Assigning Exposures to Classification Cells

Why Exposure Assignment Requires More Than a Policy Count

A classification cell is meaningful only if its exposure total represents the amount of risk assigned to that class during the experience period. A policy can cover only part of that period, insure several vehicles or locations, or move from one class to another before expiration. Counting each policy once would ignore these differences and can put exposure in a class where the risk was not actually insured.

The task here is to calculate and allocate exposure from policy records. It is important to keep three questions separate: what the exposure unit is, when that unit was covered, and which classification applied during each covered interval. A correct total answers all three.

Definition: An exposure unit is the measure of insured risk used for the ratemaking analysis, such as one vehicle-year, one location-year, or a stated amount of payroll. A class exposure is the total of those units earned in a particular classification cell during the experience period. For a time-based unit, one unit covered for a full year earns one annual exposure; one unit covered for a fraction of a year earns that fraction.

The exposure basis must be stated before calculating. For example, if the analysis uses vehicle-years, a policy covering four vehicles for half a year contributes two vehicle-years, not one policy-year. If the applicable exposure base is payroll, the exposure is the payroll amount earned or reported for the relevant period; it is not automatically the number of policies multiplied by a time fraction.

A Calculation Model for Time-Based Exposures

Suppose the records are divided into intervals during which both the covered-unit count and classification are constant. Let \(u_j\) be the number of covered units in interval \(j\), let \(d_j\) be the number of covered days in that interval, and let \(c_j\) be its classification. Using a 365-day annual basis, interval \(j\) contributes \(u_jd_j/365\) annual exposures to class \(c_j\).

E_c=\sum_{j:\,c_j=c} \frac{u_jd_j}{365}

Here \(E_c\) is the earned exposure assigned to class \(c\). If the covered-unit count is constant for one policy across a portion of the experience period, the formula reduces to units multiplied by the fraction of a year covered. With many policies, calculate the contribution for each relevant interval and sum contributions cell by cell.

Assumption and condition: The 365-day calculation treats each covered day as \(1/365\) of an annual exposure. Use the day-count and annualization convention required by the data and analysis; apply it consistently. The time-proration formula is appropriate when exposure accrues uniformly with time and the unit count and class are known for each interval. If the exposure measure changes with something other than elapsed time, such as actual payroll, use the measured exposure base rather than assuming uniform time accrual.

For date-based records, define intervals consistently. In the examples below, coverage and experience periods use half-open intervals: the start date is included and the end date is excluded. Thus, an interval from July 1 to September 1 includes July and August, and its day count is the difference between those dates. This avoids counting a boundary date twice when a record is split into consecutive intervals.

Why Time Proration Works

The proration formula follows from the definition of an annual exposure. Under uniform accrual, one unit covered for 365 days earns one annual exposure. A unit covered for \(d_j\) days earns the same proportion of an annual exposure as its covered days are of the annual basis. With \(u_j\) units, add the contribution from each unit:

\underbrace{\frac{d_j}{365}+\cdots+\frac{d_j}{365}}_{u_j\ \text{units}} =\frac{u_jd_j}{365}

When a policy changes class or unit count, split it into intervals where the relevant values are constant. The intervals do not overlap and together cover the insured time, so summing their contributions neither omits nor duplicates covered days. This is why the date of a change matters: a classification change reallocates exposure from one cell to another, while a unit-count change alters how much exposure accrues.

Worked Examples

Worked Example: Partial Terms and Multiple Vehicles

A hypothetical auto program uses vehicle-years and has a calendar-year 2026 experience period, from January 1, 2026 to January 1, 2027. Policy A covers three vehicles from October 1, 2025 to October 1, 2026. Policy B covers two vehicles from April 1, 2026 to April 1, 2027. Both policies remain in the same classification while covered. Calculate their combined 2026 exposure. The records are deterministic; no interest rate, mortality assumption, or claim distribution is used.

Method: overlap each policy term with the experience period, then multiply covered days by the number of vehicles. Policy A overlaps 2026 from January 1 to October 1. That is 273 days, so its earned exposure is \(3(273)/365=819/365=2.2438\) vehicle-years, rounded to four decimal places.

Policy B overlaps 2026 from April 1, 2026 to January 1, 2027. This is 275 days. Its earned exposure is \(2(275)/365=550/365=1.5068\) vehicle-years, rounded to four decimal places.

Adding the unrounded contributions gives \( (819+550)/365=1369/365=3.7506849\ldots\), or 3.7507 vehicle-years to four decimal places. The total is less than the five vehicles insured at some point because neither policy covered all its vehicles for the full experience year. The common mistake avoided is counting each policy once, which would give two policy exposures instead of the earned vehicle-year total.

Worked Example: Moving Units to a New Classification

A hypothetical policy covers two vehicles for all of calendar year 2026. Both vehicles are in class A from January 1 through June 30. Effective July 1, both move to class B and remain covered there through December 31. Use a 365-day basis to calculate exposure in each class and check the total. No interest rate, mortality assumption, or claim distribution is used.

Method: split the policy at the effective classification-change date. The interval from January 1 to July 1 is 181 days. Class A therefore receives \(2(181)/365=362/365=0.9918\) vehicle-years, rounded to four decimal places.

The interval from July 1, 2026 to January 1, 2027 is 184 days. Class B receives \(2(184)/365=368/365=1.0082\) vehicle-years, rounded to four decimal places.

The exact total is \((362+368)/365=730/365=2\) vehicle-years, which equals two vehicles covered for a full year. The class exposures are close to one each, but are not exactly equal because the two intervals have different numbers of days. The common mistake avoided is assigning all two vehicle-years to the class at year-end; class B applies for only the second interval.

Worked Example: Partial Policy With Unit and Class Changes

A hypothetical fleet policy uses vehicle-years and covers May 1 to November 1, 2026. The policy records show three class A vehicles and one class B vehicle from May 1 to July 1; two class A and two class B vehicles from July 1 to September 1; and two class A and one class B vehicle from September 1 to November 1. Calculate exposure by class and reconcile it to total vehicle exposure. No interest rate, mortality assumption, or claim distribution is used.

Method: divide the term at every date when the unit count or class changes. The first interval is 61 days. Class A contributes \(3(61)=183\) vehicle-days and class B contributes \(1(61)=61\) vehicle-days. The second interval is 62 days; each class contributes \(2(62)=124\) vehicle-days. The third interval is 61 days; class A contributes \(2(61)=122\) vehicle-days and class B contributes \(1(61)=61\) vehicle-days.

The class A total is \(183+124+122=429\) vehicle-days, so \(E_A=429/365=1.1753\) vehicle-years to four decimal places. The class B total is \(61+124+61=246\) vehicle-days, so \(E_B=246/365=0.6740\) vehicle-years to four decimal places.

For the reconciliation, the fleet has four vehicles for 61 days, four for 62 days, and three for 61 days. That is \(4(61)+4(62)+3(61)=244+248+183=675\) vehicle-days, or \(675/365=1.8493\) vehicle-years. The class totals also give \((429+246)/365=675/365=1.8493\), as required.

The common mistake avoided is applying the final vehicle mix, two class A and one class B, to the whole 184-day term. That would produce \(2(184)/365=368/365\) in class A and \(184/365\) in class B: only \(552\) vehicle-days in total, rather than the recorded \(675\). The error ignores both the earlier unit count and the earlier class mix.

Practical Checks and Common Traps

Exposure assignment is a data-processing task as much as an arithmetic task. Before aggregating, confirm that policy effective dates, cancellation dates, unit counts, and classification effective dates are available at the level needed to identify each interval. A policy-level start and end date alone cannot identify how much exposure belongs in each class if the classification changed midterm.

Use the defined unit. One policy is not necessarily one exposure. A policy covering several cars contributes the time-weighted number of cars when the chosen basis is vehicle-years.
Use only the overlap. A policy contributes nothing before coverage begins, after coverage ends, or outside the experience period.
Split at every relevant change. When class or unit count changes, create a new interval, even if the policy number remains the same.
Keep the convention consistent. Do not mix a 365-day basis with another day-count method across records. Follow the specified convention for leap years and partial days.
Reconcile totals. The sum across classification cells should equal the time-weighted exposure implied by the covered units and dates, provided all units belong to exactly one cell at each point in time.

Key takeaway: Assign exposure by covered unit, covered time, and classification in force during that time. Split records whenever any of those elements changes, then reconcile the cell totals to the underlying exposure represented by the policy records.

Check Your Understanding

For each problem, use a 365-day annual basis and half-open date intervals. Round final exposure answers to four decimal places.

A hypothetical policy covers three locations from March 1 to October 1, 2025. The experience period is April 1 to September 1, 2025, and all locations share one class. How many location-years are earned in that class? (Answer: 1.2575 location-years)
A hypothetical policy covers four units throughout 2026. All four are class X from January 1 to May 1, then class Y from May 1 through December 31. Find each class exposure and verify the sum. (Answer: Class X = 1.3151, class Y = 2.6849; total = 4.0000 annual exposures)
A hypothetical policy covers five vehicles from June 1 to December 1, 2026. Two vehicles are removed effective September 1; the remaining vehicles stay in the same class. Find the class exposure for the term. (Answer: 2.0082 vehicle-years)
A hypothetical policy covers three vehicles from January 1 to July 1, 2026. From January 1 to April 1, two are in class A and one in class B. From April 1 to July 1, one is in class A and two are in class B. Calculate both class exposures and check their combined total. (Answer: Class A = 0.7425, class B = 0.7452; total = 1.4877 vehicle-years)

What You'll Learn