Interpreting Count Projections For Claim Liabilities

Why Count Projections Need Interpretation

A claim count projection is useful for more than describing how many claims a cohort may ultimately produce. It can help estimate how many claims are still to be reported, how many may arrive during a future period, and how much claim-handling activity to plan for. But a count is not a dollar estimate: the same number of claims can produce very different unpaid amounts.

The key is to keep three quantities distinct: claims reported by the valuation date, claims projected to be reported eventually, and claims expected to be reported during a particular future interval. Each answers a different question. A projection also depends on the definition of a claim and on the reporting pattern appropriate to the cohort.

Definition: For a defined claim cohort, the reported count is the number of claims from that cohort reported by a stated valuation date. The ultimate count is the total number of claims expected to emerge from the cohort after its reporting is complete. A projected ultimate count estimates that eventual total. The difference between projected ultimate and reported counts is the projected number of claims still to be reported.

A count is meaningful only when the cohort and counting rules are clear. For example, the exposure period, claim type, and treatment of reopened claims should be consistent between the observed count and the reporting pattern. A cumulative reporting proportion describes the fraction of ultimate claims expected to have been reported by a specified age. It is a model input, not a guarantee about the reporting date of any individual claim.

From Reported Counts to Future Claim Activity

Suppose $R$ claims have been reported by a given age and the cumulative reporting proportion at that age is $p$. A basic projection estimates the ultimate count as $R/p$. If cumulative reporting proportions are available at two ages, their difference indicates the share of ultimate claims expected to be reported between those ages. The results are expected counts, so they can be fractional.

Model and assumptions: Treat the ultimate count $N$ for a cohort as fixed but unknown. Each claim has the same probability $p$ of being reported by the valuation age, and its reporting indicator is independent of other claims' indicators. The reporting definition is consistent, and the selected $p$ is appropriate for this cohort. Under this model, the reported count $R$ estimates the ultimate count by $R/p$. If reporting patterns vary materially by claim type, coverage, or operational conditions, one common proportion may not be appropriate.

The reason for the projection is direct. Write the reported count as the sum of $N$ indicators, one for each ultimate claim. Each indicator equals 1 if the claim has been reported by the valuation age, and 0 otherwise. Its expected value is $p$; therefore the expected reported count is $Np$. Dividing by $p$ gives an estimator whose expected value is $N$:

R=\sum_{j=1}^{N} I_j,\qquad E[R\mid N]=\sum_{j=1}^{N} E[I_j\mid N]=Np,\qquad E\!\left[\frac{R}{p}\middle|N\right]=N.

This derivation explains why the projection works under the stated assumptions. It does not show that $R/p$ is the exact number of claims that will emerge; it is an estimate based on an incomplete count and a reporting pattern. The uncertainty in both the observed count and the selected reporting proportion matters when interpreting results.

Interval projection: Let $p_a$ and $p_b$ be cumulative reporting proportions at ages $a$ and $b$, where $b$ is later than $a$. Given projected ultimate count $\widehat N$, the expected reports between those ages are $\widehat N(p_b-p_a)$. The expected count still unreported after age $b$ is $\widehat N(1-p_b)$. These calculations assume the same cohort and a cumulative, nondecreasing reporting pattern.

Worked Example: Separating Reports by Future Interval

A hypothetical equipment-insurance cohort has 63 reported claims at six months. The selected cumulative reporting proportions are 0.72 at six months, 0.90 at 12 months, and 0.96 at 18 months. Estimate the ultimate count, the reports expected in each subsequent interval, and the count expected to remain unreported after 18 months. Use the reporting-proportion projection method and explain why the projected counts need not be integers.

Step 1: Project the ultimate count. At six months, the reported count is 63 and the cumulative reporting proportion is 0.72:

\widehat N=\frac{63}{0.72}=87.5.

The estimate is 87.5 ultimate claims. This is an expected count, not a claim that half a claim exists.

Step 2: Project reports from six to 12 months. The increase in the cumulative proportion is $0.90-0.72=0.18$. Apply that increment to the projected ultimate count:

$$ 87.5(0.18)=15.75. $$

Step 3: Project reports from 12 to 18 months. The proportion increases by $0.96-0.90=0.06$, so the expected count in this interval is:

$$ 87.5(0.06)=5.25. $$

Step 4: Estimate the count still unreported after 18 months. The remaining proportion is $1-0.96=0.04$:

$$ 87.5(0.04)=3.5. $$

Check. The reported count at six months plus the later expected reports and the remaining count sum to the projected ultimate count: $63+15.75+5.25+3.5=87.5$. The method avoids treating the 0.18 interval increment as a fraction of the claims that were unreported at six months; it is a fraction of the ultimate count.

Using Projections to Anticipate Claim Work

Future report estimates can support operational planning, such as anticipating claim intake and staffing needs. They also help distinguish a cohort with many reports still expected from one that is close to fully reported. These interpretations depend on reporting behavior being sufficiently stable. A change in claim handling, notification requirements, or the mix of claims may make a historical pattern less representative.

To project a future interval, use the change in cumulative reporting proportions over that interval. Do not confuse the number still unreported with the number expected to arrive during the next period: some of the unreported claims may arrive later still.

Worked Example: Comparing Two Expected Claim Workloads

Two hypothetical property portfolios are assessed at the same point in their reporting cycles. Portfolio Harbor has 48 reported claims and a current cumulative reporting proportion of 0.60; its proportion is expected to reach 0.75 by the end of next quarter. Portfolio Ridge has 70 reported claims, with a current proportion of 0.875 expected to reach 0.95 by then. For each portfolio, estimate ultimate claims, claims still unreported now, reports expected next quarter, and claims still unreported after next quarter.

Method. For each cohort, divide the current reported count by the current cumulative proportion to estimate ultimate claims. Multiply the increase in cumulative proportion by that ultimate estimate to project next-quarter reports.

Harbor. The projected ultimate count is $48/0.60=80$. The count still unreported now is $80-48=32$. Next quarter's reporting proportion increment is $0.75-0.60=0.15$, so expected next-quarter reports are $80(0.15)=12$. The count expected to remain unreported after next quarter is $80(1-0.75)=20$.

Ridge. The projected ultimate count is $70/0.875=80$. The count still unreported now is $80-70=10$. The next-quarter increment is $0.95-0.875=0.075$, so expected next-quarter reports are $80(0.075)=6$. The count expected to remain unreported after next quarter is $80(1-0.95)=4$.

Combined interpretation and check. The two cohorts have 160 projected ultimate claims and 118 claims already reported. They therefore have 42 claims still unreported in total. The expected next-quarter reports total $12+6=18$, leaving $20+4=24$ expected claims unreported. Since $118+18+24=160$, the projection reconciles. Although both portfolios have the same projected ultimate count, Harbor is expected to generate twice as many reports next quarter. Applying the future interval proportion to the current unreported count instead of the ultimate count would understate the interval projection.

Why Counts Do Not Determine Dollar Liabilities

A claim count describes how many claims are expected, not how much money will be paid on them. Dollar amounts depend on information that a count projection does not contain: for example, the expected remaining payment per claim, payments already made, the status of reported claims, and whether expenses or other components are included in the liability measure. A projected count of claims not yet reported is therefore not itself an estimate of the unpaid liability.

Even if two cohorts have the same projected number of unreported claims, their expected unpaid dollars may differ substantially. In addition, reported claims can still have unpaid amounts. A count projection focused on claims not yet reported says nothing by itself about the remaining payments on claims already reported.

Worked Example: Same Unreported Count, Different Unpaid Dollars

Two hypothetical, fully comparable indemnity estimates are made at the same valuation date. Each portfolio has 30 projected claims not yet reported. For Portfolio Cedar, assume the expected future indemnity payment per not-yet-reported claim is $1,250. For Portfolio Mesa, assume it is $8,000. Cedar also has 12 reported open claims with an expected $900 of indemnity remaining per claim; Mesa has 5 reported open claims with an expected $4,000 remaining per claim. Estimate the unpaid indemnity for these two groups of claims. Ignore expenses and discounting.

Method and assumptions. For each group, multiply the count by the assumed expected remaining payment per claim. This is an expected-value calculation using the supplied mean payments; it does not assume that every individual claim will equal the mean. The result covers only the specified unpaid indemnity, not expenses or any other liability component.

Cedar, claims not yet reported: $30\times \$1{,}250=\$37{,}500$. Cedar, reported open claims: $12\times \$900=\$10{,}800$. Total specified unpaid indemnity is $\$37{,}500+\$10{,}800=\$48{,}300$.

Mesa, claims not yet reported: $30\times \$8{,}000=\$240{,}000$. Mesa, reported open claims: $5\times \$4{,}000=\$20{,}000$. Total specified unpaid indemnity is $\$240{,}000+\$20{,}000=\$260{,}000$.

Check and interpretation. Both portfolios have 30 projected not-yet-reported claims, but their specified unpaid indemnity estimates differ by $\$260{,}000-\$48{,}300=\$211{,}700$. The count alone cannot identify either dollar estimate. The expected payment assumptions and the reported-open-claim information are essential inputs.

Common Interpretation Errors

Calling a projection a known ultimate count. $R/p$ is an estimate under a reporting model; it is not a direct observation of all claims that will emerge.
Using the wrong base for an interval. The interval projection is $\widehat N(p_b-p_a)$, not the increment multiplied by the count currently unreported. The reporting proportion increment is defined on the ultimate-count basis.
Treating expected counts as literal claims. A result such as 15.75 represents an expectation across possible outcomes. Operational plans may need an integer, but rounding is a separate decision.
Equating an unreported count with a dollar liability. Counts do not supply payment amounts, and they do not account for remaining payments on reported open claims.
Applying a pattern without checking its relevance. A reporting proportion from a different claim definition, period, or operating environment may not describe the cohort being projected.

Key takeaway: Use reported counts and cumulative reporting proportions to estimate ultimate counts, future reports, and claims still to be reported. Treat these as count projections. A dollar liability requires additional payment information, including amounts expected on both unreported claims and reported claims that remain open.

Check Your Understanding

For each problem, distinguish the projected ultimate count, future reports, and dollar amounts where relevant.

A hypothetical liability cohort has 36 reported claims at a reporting age when the cumulative proportion is 0.60. Estimate the ultimate count and the number still unreported. (Answer: 60 ultimate claims; 24 still unreported.)
A cohort has 100 reported claims at a cumulative reporting proportion of 0.80. The proportion is expected to reach 0.92 in three months. Estimate the ultimate count, reports expected in those three months, and claims still unreported afterward. (Answer: 125 ultimate claims; 15 reports in three months; 10 still unreported.)
A hypothetical cohort has 45 reported claims at a current cumulative reporting proportion of 0.75, which is expected to reach 0.90 next quarter. Estimate the ultimate count, next-quarter reports, and claims still unreported after next quarter. (Answer: 60 ultimate claims; 9 next-quarter reports; 6 still unreported.)
A projected 25 claims not yet reported each have an expected future indemnity payment of $2,400. There are also 8 reported open claims with expected remaining indemnity of $700 each. Ignoring expenses and discounting, what is the specified unpaid indemnity estimate? (Answer: $65,600.)
Two cohorts each have a projected ultimate count of 120. Cohort A has current and next-quarter cumulative reporting proportions of 0.75 and 0.85; Cohort B has proportions of 0.90 and 0.95. For each cohort, estimate reports expected next quarter and claims still unreported afterward. (Answer: Cohort A: 12 next-quarter reports and 18 still unreported; Cohort B: 6 next-quarter reports and 6 still unreported.)

What You'll Learn