category label or vendor score is not enough. Include the source, timing, transformations, thresholds, and treatment of unavailable data.
- Extending an allowed input to an unreviewed composite. Check the proposed variable as implemented, including its inputs and combined form, rather than assuming each combination inherits permission.
- Relying on a source that does not control the filing. Confirm the jurisdiction, line, coverage, and effective period. If sources appear inconsistent or unclear, document the issue for clarification instead of making an unsupported assumption.
Check Your Understanding
For each hypothetical problem, apply only the stated fictional jurisdictional rule. The calculations are descriptive support, not a conclusion about any actual jurisdiction.
- A fictional rule permits a renewal-duration variable only for comprehensive coverage. A proposed filing also applies it to collision. State the permitted scope and the additional filing description needed. (Answer: Comprehensive only; define how renewal duration is measured and document that the factor is not applied to collision.)
- A fictional rule permits a verified annual-distance band. A hypothetical group has 250 exposures and $50,000 in losses; a second has 500 exposures and $125,000 in losses. Calculate both loss costs and their second-to-first ratio. (Answer: $200 and $250 per exposure; ratio 1.250.)
- A fictional rule allows a specified three-territory plan based on the garaging address, but a proposed factor uses postal code within each territory. What issue must be resolved before using that factor? (Answer: Whether postal code is an authorized rating variable under the controlling rule; the three-territory permission alone does not establish permission for postal-code factors.)
- A fictional rule permits an odometer-based mileage variable only when the mileage is verified. The proposed filing defines mileage bands but says nothing about verification or missing readings. Identify two documentation gaps. (Answer: The method or source used to verify mileage and the treatment of missing or invalid readings.)
- A fictional rule allows a named vehicle characteristic for collision but not liability. A filing shows a collision loss-cost ratio of 1.20 and a liability ratio of 0.95, then proposes the 1.20 factor for both coverages. State the correction. (Answer: Limit the proposed use to collision under the stated rule; the liability experience does not permit extending that factor to liability.)