RESEARCH DOSSIER
On-Level Premium Adjustments and Rate History Reconstruction
Abstract. On-level premium adjustment reconstructs historical premium at a specified common rate level so that the denominator of a loss ratio, the premium component of a ratemaking indication, or an experience comparison is interpretable across tariff changes. The underlying analytical problem is temporal: a change in filed or implemented rates affects new policies, renewals, endorsements, and earned premium on different dates. It is also compositional: the premium ultimately observed reflects exposures, classification relativities, limits, deductibles, minimum premiums, audit adjustments, individual-risk plans, and the evolution of a portfolio. This monograph develops the accounting and actuarial basis of historical premium reconstruction, distinguishes exact policy-level re-rating from aggregate on-level approximations, derives earned-premium weights under explicit earning and writing assumptions, and demonstrates single and multiple rate changes in reproducible numerical examples. It investigates the information requirements for reconstructing an auditable rate history, the mathematical limits of cumulative average rate changes, the treatment of retroactive and midterm modifications, and reconciliation across written, earned, and current-rate premium. Its original contribution is an integrated analytical comparison of estimands, information loss, reconstruction error, and decision consequences within established actuarial methods. The illustrative calculations use synthetic portfolios and confer no empirical estimate for any insurer or jurisdiction. U.S. professional standards are examined within their stated scope, while the analytical mathematics is discussed for international property/casualty insurance.
Keywords: On-level premium; rate history; earned premium; written premium; exposure extension; extension of exposures; parallelogram method; insurance ratemaking; rate-level index; policy earning; rate filing; portfolio mix; auditability; actuarial data quality; premium trend; rate adequacy.
Introduction
A premium figure belongs simultaneously to a contract, a price schedule, an exposure, and a period of risk. Without reconstructing those four attributes, an analyst can mistake a change in insurance price for a change in claim experience or can carry a historical premium denominator into a current prospective rate indication. The essential question is what the same underlying insured exposures would have generated under an explicitly selected rate schedule.
Werner and Modlin's Basic Ratemaking, chapter 5 presents the extension-of-exposures and parallelogram approaches and explains why historical premium requires adjustment before comparing it with an indicated current price. The CAS ratemaking principles connect the general task to estimating the future cost of risk transfer. More specifically, the Actuarial Standards Board's ASOP 53 historical-data guidance calls for considering consistent measurement of historical exposure and rate levels when those data support property/casualty future cost estimates.
This study uses a counterfactual definition: on-level premium for historical exposures is the premium that those exposures would have produced under a chosen reference rate structure, after preserving or explicitly restating the other pricing attributes. The reference is often the presently implemented rate level, but may instead be a precisely documented pricing date or scenario. The definition requires a declared treatment of changes in rating algorithms, classification, coverage, risk mix, and exposure bases. Consequently the calculation is a reconstruction rather than a simple multiplication performed without regard to underlying contracts.
The documentary method combines close examination of CAS educational literature, applicable U.S. actuarial standards, and selected regulatory and industry records with mathematical derivation and independently calculated synthetic examples. The analysis covers annual and other policy terms, calendar and policy-year views, homogeneous and heterogeneous books, and different implementation mechanisms. The calculations are original illustrations of established mathematics. No claim of a new actuarial theory, a new predictive model, or statistical novelty is made.
Rate Change History and Exposure
The estimand and its time coordinates
Let an experience window be denoted by H and a reference rate schedule by R*. Historical earned premium is P_E(H). An on-level earned premium P_E*(H;R*) is the amount that the same relevant earned risk and exposure would have generated if the selected rating components had been applied according to the stated counterfactual rules. The associated on-level factor is F(H;R*) = P_E*(H;R*) / P_E(H), provided P_E(H) is nonzero and the comparison is defined on compatible gross, net, and coverage bases.
A premium record contains at least three times. The transaction date is the time a system posts the policy or endorsement; the contract-effective date determines coverage and often tariff eligibility; the earning date allocates written premium to risk periods. Rate approval, implementation, and coverage-effective dates introduce additional clocks. Their separation matters because a rate approved on one day may apply only to contracts beginning or renewing later, while policies issued before implementation can continue earning their earlier premium.
The quantity reconstructed must be stated before a factor is calculated. An analyst may want the aggregate premium at a common manual tariff, the amount at current charged rates including schedule credits, or premium under a complete current rating algorithm. These are distinct estimands. A historical book whose individual-risk credits change with underwriting conditions cannot be put on a complete current-price basis merely by replacing manual base rates.
The New York filing instructions expressly distinguish the effective date of the last rate revision, its overall percentage impact, and written premium effects in their filing fields. Such regulatory descriptors help recover an implementation chronology, but their aggregate scope must be interpreted against the actual policy population and rules.
A reconstructible rate event
A rate event is represented here as a documented observation, not a newly named actuarial model. Its necessary attributes are jurisdiction and product, coverage and rating-plan scope, approval or authority record where applicable, publication or filing version, implementation date, eligible inception or renewal dates, treatment of in-force contracts, rate-table or rule version, percentage effect and its measured exposure mix, and supersession or withdrawal history.
An overall indicated change, an approved change, and a realized rate change describe different stages. An actuarial indication may be modified by a regulator or management before implementation. An approved schedule may be introduced in phases. A reported average change may be calculated on an in-force snapshot that differs from the historical experience mix. Hence a row stating "+8% effective October 1" remains insufficient unless its denominator, application mechanism, and affected policies are known.
Temporal provenance matters as much as magnitude. Evidence should preserve the original approval document, applied version, transaction log, and later corrections. When only public filings are available, the extent to which approved rates were actually applied is an open data question. It must be handled as a documented reconstruction assumption rather than silently promoted to observed fact.
Exposure, classification, and effective-date logic
Exposure is the quantity to which rates are applied. A vehicle-year, insured property-year, payroll amount, revenue amount, or other exposure base has its own measurement conventions. A policy may produce multiple coverages, layers, locations, or transaction records. A current rate schedule can map the same nominal exposure to a different premium if classifications, territories, discounts, deductibles, limits, or fees have changed.
A rate schedule can be represented generically as a function r_v(x), where v is its version and x contains the rating attributes. Written base premium for a transaction is proportional to exposure times the applicable version's rate, subject to minimums, fees, taxes, and other nonproportional terms. A coherent reconstruction stores x as of the relevant historical risk and records precisely which attributes are held fixed in the reference calculation.
The ASOP 53 exposure guidance addresses selecting workable exposure bases and considering adjustments for historical changes in the way premium charges are calculated. This is especially important if a carrier switches from a flat annual policy fee to a per-unit exposure fee or redesigns its class relativities. The resulting aggregate rate change is portfolio-dependent.
Source records and the history ledger
The minimum defensible historical ledger joins four evidence families: approved or otherwise authoritative rate documentation; production rating manuals and algorithms; policy and endorsement transactions with effective dates; and earned-premium or earning-pattern records. A fifth family consists of reconciliation totals from the accounting or actuarial data warehouse. Each family answers a different question and none is a substitute for the others.
A useful reconstruction table includes policy key, coverage key, rate-version key, inception, expiry, endorsement-effective and cancellation dates, exposure unit, historical rating attributes, originally charged manual premium, subsequent audit adjustment, earned amount by accounting period, and the reference premium under R*. Policies with multiple coverages or midterm amendments require transaction-level detail or a versioned exposure chronology. The purpose is to make the counterfactual traceable, not to collect personally identifiable information beyond legitimate analytical need.
The Actuarial Standards Board's ASOP 23 Data Quality asks actuaries to consider appropriate data, review data, and disclose significant known limitations within its professional scope. For rate reconstruction, practical review includes missing versions, unmatched transactions, negative or zero premiums, overlapping effective dates, cancellations posted late, rate indexes inconsistent with manuals, and totals that cannot be reconciled to the historical ledger.
On-Level Factors
Cumulative rate indexes and multiplicative changes
An overall rate-level index is a relative scale, conventionally normalized to 1.000 at an initial rate level. If rates change by g_1, g_2, ..., g_n in sequence, and the figures genuinely measure comparable multiplicative changes for the same target mix, then the terminal index is
I_n = ∏_(j=1)^n (1 + g_j).
A 10% increase followed by a 5% increase gives 1.10 × 1.05 = 1.155, an aggregate 15.5% increase. An additive 15% approximation loses the cross-product. A 10% increase followed by a 10% decrease gives 0.99, not a return to the original rate. Negative changes must remain greater than -100% for a positive price index.
The qualification "comparable" carries analytical weight. An overall rate impact calculated on one class mix is not generally transferable to a later portfolio with different class proportions. A company-wide change can be concentrated in a territory or coverage. Even if the original carrier filing correctly calculates an overall average, the average is a property of a rate change and its weighting population jointly, rather than a universal multiplier on all contracts.
Exact extension of exposures
Extension of exposures, also described as re-rating, recomputes premium using the chosen rate schedule for each relevant historical exposure or rating transaction. For policy i, calculate W_i*=r_*(x_i) with applicable rules and compare it with historical W_i=r_{v(i)}(x_i), using the same coverage and contract features unless the purpose requires a documented adjustment. Earn both amounts over the same exposure window using the verified earning convention.
Werner and Modlin (2016) explain the extension of exposures technique as a preferred calculation when sufficiently detailed policy records and rating capability are available. Its strength is the preservation of heterogeneous class and fee effects. Its vulnerabilities are missing historical inputs, changed rating algorithms, incompatible reference coverage, transaction duplication, and failure to reconcile the re-rated population with the historical premium totals.
A practical illustration is a portfolio of two annual policies, each having historical earned premium 1,000. Policy A's current re-rate is 1,100; policy B's is 1,300. Exact on-level earned premium is 2,400 and the factor is 1.20. If a published aggregate 10% rate increase was based on a different reference mix, applying 1.10 would produce 2,200, understating this portfolio's relevant current-price premium by 200. This illustration is synthetic; its purpose is to display the consequences of nonuniform changes.
Aggregate factors and the information they require
When re-rating is impossible, an aggregate factor estimates the common-rate premium from documented rate indexes and estimated proportions of earned premium associated with each old rate level. Let a_hj be the proportion of the historical base-rate-equivalent earned exposure in experience period h associated with rate level j, with a_hj≥0 and Σ_j a_hj=1. Then the implied average rate index is
I_bar,h = Σ_j a_hj I_j,
and the approximate on-level factor to current index I* is
F_h ≈ I* / I_bar,h.
This expression is exact under a homogeneous multiplicative price structure and correct weighting; otherwise it is an approximation. The weights must refer to the baseline rating exposure underlying premium, not simply raw policy counts unless policies are economically equivalent. For a book with rapidly changing insured values or class mix, a naïve time-weighted average can be seriously misleading.
The parallelogram method in Basic Ratemaking operationalizes such weights through policy-writing dates and earning assumptions. Its geometry is a way of estimating time weights, not a verification that the portfolio's actual new business, cancellations, seasonality, and class mix conformed to those assumptions. Monthly transaction cohorts can provide improved weights when detailed policy-level re-rating remains unavailable.
Audit controls and sensitivity
A factor is useful only when its denominator is reproducible. The audit trail should identify the initial base index, every change and implementation date, the time-zone or day convention for time boundaries, eligible policies, weights, earning curve, terminal index, and the exact applied premium basis. Data fields should store original values and reconciled revisions, preserving the reason for each change.
The ASOP 41 communication standard sets expectations for clear material assumptions, methods, limitations, and disclosures in U.S. actuarial communications within scope. The analyst should consequently disclose whether a percentage change was taken from a filing, recomputed from a manual, or estimated from realized premium. These sources have different epistemic status.
Sensitivity testing can compare the aggregate factor under alternative permissible effective dates, monthly rather than annual approximations, alternative new-business seasonality, and policy-duration distributions. A stated point factor with no tested exposure to these choices can convey precision unsupported by the source history. In the absence of a credible numeric bound, the limitation should be expressed as unresolved uncertainty instead of a fabricated confidence interval.
Parallelogram and Earning Approaches
Geometric derivation for annual policies
Assume policies are written uniformly through time, all policies have one-year terms, written exposure and base rates per policy are stationary, premium earns uniformly, and a rate change applies only to policies first effective on or after its implementation time. Express time in years. For an implemented change at τ, the fraction of earned premium in year [a,b] attributable to policies written under the new rate is the time average of the in-force new-rate proportion:
q(a,b;τ) = (1/(b-a)) ∫_a^b min(1,max(0,t-τ))dt.
For annual policies the expression min(1,max(0,t-τ)) describes how a new-business rate replaces old-rate contracts over the year after implementation. In the first calendar year, a change taking effect a fraction u of the way through the year produces q=(1-u)^2/2. In the next calendar year, while the last old contracts run off, q=1-u^2/2 when the same calendar-year origin and annual policy terms apply. These formulas follow by integrating the linear transition; they are conditional, not empirical estimates.
The geometry commonly appears as parallelograms or triangles in the earned-premium time diagram. The CAS exposition of parallelogram weights also explains how to adapt the calculation to policy-year and shorter-term records, and why the assumed uniform distribution of writings may fail for seasonal insurance.
A single July rate change
Suppose annual policies are written uniformly, premiums earn uniformly, and a 20% increase applies to policies beginning July 1, 2024, with no midterm changes. Normalize the old rate index to 1.000 and the implemented current index to 1.200. The 2024 calendar-year earned share under the new rate is (1-0.5)^2/2=0.125. The old-rate share is 0.875. Therefore I_bar,2024=0.875×1.000+0.125×1.200=1.025 and F_2024=1.200/1.025=1.1707317073.
For calendar 2025, the new-rate earned share is 1-(0.5)^2/2=0.875, so I_bar,2025=0.125×1.000+0.875×1.200=1.175. The on-level factor becomes F_2025=1.200/1.175=1.0212765957. A factor near unity in 2025 does not imply that the rate increase was small; it reflects the near-completion of its earned-premium transition.
If the hypothetical 2024 earned premium was $2,050,000, its on-level amount would be $2,050,000×1.1707317073=$2,400,000, subject to rounding. This calculation is exact within the stated continuous-time assumptions. It is not valid for a real portfolio without confirming implementation, annual-term distribution, writing pattern, earning method, and rate equivalence.
Two changes in one year
Now assume the same annual-policy mechanism with a 10% increase effective April 1, 2024, followed by a 5% increase effective October 1, 2024. All policies written or renewed thereafter use the applicable latest rate, and changes are multiplicative. The rate indexes are 1.000 before April, 1.100 from April through September, and 1.155 from October onward.
For 2024 calendar-year earned premium, the share subject to at least the April change is (0.75)^2/2=0.28125. The share subject to both changes is (0.25)^2/2=0.03125. Thus the mutually exclusive earned exposure weights are 0.71875 at index 1.000, 0.25000 at index 1.100, and 0.03125 at index 1.155. Their weighted rate index is 1.02984375.
For 2025, after allowing the original policies to expire, the share affected by at least the April change is 0.96875 and the share affected by both changes is 0.71875. The corresponding mutually exclusive weights are 0.03125, 0.25000, and 0.71875. The weighted index is 1.13640625. These two years jointly illustrate the transition in which a change announced long before the end of the experience year still enters the earned-premium denominator gradually.
| Calendar experience year | Old-index weight | Intermediate-index weight | Current-index weight | Weighted rate index | On-level factor | Historical earned premium | On-level earned premium |
|---|---|---|---|---|---|---|---|
| 2024 | 0.71875 | 0.25000 | 0.03125 | 1.02984375 | 1.12152936 | $2,000,000 | $2,243,058.72 |
| 2025 | 0.03125 | 0.25000 | 0.71875 | 1.13640625 | 1.01636189 | $2,500,000 | $2,540,904.72 |
The arithmetic is 1.155/1.02984375=1.1215293582 for 2024 and 1.155/1.13640625=1.0163618864 for 2025. The figures permit independent recalculation from the displayed assumptions and weights. They do not incorporate risk growth, class mix, cancellations, earned exposure nonuniformity, or rounding practices of any insurer.
Alternative earnings, terms, and midterm effects
If the writing density is w(s) instead of a constant, the proportion of earned premium under a rate change at τ requires weighting each inception s by both its exposure value and the fraction of its coverage earned within H. For a one-year policy with earnings kernel k(t|s), a general share may be written as the ratio of the double integral over inception and earnings time for s≥τ to the corresponding integral over all eligible inception dates. This formulation accommodates seasonality when writing-cohort data are available.
Six-month policies produce a faster transition than twelve-month policies. Multi-year policies can retain old-rate premium beyond the next year. A rate change that immediately applies to the unexpired portion of existing contracts uses a different transition: the new rate may influence earned premium without waiting for renewal. Such a change cannot be reconstructed by merely substituting another percentage into the annual new-business parallelogram.
Industry history also shows why implementation details matter. The NCCI premium-level-change exhibit distinguishes revised-rate effective dates from legislative benefit-change dates and discusses changes applicable to new/renewal and unexpired policies. That distinction is illustrative for workers compensation systems within the exhibit's scope; it should not be generalized as a rule for all insurance products.
Multi-Change Rate Histories
Reconstruction from heterogeneous implementation records
An implementable rate history consists of ordered transitions with effective-date eligibility, affected rating components, and recoverable initial conditions. For two rate changes affecting different coverages, a single product-wide chain I_n can be misleading. Instead, compute a distinct rate history for each stable segment or perform policy-level re-rating from the manual version applying to the contract.
When rate filings are incomplete, the historian should separate three layers: documentary observations, inferred transition logic, and unobserved implementation behavior. A published tariff with an approval letter documents a rate plan; a production implementation log documents actual deployment; the measured premium realized on policies also includes underwriting and market actions. These evidence types can be mutually informative while remaining conceptually different.
The CAS 2021 reinstatement notice explains that the ratemaking Statement of Principles was reinstated specifically for reference in U.S.-regulated ratemaking, while actuarial practice varies internationally. Reconstructing a global insurance book should therefore identify country-specific rate authority and implementation rules rather than assuming uniform approval requirements.
Class, territory, fee, and rating-rule changes
Consider a simplified premium algorithm P=B×E×C+F, where B is a base rate, E an exposure measure, C a classification factor, and F a fixed policy fee. A change to B by 10% changes the rate-sensitive component of premium by 10%, but it changes total P by less than 10% when a positive fixed fee remains unchanged. A simultaneous 10% change in B and a 20% change in C produces a 32% change to the multiplicative component, before accounting for fees or changes in the rating mix.
If a rate plan replaces a fee with a premium minimum, the old-to-current ratio can vary nonlinearly with exposure size. A portfolio-wide average change estimated on a particular sample is then dependent on the sample's size distribution. The same reasoning applies to changes in deductible credits, policy limits, territory definitions, underwriting tiers, and telematics discounts. Aggregate reconstruction becomes most defensible when segmentation preserves the material structure of the change.
The professional concern follows from the ASOP 56 modeling requirements concerning purpose-fit models, inputs, validation, reconciliation, and material limitations within their scope. Even a spreadsheet reproducing historical indexes can involve meaningful model risk if it determines a material rate indication. A complex algorithm does not exempt the reviewer from checking basic effective dates and manual versions.
Rate-change versus observed-price identification
A rate filing's overall average change is a controlled comparison using stated rating assumptions. Observed average premium per policy also changes when customers alter coverage, insured values, class, limits, deductibles, or optional protections. Consequently a premium per exposure increase of 12% cannot establish a 12% manual rate increase without controlling for compositional effects.
Formally, write observed premium P_t as a functional of rate schedule R_t and exposure distribution D_t. Comparing P_1 and P_0 combines movement in R and D. An on-level counterfactual holds the chosen D at the historical composition and evaluates a different R. This identifies a price-schedule effect conditional on a specified exposure representation, not a causal response of policyholder behavior to rate changes. Selection, retention, and shopping reactions remain outside that static counterfactual.
An important consequence arises when a rate increase changes the book itself. Re-rating retained exposures at the new tariff can be internally correct while still failing to forecast the premium volume of a future book whose risk distribution changes in response to the price. That forecast is a separate actuarial and economic problem. On-level premium and premium demand should therefore be maintained as separate analytic quantities.
An executable reconstruction specification
A disciplined implementation can be summarized as a reproducible sequence: select the experience cohort and earning basis; fix the reference rate date and tariff; extract every historical policy or homogeneous rate segment; join rate-version and policy-effective-date histories; derive reference premium; earn historical and reference amounts consistently; aggregate; reconcile to source books; and document overrides and exceptions. Each decision should be data-versioned for review.
The following pseudocode expresses the policy-level calculation without asserting that this dossier has processed a real carrier file:
for transaction in historical_policy_exposures:
historical_version = resolve_rate_version(
transaction.jurisdiction,
transaction.product,
transaction.effective_date,
transaction.coverage,
transaction.implementation_rules
)
historical_premium = apply_rating(
historical_version, transaction.eligible_rating_attributes
)
reference_premium = apply_rating(
specified_reference_version, transaction.eligible_rating_attributes
)
earned_fraction = earn(
transaction.coverage_interval, experience_period, documented_earning_rule
)
sum_historical += historical_premium * earned_fraction
sum_reference += reference_premium * earned_fraction
on_level_factor = sum_reference / sum_historical
reconcile(sum_historical, general_ledger_and_actuarial_experience)
In real implementations the code must handle endorsements, cancellation reversals, premium taxes if included in the denominator, minimums, experience-rating adjustments, rounding, applicable coverage conversions, and partial data. The ASOP 23 review framework and ASOP 41 disclosure framework supply relevant professional reference points for documenting the resulting data choices and communication.
Validation and Misstatement Risks
Method comparison and error decomposition
Policy-level extension is generally the strongest representation of the declared counterfactual when detailed data and a reliable reference rating engine exist. The parallelogram method is an analytically transparent approximation when the rate changes are sufficiently homogeneous and the writing and earning assumptions are credible. Transaction-cohort weighting occupies an intermediate position: it uses actual timing and approximate rate indexes without requiring every old record to be re-rated.
The potential difference between methods can be analyzed without inventing a statistical confidence measure. Let true baseline premium weights for rate group j be a_j, assumed weights be a'_j, and true rate indexes be I_j while assumed indexes are I'_j. The difference in reconstructed average index includes a term from timing and exposure weights, another from rate-index error, and an interaction term. Specifically,
Σ_j a'_j I'_j − Σ_j a_j I_j = Σ_j (a'_j−a_j) I_j + Σ_j a_j (I'_j−I_j)
- Σ_j (a'_j−a_j)(I'_j−I_j).
This exact algebraic identity does not create a new modeling framework. It clarifies three established sources of error: incorrect composition or earning, incorrect rate history, and their interaction. A fourth source arises if an aggregate index cannot represent heterogeneous policies at all. In that case the mismatch between the modeled scalar index and policy-level price mechanics must be measured separately.
Reconciliation and governance
Control totals provide the starting point for validation. Compare the sum of extracted historical written premiums with policy-system totals by period and coverage; reconcile earned premium against actuarial triangles or accounting balances after documenting differences in basis; compare rate-event versions with filed or published manuals; and cross-check the final reference-rate premiums with an independently rated sample.
A comprehensive review also compares earned-premium factors across adjacent periods. Sudden jumps can be appropriate around major changes, but they require documentary explanation. Factors should converge toward one after all old-rate policies have expired under a stable tariff, if the method's assumptions apply. Persistent deviations can indicate long-term contracts, delayed renewals, nonuniform writings, actual re-pricing differences, or data errors.
The ASB model validation guidance addresses checking and reconciliation appropriate to the intended application. Here, a reproducible calculation should retain data extracts, rate tables, program versions, test cases, exception logs, time conventions, reviewer decisions, and a checksum of the approved analytical package. Hashes establish that a particular artifact is unchanged; they do not establish that its historical rate assumptions are correct.
International relevance and jurisdictional limits
The time-weighting mathematics is broadly useful wherever contracts are priced at inception or renewal and premium earns through a coverage period. Yet local regulatory meanings of rate approval, filing, change in policy benefits, taxes and levies, individual risk rating, and recognized premium are jurisdiction-specific. Even the choice between booked, earned, billed, audited, written, net, and gross premium can substantially affect the proper data set.
ASOP 53 is U.S. professional guidance for defined property/casualty future cost estimation work; it is not a global statutory code. The ASOP 53 scope statement also makes clear that estimated future costs, pricing decisions, and rate filings are not equivalent concepts. The CAS educational material is international in pedagogical reach but is not itself an enacted law. An international application should first establish its relevant local legal, accounting, and actuarial requirements.
The ASOP 13 distinctions are also important: trending addresses temporal movements in future values, while premium development and on-level rate adjustment serve different purposes. Treating their outputs as interchangeable can conceal an analytical misstatement even if every arithmetic operation is correct.
Original synthesis and practical implications
The main interpretive result is that on-level premium is meaningful only relative to a declared counterfactual. Its numerator must specify what is held constant, which current prices replace historical prices, how the earning period is preserved, and which transactions belong to the book. This observation unifies the apparently different extension and parallelogram procedures under their common practical objective while respecting the information sacrificed by aggregation.
A second conclusion concerns historical provenance. A rate history is a history of applicable rules and implementation, not simply a succession of published percentage changes. Aggregating such changes without the policy population for which they were measured creates an apparent numeric precision that can mask class-specific bias. Conversely, an exact-looking re-rating based on incomplete historical attributes can reproduce the wrong contract.
A third conclusion concerns decision use. A valid on-level historical denominator improves the interpretation of experience, but a prospective rate estimate additionally depends on forward losses, expenses, capital, risk mix, and market and regulatory conditions. The transformation is one auditable step in ratemaking. Its results should never be presented as a complete prospective price or as evidence that a particular insurer's existing rates are adequate.
Finally, source completeness is a graded evidentiary matter. A public filing may justify the reported approved change but not the realized rate for a historical insured. A rating-manual archive may justify the algorithm but not the exposure values actually used. A policy-system replay may establish computational consistency but not independence from shared data errors. Good scholarly and actuarial practice assigns each source precisely the claim it supports and identifies the remaining gaps.
Conclusion
On-level premium reconstruction translates historical earned or written premium to a common rate basis by combining the sequence of applicable rate changes with the actual or modeled temporal allocation of insured exposures. Under fully specified, homogeneous conditions the familiar factor is the terminal rate index divided by a weighted historical index. For heterogeneous risks, exact extension of exposures more faithfully preserves the class, fee, and rating-rule changes that aggregate percentages conceal. The parallelogram is valuable precisely because its assumptions are explicit and its arithmetic can be independently checked.
The worked examples show the essential time lag. A July increase affects only a small portion of the first year's earned premium under annual new-business and uniform earning assumptions, while its effect approaches full recognition in the next. Two changes in the same year require sequential multiplicative rate indexes and disjoint earned exposure weights. The resulting premium amounts are conditional numerical illustrations, not observations about an actual portfolio.
The analysis establishes a disciplined sequence of scientific tasks: identify the intended current-rate counterfactual; reconstruct approval and implementation chronology; verify contract-level eligibility, exposures, and earning patterns; select re-rating or a properly supported approximation; reconcile data and arithmetic; evaluate uncertainty from aggregation and missing records; and communicate the result with its precise boundaries. These conclusions are derived from established actuarial methods and verified primary professional materials. They provide a source-based, internationally interpretable account of rate-history reconstruction without establishing a new actuarial theory.
References
Actuarial Standards Board. (2009; last revised 2011). Actuarial Standard of Practice No. 13: Trending Procedures in Property/Casualty Insurance. Effective November 1, 2009. Official ASOP 13.
Actuarial Standards Board. (2010). Actuarial Standard of Practice No. 41: Actuarial Communications. Effective May 1, 2011. Official ASOP 41.
Actuarial Standards Board. (2016). Actuarial Standard of Practice No. 23: Data Quality. Official ASOP 23.
Actuarial Standards Board. (2017). Actuarial Standard of Practice No. 53: Estimating Future Costs for Prospective Property/Casualty Risk Transfer and Risk Retention. Effective August 1, 2018. Official ASOP 53.
Actuarial Standards Board. (2019). Actuarial Standard of Practice No. 56: Modeling. Effective October 1, 2020. Official ASOP 56.
Casualty Actuarial Society. (1988; reinstated 2021 for reference for U.S.-regulated ratemaking). Statement of Principles Regarding Property and Casualty Insurance Ratemaking. CAS principles document.
Casualty Actuarial Society. (2021, May 7). CAS Board Agrees to Reinstate Ratemaking Principles. CAS board notice.
National Council on Compensation Insurance. (2025). Annual Statistical Bulletin, Exhibit 2, Premium Level Changes. NCCI statistical bulletin.
New York State Department of Financial Services. (n.d.; accessed November 14, 2026). Property Insurers: Filing Instructions for Paper Submissions. DFS filing instructions.
Werner, Geoff, and Claudine Modlin. (2016). Basic Ratemaking. Fifth edition, Casualty Actuarial Society, especially chapters 4–5 and chapter 8. CAS complete textbook.