Exposure Changes Versus Exposure Trend

Why Exposure Changes and Exposure Trend Must Be Separated

Expected losses can rise because an insurer covers more exposure, because the expected cost of each exposure unit has increased, or because both have happened. Those explanations are not interchangeable. A portfolio that adds policies may have higher total expected losses even if its expected loss cost per policy is unchanged. Calling the entire increase “trend” would confuse the amount of business insured with a change in expected cost per unit.

An exposure unit is the measure of insured business used as the denominator for a loss measure. It should match the coverage and rating basis being analyzed. Examples include a vehicle-year, a policy-year, a payroll amount, or a fixed amount of total insured value. The appropriate unit depends on the coverage; there is no single exposure measure suitable for every line.

Definition: Let $E$ denote the amount of exposure and $c$ the expected loss cost per exposure unit, measured on a specified coverage and loss basis. If the expected cost is homogeneous across the exposure units, expected aggregate losses are $E[L]=Ec$. A change in $E$ is an exposure change. A change in $c$, with the exposure and coverage basis held consistent, is a change in expected loss cost per unit and may be evidence of loss trend.

The word “may” matters. A changed average cost per unit can reflect a trend in claim frequency or severity, but it can also reflect changed coverage, a different mix of risks, or a changed measurement basis. The distinction developed in the earlier material on loss measures for trend analysis applies here: comparisons require a consistent loss measure and denominator.

Separating the Two Effects

For two periods, write expected aggregate losses as $E_1c_1$ and $E_2c_2$. Their ratio can be factored into an exposure ratio and a unit-cost ratio:

\frac{E_2c_2}{E_1c_1} = \frac{E_2}{E_1}\times\frac{c_2}{c_1}.

This identity gives a direct diagnostic. If the number or amount of exposure changes but $c$ does not, the aggregate loss change is due to exposure alone. If $E$ is unchanged but $c$ changes, the aggregate change is in expected cost per unit. If both change, both factors contribute. The combined percentage change is not generally the sum of the two percentage changes; the multiplicative factors must be combined.

The identity follows from the expected aggregate loss model. Suppose there are $E$ homogeneous exposure units and the expected loss for each unit is $c$. If $L_j$ is the loss associated with unit $j$, then linearity of expectation gives:

E[L] = E\!\left[\sum_{j=1}^{E}L_j\right] = \sum_{j=1}^{E}E[L_j] = \sum_{j=1}^{E}c = Ec.

This derivation does not require the losses from different units to be independent; linearity of expectation is sufficient. It does require that the same expected cost $c$ applies to each unit for the simple product $Ec$ to describe the portfolio. If units differ, use $E[L]=\sum_j c_j$, or divide the portfolio into sufficiently consistent groups and sum their expected losses. A shift in the mix of units can change the portfolio average $c$ even when the expected cost for every fixed risk type is unchanged.

Worked Examples

Worked Example: Policy Growth With Stable Cost per Policy-Year

A hypothetical insurer has 10,000 policy-years in period 1 and 12,000 policy-years in period 2. In both periods, claim counts per policy-year follow a Poisson distribution with mean $0.06$. Individual claim severities are independent of claim counts and have a Gamma distribution with shape $5$ and scale $\$1{,}000$, so the mean severity is $5\times \$1{,}000=\$5{,}000$. Assume the same coverage in both periods, no discounting ($i=0$), and no mortality effect. Find expected aggregate losses and separate the exposure change from the unit-cost change.

Method. Use frequency times mean severity to obtain expected loss cost per policy-year, then multiply by policy-years. The Poisson frequency parameter is $0.06$ claims per policy-year.

Period 1 unit cost. The expected claim count per policy-year is $0.06$, and the expected severity is $\$5{,}000$. Therefore:

c_1=0.06\times \$5{,}000=\$300\text{ per policy-year}.

Expected aggregate losses in period 1 are $10{,}000\times \$300=\$3{,}000{,}000$.

Period 2 unit cost. The claim-count and severity assumptions are unchanged, so $c_2=0.06\times \$5{,}000=\$300$ per policy-year. Expected aggregate losses in period 2 are $12{,}000\times \$300=\$3{,}600{,}000$.

Separate the effects. Exposure increased by $12{,}000/10{,}000=1.20$, or 20%. Unit cost changed by $\$300/\$300=1.00$, or 0%. Thus the aggregate expected loss ratio is $1.20\times1.00=1.20$: expected losses rose by 20%, entirely because the insurer wrote more policy-years.

Check and trap avoided. The difference is $\$3{,}600{,}000-\$3{,}000{,}000=\$600{,}000$, which is 20% of $\$3{,}000{,}000$. Treating that increase as a 20% increase in loss cost per policy-year would be wrong: the calculated cost remains $\$300$ in both periods. This is an exposure change, not evidence of exposure trend.

Worked Example: A Coverage Change Alters Unit Cost

A hypothetical insurer compares two groups of policy-years. The first has 800 policy-years under a narrower coverage; the second has 1,000 policy-years under broadened coverage. For both groups, the claim count per policy-year is Poisson with mean $0.06$. Under the narrower coverage, covered claim severity has a Gamma distribution with shape $4$ and scale $\$1{,}000$, giving mean $\$4{,}000$. Under the broadened coverage, it has a Gamma distribution with shape $4.6$ and the same scale, giving mean $\$4{,}600$. Assume claims are independent of counts, no discounting ($i=0$), and no mortality effect. Compare expected aggregate losses and explain whether the unit-cost increase alone can be called trend.

Method. Calculate expected covered loss per policy-year under each coverage, then factor the aggregate change into exposure and unit-cost changes. The distinct severity distributions represent the distinct hypothetical coverage terms; they are not a claim that coverage changes always alter severity in this exact way.

Under the narrower coverage, expected loss cost is $0.06\times \$4{,}000=\$240$ per policy-year. Expected aggregate losses are $800\times \$240=\$192{,}000$.

Under the broadened coverage, the expected severity is $4.6\times \$1{,}000=\$4{,}600$, so expected loss cost is $0.06\times \$4{,}600=\$276$ per policy-year. Expected aggregate losses are $1{,}000\times \$276=\$276{,}000$.

Exposure changed by $1{,}000/800=1.25$, or 25%. Expected cost per policy-year changed by $\$276/\$240=1.15$, or 15%. The aggregate ratio is:

1.25\times1.15=1.4375.

Thus expected aggregate losses increased by 43.75%. The two percentage increases do not simply add: $25\%+15\%=40\%$, but the additional interaction is $25\%\times15\%=3.75\%$.

Check and trap avoided. The direct difference is $\$276{,}000-\$192{,}000=\$84{,}000$, and $\$84{,}000/\$192{,}000=43.75\%$. It would be misleading to label the full 43.75% an increase in trend: 25% comes from more policy-years, and the per-policy-year cost comparison is also affected by broader coverage. To estimate trend on a consistent basis, the coverage change must be accounted for rather than interpreted as a change in cost under unchanged terms.

Worked Example: Insured Values as the Exposure Base

A hypothetical property portfolio has total insured value (TIV) of $\$100$ million in period 1 and $\$110$ million in period 2. Measure exposure in units of $\$1{,}000$ TIV. For each such unit, claim count is Poisson. In period 1 its mean is $0.01$, and claim severity has a Gamma distribution with shape $4$ and scale $\$20$. In period 2 the count mean remains $0.01$, while severity has a Gamma distribution with shape $4.4$ and scale $\$20$. Assume independent counts and severities, unchanged coverage, no discounting ($i=0$), and no mortality effect. Calculate the exposure and unit-cost contributions to expected losses.

Method. Convert TIV to $\$1{,}000$ units. For each unit, multiply expected claim count by mean severity. Then apply the exposure-times-unit-cost decomposition.

Period 1 has $\$100{,}000{,}000/\$1{,}000=100{,}000$ exposure units. Its mean severity is $4\times \$20=\$80$, so expected loss cost per unit is $0.01\times \$80=\$0.80$. Expected aggregate losses are $100{,}000\times \$0.80=\$80{,}000$.

Period 2 has $\$110{,}000{,}000/\$1{,}000=110{,}000$ exposure units. Its mean severity is $4.4\times \$20=\$88$, so expected loss cost per unit is $0.01\times \$88=\$0.88$. Expected aggregate losses are $110{,}000\times \$0.88=\$96{,}800$.

TIV exposure increased by $110{,}000/100{,}000=1.10$, or 10%. Expected cost per $\$1{,}000$ TIV increased by $\$0.88/\$0.80=1.10$, also 10%. Therefore the aggregate ratio is $1.10\times1.10=1.21$, a 21% increase. The interaction contributes the additional 1% beyond adding the two 10% changes.

Check and trap avoided. The aggregate difference is $\$96{,}800-\$80{,}000=\$16{,}800$, which is 21% of $\$80{,}000$. If severity rose only in proportion to the insured value represented by the exposure measure, the cost per $\$1{,}000$ TIV would remain constant; the aggregate increase would then be attributable to exposure alone. Here, the stated model also increases expected cost per TIV unit. A change in average loss per property, without accounting for the change in insured values, would not isolate that unit-cost change.

Interpreting Exposure Measures Carefully

The choice of exposure denominator determines what “cost per unit” means. Policy-years are convenient when the insured unit is a policy over time, but two policy-years may represent substantially different amounts of insured property or different coverage terms. Payroll can be appropriate for some liability coverages, while TIV can be meaningful for property coverage. Comparisons are informative only when the exposure measure is defined consistently and is suitable for the loss being studied.

A change in exposure does not necessarily mean that the count of policies changed. Exposure can change because policies were added or removed, terms in force changed, the amount of payroll changed, or insured values were updated. For a TIV-based analysis, updating a building’s insured value increases measured exposure. Whether that update also corresponds to a change in expected loss cost per unit depends on the loss model and the consistency of the valuation basis.

Likewise, a portfolio’s average cost per exposure unit can change when its composition changes. For example, adding a larger share of high-hazard risks raises the portfolio average even if each risk group’s own expected cost is stable. A comparison meant to isolate change in expected cost under a fixed risk mix must account for composition as well as exposure volume.

Key takeaway: First identify the exposure measure and hold its definition, coverage, and comparison basis clear. Then compare the amount of exposure separately from expected loss cost per unit. An increase in total expected losses by itself does not establish an increase in loss trend.

Common Exam Trap

A common error is to calculate the percentage change in aggregate expected losses and report it as the percentage change in unit loss cost. The calculation is valid only if exposure is unchanged. If exposure changes, use:

\text{aggregate loss ratio} = \text{exposure ratio}\times\text{unit-cost ratio}.

For example, if exposure grows by 10% and cost per unit grows by 10%, aggregate expected losses grow by $1.10\times1.10-1=21\%$, not 20%. Another trap is to call every change in observed cost per unit “trend.” A changed deductible, limit, coverage, or portfolio mix may alter the unit cost being compared. State the basis of comparison before interpreting the result. These distinctions are aligned with the CAS Exam 5 (Basic Ratemaking and Estimated Claim Liabilities) syllabus.

Check Your Understanding

For each problem, separate the change in exposure from the change in expected loss cost per unit. Monetary answers are in dollars unless stated otherwise.

A hypothetical portfolio grows from 5,000 to 6,000 policy-years. Expected loss cost remains $\$72$ per policy-year under unchanged coverage. Find expected aggregate losses in both periods and state the percentage change attributable to exposure. (Answer: $\$360{,}000$ and $\$432{,}000$; exposure rises 20%, unit cost is unchanged.)
A portfolio increases from 2,000 to 2,200 exposure units. Expected loss cost rises from $\$150$ to $\$165$ per unit. Calculate the aggregate expected losses and their percentage change. (Answer: $\$300{,}000$ and $\$363{,}000$; a 21% increase, comprising 10% more exposure and 10% higher unit cost.)
A property portfolio’s TIV increases from $\$60$ million to $\$66$ million. Expected loss cost is $\$0.70$ and then $\$0.77$ per $\$1{,}000$ TIV. Calculate expected aggregate losses and separate the two ratios. (Answer: $\$42{,}000$ and $\$50{,}820$; exposure ratio $1.10$, unit-cost ratio $1.10$, aggregate increase 21%.)
A hypothetical insurer has 900 policy-years in each period. The expected cost per policy-year falls from $\$200$ to $\$190$, with coverage and mix held constant. Find aggregate expected losses and the unit-cost change. (Answer: $\$180{,}000$ and $\$171{,}000$; exposure is unchanged and unit cost falls 5%, so aggregate expected losses also fall 5%.)
Expected aggregate losses rise from $\$500{,}000$ to $\$605{,}000$ while exposure rises from 2,500 to 2,750 units. Find the unit cost in each period and decompose the aggregate ratio. (Answer: $\$200$ and $\$220$ per unit; exposure ratio $1.10$, unit-cost ratio $1.10$, aggregate ratio $1.21$, or a 21% increase.)

What You'll Learn