Reconciling Exposure Totals

Why Exposure Reconciliation Requires More Than a Grand Total

Exposure totals are inputs to ratemaking analyses, so a total that cannot be tied to a defined population and unit is difficult to interpret. A useful reconciliation asks more than whether two grand totals are equal. It asks whether both totals cover the same policies and coverages, use the same exposure basis and units, and assign exposure to the same periods. Even when those aggregate checks agree, record-level errors can cancel one another.

For this tutorial, an exposure record is a policy-coverage record, or another defined observational unit, with an exposure amount assigned to a specified period. The correct unit depends on the exposure basis—for example, vehicle-years, payroll, sales, or insured value. The earlier material on exposure bases, period assignment, identifiers, duplicates, and exposure corrections establishes how those amounts and records are constructed. Reconciliation checks whether the resulting data agree with independent controls and with their underlying source records.

Definition: An exposure reconciliation compares calculated exposure records or aggregates with independently defined control records or totals. A reconciliation must specify the population, period, coverage or other grouping, exposure basis, unit and scale, and the method used to compare the two sides. A total is reconciled only to the extent supported by checks at the stated level of detail.

Set the Reconciliation Level Before Comparing

A control total is meaningful only when its definition matches the calculated total. Before calculating a difference, establish which policies and coverages are in scope; which date or period convention applies; what one exposure unit represents; and whether monetary amounts are expressed in dollars, thousands of dollars, or another scale. If, for example, a control covers policy-year exposure but the calculation covers calendar-year exposure, a difference may reflect the period definitions rather than an error in arithmetic.

The two sides also need a common grouping. A total by year cannot directly establish that each policy was assigned correctly. A total by coverage cannot establish that policy identifiers match. A practical reconciliation therefore moves from broad aggregates to finer detail: first overall totals, then period and exposure-unit totals, then coverage or other relevant groups, and finally individual identifiers and source records.

Keep counts and exposure amounts separate. A policy count measures records or policies under a stated counting rule; an exposure total measures amounts in the chosen exposure unit. One policy can have several coverage records, and different policies can contribute different exposure amounts. Therefore, equal policy counts do not prove equal exposure totals, and equal exposure totals do not prove equal policy or coverage counts.

Definition: For a group $g$, let $E^{\mathrm{calc}}_g$ be the calculated exposure total and $E^{\mathrm{ctrl}}_g$ the corresponding control total. Define the signed difference as $D_g=E^{\mathrm{calc}}_g-E^{\mathrm{ctrl}}_g$. A positive value means the calculation exceeds the control; a negative value means it is below the control. The difference is interpretable only if both totals use the same population, grouping, period, basis, and units.

If the control is nonzero, a relative difference can also be useful: $D_g/E^{\mathrm{ctrl}}_g$. For example, a difference of 2 exposure units is more material against a control of 20 than against a control of 20,000. Do not calculate this ratio when the control is zero; report the signed difference and investigate the underlying records instead.

A Reliable Reconciliation Sequence

Define the control.
Record the control’s population, period, exposure basis, units, grouping, and counting rule. Confirm that these match the calculation being checked.

Compare aggregate exposure and counts.
Calculate differences separately for exposure amounts and for policy or policy-coverage counts. A count difference is not an exposure difference.

Break differences into useful groups.
Reconcile by period, coverage, exposure unit, or another relevant attribute. Group totals show where to investigate but do not identify which records caused the difference.

Compare identifiers and source values.
Match records using stable identifiers and compare the amount, period, coverage, and unit for each matched record. Identify records present on only one side as well as matched records with different values.

Resolve and document.
Trace mismatches to source records or processing rules, correct only with support, and rerun the aggregate and record-level checks. Preserve the original difference and the reason for its resolution.

Why a Matching Total Does Not Prove Matching Records

The reason to inspect records, even when an aggregate ties, follows directly from addition. Suppose each record $j$ has a calculated amount $e^{\mathrm{calc}}_j$ and a control amount $e^{\mathrm{ctrl}}_j$. The aggregate difference is the sum of the record-level differences:

D=\sum_{j=1}^{n}e^{\mathrm{calc}}_j-\sum_{j=1}^{n}e^{\mathrm{ctrl}}_j =\sum_{j=1}^{n}\left(e^{\mathrm{calc}}_j-e^{\mathrm{ctrl}}_j\right).

If one record is understated by 0.50 and another is overstated by 0.50, their differences sum to zero. Thus $D=0$ does not imply that every record-level difference is zero. It establishes only that the net aggregate difference is zero for the records and definitions included in the comparison. This is why aggregate agreement is a useful check, but not a substitute for identifier-level verification when record accuracy matters.

Worked Example: A Period Total Ties While a Record Is Misallocated

A hypothetical insurer checks exposure in vehicle-years for two policy years. The year-two control contains three policy-coverage records: J has 0.50, K has 1.50, and L has 2.00 vehicle-years. The calculated file shows J with 0.00, K with 2.00, and L with 2.00. Year one agrees at 4.00 vehicle-years on both sides. Reconcile the totals and determine what the year-two agreement does and does not establish.

Assumptions. This is a deterministic data reconciliation, not a probabilistic model: no probability distribution, mortality assumption, or interest rate is used. Both sides are intended to measure vehicle-years for the same policy-coverage records and policy-year definitions.

Step 1: Calculate year-two control exposure. The control total is $0.50+1.50+2.00=4.00$ vehicle-years.

Step 2: Calculate year-two calculated exposure. The calculated total is $0.00+2.00+2.00=4.00$ vehicle-years. The signed aggregate difference is $4.00-4.00=0.00$.

Step 3: Compare by record. J is understated by $0.00-0.50=-0.50$; K is overstated by $2.00-1.50=+0.50$; and L has difference $2.00-2.00=0.00$. The record differences sum to $-0.50+0.50+0.00=0.00$, explaining why the year-two aggregate ties.

Conclusion and trap avoided. The year-two total ties the control, but that total does not establish that J’s 0.50 allocation is present. Comparing the records shows both J’s shortfall and K’s offsetting excess. Saying “the year-two total ties, so J’s allocation is present” would confuse aggregate agreement with record-level agreement.

Check Units Before Treating a Difference as an Exposure Error

A numerical difference can arise from inconsistent scales rather than missing or extra exposure. This is especially important for monetary exposure bases such as insured value or payroll, where one file may store dollars and another may store thousands of dollars. Convert to one unit before comparing totals; never infer a conversion merely because the totals look different.

Worked Example: Finding a Scale Mismatch in Insured Values

A hypothetical property portfolio has three in-scope locations with insured values of $120,000, $80,000, and $100,000. The control reports insured value in thousands of dollars. The calculated file is intended to use the same unit, but its records show 120, 80, and 100,000. Reconcile the amount and identify what should be checked.

Assumptions. The values describe the same three locations, period, and coverage on both sides. The control unit is $1,000 of insured value. No interest rate, mortality assumption, or probability distribution is involved; this is a deterministic comparison.

Step 1: Convert source values to the control unit. Divide each dollar amount by $1,000: $120{,}000/1{,}000=120$, $80{,}000/1{,}000=80$, and $100{,}000/1{,}000=100$. The control total is $120+80+100=300$ thousand dollars.

Step 2: Sum the calculated file as stored. Its total is $120+80+100{,}000=100{,}200$ thousand dollars. The signed difference is $100{,}200-300=99{,}900$ thousand dollars.

Step 3: Trace the large difference. The first two records match their converted source values. The third record is stored as 100,000 where 100 is expected in thousands of dollars. Its excess is $100{,}000-100=99{,}900$, exactly the aggregate difference.

Conclusion and trap avoided. The record comparison points to a scale mismatch on the third value, not a small rounding difference. Dividing the entire calculated total by 1,000 would be the wrong correction: it would also change the two records already in the correct unit. Confirm the source field and conversion rule, correct the affected record, and then rerun the reconciliation.

Counts and Identifier Sets Reveal Different Problems

A count comparison is a useful screening check, but its interpretation must be precise. If the calculated file has six policy-coverage combinations and the control has five, the count difference is a net excess of one combination. The counts alone do not establish that exactly one particular combination is extra. The calculated set could contain several combinations absent from the control while also omitting combinations that appear in the control. Comparing identifiers reveals which combinations are present on one side but not the other.

Worked Example: Interpreting a Net Excess in Coverage Records

A hypothetical fleet program has the following control policy-coverage combinations and exposures, in vehicle-years: P1-A, 10; P1-B, 5; P2-A, 8; P3-A, 12; and P4-A, 15. The calculated file contains P1-A, 10; P2-A, 8; P2-B, 4; P3-A, 12; P4-A, 15; and P5-A, 6. Reconcile both record counts and exposure totals.

Assumptions. Each listed pair is one policy-coverage combination; both sides use vehicle-years and the same period. No probability distribution, mortality assumption, or interest rate is used.

Step 1: Compare counts. The control has five combinations and the calculation has six. The count difference is $6-5=1$: a net excess of one calculated combination.

Step 2: Compare identifier sets. P1-B appears in the control but not the calculation. P2-B and P5-A appear in the calculation but not the control. Therefore, the comparison identifies one control-only combination and two calculated-only combinations; it does not identify a single extra combination.

Step 3: Compare exposure totals. The control total is $10+5+8+12+15=50$ vehicle-years. The calculated total is $10+8+4+12+15+6=55$ vehicle-years. The signed exposure difference is $55-50=5$ vehicle-years.

Step 4: Explain the net difference. The calculation omits 5 vehicle-years for P1-B and includes 4 for P2-B plus 6 for P5-A. The net difference is $-5+4+6=5$, agreeing with the aggregate comparison.

Conclusion and trap avoided. The count difference of one is a net result, not proof of exactly one extra record. Identifier matching finds two calculated-only combinations and one control-only combination. A reviewer who labels one combination “the extra record” based only on the count difference would miss the omission and the second addition.

Trace Differences to Evidence, Not Guesswork

Once a difference is localized, inspect the relevant source record and processing path. Depending on the discrepancy, the investigation may check whether a record was omitted or duplicated, whether an exposure correction was applied on only one side, whether a period assignment differs, or whether the exposure unit was converted consistently. The earlier material on exposure corrections and restatements is relevant when a valid source change has occurred: the reconciliation should identify which version or adjustment each side includes.

Do not force a tie by changing a value without evidence. A difference may be valid if the two controls deliberately use different populations or definitions, but that is a scope difference to document—not a data correction. Conversely, a zero aggregate difference is not enough to close the review if the task requires verifying the record population or allocations.

Key takeaway: Reconcile exposure in layers. Confirm definitions and units, compare exposure totals and counts separately, inspect differences by period and other relevant groups, and use identifiers and source records to establish what is actually present. A tied aggregate demonstrates a tied aggregate—not necessarily a correct record set or allocation.

Check Your Understanding

For each question, state what the aggregate comparison establishes and what additional check, if any, is needed.

A control reports 18.0 exposure units for a period. Two calculated records are 7.0 and 11.0 units. What is the signed difference, and does the total alone prove that both records are correct? (Answer: The difference is $18.0-18.0=0.0$; no, compare the two records with their controls or source records.)
A payroll control is $2,400 thousand. A calculated total is $2,520 thousand. Find the signed and relative differences, using the control as the denominator. (Answer: $+120$ thousand dollars and $120/2{,}400=0.05$, or $+5\%$.)
A control has four policy-coverage combinations; the calculated file has five. What does the count difference establish, and what does it not establish? (Answer: It establishes a net excess of one calculated combination; it does not establish that exactly one specific combination is extra or rule out additions and omissions.)
For one period, control exposures for records A, B, and C are 2.0, 3.0, and 5.0. Calculated values are 2.5, 2.5, and 5.0. Find the aggregate difference and the record differences. (Answer: Aggregate difference $=10.0-10.0=0.0$; record differences are A $+0.5$, B $-0.5$, and C $0.0$.)
A control total is 240,000 dollars, while the calculated total is 240 in thousands of dollars. Are the totals reconciled? Give the common unit and the difference. (Answer: Yes, after unit conversion both equal $240,000; the difference is $0.)

What You'll Learn