Protected Traits and Ratemaking Restrictions

Why a Predictive Factor May Still Be Restricted

A rating characteristic can be associated with expected losses and still be unavailable for pricing. Laws and regulations may prohibit or limit the use of particular protected traits, and the restrictions can differ by jurisdiction, insurance line, product, and rating purpose. A predictive model is not permission to use every variable it can measure.

The practical task is to identify whether the trait itself enters the rating decision, determine which specific rules apply, and confirm that both the proposed method and its implementation satisfy those rules. This is distinct from the question of whether risk differences are actuarially supportable, addressed in the earlier material on unfair discrimination versus legitimate risk differentiation.

Definition: Direct use of a protected trait occurs when the trait itself is an input to a rating calculation or rating rule. Examples include assigning different relativities by trait value, using the trait to select a rating tier, or including it in an interaction that changes the premium. Whether a trait is legally protected, and whether its direct use is restricted, must be determined under the applicable law and regulation.

A direct-use review asks about more than the name of a field in a model. A trait can be embedded in a lookup table, a formula, a manual underwriting rule, or an eligibility decision that affects the price offered. Review the whole decision path, including the data fields used to assign a rating category. Conversely, collection of a characteristic for a legally required purpose, such as reporting or monitoring, does not by itself establish that it is being used to set a rate. The purpose, access controls, and flow of the data into pricing must be understood.

Permission and Predictive Support Are Separate Questions

Actuarial evidence addresses whether a characteristic helps estimate expected cost under the model and data assumptions. Legal review addresses whether the characteristic may be used for this product and purpose, and under what conditions. Passing one test does not establish that the other is passed.

The distinction can be expressed as a decision rule. Let $P$ mean that a proposed use is legally permitted and $A$ mean that it is actuarially supported and properly implemented. Proceeding with the use requires both conditions, $P \land A$. If the use is prohibited, strong predictive evidence does not make it permissible. If the use is permitted, that permission alone does not demonstrate that the factor is sound, adequately supported, or correctly applied.

\text{Proceed with a proposed direct use only if } \text{legal permission} \land \text{actuarial support} \land \text{compliant implementation}.

This is a screening rule, not a substitute for interpreting the applicable law. A restriction may be categorical, may allow use only under stated conditions, or may leave an issue requiring regulatory or legal clarification. The relevant answer can change across states or provinces, lines of insurance, and coverage forms. A rule for one product should not be generalized to another.

Key distinction: A finding that a trait predicts losses answers a statistical question. It does not answer whether the insurer may use that trait in rating. Check both questions independently, and record the scope of the legal conclusion.

A Practical Review Sequence

Before a protected characteristic is used in a proposed rate, apply a review that traces the characteristic from the governing rule to the actual premium calculation. The steps below are useful both for a new rating factor and for checking an existing rating plan.

Define the proposed use.
Identify the exact characteristic, rating decision, product, coverage, and jurisdiction. State whether it changes a premium, tier, eligibility outcome, or another rating-related decision.

Identify the applicable restriction.
Review current statutes, regulations, orders, approved or filed rules, and relevant regulatory guidance. Determine whether the rule prohibits the use, permits it conditionally, or requires further interpretation.

Check the conditions and boundaries.
If use is conditional, identify the allowed product, permitted categories, required evidence, filing or approval requirements, and any limits on how the factor may operate.

Trace the factor through the rating system.
Check model inputs, transformations, interactions, rating tables, manual rules, and production code. Confirm that the system does not use the trait outside the permitted scope.

Document the decision and controls.
Record the authority reviewed, the conclusion and its scope, supporting actuarial evidence, implementation checks, and any follow-up needed if the rule or product changes.

The sequence matters. Starting with a model result and then searching for a rationale can encourage confirmation bias. Defining the use and the applicable rule first makes it clearer what evidence is relevant and what conditions must be met. If the rule is ambiguous, the appropriate response is to seek qualified legal or regulatory guidance rather than infer permission from silence.

Worked Examples

Worked Example: A Direct Factor Is Explicitly Prohibited

A hypothetical personal auto program has a base annual premium of $800 before a proposed sex-based factor. The proposed relativities are 0.90 for group A and 1.10 for group B. For this example, assume the applicable jurisdiction expressly prohibits direct use of this trait in personal auto rating. Determine the indicated premiums under the proposal and the compliant treatment of this direct factor.

Step 1: Calculate the proposed premiums. The proposed rating rule is base premium multiplied by the group relativity.

\begin{aligned} \text{Group A premium} &= \$800 \times 0.90 = \$720,\\ \text{Group B premium} &= \$800 \times 1.10 = \$880. \end{aligned}

Step 2: Measure the direct difference. The proposed premium difference is $\$880-\$720=\$160$. Relative to the $800 base, the proposed factors produce a decrease of 10% for group A and an increase of 10% for group B.

Step 3: Apply the stated restriction. Because the hypothetical rule prohibits direct use in this product, the proposed factor cannot be used, even if separate analysis showed that the groups had different observed losses. For applicants identical on the other rating inputs, the direct trait multiplier must not create different premiums. If the plan uses only the stated $800 base and this proposed factor, removing the prohibited factor leaves a $800 premium for each group.

Check. The resulting direct-factor difference is $\$800-\$800=\$0$. This does not mean every policyholder must have the same premium: other permitted rating characteristics may still produce different rates.

Method and common mistake avoided. Apply the explicit restriction to the rating rule, then recompute without the prohibited direct factor. The mistake is to retain the factor because its observed loss differential appears credible; actuarial support does not override the assumed prohibition.

Worked Example: Conditional Use With Prescribed Categories

A hypothetical jurisdiction permits a protected characteristic called category Q for a specified personal-lines product, but only through three filed age bands: 18–24, 25–64, and 65 or older. The approved relativities are 1.20, 1.00, and 1.10, respectively. The base premium is $1,000. A proposed system assigns ages 20, 42, and 70 to the filed bands, but a developer has also created a separate 18–20 sub-band with relativity 1.30. Assess the proposed system and calculate the premiums for the three stated ages using the approved categories.

Step 1: Check the rule before calculating. The hypothetical authorization is limited to the three filed bands and their approved relativities. The extra 18–20 category is outside those stated conditions; it should not be used as an additional direct distinction unless the required authority and filing are obtained.

Step 2: Map each age to an approved band. Age 20 falls in 18–24, age 42 in 25–64, and age 70 in 65 or older.

Step 3: Calculate using the approved factors.

\begin{aligned} \text{Age 20} &: \$1{,}000 \times 1.20 = \$1{,}200,\\ \text{Age 42} &: \$1{,}000 \times 1.00 = \$1{,}000,\\ \text{Age 70} &: \$1{,}000 \times 1.10 = \$1{,}100. \end{aligned}

Step 4: Check the implementation boundary. The age-20 premium is $1,200 under the authorized 18–24 factor, not $1,300 under the unapproved sub-band. The approved table and the system's actual mapping must agree.

Check. The calculated premiums use only the three stated filed relativities, and none uses the extra 1.30 factor. The premium calculations are internally consistent with the hypothetical authorization.

Method and common mistake avoided. Treat the permission as conditional and enforce its category boundaries. A common error is to assume that authorization to use a characteristic permits arbitrary refinements or new relativities.

Worked Example: A Narrow Exception Does Not Transfer Across Products

Suppose a hypothetical jurisdiction generally restricts direct use of protected characteristic R. It expressly allows the use for Product M only when the factor is supported, included in an approved filing, and applied as specified in that filing. Product N has no stated exception. A proposed factor has a relativity of 1.15 for one value of R and 1.00 for another. The filing for Product M includes those factors, and the implementation matches it. A product team proposes copying the same factor into Product N. Decide what can be concluded.

Step 1: Evaluate Product M within the exception. The described use is for Product M, the factor has the stipulated support, it appears in the approved filing, and the implementation matches the filed values. On the stated facts, the conditions in the hypothetical exception are satisfied for Product M. The factor changes the premium by 15% relative to the 1.00 category, since $1.15-1.00=0.15$.

Step 2: Test the proposed transfer separately. Product N is a different product, and the facts provide no exception for it. The Product M approval is not evidence of permission for Product N. The proposed team should not copy the factor into Product N on the basis of the Product M filing.

Step 3: State the appropriate next action. Keep the use within Product M's specified scope. For Product N, obtain a product-specific legal and regulatory determination before proposing direct use. If the applicable rule prohibits it, omit the factor; if an exception exists, satisfy its conditions and any required filing or approval before implementation.

Check. The conclusion is limited to the facts given: Product M meets the hypothetical conditions, while permission for Product N has not been established. This avoids making a broader claim about the treatment of characteristic R in real jurisdictions.

Method and common mistake avoided. Match each use to its product-specific authority and conditions. The mistake is treating a narrow exception or approval as a general authorization across a company, line, or jurisdiction.

Implementation and Documentation

A conclusion about the filed rating plan is not enough if production systems behave differently. Testing should compare the governing rating rule with the data fields, transformations, lookup tables, and code that actually generate premiums. For a conditionally permitted factor, the test should verify both the allowed scope and the stated values. A change to a product, territory, rating algorithm, or filing may require the review to be revisited.

A review record should be precise rather than simply stating that a factor is “legal” or “approved.” It should identify the applicable jurisdiction and product, the authority and version reviewed, the exact proposed use, any conditions, and the controls used to confirm compliance. It should also distinguish a regulatory determination from an actuarial conclusion about predictive support. If the relevant rule cannot be interpreted confidently, escalate it to qualified counsel or the appropriate regulatory contact.

Keep data collection and rating use conceptually separate. A characteristic may be collected for a legally required report or another specified purpose without being available to the pricing algorithm. Access restrictions, data dictionaries, and system testing can help confirm that information collected for one purpose does not unintentionally flow into a rating decision. This review is about direct use; a variable that may stand in for a protected trait requires a separate analysis.

Common Exam Traps

Equating prediction with permission. A statistically supported difference does not establish legal authorization to use the trait.
Assuming one rule applies everywhere. Confirm jurisdiction, line, product, and purpose rather than generalizing from another filing or market.
Ignoring conditions on an exception. A permitted use may be limited to specified categories, evidence, filings, or implementation rules.
Reviewing only model variables. Direct use can also occur in tables, tiers, underwriting or eligibility rules, and manual adjustments that affect the offered rate.
Overstating the conclusion. A decision for one product or a particular version of a rule should not be described as a universal legal rule.

Key takeaway: For direct use of a protected trait, first identify the exact rating decision and its jurisdiction-specific restrictions. Then verify any conditions, actuarial support, filed treatment, and production implementation. Predictive evidence cannot substitute for legal permission, and permission in one setting does not automatically transfer to another.

Check Your Understanding

For each hypothetical, distinguish the legal question from the actuarial question and apply only the facts stated.

A hypothetical state prohibits a protected trait from changing homeowners premiums. A proposed model estimates a loss ratio of 1.18 for one trait category and 0.92 for another. What is the direct-rating conclusion under the stated rule? (Answer: Do not use the trait directly to change the homeowners premium; the modeled difference does not override the stated prohibition.)
A hypothetical regulation permits a protected characteristic for Product A only using two filed categories, with relativities 0.95 and 1.05. A policy in the second category has a $600 base premium. Calculate its premium and state one implementation check. (Answer: $630; verify the system uses the filed second category and 1.05 relativity, not an unfiled subdivision.)
A factor is allowed in a hypothetical jurisdiction for one coverage only if it is included in an approved filing. The factor is predictive, but the filing does not include it. May the insurer use it immediately? (Answer: No; predictive support does not satisfy the stated filing condition.)
A company has an approved direct-use factor for Product P in one jurisdiction and wants to use it for the same product in another jurisdiction. What must it establish first? (Answer: Whether the second jurisdiction permits that specific use and what conditions apply there; the first jurisdiction's approval does not establish permission in the second.)
A rating system collects a protected characteristic for a required report. The pricing code is said not to use it. Name two checks that help establish the distinction. (Answer: Review data flows and system inputs, and test rating outputs or code to confirm the field does not affect the premium or rating tier.)

What You'll Learn