Why the Record Type Matters
Exposure calculations depend on when coverage was actually in force, not on how many database rows describe a policy. A policy may generate several records during its term: an inception, an endorsement, a cancellation, and perhaps a reinstatement. Those records describe events or changes. They do not necessarily represent separate policies or separate blocks of time.
A policy snapshot answers a different question: what did the policy look like at a particular point in time? It may show that a policy was active on the snapshot date, but it generally does not show every earlier interval of active and inactive coverage. Treating a snapshot as a complete history can therefore produce incorrect exposure.
For historical exposure, the effective date usually determines where an event belongs in the coverage timeline. The processing date still matters for data controls: a late-entered transaction may mean that an earlier snapshot did not contain the event. When the purpose is to reconstruct actual historical coverage, the transaction history must be interpreted using the effective dates and any applicable corrections, rather than simply reading the latest snapshot.
Reconstructing the Coverage Timeline
A reliable reconstruction starts with a policy’s coverage term and its effective-dated changes. Sort events by effective date, determine the status immediately after each event, and join consecutive dates over which coverage status is unchanged. An endorsement that changes policy information without ending coverage does not, by itself, create a new exposure interval or a new policy.
For the examples in this tutorial, use a monthly pro rata convention: each month of active coverage contributes \(1/12\) policy-year, and each month of inactive coverage contributes zero. Dates are at month boundaries, and an interval includes its start but excludes its end, written \([a,b)\). This convention prevents double counting at a change date: the time immediately before the date belongs to the earlier interval, and the time from the date onward belongs to the new status. These are deterministic exposure calculations; no interest rate, mortality assumption, probability distribution, or distribution parameter is used.
The reconstruction can be justified by partitioning the measurement period at every effective date that changes coverage status. Let \(I(t)\) equal 1 when coverage is active at time \(t\), and 0 otherwise. Exposure is the total time for which \(I(t)=1\). Because the effective dates partition the timeline into intervals of constant status, total active time is the sum of the lengths of active intervals:
Here, \([a_j,b_j)\) are the partitioned intervals, and the sum includes only intervals where coverage is active. Dividing active months by 12 converts this time to policy-years under the monthly convention. The partition is important: intervals neither overlap nor leave an unclassified gap, unless the policy is genuinely inactive during that gap.
Worked Examples
Worked Example: Several Transactions, One Policy
A hypothetical policy has a term from January 1, 2025, through January 1, 2026. Its transaction history shows inception effective January 1, an endorsement effective April 1, and a second endorsement effective June 1. Both endorsements update policy information but do not cancel or interrupt coverage. A July 1 snapshot shows the policy active. Calculate its earned policy-year exposure from January 1 through July 1, and state how many policies the records represent.
Method. Reconstruct the active timeline from the effective dates, then total active months. The processing dates are not needed because the problem provides the effective dates and says coverage was uninterrupted.
The coverage interval is \([\text{January 1},\text{July 1})\), which contains six months: January through June. The April 1 and June 1 endorsements divide the record history into transactions, but neither date changes the policy’s active status.
Active exposure is \(6/12=0.5\) policy-years. The transaction history represents one policy, not three policies: the inception and two endorsements all relate to the same policy.
Check. The snapshot confirms the policy is active on July 1, but the effective-dated history establishes the six months of coverage before that date. The calculation counts no month twice and does not treat either endorsement as a new policy. The common mistake avoided is counting three transaction rows as three policies or as three separate exposures.
Worked Example: Cancellation and Reinstatement Create a Gap
A hypothetical annual policy begins January 1, 2025, and is scheduled to end January 1, 2026. It is cancelled effective May 1 and reinstated effective August 1, with coverage continuing through the scheduled end date. Calculate its earned exposure for the full policy term. A September 1 snapshot shows the policy active.
Method. Use the effective dates to form active and inactive intervals. The September snapshot describes status on that date; it does not establish uninterrupted coverage since January.
The first active interval is \([\text{January 1},\text{May 1})\), or four months. From May 1 through August 1, coverage is inactive for three months. The second active interval is \([\text{August 1},\text{January 1, 2026})\), or five months.
Total active time is \(4+5=9\) months. Therefore, earned exposure is \(9/12=0.75\) policy-years. The inactive gap contributes zero exposure, even though the policy is active on September 1.
Check. The intervals cover the full 12-month term: four active months, three inactive months, and five active months, for \(4+3+5=12\). The active and inactive durations are consistent, and the total active exposure is less than one policy-year. A common incorrect calculation is to count January through September as eight active months because the policy is active on September 1. That ignores the cancellation gap. The corrected calculation through September 1 is four months before cancellation plus one month after reinstatement, or \(5/12\) policy-years.
Worked Example: Aggregate Exposure Across Two Policies
A hypothetical book contains two policies. Policy A is effective January 1, 2025, through January 1, 2026. It has a non-cancelling endorsement effective June 1, is cancelled effective August 1, and is reinstated effective October 1. Policy B is effective April 1, 2025, through October 1, 2025, with no interruption. Calculate total earned policy-year exposure for both policies through January 1, 2026, and distinguish the number of policies from the number of listed transactions.
Method. Reconstruct each policy’s active intervals separately, calculate each policy’s active months, and then add the exposures. An endorsement without an interruption changes the transaction history but not active status.
For Policy A, coverage is active from January 1 through August 1: January through July, or seven months. It is inactive from August 1 through October 1, or two months. It is active again from October 1 through January 1, 2026: October through December, or three months. Thus Policy A has \(7+3=10\) active months and exposure \(10/12\) policy-years.
For Policy B, the interval from April 1 through October 1 is six months. Its exposure is \(6/12\) policy-years.
Adding the two policies gives \(10+6=16\) active policy-months. Total exposure is \(16/12=1.3333\) policy-years, rounded to four decimal places. The record list describes two distinct policies. Policy A has four listed transactions, and Policy B has one inception transaction, so there are five listed transaction records.
Check. Policy A’s full 12-month term comprises \(7+2+3=12\) months, of which 10 are active. Policy B contributes six active months. Total exposure is \(10/12+6/12=16/12\), as calculated. The common mistake avoided is treating the five transaction records as five policies or counting Policy A’s endorsement as extra exposure.
Data Checks and Common Traps
Before aggregating exposure, verify that the records describe a coherent timeline. The following checks help distinguish a true coverage change from a data artifact:
- Separate effective date from entry date. A transaction entered late may change the reconstructed history from its effective date. Do not automatically assign it to the entry date.
- Sort by effective date. Records may have been entered in a different order from the order in which their changes took effect.
- Do not turn every transaction into a policy. An endorsement or correction can relate to an existing policy without creating a new policy or a new active interval.
- Represent gaps explicitly. If cancellation and reinstatement leave a period without coverage, that period contributes no earned exposure.
- Do not infer a full history from a snapshot. An active status at one date does not prove coverage was active throughout the preceding period. A snapshot might not retain earlier cancellations, reinstatements, or other changes.
- Check the measurement window. Count only active time that falls inside the requested period; do not include future coverage just because it appears in the policy’s full term.
A snapshot is useful when the question concerns the portfolio’s state at a specific date. A transaction history is generally needed when the question concerns exposure earned across a period. If the transaction history is incomplete or conflicting, the correct response is not to manufacture continuity from snapshots: the missing coverage history must be resolved or the uncertainty clearly recognized.
Check Your Understanding
For each problem, use the monthly pro rata convention and the half-open interval convention described above.
- A hypothetical policy runs from January 1 through January 1 of the following year. It has two non-cancelling endorsements effective March 1 and July 1. What is its exposure through September 1, and how many policies do the three transaction records represent? (Answer: \(8/12=0.6667\) policy-years; 1 policy.)
- A policy is active from January 1 through June 1, cancelled from June 1 through September 1, and reinstated from September 1 through January 1 of the next year. What is its exposure for the term? (Answer: \(5+4=9\) active months, or \(0.75\) policy-years.)
- A policy is active from January 1 through May 1, cancelled through August 1, and active again through January 1 of the next year. A snapshot on July 1 shows it inactive. How much of the full term is earned exposure? (Answer: \(4+5=9\) active months, or \(0.75\) policy-years.)
- Two distinct policies are each active without interruption for six months within a measurement period. One has four endorsement records and the other has none. What is their combined exposure, and how many policies are represented? (Answer: \(6/12+6/12=1.0\) policy-year; 2 policies.)
- A snapshot shows a policy active on January 1 and active again on July 1. No transaction history or other dates are available. Can the earned exposure between those dates be determined from these two snapshots alone? (Answer: No. The snapshots do not establish whether coverage was uninterrupted between January 1 and July 1.)