Diagnostic Checks For Expected Loss Calculations

Why Expected Loss Estimates Need Diagnostics

An expected loss estimate can be calculated correctly from the numbers entered and still be unreliable. A misplaced decimal, an annual loss cost applied to monthly exposure, or a selected loss ratio inconsistent with the premium input can move the result substantially. A diagnostic review traces the estimate back to its inputs and tests whether the arithmetic, units, period, and implied magnitude fit together.

These checks do not establish that a selected loss ratio, loss cost, or emergence pattern is appropriate. They identify calculations and assumptions that deserve review. In particular, a difference between expected and observed losses is not automatically an error: actual claims fluctuate around expectations. The diagnostic question is whether the estimate is internally coherent and whether its differences have a reasonable explanation.

Definition: A diagnostic check for an expected loss estimate is a calculation or comparison used to test the estimate's arithmetic, input units, period alignment, magnitude, or consistency with its stated premium, exposure, and selected assumptions. A failed check is a signal to investigate, not by itself proof that the estimate is wrong.

Trace the Calculation and Reconcile Its Parts

For a premium-based estimate, expected ultimate losses are earned premium multiplied by the selected expected loss ratio. For an exposure-based estimate, expected ultimate losses are exposure multiplied by the expected loss cost per exposure unit. If the estimate is divided into segments, calculate each segment and add the results; do not assume an unweighted average of segment assumptions will reproduce the total.

\widehat{U}_{P} = P \times r \qquad\text{and}\qquad \widehat{U}_{E} = E \times c

Here $P$ is the earned premium used in the estimate, $r$ is the selected expected loss ratio, $E$ is the relevant exposure, and $c$ is the selected expected loss cost per exposure unit. The hat indicates an estimate. Before multiplying, confirm that the premium, ratio, exposure, and loss cost refer to compatible periods and definitions.

\widehat{U} = \sum_{j=1}^{n} P_j r_j \qquad\text{or}\qquad \widehat{U} = \sum_{j=1}^{n} E_j c_j

These aggregate formulas follow by adding the expected losses across segments. If segment $j$ has premium $P_j$ and selected loss ratio $r_j$, its expected losses are $P_jr_j$; adding those amounts gives the total. The same reasoning applies to exposure and loss cost. If all segments use a common loss ratio $r$, then $\sum_j P_jr = r\sum_jP_j$. This is why multiplying total premium by a common ratio should agree with summing segment results, apart from rounding.

Recalculate from the stated inputs.
Write down the exact premium or exposure, selected assumption, and operation. Recompute independently rather than checking only a reported total.

Check units and period.
Confirm that exposure units and loss costs have matching time bases, and that the premium represents the period intended by the estimate.

Reconcile components.
Add segment estimates and, when relevant, check that expected reported and expected unpaid amounts sum to expected ultimate losses.

Assess magnitude and consistency.
Compare the result with the scale of its inputs, implied loss ratio or loss cost, and stated emergence assumptions. Investigate unexplained differences.

Worked Checks

Worked Example: Auditing a Premium-Based Estimate

A hypothetical insurer estimates expected ultimate losses for a book using earned premium of $2,400,000 and a selected expected loss ratio of 62%. The selected emergence pattern indicates that 75% of expected ultimate losses are expected to be reported by the evaluation date. Recalculate expected ultimate, expected reported, and expected unpaid losses, then verify the results.

Method: premium-based estimate followed by an emergence check. First calculate expected ultimate losses:

\widehat{U} = \$2{,}400{,}000 \times 0.62 = \$1{,}488{,}000.

Expected reported losses are 75% of that estimate:

\widehat{R} = \$1{,}488{,}000 \times 0.75 = \$1{,}116{,}000.

The expected unpaid share is $1-0.75=0.25$, so:

\widehat{U}_{\text{unpaid}} = \$1{,}488{,}000 \times 0.25 = \$372{,}000.

Check the split: $\$1{,}116{,}000+\$372{,}000=\$1{,}488{,}000$, which equals expected ultimate losses. Also check the implied ratios: expected reported losses are $0.75\times0.62=0.465$, or 46.5% of premium; expected unpaid losses are $0.25\times0.62=0.155$, or 15.5% of premium. Together they reproduce the selected 62% loss ratio.

Common mistake avoided: applying the 75% reported share to premium rather than to expected ultimate losses would give $\$2{,}400{,}000\times0.75=\$1{,}800{,}000$. That is not expected reported loss; it incorrectly treats the reported share as a loss ratio. The correct calculation applies 75% to the $1,488,000 expected ultimate amount.

Worked Example: Checking a Segmented Exposure Estimate

A hypothetical commercial vehicle portfolio has two segments. Segment A has 6,000 vehicle-years and a selected expected loss cost of $150 per vehicle-year. Segment B has 2,400 vehicle-years and a selected expected loss cost of $255 per vehicle-year. Calculate expected losses by segment and in total, then check the portfolio-level loss cost.

Method: exposure-based estimate with segment aggregation. For Segment A:

6{,}000 \times \$150 = \$900{,}000.

For Segment B:

2{,}400 \times \$255 = \$612{,}000.

The total exposure is $6{,}000+2{,}400=8{,}400$ vehicle-years. The aggregate expected losses are:

\$900{,}000+\$612{,}000=\$1{,}512{,}000.

The implied portfolio-level expected loss cost is $\$1{,}512{,}000/8{,}400=\$180$ per vehicle-year. As a check, the exposure-weighted calculation is $(6{,}000\times\$150+2{,}400\times\$255)/8{,}400=\$180$. Multiplying $8{,}400$ by $\$180$ returns $\$1{,}512{,}000$.

Common mistake avoided: the simple average of the two segment loss costs is $(\$150+\$255)/2=\$202.50$, but that gives the smaller Segment B the same weight as Segment A. Applying it to total exposure would produce $\$1{,}701{,}000$, not the segment-based total. Use exposure weights when deriving an aggregate loss cost.

Worked Example: Detecting a Time-Unit Error

A hypothetical liability portfolio has 21,600 vehicle-months of exposure. Its selected expected loss cost is $480 per vehicle-year. At the evaluation date, the selected emergence pattern indicates that 60% of expected ultimate losses are expected to be reported. Calculate expected ultimate, expected reported, and expected unpaid losses. Then explain the result of a calculation that mistakenly uses vehicle-months as vehicle-years.

Method: convert exposure to the unit used by the loss cost, then apply emergence. There are 12 vehicle-months in one vehicle-year, so the exposure is:

21{,}600 \div 12 = 1{,}800 \text{ vehicle-years}.

Expected ultimate losses are $1{,}800\times\$480=\$864{,}000$. Expected reported losses are $0.60\times\$864{,}000=\$518{,}400$. The expected unpaid share is $1-0.60=0.40$, giving expected unpaid losses of $0.40\times\$864{,}000=\$345{,}600$.

Check the split: $\$518{,}400+\$345{,}600=\$864{,}000$. Also, $ \$518{,}400/\$864{,}000=0.60$, matching the selected reported share.

Common mistake avoided: multiplying 21,600 vehicle-months directly by $480 per vehicle-year gives $10,368,000. That result is 12 times the correct estimate because it treats each vehicle-month as a vehicle-year. Converting the exposure first gives the correct $864,000. If observed reported losses were, for example, $520,000, their difference from the $518,400 expected reported amount would be $1,600; that small difference merits interpretation, not automatic correction of the expected estimate.

Interpreting Diagnostic Results

A diagnostic review is most useful when it separates mechanical failures from judgmental differences. Arithmetic errors, inconsistent units, omitted segments, or totals that fail to reconcile should be corrected or explained. A difference between actual and expected losses may instead reflect random variation, changes in the underlying portfolio, or assumptions that warrant review. The size and direction of the difference help prioritize investigation, but do not by themselves identify its cause.

For magnitude checks, calculate an implied assumption from the result. For a premium-based estimate, divide estimated losses by the premium to recover the implied loss ratio. For an exposure-based estimate, divide by exposure to recover the implied loss cost. Compare those implied values with the selected assumptions. A result many times larger or smaller than expected often points to a unit, decimal, or period error; a modest difference may be consistent with rounding or experience variation.

Period checks matter even when every multiplication is correct. A loss cost stated per vehicle-year cannot be applied directly to vehicle-months. Likewise, the emergence percentage must describe the same evaluation point as the expected reported and unpaid amounts being checked. A correct formula with inputs from different periods does not produce a meaningful estimate.

Finally, keep the scope of each calculation visible. Expected ultimate losses, expected reported losses, and expected unpaid losses are related but not interchangeable quantities. Under an assumed emergence share $q$, expected reported losses equal $q\widehat{U}$, while expected unpaid losses equal $(1-q)\widehat{U}$; these two expected components reconcile to $\widehat{U}$. Actual reported losses need not equal the expected reported amount, so substituting actual experience into one part of that expected split changes what is being measured.

Key takeaway: Recompute from source inputs, confirm compatible units and periods, reconcile segment and emergence components, and recover the implied loss ratio or loss cost. Treat a failed check as a prompt to investigate; do not confuse an actual-versus-expected difference with an arithmetic error.

Check Your Understanding

For each problem, check both the calculation and the stated units or assumptions.

A hypothetical book has earned premium of $1,250,000 and an expected loss ratio of 64%. What is expected ultimate loss? (Answer: $800,000)
A property portfolio has 3,600 insured-location years and an expected loss cost of $125 per location-year. Calculate expected ultimate losses and the implied loss cost if the estimate is $450,000. (Answer: Expected ultimate losses are $450,000; the implied loss cost is $125 per location-year.)
A hypothetical fleet has 14,400 vehicle-months of exposure and a selected loss cost of $310 per vehicle-year. Calculate expected ultimate losses after converting the exposure unit. (Answer: 1,200 vehicle-years and $372,000 expected ultimate losses.)
Segment A has 2,500 exposure units at an expected loss cost of $90; Segment B has 1,500 exposure units at $150. Calculate total expected losses and the exposure-weighted aggregate loss cost. (Answer: $450,000 total expected losses and $112.50 per exposure unit.)
Expected ultimate losses are $2,200,000, and 80% are expected to be reported by the evaluation date. Calculate expected reported and expected unpaid losses, then check their sum. (Answer: $1,760,000 expected reported, $440,000 expected unpaid, and $2,200,000 combined.)

What You'll Learn