Why Detecting Practice Changes Requires More Than a Triangle
Case reserves are estimates of the unpaid amounts on reported claims. If adjusters become more or less conservative, the case reserves recorded at a given claim age may change even when the underlying claims do not. That change can alter incurred-loss development patterns and, in turn, affect reserve analyses. The analytical task here is to identify evidence of a change in reserving practice—not to assume every movement in a case reserve ratio has that cause.
A useful first screen compares accident-year cohorts at the same development age. A second screen examines claim-level histories: whether reserve changes are broad across comparable claims, and whether they coincide with new information, payments, settlements, or a calendar-time shift in review practice. Neither screen proves a cause by itself. Claim severity, claim mix, settlement behavior, data handling, and random variation can all move aggregate ratios.
The examples use hypothetical nominal dollar amounts as of the stated valuation dates. No interest discounting or mortality model is used; the quantities are observed claim payments and case reserves, not present values or outputs from a probability distribution.
Ratios and Their Interpretation
The ratios are age-specific summaries. For example, \(R_{y,12}\) describes case outstanding relative to cumulative paid at 12 months of development; it is not the same quantity as \(R_{y,24}\). Comparisons intended to detect a practice shift should therefore align development age and use consistent definitions of paid and case outstanding. If a cell has no paid loss, \(R_{y,k}\) and \(Q_{y,k}\) are undefined. If paid loss is very small, the ratios can be unstable and should not be treated as precise signals.
The incurred-to-paid ratio is directly related to the case outstanding-to-paid ratio:
This identity is useful as a consistency check. If both ratios are calculated from the same paid and case amounts, their difference must be 1. A failure of that check points to inconsistent definitions, units, or data—not a subtle actuarial interpretation.
Worked Example: Comparing Cohorts at the Same Age
A hypothetical insurer summarizes three accident years at 12 months of development. Amounts are in thousands of dollars. For accident years 2022, 2023, and 2024, respectively, paid losses are 600, 700, and 800; case outstanding is 480, 560, and 360. Calculate both ratios for each year and state what the pattern suggests.
Method and assumptions. Use cumulative paid and case outstanding on the same loss basis, at the same 12-month age. Treat the supplied amounts as comparable nominal amounts; do not apply an interest adjustment. This ratio screen alone does not establish why reserves differ.
Accident year 2022. \(R_{2022,12}=480/600=0.800\). Incurred is \(600+480=1{,}080\), so \(Q_{2022,12}=1{,}080/600=1.800\). The identity check gives \(1+0.800=1.800\).
Accident year 2023. \(R_{2023,12}=560/700=0.800\). Incurred is \(700+560=1{,}260\), so \(Q_{2023,12}=1{,}260/700=1.800\). Again, \(1+R=1.800\).
Accident year 2024. \(R_{2024,12}=360/800=0.450\). Incurred is \(800+360=1{,}160\), so \(Q_{2024,12}=1{,}160/800=1.450\). The identity check gives \(1+0.450=1.450\).
The newest cohort has a materially lower case outstanding relative to paid at the same age than the two earlier cohorts. That is a signal to investigate possible changes in case-reserving practice. It is not a conclusion: changes in claim mix, claim severity, payment timing, or data treatment could also explain the difference. The common mistake is to call the lower ratio proof of less adequate reserves. The correction is to treat it as a diagnostic and seek corroborating evidence.
What Makes a Ratio Comparison Informative?
A ratio comparison is most informative when the cohorts are comparable in development age, coverage, limits, claim status rules, and the definition of incurred loss. Check that the case outstanding amount has not changed definition—for example, by a change in whether particular expense amounts are included. Also examine the underlying paid amounts: a large increase in paid loss can reduce \(C/P\) even if case outstanding is unchanged.
The ratio is an aggregate, so it is sensitive to the composition of the claims contributing to it. It does not say whether every claim has a smaller reserve, or whether a few large claims dominate the difference. If possible, compare meaningful claim segments and review individual claims. A pattern that appears in several comparable segments is more persuasive than a shift confined to one segment, but still requires interpretation.
A derivation makes the weighting issue clear. If a cohort is split into segments \(s\), then its aggregate ratio is
Thus the aggregate ratio is a paid-weighted average of segment ratios, not generally their simple average. It can change when the paid-weighted mix changes even if every segment’s own ratio stays constant. The equation assumes positive paid amounts in the segments used; a segment with zero paid cannot have a defined \(C_s/P_s\) ratio and must be handled separately.
Worked Example: Separating a Mix Shift from a Practice Signal
At a common development age, a hypothetical insurer has two claim segments. Segment A has case outstanding equal to 0.50 times paid loss; segment B has case outstanding equal to 1.50 times paid loss. In an earlier cohort, paid loss is 800 in A and 200 in B. In a later cohort, it is 200 in A and 800 in B. All amounts are in thousands. Calculate each aggregate ratio and decide whether the aggregate change alone demonstrates a practice change.
Method and assumptions. The segment ratios are assumed unchanged across the cohorts, and paid and case amounts use consistent definitions. Calculate total case outstanding divided by total paid; no interest adjustment or probability model is involved.
Earlier cohort. Segment A case outstanding is \(0.50(800)=400\). Segment B case outstanding is \(1.50(200)=300\). Total paid is \(800+200=1{,}000\), and total case outstanding is \(400+300=700\). Therefore, the aggregate ratio is \(700/1{,}000=0.700\).
Later cohort. Segment A case outstanding is \(0.50(200)=100\). Segment B case outstanding is \(1.50(800)=1{,}200\). Total paid is \(200+800=1{,}000\), and total case outstanding is \(100+1{,}200=1{,}300\). Therefore, the aggregate ratio is \(1{,}300/1{,}000=1.300\).
The aggregate ratio increased from 0.700 to 1.300, although neither segment ratio changed. The paid mix shifted toward segment B, which has the higher ratio. The earlier incurred-to-paid ratio is \(1+0.700=1.700\); the later one is \(1+1.300=2.300\).
Common mistake. Taking the unweighted average of the segment ratios gives \((0.50+1.50)/2=1.000\) in both cohorts and misses the actual aggregate ratios. For the earlier cohort, the paid weights are \(800/1{,}000=0.80\) and \(200/1{,}000=0.20\), giving \(0.80(0.50)+0.20(1.50)=0.700\). For the later cohort, the weights are 0.20 and 0.80, giving \(0.20(0.50)+0.80(1.50)=1.300\). The aggregate movement is explained by mix in this example, so it is not evidence by itself of changed reserving practice.
Using Claim-Level Reserve Histories
Aggregate ratios identify where to investigate; claim-level histories help explain what happened. For each claim, the history should preserve valuation dates, cumulative paid amounts, case outstanding amounts, claim status, and—where available—the timing and reason for material reserve changes. Reviewing only the latest reserve value discards the sequence needed to distinguish a broad calendar-time change from ordinary claim-by-claim development.
A practical review compares claims that are reasonably similar in development age, coverage, severity, and status. Ask whether reserve reductions or increases cluster around a common calendar date, whether they affect many comparable claims, and whether recorded claim events explain them. A single large claim can move a segment ratio substantially; a repeated directional change across many claims is more consistent with a broad practice signal, though it is still not proof of cause.
Recalculate the ratios from the underlying paid and case amounts and check \(Q=1+R\). Verify that the comparison uses the same development age and data definitions.
Break the comparison into relevant segments and identify whether the movement is broad or driven by a small number of claims or a changing mix.
For comparable claims, inspect the timing and size of reserve changes alongside payments, settlements, new information, and status changes.
Describe whether the evidence is consistent with a practice change, ordinary development variation, a mix shift, or an unresolved combination. Do not treat a ratio as causal evidence on its own.
Worked Example: Interpreting a Cluster of Claim-Level Changes
A hypothetical insurer reviews five comparable claims that remain open at two successive valuation dates. At the first date, total cumulative paid is 200 and total case outstanding is 300. At the second date, three months later, cumulative paid is 220 and case outstanding is 230. The claim histories show that every claim’s case reserve decreased; the file records show no new medical or liability information during the interval. Calculate the ratios and assess the evidence.
Method and assumptions. Use the stated aggregate amounts in the same units and definitions at each date. The claims remain open, but they have aged three months; therefore, the comparison is not a same-development-age cohort comparison. The absence of recorded new information is part of the hypothetical facts, not proof that no other cause exists.
First valuation. Incurred is \(200+300=500\). Thus \(R=300/200=1.500\) and \(Q=500/200=2.500\).
Second valuation. Incurred is \(220+230=450\). Thus \(R=230/220=1.04545\ldots\), or 1.045 to three decimals, and \(Q=450/220=2.04545\ldots\), or 2.045 to three decimals. The identity check gives \(1+1.04545\ldots=2.04545\ldots\).
Paid increased by \(220-200=20\), while case outstanding decreased by \(300-230=70\). Total incurred consequently decreased by \(450-500=-50\), which also equals the net change \(20-70=-50\). The across-claim direction, combined with the absence of recorded new information, warrants investigation of a common review or reserving change. However, three months of development have passed, and a reserve reduction may reflect ordinary claim evaluation or information not captured in the summary. The evidence supports a diagnostic concern, not a definitive causal finding.
Common mistake. Calling the 70 decrease a practice change solely because it is larger than the 20 of new paid loss ignores the passage of time and claim-specific development. The correction is to examine the reserve-change reasons and compare with other claims at comparable ages and statuses before attributing the pattern to practice.
Common Traps and Practical Limits
- Comparing different ages. A 12-month ratio and a 24-month ratio describe different stages of claim development. Align ages before interpreting cohort differences.
- Reading the ratio as a reserve adequacy score. A lower \(C/P\) may reflect higher paid losses, different claim mix, or case-reserving changes. It does not by itself establish that ultimate losses are more or less adequately reserved.
- Ignoring the denominator. Ratios with low or zero cumulative paid can be highly unstable or undefined. Review the underlying amounts and avoid over-interpreting such cells.
- Using a simple average across segments. The aggregate is weighted by paid amounts. Calculate from total case outstanding divided by total paid, or use the correct paid weights.
- Treating claim histories as self-explanatory. Reserve changes can arise from new facts, routine review, settlement, or changes in recording practice. The timing and reasons matter.
The purpose of detecting a possible practice change is to identify an issue requiring explanation before relying on observed incurred development. A credible investigation combines age-specific ratios, underlying claim composition, and the chronology of reserve decisions. These diagnostics are aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.
Check Your Understanding
For each problem, use the definitions in this tutorial and distinguish a diagnostic signal from proof of cause.
- At 18 months, a hypothetical cohort has paid loss of 500 and case outstanding of 350. Calculate \(R\) and \(Q\), and state the identity check. (Answer: \(R=0.700\), \(Q=1.700\), and \(Q=1+R\).)
- At the same age, an earlier cohort has paid 400 and case outstanding 320; a later cohort has paid 600 and case outstanding 360. Calculate both cohorts’ ratios and identify the direction of the change. (Answer: earlier \(R=0.800,\ Q=1.800\); later \(R=0.600,\ Q=1.600\); both ratios decreased.)
- Two stable segments have ratios 0.40 and 1.20. In a cohort, their paid amounts are 900 and 100. Calculate the aggregate ratio. (Answer: \((900(0.40)+100(1.20))/1{,}000=0.480\).)
- A segment has paid loss of 250 and case outstanding of 175 at one valuation. At the next valuation, paid is 300 and case outstanding is 150. Calculate both sets of ratios and the change in incurred loss. (Answer: first \(R=0.700,\ Q=1.700\); second \(R=0.500,\ Q=1.500\); incurred changes from 425 to 450, an increase of 25.)
- At a common age, a cohort ratio falls from 0.90 to 0.55. Claim-level review shows the lower ratio is concentrated in a newly larger-paid segment, while comparable within-segment ratios are stable. What is the best-supported interpretation? (Answer: the aggregate decline is consistent with a mix shift and does not, by itself, demonstrate a change in case-reserving practice.)