Estimating Claim Severity Trends

Why Estimate Claim Severity Trends?

Claim severity is the average loss amount per claim. Its trend helps describe how the cost of an individual claim changes over time. That change may reflect repair prices, medical costs, changes in the claims process, or differences in which claims are included in the data. It is therefore important to calculate severity consistently and to distinguish an observed change from a change in the underlying cost of comparable claims.

The basic calculation is simple, but its interpretation is not automatic. A period with more large claims may have higher average severity even if the cost of each type of claim has not changed. A changed reporting threshold can also alter the average by adding or removing small claims. The analysis should make the claim definition, loss basis, and time interval explicit. This tutorial develops those calculations and interpretation checks, aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Definition: For a specified period and a consistently defined set of claims, average claim severity is aggregate loss divided by claim count. If $L_t$ is the aggregate loss for period $t$ and $N_t$ is its corresponding claim count, then $S_t=L_t/N_t$. The loss measure and count must refer to the same claims, coverage, valuation basis, and inclusion rules.

The loss numerator might be paid loss, incurred loss, or another specified loss measure. Do not combine a numerator that includes one set of claims with a denominator that counts a different set. For example, losses for claims above a reporting threshold should be divided by the number of claims above that same threshold, not by the count of all claims.

Calculating and Annualizing Severity Trend

For two observations, the percentage change in severity is the later average divided by the earlier average, less one. When the observations are separated by $t$ years, a constant compound annual trend assumption expresses that total change as the same multiplicative change in each year. This assumption is a convenient way to annualize the observed change; it does not establish that severity actually changed at a constant rate during the interval.

S_1=\frac{L_1}{N_1},\qquad S_2=\frac{L_2}{N_2},\qquad \text{annualized trend}= \left(\frac{S_2}{S_1}\right)^{1/t}-1

The exponent $1/t$ matters whenever the interval is not one year. For instance, if the later severity is 21% higher after two years, the annualized trend is $(1.21)^{1/2}-1=10\%$, not 21% per year. The calculation assumes that the measured severities are comparable; annualizing a change does not correct inconsistent claim definitions, changing mix, or an unusual large loss.

Derivation: Under a constant compound annual trend $g$, severity after $t$ years is $S_2=S_1(1+g)^t$. Dividing by $S_1$ gives $S_2/S_1=(1+g)^t$. Taking both sides to the power $1/t$ gives $(S_2/S_1)^{1/t}=1+g$, so $g=(S_2/S_1)^{1/t}-1$. This derivation depends on the constant compound trend assumption; it does not imply that the observed data followed a smooth path.

Worked Example: Annualizing a Change Over 18 Months

A hypothetical insurer observes $1,080,000 of losses on 240 claims in an earlier period. Eighteen months later, it observes $1,485,000 on 300 claims. Assume both periods use the same loss basis and claim definition. Calculate each average severity, the total percentage change, and the annualized compound trend. Use the method of dividing period losses by the matching counts and then annualizing the ratio.

Earlier severity. Divide aggregate loss by the claim count:

S_1=\frac{\$1{,}080{,}000}{240}=\$4{,}500

Later severity. Apply the same calculation:

S_2=\frac{\$1{,}485{,}000}{300}=\$4{,}950

The observed change over the full interval is $(4{,}950/4{,}500)-1=0.10$, or 10.00%. Eighteen months is $18/12=1.5$ years, so the annualized trend is:

\left(\frac{4{,}950}{4{,}500}\right)^{1/1.5}-1 =1.1^{2/3}-1 \approx 0.0656

The annualized compound trend is approximately 6.56%. As a check, compounding this rate for 1.5 years gives approximately $1.0656^{1.5}=1.100$, the observed severity ratio. The common mistake is to report 10% per year: 10% is the total increase over 18 months, not the annualized increase.

Claim Mix and Comparable Severity

An aggregate average is a weighted average of the severities of the claim groups it contains. The weights are the proportions of claims in each group, not the proportions of losses. If group $j$ has $N_{tj}$ claims and average severity $S_{tj}$ in period $t$, then aggregate severity is $\sum_j (N_{tj}/N_t)S_{tj}$, where $N_t=\sum_j N_{tj}$. This identity follows because group loss is $N_{tj}S_{tj}$, so total loss divided by total claims is $\sum_j N_{tj}S_{tj}/N_t$.

If group proportions change, aggregate severity can change even when every group’s average stays the same. A useful diagnostic is to apply one period’s claim proportions to the other period’s group severities. This produces a standardized comparison: it answers what the later severity would have been under the selected reference mix. It does not replace the observed aggregate; it separates a mix effect from changes within the groups included in the calculation.

Worked Example: Separating Within-Group Change from Mix Change

A hypothetical liability portfolio has two claim types, L and H. In Period A, there are 90 Type L claims averaging $800 and 10 Type H claims averaging $4,000. In Period B, Type L averages $880 and Type H averages $4,400; 70% of Period B claims are Type L and 30% are Type H. Calculate aggregate severity in each period, its observed percentage change, and Period B severity standardized to Period A’s mix. Use claim proportions as weights.

Period A aggregate. The claim proportions are 90/100=90% Type L and 10/100=10% Type H. Thus:

S_A=0.90(\$800)+0.10(\$4{,}000) =\$720+\$400=\$1{,}120

Period B aggregate. Use the stated Period B proportions, 70% and 30%:

S_B=0.70(\$880)+0.30(\$4{,}400) =\$616+\$1{,}320=\$1{,}936

The observed aggregate increase is $(1{,}936/1{,}120)-1=0.728571$, or 72.86%. That comparison reflects both within-type severity changes and the shift toward Type H claims.

Standardize Period B to Period A’s mix. Apply the Period A weights, 90% and 10%, to Period B’s group averages:

S_{B\mid A\text{ mix}} =0.90(\$880)+0.10(\$4{,}400) =\$792+\$440=\$1{,}232

Compared with Period A’s $1,120, this standardized value is a 10.00% increase: $(1{,}232/1{,}120)-1=0.10$. The method shows a 10% increase within the fixed Period A mix, while the much larger observed change also reflects the changed mix. A common error is to use Period B’s 70%/30% weights for this standardization; those weights calculate the observed Period B aggregate of $1,936, not Period B at Period A’s mix. The calculation checks because the standardized result uses exactly the 90%/10% reference proportions.

Thresholds and Large Claims

A reporting or analysis threshold changes which claims enter both the numerator and denominator. A severity calculated for claims above a threshold is a conditional average for that selected group; it is not directly comparable with an average that includes all claims unless the effect of the changed selection is addressed. Even with the same stated threshold, changes in reporting practice or in the timing of claim emergence can affect which claims are observed. Comparisons should therefore state the threshold, the loss basis, and the valuation date.

Large claims can have a substantial effect on aggregate severity because each dollar of loss enters the numerator while each claim contributes only one unit to the denominator. Analysts may examine results with and without large claims, or apply a consistent cap or other specified treatment. Such adjustments answer different questions: excluding large claims describes the remaining selected claims, while an uncapped result includes the full loss experience. The treatment must be applied consistently across periods and disclosed; removing a large claim from the numerator but leaving it in the count is not a consistent severity calculation.

Worked Example: Testing the Effect of One Large Claim

A hypothetical portfolio has 200 claims and $1,200,000 of total loss in Year 1, including one $300,000 claim. In Year 3 it has 220 claims and $1,452,000 of total loss, including one $400,000 claim. Calculate gross average severity and its annualized trend over the two-year interval. Then calculate severity after excluding the identified large claim from both loss and count in each year, and annualize that comparison. Assume all other claims remain included and the same exclusion rule is used in both periods.

Gross severity. Keep each large claim in both the loss total and claim count:

S_1=\frac{\$1{,}200{,}000}{200}=\$6{,}000,\qquad S_3=\frac{\$1{,}452{,}000}{220}=\$6{,}600

The gross severity ratio is $6{,}600/6{,}000=1.10$. Over two years, the annualized gross trend is $\sqrt{1.10}-1\approx 4.88\%$.

Severity excluding the identified large claims. Remove each large loss and its corresponding claim from the calculation:

S_{1,\text{without large}}= \frac{\$1{,}200{,}000-\$300{,}000}{200-1} =\frac{\$900{,}000}{199} \approx \$4{,}522.61

S_{3,\text{without large}}= \frac{\$1{,}452{,}000-\$400{,}000}{220-1} =\frac{\$1{,}052{,}000}{219} \approx \$4{,}803.65

The ratio of the unrounded severities is approximately $1.06214$. Therefore the two-year annualized trend is $\sqrt{1.06214}-1\approx 3.06\%$. Rounded to cents, the adjusted severities are $4,522.61 and $4,803.65; the trend is reported to two decimals. The gross trend is higher than the adjusted trend because the two identified large claims are larger in Year 3. This comparison does not prove the remaining claims have a stable mix. The common mistake is to subtract each large loss but leave the claim count unchanged; that would not calculate average severity for the claims that remain.

Interpreting a Severity Estimate

Average severity is an aggregate measure, not a direct estimate of the change in cost for an identical claim. Before treating it as a trend indication, check whether the periods use a common coverage definition, claim count convention, loss basis, threshold, and valuation approach. Consider whether claim types or limits have shifted, and whether one or a few large losses dominate the result. If multiple periods are available, a fitted time trend may summarize the observations, but the same comparability concerns remain.

The main exam trap is confusing a change in the observed average with a change in the cost of comparable claims. The calculations are necessary, but they do not by themselves explain causation. State what the measure includes, calculate the average using matching losses and counts, annualize only after identifying the interval, and interpret the result in light of claim selection and mix.

Key takeaway: A severity trend is only as comparable as the claim populations being divided. Match losses to counts, use a consistent claim definition, annualize using the actual elapsed time, and investigate thresholds, large losses, and mix before interpreting the result.

Check Your Understanding

For each problem, calculate using the stated claim definition and interval. Round annualized trends to two decimal places.

A hypothetical program records $936,000 on 240 claims in one period and $1,134,000 on 270 claims exactly two years later. Find both average severities and the annualized compound trend. (Answer: $3,900 and $4,200; 3.78% per year.)
A portfolio has 150 claims with $1,050,000 total loss in one period and 180 claims with $1,404,000 total loss 18 months later. Find the average severities and annualized trend. (Answer: $7,000 and $7,800; 7.48% per year.)
In Period 1, 75% of claims are Group A averaging $500 and 25% are Group B averaging $2,000. In Period 2, the group averages are $600 and $2,600. Calculate Period 1 aggregate severity, Period 2 severity standardized to Period 1’s mix, and the standardized percentage change. (Answer: $875; $1,100; 25.71%.)
In Year 1, 50 claims total $300,000, including one $100,000 claim. In Year 2, 60 claims total $390,000, including one $150,000 claim. Calculate the annual severity trend after excluding each identified large claim from both the loss total and count. (Answer: Year 1 severity $200,000/49 = $4,081.63; Year 2 severity $240,000/59 = $4,067.80; approximately −0.34%.)
Two years apart, a portfolio’s aggregate severity rises from $5,000 to $6,050. A consistently defined subset that excludes large claims rises from $4,000 to $4,356 over the same interval. Find the annualized compound trend for each measure and state which has the higher indicated trend. (Answer: Aggregate 10.00% per year; subset approximately 4.36% per year; the aggregate measure has the higher indicated trend.)

What You'll Learn