Why Underwriting Rules Must Match the Rating Plan
A rating plan assigns premiums to risks using specified variables, factors, and effective dates. Underwriting rules separately determine which risks may be quoted or bound automatically, which require review, and which are ineligible. These decisions interact: a risk can have a valid rate in the rating plan but never reach that rate if an eligibility rule excludes it. Conversely, a rule change can cause risks to enter the portfolio even when the rating plan was not designed or reviewed for that population.
The central deployment question is therefore not only whether a rating calculation is correct. It is also whether the rules governing access to that calculation are consistent with the filed and approved plan and with the intended appetite. A useful review compares the variables, thresholds, exceptions, and effective dates in the two specifications. This operational control is aligned with the CAS Exam 8 (Advanced Ratemaking) syllabus.
For example, a roof-age factor can assign different premiums to roofs of different ages. An underwriting rule may separately state that roofs above a specified age are ineligible, or that a range of ages requires inspection. If the factor table includes a range that an eligibility rule excludes, the plan contains a nominal rate for risks the product will not actually accept. That may be intentional, but it should be an explicit, reviewed choice rather than an unnoticed inconsistency.
A Practical Rule-to-Rate Review
A crosswalk makes the relationship visible. For each relevant variable, record how the rating plan uses it and how underwriting uses it. Check that definitions and units agree: a rating variable recorded in completed years is not automatically equivalent to an underwriting rule expressed in months or at a different observation date. Check boundaries as well. “More than 20 years” and “20 years or more” differ for a risk exactly 20 years old.
| Review item | Question to resolve |
|---|---|
| Variable definition | Do rating and underwriting use the same field, units, and time reference? |
| Threshold and boundary | Are the less-than, greater-than, and equality cases stated consistently? |
| Decision outcome | Does the case rate, refer, or become ineligible? Are those outcomes unambiguous? |
| Exceptions and overrides | Who can override the rule, under what authority, and how is the decision recorded? |
| Effective date and population | Which transactions and policy cohort use the new rule and the new rating plan? |
A complete test includes representative cases on both sides of every threshold, values exactly at the threshold, and relevant combinations of characteristics. Testing only typical cases can miss a boundary error or an interaction between two individually correct rules. The decision table should distinguish “ineligible” from “refer”: a referred risk has not necessarily been declined, and its eventual disposition should not be treated as an automatic acceptance.
How Eligibility Changes Portfolio Composition
A rule can change the average premium of accepted risks even when every individual rating factor is unchanged. Let \(Z\) denote a risk class, \(P(Z)\) its premium under the rating plan, and \(A\) the event that the risk is accepted. The average premium among accepted risks is the conditional expectation \(E[P(Z)\mid A]\). If \(A\) depends on \(Z\), changing the eligibility rule changes the distribution of \(Z\) among accepted risks.
This expression follows by partitioning the accepted portfolio into risk classes. The probability of both being in class \(z\) and being accepted is \(P(A\mid Z=z)P(Z=z)\). Multiplying that quantity by the class premium and summing gives the expected premium contribution per applicant. Dividing by the total probability of acceptance gives the mean premium conditional on acceptance. Thus, an eligibility change may affect both the number accepted and their average premium.
Worked Example: A Roof-Age Rule Excludes Rated Risks
A hypothetical property product has a base annual premium of $800 for roofs no more than 15 years old and a factor of 1.25 for roofs 16 through 25 years old. The intended eligibility rule accepts both age bands. A legacy rule still rejects roofs over 20 years old. Among 100 otherwise eligible applicants, 60 have roofs no more than 15 years old, 25 have roofs 16 through 20 years old, and 15 have roofs 21 through 25 years old. Quantify the portfolio effect, assuming every eligible applicant is written.
Method and rule comparison. The rating plan gives the 16–25 band an annual premium of \(800\times1.25=\$1{,}000\). The intended rule accepts all three groups. The legacy rule accepts the first two groups but excludes the 15 applicants in the 21–25 band, even though the rating plan assigns that band a premium.
Intended accepted portfolio. The total premium is \(60\times\$800=\$48{,}000\) for the first group and \((25+15)\times\$1{,}000=\$40{,}000\) for the older group. Total premium is \(\$48{,}000+\$40{,}000=\$88{,}000\) over 100 risks, so average premium is \(\$88{,}000/100=\$880\).
Portfolio under the legacy rule. The accepted risks generate \(\$48{,}000+25\times\$1{,}000=\$73{,}000\) over \(60+25=85\) risks. Their average premium is \(\$73{,}000/85=\$858.8235\ldots\), or $858.82 to the nearest cent. The rule excludes 15 risks and lowers the accepted portfolio’s average premium by \(\$880-\$858.8235\ldots=\$21.1764\ldots\), or $21.18 to the nearest cent.
Check. The legacy-rule average lies between $800 and $1,000, as it should. It is below the intended average because the legacy rule removes risks in the higher-premium band. The common mistake is to compare only the premiums of risks that remain and miss the 15 excluded risks; the correction is to compare both the accepted counts and the resulting conditional averages.
Referral Rules and Unreviewed Risk
Referral thresholds deserve the same precision as eligibility limits. A rule that routes a class for review should identify the exact condition, the reviewer or authority, and the permissible outcomes. If a deployed threshold is narrower than intended, cases may be issued without the planned review. If it is broader, more cases may be delayed or reviewed, potentially changing the practical access to the product. A referral is not itself a premium surcharge or a decline.
Worked Example: A Referral Threshold Is Too Narrow
A hypothetical specialty vehicle program intends to refer score levels A and B, while levels C and D may proceed automatically. A production rule refers only level A. The quote population is 10 level-A risks, 20 level-B risks, 50 level-C risks, and 20 level-D risks. For this illustration, each class has the following annual premium and expected claim cost per risk: A, $3,000 and $2,400; B, $2,200 and $1,800; C, $1,400 and $1,000; D, $800 and $500. Assume that every case not referred is issued and that referred cases are held out pending review.
Method and intended routing. Under the intended rule, A and B are referred: \(10+20=30\) risks. C and D proceed automatically: \(50+20=70\) risks. Their premium is \(50\times\$1{,}400+20\times\$800=\$70{,}000+\$16{,}000=\$86{,}000\). Their expected claim cost is \(50\times\$1{,}000+20\times\$500=\$50{,}000+\$10{,}000=\$60{,}000\).
Production routing. Referring only A leaves B, C, and D to proceed automatically: \(20+50+20=90\) risks. Their premium is \(20\times\$2{,}200+\$70{,}000+\$16{,}000=\$130{,}000\). Their expected claim cost is \(20\times\$1{,}800+\$50{,}000+\$10{,}000=\$96{,}000\). Thus, 20 additional risks, all level B, bypass the intended review; the automatically issued segment has $44,000 more premium and $36,000 more expected claim cost than under intended routing.
Check. The expected claim-cost-to-premium ratios are \(\$60{,}000/\$86{,}000=0.697674\ldots\), or 0.6977, under intended routing and \(\$96{,}000/\$130{,}000=0.738461\ldots\), or 0.7385, under the production rule. These are illustrative ratios before expenses and other adjustments, not profitability conclusions. The common mistake is to count the extra $44,000 of premium as the whole impact. The correction is to assess the associated risk and review workload as well as the premium.
Test Rule Interactions, Not Just Individual Rules
A rule set can produce unintended results even when each rule appears reasonable in isolation. Suppose one rule allows a property with a monitored alarm and another excludes properties with a particular occupancy. The combined decision must specify whether the alarm exception overrides the occupancy rule. Similar issues arise when rating-plan tiers use combinations of variables but eligibility rules test only one variable. Testing combinations helps reveal whether the deployed portfolio matches the reviewed appetite.
Worked Example: A Blanket Exclusion Overrides a Rated Class
A hypothetical rental-property product has a $1,000 base premium. It applies a factor of 1.30 to seasonal properties with verified central monitoring. Seasonal properties without monitoring are ineligible. Primary residences have a factor of 1.00. In a proposed cohort there are 30 primary residences, 20 monitored seasonal properties, and 10 unmonitored seasonal properties. Expected annual claim costs per risk are, respectively, $600, $900, and $1,500. A legacy eligibility rule excludes every seasonal property. Compare the intended and legacy outcomes, assuming accepted risks are written.
Method and intended outcome. The rating plan assigns primary residences a premium of \(1{,}000\times1.00=\$1{,}000\), monitored seasonal properties \(\$1{,}000\times1.30=\$1{,}300\), and no premium to the ineligible unmonitored group. The intended rule accepts the first two groups: \(30+20=50\) risks. Their total premium is \(30\times\$1{,}000+20\times\$1{,}300=\$30{,}000+\$26{,}000=\$56{,}000\), for an average of \(\$56{,}000/50=\$1{,}120\). Expected claim cost is \(30\times\$600+20\times\$900=\$18{,}000+\$18{,}000=\$36{,}000\).
Legacy outcome. The blanket exclusion accepts only the 30 primary residences. Total premium is \(30\times\$1{,}000=\$30{,}000\), average premium is \(\$30{,}000/30=\$1{,}000\), and expected claim cost is \(30\times\$600=\$18{,}000\). The legacy rule therefore excludes 20 risks that the intended plan rates and accepts, reducing this cohort’s premium by $26,000 and expected claim cost by $18,000.
Check. The intended average is $1,120, between the $1,000 primary-residence premium and the $1,300 monitored seasonal premium. The legacy rule lowers the average by excluding the higher-premium group, not by changing any rating factor. The common mistake is to treat the blanket seasonal exclusion as harmless because the monitoring factor is correctly configured; the correction is to test the interaction between the exception and the eligibility decision.
Operational Controls and Common Traps
Before deployment, translate the approved intent into explicit decision cases and expected outcomes. The cases should cover eligibility, referral, and rating together, including boundaries and exceptions. For each case, record the input values, expected decision, expected rate class, and whether a human decision is required. Compare that result with the deployed behavior. This makes a mismatch diagnosable: a wrong data mapping, rule threshold, exception order, or date condition leads to a different pattern of failures.
- Do not assume a rated class is automatically eligible. Confirm that eligibility rules allow the intended population to reach the rating plan.
- Do not equate referral with rejection. Track the referral outcome separately from automatic eligibility and final acceptance.
- Do not overlook equality cases. Test exactly-at-threshold values, not only values clearly above or below them.
- Do not evaluate a rule change only by total premium. Compare counts and risk mix, and use expected cost or other appropriate measures when available.
- Do not ignore date alignment. Confirm that the rule version and rating version apply to the intended transactions and policy population, consistent with the applicable effective-date specification.
Check Your Understanding
For each hypothetical situation, compare the intended rule with the deployed outcome. Each problem is self-contained; report counts, premiums, or rates as requested.
- A product rates equipment with a base premium of $500 and a factor of 1.20 for ages 6–10. Intended eligibility includes ages 1–10, but a legacy rule accepts only ages 1–8. Of 40 applicants, 20 are age 1–5, 12 are age 6–8, and 8 are age 9–10. Assuming all eligible applicants are written, find the intended and legacy accepted counts and average premiums. (Answer: Intended: 40 risks and $550.00 average premium; legacy: 32 risks and $537.50 average premium.)
- A rule should refer scores 3 or lower. The deployed rule refers scores 2 or lower. There are 7 score-1 risks, 9 score-2 risks, 11 score-3 risks, and 23 score-4 risks. How many risks that should be referred proceed without referral under the deployed rule? (Answer: 11 score-3 risks.)
- A proposed rule accepts 24 low-hazard risks at $900 premium each and 16 moderate-hazard risks at $1,200 each. A stale rule accepts only the low-hazard group. Find each rule’s accepted count and average premium, and the premium omitted from the stale-rule cohort. (Answer: Proposed: 40 risks and $1,020.00 average; stale: 24 risks and $900.00 average; $19,200 omitted premium.)
- A program intends to refer all risks with a protection score of 1 or 2. Scores have counts 5, 15, 30, and 50 for levels 1, 2, 3, and 4. The deployed rule refers only level 1. How many additional cases are automatically routed rather than referred, and what proportion of the intended referral set is that? (Answer: 15 additional cases; \(15/20=0.7500\), or 75.00%.)
- A rating plan accepts monitored seasonal properties at $1,400 and primary properties at $1,000. Unmonitored seasonal properties are ineligible. A cohort contains 18 primary, 12 monitored seasonal, and 5 unmonitored seasonal properties. A deployed blanket rule excludes all seasonal properties. Assuming all accepted risks are written, find the intended and deployed counts and total premiums. (Answer: Intended: 30 risks and $34,800; deployed: 18 risks and $18,000.)