Selecting a Standard Case Reserve Adequacy Pattern

Introduction

A case-reserve adequacy pattern describes case outstanding reserves relative to paid losses at successive development ages. Selecting the pattern is not simply a matter of choosing the latest period or the smoothest-looking ratios. The candidate data must reflect a reasonably stable reserving process, contain enough credible experience, and represent the claims to which the pattern will be applied.

The selection is a foundation for Berquist-Sherman analysis: a reference pattern must be chosen before comparisons can be made against it. This tutorial focuses on that selection. It does not apply the selected pattern to adjust historical case reserves.

Definition: For cohort $i$ observed at development age $j$, let $P_{ij}$ be cumulative paid losses and $C_{ij}$ be case outstanding reserves, measured at the same valuation date and on the same claim, currency, and gross-or-net basis. When $P_{ij}>0$, the cohort’s case-reserve-to-paid ratio is $$ r_{ij}=\frac{C_{ij}}{P_{ij}}. $$ A case-reserve adequacy pattern is a selected sequence of these ratios, or of consistently pooled ratios, across development ages. It is a reference description of reserves relative to paid losses, not a measure of ultimate losses by itself.

What Makes a Candidate Pattern Suitable?

A candidate is typically a set of accident or report cohorts observed at comparable development ages. The selection should consider both the numerical pattern and the conditions that produced it. In particular, assess whether the cohorts experienced the same case-reserving rules, claim-handling approach, and operational environment. A documented change in reserve authority, reserving guidelines, or case review practice can make periods on opposite sides of the change unsuitable for direct combination.

Also assess credibility and claim mix. A ratio based on substantial paid losses and many claims is generally less sensitive to a few large claims than one based on a small denominator. But volume alone does not establish suitability: a large data set can still represent a different claim population, or span materially different reserving practices. Review claim counts, paid-loss volume, reserve concentration, and relevant mix indicators together. These are diagnostics, not a universal formula that automatically determines the winning period.

Selection consideration	Question to ask	Possible warning sign
Operational stability	Were reserving practices reasonably consistent throughout the candidate period?	A documented change in case-reserve rules or handling approach
Data credibility	Are there enough claims and paid losses to make the ratios reasonably stable?	A small denominator or a ratio dominated by a few claims
Claim mix	Does the candidate represent the claims relevant to the intended use?	Material changes in severity, coverage, territory, or claim type
Comparability	Are paid losses and case reserves measured consistently at each age?	Different valuation, currency, or gross/net bases
Pattern behavior	Are age-to-age differences plausible and explainable?	A sharp isolated jump without a supported operational or mix explanation

A stable reference period need not be the most recent period. If recent experience includes a process change, the latest ratios may describe a new reserving regime rather than a stable basis. Conversely, an older period may be poorly matched to the current claim population. Selection requires weighing these considerations and documenting why the chosen period is appropriate.

Pooling Ratios Correctly

For a group of cohorts at the same development age, one common summary is the ratio of total case reserves to total paid losses. When every included cohort has positive paid losses, this pooled ratio is exactly a paid-loss-weighted mean of the individual ratios.

R_j = \frac{\sum_i C_{ij}}{\sum_i P_{ij}} = \sum_i \left( \frac{P_{ij}}{\sum_k P_{kj}} \right)r_{ij}, \qquad P_{ij}>0\ \text{for every included } i.

The condition on every included cohort matters. If an included cohort has zero paid losses, its ratio $C_{ij}/P_{ij}$ is undefined. If that cohort has positive case reserves, those reserves still affect the ratio of totals, but they cannot be represented as a weighted contribution from a defined cohort ratio.

Derivation: Assume that each included cohort at age $j$ has $P_{ij}>0$, and define $r_{ij}=C_{ij}/P_{ij}$. Then $C_{ij}=P_{ij}r_{ij}$, so $$ \frac{\sum_i C_{ij}}{\sum_i P_{ij}} = \frac{\sum_i P_{ij}r_{ij}}{\sum_i P_{ij}} = \sum_i\left(\frac{P_{ij}}{\sum_k P_{kj}}\right)r_{ij}. $$ Each weight is nonnegative and the weights sum to one. Thus the pooled ratio is a paid-loss-weighted mean of the defined cohort ratios. If any included $P_{ij}=0$, this substitution and weighted-mean representation do not apply to that cohort.

The pooled ratio and the unweighted average answer different questions. The pooled ratio gives greater influence to cohorts with more paid losses. The unweighted average gives each cohort the same influence regardless of its paid volume. Neither summary resolves a claim-mix or operational comparability problem; pooling only describes how the selected data combine.

Worked Example: Comparing Two Operational Periods

A hypothetical insurer is evaluating two candidate periods for a stable reference pattern. The first period had consistent case-review guidance. The second followed a documented change in reserve authority. At age 12 months, the first period has three cohorts with $(P,C)$ values of $(400,80)$, $(600,150)$, and $(500,115)$, in thousands of dollars. At age 24 months, the values are $(900,180)$, $(1{,}000,140)$, and $(800,120)$. The second period has two cohorts at each age: at 12 months, $(700,280)$ and $(800,320)$; at 24 months, $(1{,}300,260)$ and $(1{,}400,280)$. Compare the pooled patterns and make a selection.

Method. Calculate the ratio of total case reserves to total paid losses separately for each period and age. All paid amounts are positive, so the paid-weighted representation is valid.

First period, age 12 months. Total paid losses are $400+600+500=1{,}500$. Total case reserves are $80+150+115=345$. The ratio is $345/1{,}500=0.230$.

First period, age 24 months. Total paid losses are $900+1{,}000+800=2{,}700$. Total case reserves are $180+140+120=440$. The ratio is $440/2{,}700=0.1629629\ldots$, or $0.163$ to three decimals.

Second period, age 12 months. Total paid losses are $700+800=1{,}500$, and total case reserves are $280+320=600$. The ratio is $600/1{,}500=0.400$.

Second period, age 24 months. Total paid losses are $1{,}300+1{,}400=2{,}700$, and total case reserves are $260+280=540$. The ratio is $540/2{,}700=0.200$.

The first candidate pattern is therefore $0.230$ at 12 months and $0.163$ at 24 months; the second is $0.400$ and $0.200$. The later period’s higher ratios are consistent with the documented change in reserve authority, so averaging the two periods would blur two operating regimes. If the intended reference is the stable pre-change regime and its claim mix is appropriate, the first period is the more defensible candidate. The selection should still be checked against claim counts, reserve concentration, and mix.

Common mistake avoided: Do not select the second period simply because it is more recent, or average its ratios with the first period without addressing the practice change. The arithmetic would be possible, but the resulting pattern would not represent either consistent regime.

Assessing Claim Mix and Credibility

Aggregate ratios can change even when the ratios within claim segments do not. For example, a population with a larger share of high-severity claims may have a higher overall case-reserve-to-paid ratio than a population with a larger share of low-severity claims. The difference may reflect composition rather than a change in reserve adequacy. Relevant segments depend on the portfolio; possible examples include coverage, claim type, severity group, or territory.

A useful diagnostic is to compare candidate periods within important segments, as well as in aggregate. If segment ratios are stable but aggregate ratios differ, mix is a plausible explanation. If ratios differ within segments and the change coincides with a documented operational shift, practice is a plausible explanation. These checks do not prove causation, but they make the selection more defensible.

Worked Example: Separating Mix From Within-Segment Results

Two hypothetical periods have low-severity and high-severity segments. In Period A, low-severity claims have paid losses of $800$ and case reserves of $160$, while high-severity claims have paid losses of $200$ and case reserves of $80$, all in thousands of dollars. In Period B, the corresponding values are $400$ and $80$ for low severity, and $600$ and $240$ for high severity. Calculate the aggregate ratios and assess whether the difference alone establishes a change in adequacy.

Period A. The low-severity ratio is $160/800=0.200$, and the high-severity ratio is $80/200=0.400$. Total paid losses are $800+200=1{,}000$; total case reserves are $160+80=240$. The aggregate ratio is $240/1{,}000=0.240$.

Period B. The low-severity ratio is $80/400=0.200$, and the high-severity ratio is $240/600=0.400$. Total paid losses are $400+600=1{,}000$; total case reserves are $80+240=320$. The aggregate ratio is $320/1{,}000=0.320$.

Within each segment, the ratio is unchanged. The high-severity share of paid losses rose from $200/1{,}000=0.20$ in Period A to $600/1{,}000=0.60$ in Period B. Thus the aggregate ratio increased from $0.240$ to $0.320$ alongside a substantial change in mix. The aggregate difference alone does not establish that case reserves became less adequate. A selection should examine segment-level consistency and determine whether the candidate represents the intended claim mix.

Common mistake avoided: Do not interpret an aggregate ratio change automatically as a reserving-practice change. Here the segment ratios are identical, while the composition of paid losses differs.

Zero-Paid Cohorts and Denominator Sensitivity

A cohort with $P_{ij}=0$ and $C_{ij}=0$ contributes nothing to either pooled total and can be omitted from that calculation. A cohort with $P_{ij}=0$ and $C_{ij}>0$, however, contributes case reserves to the numerator but nothing to the paid denominator. The ratio of totals can still be calculated if total paid losses are positive, but it is not a paid-weighted mean of all cohort ratios. Report this condition and investigate whether the cohort belongs in the intended comparison.

Even with positive denominators, a small amount of paid loss can make a cohort ratio highly sensitive to a modest reserve change. Inspect the underlying values before relying on an extreme ratio. A high ratio may be a genuine signal, a small-denominator effect, an unusual claim, or a data issue; the ratio alone cannot distinguish among these explanations.

Worked Example: A Zero-Paid Cohort Changes the Pooled Ratio

At a given age, a candidate period contains three cohorts with positive paid losses: $(P,C)=(1{,}000,200)$, $(800,160)$, and $(200,40)$, in thousands of dollars. It also contains a cohort with $P=0$ and $C=30$. Calculate both the ratio using all cohorts’ totals and the paid-weighted mean for the positive-paid cohorts. Explain the difference.

Positive-paid cohorts. Their total paid losses are $1{,}000+800+200=2{,}000$. Their total case reserves are $200+160+40=400$. The ratio for this subset is $400/2{,}000=0.200$. Each individual positive-paid ratio is also $0.200$, so their paid-weighted mean is $0.200$.

All cohorts’ totals. Including the zero-paid cohort adds $30$ to case reserves and $0$ to paid losses. The totals are therefore $P=2{,}000$ and $C=430$, giving $430/2{,}000=0.215$.

The all-cohort ratio is $0.215$, while the positive-paid weighted mean is $0.200$. The zero-paid cohort has no defined $C/P$ ratio, so it cannot be inserted into the weighted-mean identity. If the intended measure is the ratio of all case reserves to all paid losses, retain and disclose the cohort. If the intended comparison is a paid-weighted mean of defined cohort ratios, report the positive-paid result separately and explain the exclusion. Apply the same rule consistently across candidates.

Common mistake avoided: Do not call $0.215$ the paid-weighted mean of the four cohort ratios. The fourth ratio would be $30/0$, which is undefined.

A Practical Selection Process

The following sequence helps keep the selection focused on comparability rather than on a single attractive statistic.

Fix the measurement basis.
Confirm development ages, valuation dates, claim population, currency, and gross-or-net treatment. Use consistent definitions of paid losses and case outstanding reserves.

Identify operational regimes.
Use documented changes in guidelines, authority, or claim handling to define periods that can reasonably be evaluated separately.

Calculate age-specific summaries.
For each candidate and age, calculate the ratio of totals. Verify positive denominators, and separately identify zero-paid cohorts with positive reserves.

Test credibility and mix.
Review claim counts, paid volume, concentration, and important claim segments. Do not treat volume as a substitute for representativeness.

Select and document.
Choose the stable, sufficiently supported candidate that best represents the intended use. State the period, rationale, exclusions, and known limitations.

The pattern should be judged across development ages, not just at one age. A candidate need not be perfectly smooth: genuine claim dynamics can produce changing ratios. But an abrupt or isolated movement deserves explanation before it is accepted as a reference feature. A simple average across ages or periods can conceal such a movement rather than resolve it.

Common Selection Traps

Choosing the newest period automatically. Recency does not outweigh a documented process change or a poor match to the intended claims.
Equating a large data set with a suitable pattern. High volume supports credibility but does not correct a different claim mix or inconsistent reserving practices.
Confusing a pooled ratio with an unweighted average. The ratio of totals weights cohorts in proportion to paid losses, provided every included cohort has positive paid losses.
Ignoring zero-paid cohorts. A positive reserve with no paid loss changes the ratio of totals but cannot be expressed as a finite cohort ratio or included in the paid-weighted identity.
Comparing inconsistent data. Differences in valuation dates, claim definitions, or gross/net basis can create apparent pattern changes that are not comparable.

Key takeaway: Select a reference case-reserve adequacy pattern by combining age-specific calculations with evidence about operating practices, data credibility, and claim mix. A ratio is meaningful for selection only when its data basis and pooling method are clear.

Check Your Understanding

For each problem, calculate or assess the candidate pattern before checking the answer.

At one age, two cohorts have $(P,C)=(300,60)$ and $(700,140)$, in thousands of dollars. Calculate the pooled ratio and state whether it equals the paid-weighted mean. (Answer: $200/1{,}000=0.200$; yes, both cohort ratios are $0.200$.)
At one age, cohort A has $P=400,C=100$, and cohort B has $P=600,C=120$, in thousands. Calculate the pooled ratio and the unweighted average of the cohort ratios. (Answer: pooled $220/1{,}000=0.220$; unweighted average $(0.250+0.200)/2=0.225$.)
Two periods have identical low-severity ratios of $0.15$ and high-severity ratios of $0.35$, but the high-severity share of paid losses rises from $20\%$ to $60\%$. Calculate each aggregate ratio using the paid-loss shares as weights. (Answer: first $0.8(0.15)+0.2(0.35)=0.190$; second $0.4(0.15)+0.6(0.35)=0.270$.)
At one age, positive-paid cohorts total $P=1{,}500,C=300$, in thousands. A separate cohort has $P=0,C=45$. Calculate the all-cohort ratio and explain why it is not the paid-weighted mean of all cohort ratios. (Answer: $345/1{,}500=0.230$; the zero-paid cohort has an undefined reserve-to-paid ratio and cannot enter the weighted-mean identity.)
A recent candidate has more claims than an older candidate, but it spans a documented reserve-guideline change and has a substantially different coverage mix. The older candidate has stable practices and a mix closer to the intended use. Which should be selected without further evidence? (Answer: neither solely on these facts; investigate and quantify credibility and mix, and do not pool across the change automatically. The older candidate is a stronger initial reference candidate if its volume is adequate, but its suitability still requires support.)

What You'll Learn