From Loss Experience to Capital Need

From Historical Losses to a Decision Basis

A pricing decision often requires more than a loss-cost calculation. Historical losses may be immature, recorded at an earlier cost level, or drawn from a different exposure volume than the portfolio being priced. A rate indication must first put losses and exposures on a prospective basis. A capital calculation then needs a separate, explicit model for the aggregate losses over a stated period.

The key discipline is to keep the bases aligned. State the period, exposure count, claim-loss definition, development and trend assumptions, expense treatment, and risk measure. In particular, an indicated premium and a capital requirement answer different questions: the premium funds expected costs under a pricing rule, while the capital measure describes adverse aggregate outcomes relative to an expected-loss benchmark.

Definition: For an experience period, let reported losses be adjusted to ultimate using a loss development factor, then adjusted to the prospective cost level using a trend factor. The resulting prospective-level losses divided by the corresponding earned exposure give an experience-based prospective pure premium per exposure unit. A rate indication is the premium per exposure unit implied by a stated pricing equation, including any specified expenses or loads.

A loss development factor (LDF) estimates the ratio of ultimate losses to losses currently reported for a defined cohort. A trend factor adjusts ultimate losses from their historical cost level to the prospective cost level. These adjustments are applied to losses before they are divided by the exposure that generated them. If several experience periods are combined, sum the adjusted losses and divide by the sum of their corresponding exposures; this is an exposure-weighted average, not necessarily an unweighted average of period loss costs.

Developing Losses and Indicating a Rate

Consider a hypothetical annual policy with one exposure unit per policy-year. Suppose reported losses for a historical accident period are $800,000, the selected LDF to ultimate is 1.25, and the selected trend factor to the prospective period is 1.00. There are 1,000 earned exposure units. The prospective-level pure premium is

\frac{800{,}000(1.25)(1.00)}{1{,}000} = \$1{,}000 \text{ per exposure unit.}

The factor 1.00 represents an explicit assumption of no change in the loss cost level over the stated interval; it is not an omitted adjustment. In practice, an actuary would select development and trend assumptions appropriate to the coverage, experience period, and prospective period.

For a simple gross rate indication, let $L$ be expected loss cost per exposure unit, $e_f$ fixed expense per exposure unit, $e_v$ the variable expense proportion of premium, and $R$ the indicated premium per exposure unit. If expenses are the only additions and variable expenses are a proportion of premium, the pricing equation is

R=L+e_f+e_vR, \qquad R=\frac{L+e_f}{1-e_v}, \quad e_v<1.

This equation does not include a capital charge, investment income adjustment, or profit provision unless the pricing basis explicitly specifies one. Keeping such items separate avoids adding an unstated load or counting the same provision twice.

Worked Example: Adjusting Loss Experience for a Gross Rate

A hypothetical book has $800,000 of reported losses for a historical period, a selected LDF of 1.25, and a prospective trend factor of 1.00. The period has 1,000 earned exposure units. The proposed rate basis includes fixed expenses of $50 per exposure and variable expenses equal to 20% of premium. Calculate the prospective pure premium and gross rate indication. Ignore all other provisions.

Method. Develop and trend losses, divide by earned exposure, and solve the premium equation for the variable expense proportion. This keeps the fixed expense per exposure distinct from the variable expense percentage of premium.

Step 1: Adjust the losses. Ultimate losses are $800{,}000(1.25)=1{,}000{,}000$. Applying the trend factor gives $1{,}000{,}000(1.00)=1{,}000{,}000$ at the prospective level.

Step 2: Calculate the pure premium. Divide adjusted losses by the 1,000 earned exposure units: $L=1{,}000{,}000/1{,}000=\$1{,}000$ per exposure.

Step 3: Solve for the rate. The rate equation is $R=1{,}000+50+0.20R$. Therefore $0.80R=1{,}050$, so $R=\$1{,}312.50$ per exposure.

Check. Variable expenses are $0.20(1{,}312.50)=\$262.50$. The premium less variable expenses is $1{,}312.50-262.50=\$1{,}050$, exactly the $1,000 loss cost plus $50 fixed expense. The common mistake avoided is dividing by $1+e_v$, or adding 20% of the loss cost as though variable expenses were a fixed dollar amount.

Aggregate Loss and Capital

A rate calculation usually focuses on expected losses per exposure. A capital calculation instead considers the distribution of total losses across a portfolio and a defined period. For a one-year compound Poisson model, let $N$ be the number of claims, with $N\sim\operatorname{Poisson}(\lambda)$. Let $X_1,X_2,\ldots$ be independent claim severities with the same distribution as loss $X$, independent of $N$. The annual aggregate loss is $S=\sum_{j=1}^{N}X_j$, with $S=0$ if $N=0$.

Result: Under the compound Poisson assumptions, if $E[X]$ and $E[X^2]$ are finite, then $E[S]=\lambda E[X]$ and $\operatorname{Var}(S)=\lambda E[X^2]$. For $n$ independent, identically distributed exposure units with Poisson claim frequency $\lambda$ per unit, the portfolio count is Poisson with mean $n\lambda$, so the aggregate mean and variance are $n\lambda E[X]$ and $n\lambda E[X^2]$.

The variance uses the second moment $E[X^2]$, not just the severity variance. This distinction matters because randomness in claim count contributes to aggregate variability as well as randomness in claim sizes.

Why the Compound Poisson Variance Has This Form

Condition on the claim count. Given $N$, the aggregate is a sum of $N$ independent severities, so $E[S\mid N]=NE[X]$ and $\operatorname{Var}(S\mid N)=N\operatorname{Var}(X)$. The law of total expectation gives

E[S]=E\big[E[S\mid N]\big]=E[N]E[X]=\lambda E[X].

By the law of total variance, the aggregate variance is the expected conditional variance plus the variance of the conditional mean:

\begin{aligned} \operatorname{Var}(S) &=E\big[\operatorname{Var}(S\mid N)\big] +\operatorname{Var}\big(E[S\mid N]\big)\\ &=E[N]\operatorname{Var}(X)+\operatorname{Var}(N)(E[X])^2\\ &=\lambda\operatorname{Var}(X)+\lambda(E[X])^2\\ &=\lambda E[X^2]. \end{aligned}

The last equality follows from $E[X^2]=\operatorname{Var}(X)+(E[X])^2$, and a Poisson random variable has both mean and variance $\lambda$. The result relies on the stated frequency-severity assumptions; it does not automatically hold for a non-Poisson count, dependent claim sizes, or dependence between count and severity.

Worked Example: A One-Year Aggregate-Loss Quantile

A hypothetical portfolio has 1,000 policies. For each policy, the annual claim count is Poisson with mean 1.25. Claim severity is $400 or $1,200, each with probability 0.5, independently across claims and independently of frequency. Use a normal approximation to estimate the 99th percentile of annual aggregate claim losses and the capital defined as that percentile less expected aggregate losses.

Method. Calculate the severity first and second moments, aggregate mean, and aggregate variance. Then use a normal quantile approximation. The approximation is an estimate, not an exact quantile of the discrete aggregate distribution.

Step 1: Severity moments. The mean is $E[X]=0.5(400)+0.5(1{,}200)=800$. The second moment is $E[X^2]=0.5(400^2)+0.5(1{,}200^2)=0.5(160{,}000)+0.5(1{,}440{,}000)=800{,}000$.

Step 2: Portfolio claim frequency. The total count has mean $1{,}000(1.25)=1{,}250$. Thus $E[S]=1{,}250(800)=\$1{,}000{,}000$, and $\operatorname{Var}(S)=1{,}250(800{,}000)=1{,}000{,}000{,}000$ dollars squared.

Step 3: Approximate the quantile. The standard deviation is $\sqrt{1{,}000{,}000{,}000}=\$31{,}622.7766$. Using the standard normal 99th-percentile value $z_{0.99}=2.3263$, the approximate 99th percentile is $1{,}000{,}000+2.3263(31{,}622.7766)=\$1{,}073{,}564.07$.

Step 4: Calculate capital. With capital defined as VaR less expected loss, the estimate is $1{,}073{,}564.07-1{,}000{,}000=\$73{,}564.07$. Amounts are rounded to the nearest cent at the final step.

Check. The estimated quantile is above the mean, as required for a 99th percentile in this model. The common mistake avoided is using $\operatorname{Var}(X)$ in place of $E[X^2]$: here $\operatorname{Var}(X)=160{,}000$, which would substantially understate aggregate variance.

An Integrated Experience-to-Capital Case

The following case joins the experience adjustment, rate indication, and aggregate risk calculation. The model assumptions are selected so that the experience-based expected loss cost and the prospective frequency-severity model agree. This reconciliation is a useful check, but it does not make the estimated capital a component of the rate unless a pricing rule explicitly says to include a capital charge.

Worked Example: Reconciling a Rate Indication and Capital Need

A hypothetical annual book has 1,000 earned exposure units and $800,000 of reported losses for its historical accident period. The selected LDF is 1.25 and the trend factor to the prospective year is 1.00. For the prospective year, claim frequency per exposure is Poisson with mean 1.25, and severity is $400 or $1,200 with equal probability. The rate basis has fixed expenses of $50 per exposure and variable expenses of 20% of premium. Estimate the gross rate per exposure, annual aggregate expected loss, 99th-percentile capital, and total indicated premium for the 1,000 exposures. Define capital as the 99th percentile of annual aggregate claim losses less their mean, using a normal approximation. Ignore expenses in the capital measure.

Method. First derive the experience pure premium and rate, then independently derive aggregate risk from the prospective model. Finally compare expected losses per exposure to the historical indication and state exactly which amounts the capital measure includes.

Step 1: Experience-based pure premium. Ultimate, prospective-level losses are $800{,}000(1.25)(1.00)=\$1{,}000{,}000$. Dividing by 1,000 exposures gives $\$1{,}000$ per exposure.

Step 2: Gross rate. Solve $R=1{,}000+50+0.20R$. Then $R=1{,}050/0.80=\$1{,}312.50$ per exposure. For 1,000 exposures, total indicated premium is $1{,}000(1{,}312.50)=\$1{,}312{,}500$.

Step 3: Check the model's expected loss. The prospective severity mean is $\$800$, so expected loss per exposure is $1.25(800)=\$1{,}000$, matching the experience pure premium. The portfolio aggregate expected loss is $1{,}000(1{,}000)=\$1{,}000{,}000$.

Step 4: Aggregate variance and capital. From the severity calculation, $E[X^2]=800{,}000$. Therefore $\operatorname{Var}(S)=1{,}000(1.25)(800{,}000)=1{,}000{,}000{,}000$. The standard deviation is $\$31{,}622.7766$. Using $z_{0.99}=2.3263$, the approximate 99th percentile is $\$1{,}073{,}564.07$, and capital is $\$73{,}564.07$.

Check and interpretation. The indicated premium of $1,312,500 covers the model's $1,000,000 expected claim cost, $50,000 fixed expenses, and $262,500 variable expenses; these sum to $1,312,500. The capital estimate is calculated on annual claim losses only, before expenses, and is not added to premium in this example. The common mistake avoided is mixing bases—for example, calculating capital on claim losses after expenses while subtracting an expected loss that excludes expenses, or adding capital to the rate without a stated capital-pricing rule.

Basis Checks Before Using the Result

A calculation can be arithmetically correct and still be unsuitable if its components refer to different periods or quantities. Before relying on a rate or capital result, verify the following:

Loss maturity: the development factor applies to the reported loss measure and cohort used in the analysis.
Cost level: trend runs from the historical loss period to the intended prospective period, not to an unrelated valuation date.
Exposure denominator: adjusted losses are divided by the matching earned exposure units, and any portfolio scaling uses the same exposure definition.
Expense basis: fixed expense per exposure and variable expense as a proportion of premium are not interchanged.
Aggregate period and scope: the claim count model, severity model, portfolio size, and quantile all refer to the same period and loss definition.
Capital definition: state the confidence level and whether capital is VaR itself, VaR less expected loss, or another specified measure.

A normal approximation can be useful for a large aggregate, but it remains an approximation. It can be less reliable when the count is small, the severity distribution is highly skewed, or a small number of very large claims dominate. Whatever method is used, report the modeled risk measure and its assumptions rather than presenting an approximate quantile as an exact outcome.

Key takeaway: Build the decision in sequence: bring losses to ultimate and prospective levels, normalize by earned exposure, solve the stated pricing equation, then calculate aggregate risk from an explicit frequency-severity model. Reconcile expected loss across the experience and model bases, and keep capital separate from premium unless the pricing rule explicitly connects them.

Check Your Understanding

For each hypothetical case, carry the stated units through the calculation and use the definition of capital given in the question.

A book has $600,000 of reported losses, an LDF of 1.10, a trend factor of 1.00, and 660 earned exposures. Calculate the prospective pure premium. (Answer: $1,000 per exposure.)
A prospective pure premium is $900 per exposure. Fixed expenses are $30 per exposure and variable expenses are 10% of premium. Find the gross indicated rate. (Answer: $1,033.33 per exposure, rounded to the nearest cent.)
A portfolio has 400 exposures, each with Poisson claim frequency 2.0. Severity is $100 or $500 with equal probability. Calculate the aggregate expected loss and variance. (Answer: Expected loss $240,000; variance $104,000,000 dollars squared.)
A compound Poisson portfolio has annual aggregate mean $500,000 and standard deviation $20,000. Under a normal approximation with $z_{0.95}=1.6449$, calculate 95th-percentile VaR and capital defined as VaR less mean. (Answer: VaR $532,898; capital $32,898, rounded to the nearest dollar.)
A hypothetical book has expected loss cost of $800 per exposure, fixed expense of $40 per exposure, and variable expense of 20% of premium. The annual aggregate claim-loss mean is $800,000 and its 99th-percentile VaR is $860,000. Find the gross rate per exposure and the capital under the definition VaR less expected loss; do not add capital to the rate. (Answer: Gross rate $1,050 per exposure; capital $60,000.)

What You'll Learn