Projecting Ultimate Claim Counts From Reporting Patterns

Why Reporting Patterns Matter

An accident period’s reported claim count is usually incomplete while claims are still emerging. The count observed at a valuation date therefore reflects both the period’s ultimate claim volume and how much time has elapsed for claims to be reported. A historical reporting pattern helps separate those two effects: it estimates what fraction of the eventual claims would typically have been reported by a given age.

This tutorial focuses on projecting claim counts, not claim dollars. A count reporting pattern concerns when claims are first reported. It is not a paid-loss pattern, a closure pattern, or a measure of how much of a claim’s cost has been paid. The method is aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Definition: For an accident period observed at development age \(t\), let \(R(t)\) be its cumulative number of claims reported by age \(t\), and let \(U\) be its ultimate claim count after the reporting process is substantially complete. The cumulative reporting proportion at age \(t\) is \(p(t)=R(t)/U\), when \(U\) is known. It represents the fraction of ultimate claims reported by that age. The corresponding count development factor is \(1/p(t)\).

Estimating the Historical Reporting Proportion

A historical accident period can be used to estimate a reporting proportion at age \(t\) if its ultimate count is sufficiently mature and can be estimated reliably. For several comparable historical periods, a standard pooled estimate is the total reported count at age \(t\) divided by the total ultimate count:

\widehat{p}(t) = \frac{\sum_{i=1}^{n} R_i(t)} {\sum_{i=1}^{n} U_i}

Here, \(R_i(t)\) is the cumulative count reported for historical period \(i\) by age \(t\), and \(U_i\) is that period’s ultimate count. Pooling this way weights each period in proportion to its ultimate count. It is generally different from taking an unweighted average of the individual ratios \(R_i(t)/U_i\).

The estimated proportion should be between zero and one. When the same claims are counted cumulatively and the claim definition is unchanged, the proportion should not decrease as development age increases: a claim reported by an earlier age remains part of the later cumulative count. A decrease in a constructed pattern can signal inconsistent data, a changed claim definition, or a calculation error.

From a Reported Count to an Ultimate Count

Suppose an accident period has \(R(t)\) claims reported by age \(t\), and a suitable historical pattern gives an estimated reporting proportion \(\widehat{p}(t)\). The basic projection divides the observed count by that proportion. The estimated number not yet reported is the estimated ultimate count less the observed count.

\widehat{U}=\frac{R(t)}{\widehat{p}(t)} \qquad\text{and}\qquad \widehat{U}_{\text{unreported}} =\widehat{U}-R(t) =R(t)\left(\frac{1-\widehat{p}(t)}{\widehat{p}(t)}\right)

The estimated ultimate count need not be an integer. It is an estimate, not a claim-by-claim prediction that a fractional claim will occur. Keep the observed reported count distinct from the estimated count still to be reported.

Why the Projection Formula Works

A simple probability model gives the rationale. Assume the ultimate count \(U\) is fixed and each ultimate claim has the same probability \(p(t)\) of having been reported by age \(t\), independently of the reporting status of other claims. Then \(R(t)\), conditional on \(U\), has a binomial distribution with \(U\) trials and probability \(p(t)\). In particular:

E[R(t)\mid U]=U p(t)

The observed count is therefore expected to be the ultimate count multiplied by the reporting proportion. Solving this relationship for \(U\) gives \(U=E[R(t)\mid U]/p(t)\). Replacing the expected reported count with the actual reported count and the unknown proportion with its historical estimate yields \(\widehat{U}=R(t)/\widehat{p}(t)\). If \(p(t)\) were known, \(R(t)/p(t)\) would be unbiased for fixed \(U\) under this model, because \(E[R(t)/p(t)\mid U]=U\). In practice, the reporting proportion is estimated too, so the formula is a projection whose quality depends on the adequacy of the historical pattern.

The equal-probability model is a justification, not a guarantee that all claims really report independently at the same rate. If reporting speed varies substantially by claim type or operating condition, a single pooled proportion may be inappropriate. Segment-specific patterns or a different model may be needed.

Worked Example: Pooling Mature Cohorts at One Development Age

A hypothetical insurer has three sufficiently mature accident periods. At 12 months, they had 84, 105, and 91 claims reported, respectively; their estimated ultimate counts are 120, 150, and 130. A newer accident period has 98 claims reported at 12 months. Estimate its ultimate count and the number still unreported. Use a pooled historical proportion.

Method: pooled cumulative reporting proportion. First total the historical reported counts at the matching age:

$$ 84+105+91=280 $$

Then total the historical ultimate counts:

$$ 120+150+130=400 $$

The estimated 12-month reporting proportion is therefore:

\widehat{p}(12)=\frac{280}{400}=0.70

Apply the proportion to the newer period’s 98 reported claims:

\widehat{U}=\frac{98}{0.70}=140

Subtract the claims already reported:

\widehat{U}_{\text{unreported}}=140-98=42

Check. If 70% of 140 ultimate claims are reported by 12 months, the expected reported count is \(140(0.70)=98\), matching the observation. The projection is 140 ultimate claims, of which 42 are estimated to remain unreported. The common mistake avoided is averaging the three cohort proportions without regard to cohort size; here, pooled counts produce the stated 70% directly.

Selecting and Applying a Development Age

The historical proportion must correspond to the same development age as the current reported count. A 12-month reported count cannot be divided by a 6-month reporting proportion. Development age is measured consistently from the chosen origin, commonly the accident period, to the valuation date. Use cumulative reported counts at that age, rather than incremental counts from only the latest interval.

For an accident period with counts available at multiple ages, a more developed count will often provide a more stable projection because it uses more observed information. However, comparisons across ages are only meaningful when the historical proportions were constructed on consistent definitions and the current counts are cumulative. The following example calculates proportions at several ages and applies the one matching the current count.

Worked Example: Choosing the Matching Age

Three hypothetical mature accident periods have ultimate counts of 200, 250, and 150. Their cumulative reported counts at 6 months are 100, 125, and 75; at 12 months they are 150, 200, and 120; and at 24 months they are 190, 240, and 147. A current period has 188 claims reported at 12 months. Estimate the ultimate and unreported counts.

Method: construct cumulative proportions by age, then use the 12-month proportion. The total historical ultimate count is:

$$ 200+250+150=600 $$

At 6 months, the historical reported total is \(100+125+75=300\), so:

\widehat{p}(6)=\frac{300}{600}=0.5000

At 12 months, the reported total is \(150+200+120=470\), giving:

\widehat{p}(12)=\frac{470}{600}=0.7833\overline{3}

At 24 months, the reported total is \(190+240+147=577\), giving:

\widehat{p}(24)=\frac{577}{600}=0.9616\overline{6}

The proportions rise with development age, as expected for cumulative counts. Since the current count is measured at 12 months, use \(\widehat{p}(12)=470/600=47/60\):

\widehat{U} =\frac{188}{47/60} =\frac{188(60)}{47} =4(60) =240

Thus:

\widehat{U}_{\text{unreported}}=240-188=52

Check. The 12-month proportion applied to 240 ultimate claims gives \(240(47/60)=188\) reported claims. The 24-month proportion is not appropriate merely because it is closer to 100%; it describes a different age. Dividing 188 by the 24-month proportion would mix a 12-month observation with a 24-month pattern.

When Reporting Patterns Differ by Segment

A pooled pattern is useful only when historical and current business have sufficiently similar reporting behavior. If one segment tends to report promptly and another reports more slowly, changes in the relative sizes of those segments can make an overall historical proportion misleading. Apply segment-specific proportions when reliable patterns are available, then sum the segment estimates.

Worked Example: Projecting With Different Segment Patterns

A hypothetical portfolio has two claim segments. Historical mature data support a reporting proportion of 0.80 at the selected age for Segment A and 0.50 for Segment B. The current accident period has 120 claims reported in Segment A and 60 in Segment B. Estimate the portfolio ultimate and unreported counts. Assume the segment-specific historical patterns remain applicable.

Method: project separately by segment. For Segment A:

\widehat{U}_A=\frac{120}{0.80}=150, \qquad \widehat{U}_{A,\text{unreported}}=150-120=30

For Segment B:

\widehat{U}_B=\frac{60}{0.50}=120, \qquad \widehat{U}_{B,\text{unreported}}=120-60=60

Add the segment results:

\widehat{U}=150+120=270, \qquad \widehat{U}_{\text{unreported}}=30+60=90

The two reported segment counts sum to \(120+60=180\), and \(270-180=90\), which checks the unreported total. A common shortcut would be to take the unweighted average proportion, \((0.80+0.50)/2=0.65\), and project \(180/0.65=276.9231\). That calculation treats the current reported count as though it had a single 65% reporting proportion, despite its unequal segment composition. It does not reproduce the segment-specific estimate of 270. The appropriate method is to use each segment’s pattern when the segment mix or reporting behavior makes one aggregate pattern unsuitable.

Assumptions, Limitations, and Common Traps

A count projection from reporting proportions depends on the data and pattern being comparable. In practice, assess whether the historical periods and the current period use the same definition of a claim, the same accident-period basis, and consistent rules for recording reports. Material operational changes, claim-system changes, or shifts in the types of claims can alter reporting speed. The historical proportion should also be based on sufficiently mature ultimate counts; if those “ultimate” counts are still incomplete, the estimated proportion may be biased.

The formula is especially sensitive when \(\widehat{p}(t)\) is small. A small change in the proportion can then create a large change in \(\widehat{U}\). Recent periods observed at very young ages may therefore produce unstable projections. The estimate also does not describe the uncertainty of the ultimate count by itself; it is a point projection conditional on the selected pattern.

Mixing ages: Match the reporting proportion to the age of the observed cumulative count.
Using incremental counts: Convert interval counts to cumulative reported counts before applying a cumulative proportion.
Averaging cohort ratios blindly: A simple average gives small and large cohorts equal weight; pooled counts weight them by their ultimate counts.
Ignoring changing composition: If segments have different reporting speeds, an aggregate proportion can shift with the mix.
Confusing reported and unreported: \(R(t)/\widehat{p}(t)\) estimates the ultimate count; subtract \(R(t)\) to estimate the unreported count.

Key takeaway: Estimate a cumulative reporting proportion from comparable, mature experience; apply it at the matching development age; and calculate the unreported count as projected ultimate less the count already reported. Segment the projection when reporting patterns materially differ.

Check Your Understanding

For each problem, use the stated reporting proportion or derive it from the historical counts. Treat the patterns as applicable to the current period.

At a given age, a current accident period has 72 reported claims and an estimated reporting proportion of 0.60. Calculate the ultimate and unreported counts. (Answer: ultimate 120; unreported 48.)
Two mature historical periods have ultimate counts of 240 and 360, with 144 and 276 claims reported at 12 months. A current period has 105 claims reported at 12 months. Use the pooled proportion to estimate its ultimate and unreported counts. (Answer: pooled proportion 0.70; ultimate 150; unreported 45.)
At the matching age, Segment A has proportion 0.90 and 90 reported claims; Segment B has proportion 0.60 and 72 reported claims. Estimate the combined ultimate and unreported counts using segment-specific projections. (Answer: ultimate 220; unreported 58.)
A historical pattern has cumulative proportions of 0.40 at 6 months, 0.70 at 12 months, and 0.90 at 24 months. A current period has 154 claims reported at 12 months. Estimate ultimate and unreported counts. (Answer: ultimate 220; unreported 66.)
Two mature cohorts have ultimate counts of 100 and 300, and reported counts of 50 and 240 at a selected age. A current period has 145 claims reported at that age. Calculate the pooled historical proportion and the projected ultimate and unreported counts. (Answer: pooled proportion 0.725; ultimate 200; unreported 55.)

What You'll Learn