Variability-Based Contingency Loads

Why Measure a Contingency Load by Variability?

Expected losses and expenses describe an average, but an insurer’s actual underwriting result for a period can differ from that average. A contingency load provides an additional cushion for this variation. In this tutorial, the load is calculated as a selected multiple of the standard deviation of underwriting results. The multiple controls the size of the cushion; it does not change the underlying uncertainty in the results.

The calculation depends on the quantity whose variability is being measured. For example, if premium and expenses are fixed for the period, variation in the underwriting result comes from variation in losses. If expenses also vary, their variability and their relationship with losses must be included. The assumed model and the treatment of dependencies should therefore be stated before calculating a load.

Definition: Let $R$ be the underwriting result before a contingency load, and let $\sigma_R=\sqrt{\operatorname{Var}(R)}$ be its standard deviation. A variability-based contingency load is $C=k\sigma_R$, where $k$ is a nonnegative selected multiple. If the load is collected as additional premium, $C$ is a dollar amount for the modeled period. Relative to a specified premium base $P$, its percentage is $C/P$.

For a one-period model with fixed premium $P$, fixed expenses $E$, and random aggregate losses $L$, define $R=P-E-L$. Then $\operatorname{Var}(R)=\operatorname{Var}(L)$: subtracting a random quantity changes the direction of deviations, not their squared size. If expenses are random too, then $\operatorname{Var}(R)=\operatorname{Var}(E+L)$, which includes the covariance between expenses and losses.

This definition separates two ideas. Expected underwriting profit is based on the expected result, $E[R]$; the contingency load is based on the spread of possible results, $\sigma_R$. A load collected as additional premium increases the result by a fixed amount in this model. It does not reduce the standard deviation of future losses.

Why the Multiple Scales the Cushion

The relationship $C=k\sigma_R$ follows directly from expressing the cushion in units of the result’s standard deviation. A load of $0.5\sigma_R$ is half a standard deviation; a load of $1.5\sigma_R$ is one and a half standard deviations. Holding the risk model fixed, multiplying $k$ by a factor multiplies the dollar load by the same factor.

C(k)=k\sigma_R,\qquad C(k_2)-C(k_1)=(k_2-k_1)\sigma_R

If $C$ is added to premium, the post-load result is $R+C$. Since $C$ is fixed, $E[R+C]=E[R]+C$, while $\operatorname{Var}(R+C)=\operatorname{Var}(R)$. Thus, a larger multiple increases the amount available to absorb an unfavorable result relative to the original expected result, but it does not make the underlying loss distribution less variable.

A selected multiple alone does not establish a particular probability of avoiding an underwriting loss. That interpretation would require a specified distributional model and a precise definition of the loss event. Here, $k$ is a chosen scale for the cushion, not a guarantee that adverse results will be covered.

Calculating Variability from Loss Models

When losses are the only random component, a useful first step is to calculate the aggregate-loss variance. The model determines the appropriate formula. For a fixed number of independent, identical claim opportunities with a fixed payment per claim, a binomial claim count gives a direct calculation. For a Poisson claim count with random severities, the compound Poisson variance uses the severity’s second moment, not just its variance.

Result: If $L=\sum_{j=1}^{N}X_j$, where $N$ has a Poisson distribution with parameter $\lambda$, the severities $X_j$ are independent and identically distributed with finite second moment, and the severities are independent of $N$, then $E[L]=\lambda E[X]$ and $\operatorname{Var}(L)=\lambda E[X^2]$.

To see why the variance formula holds, condition first on the number of claims. Given $N=n$, the aggregate is a sum of $n$ independent severities, so its conditional mean is $nE[X]$ and its conditional variance is $n\operatorname{Var}(X)$. The law of total variance gives $\operatorname{Var}(L)=E[N]\operatorname{Var}(X)+\operatorname{Var}(N)(E[X])^2$. For a Poisson variable, both $E[N]$ and $\operatorname{Var}(N)$ equal $\lambda$. Since $E[X^2]=\operatorname{Var}(X)+(E[X])^2$, the result follows. This formula does not apply unchanged if claim severities depend on the count or on one another.

Worked Example: A Fixed Payment with a Binomial Claim Count

A hypothetical annual book has 100 independent policies. Each policy has probability $0.02$ of producing a claim, and each claim costs exactly \$5,000. Premium is \$18,000 and fixed expenses are \$3,000. Calculate the contingency load for $k=0.75$, and express it as a percentage of the original premium. Use a one-period model with no discounting (interest rate $0\%$); mortality is not applicable. The claim count is binomial with parameters $n=100$ and $p=0.02$.

Method and claim-count moments. Let $N$ be the number of claims. Then $N\sim\operatorname{Binomial}(100,0.02)$, so $E[N]=np=100(0.02)=2$, and $\operatorname{Var}(N)=np(1-p)=100(0.02)(0.98)=1.96$.

Expected losses and result. Aggregate losses are $L=5{,}000N$. Therefore, $E[L]=5{,}000(2)=\$10{,}000$, and the expected underwriting result before the load is $\$18{,}000-\$3{,}000-\$10{,}000=\$5{,}000$.

Standard deviation and load. Since the payment per claim is fixed, $\operatorname{Var}(L)=5{,}000^2(1.96)=25{,}000{,}000(1.96)=49{,}000{,}000$. Thus $\sigma_R=\sigma_L=\sqrt{49{,}000{,}000}=\$7{,}000$, and $C=0.75(\$7{,}000)=\$5{,}250$.

Premium percentage and check. Relative to the original premium, the load is $\$5{,}250/\$18{,}000=0.291666\ldots$, or 29.17% to two decimal places. The expected result after collecting the load is $\$5{,}000+\$5{,}250=\$10{,}250$, while the standard deviation remains \$7,000. The calculation avoids confusing the expected result with the variability-based load.

Worked Example: A Compound Poisson Aggregate

A hypothetical one-year coverage has claim count $N\sim\operatorname{Poisson}(40)$. Each claim independently costs \$1,000 or \$2,000, each with probability $0.5$. The premium is \$90,000 and fixed expenses are \$12,000. Calculate a $1.2$-standard-deviation contingency load and its percentage of premium. Assume one-period cash flows with interest rate $0\%$; mortality is not applicable.

Method and severity moments. For severity $X$, $E[X]=0.5(\$1{,}000)+0.5(\$2{,}000)=\$1{,}500$. Its second moment is $E[X^2]=0.5(\$1{,}000^2)+0.5(\$2{,}000^2)=0.5(\$1{,}000{,}000)+0.5(\$4{,}000{,}000)=\$2{,}500{,}000$.

Aggregate moments. By the compound Poisson result, $E[L]=40(\$1{,}500)=\$60{,}000$, and $\operatorname{Var}(L)=40(\$2{,}500{,}000)=\$100{,}000{,}000$. Thus, $\sigma_R=\sigma_L=\sqrt{\$100{,}000{,}000}=\$10{,}000$.

Load and check. The load is $C=1.2(\$10{,}000)=\$12{,}000$, or $\$12{,}000/\$90{,}000=0.133333\ldots=$13.33% of premium to two decimal places. Before the load, the expected result is $\$90{,}000-\$12{,}000-\$60{,}000=\$18{,}000$. After adding the load, it is \$30,000, and the standard deviation remains \$10,000.

Common mistake avoided. The severity variance is $0.5(\$1{,}000-\$1{,}500)^2+0.5(\$2{,}000-\$1{,}500)^2=\$250{,}000$. Using only this variance would omit the variability in the number of claims. The compound Poisson variance uses $E[X^2]=\$2{,}500{,}000$, giving the correct aggregate variance of \$100,000,000.

Combining Multiple Lines

For multiple components, standard deviations generally cannot be added directly. If total losses are $L_1+L_2$, their variance is $\operatorname{Var}(L_1)+\operatorname{Var}(L_2)+2\operatorname{Cov}(L_1,L_2)$. The covariance term reflects whether the components tend to be high or low together. Independence makes covariance zero; positive covariance increases total variability, while negative covariance reduces it. A valid calculation must use the dependence assumption appropriate to the model.

Worked Example: Correlated Losses from Two Lines

A hypothetical insurer models the annual losses of two lines as a bivariate normal approximation. Line 1 has mean \$40,000 and standard deviation \$3,000; line 2 has mean \$20,000 and standard deviation \$4,000. Their correlation is $0.5$. Premium is \$100,000 and fixed expenses are \$10,000. Calculate a $1.5$-standard-deviation load and its percentage of premium. Use a one-year model with interest rate $0\%$; mortality is not applicable. The normal model is an approximation for aggregate results; only the stated means, standard deviations, and correlation are used below.

Method and covariance. The covariance is $\operatorname{Cov}(L_1,L_2)=0.5(\$3{,}000)(\$4{,}000)=\$6{,}000{,}000$. The variance of total losses is $\$3{,}000^2+\$4{,}000^2+2(\$6{,}000{,}000) =\$9{,}000{,}000+\$16{,}000{,}000+\$12{,}000{,}000 =\$37{,}000{,}000$.

Standard deviation and load. The total-loss standard deviation is $\sqrt{\$37{,}000{,}000}=\$6{,}082.7625\ldots$. Hence $C=1.5(\$6{,}082.7625\ldots)=\$9{,}124.1438\ldots$, or \$9,124.14 to the nearest cent.

Premium percentage and check. The load is $\$9{,}124.14/\$100{,}000=9.12414\ldots\%$, or 9.12% to two decimal places. Expected total losses are \$60,000, so the pre-load expected result is $\$100{,}000-\$10{,}000-\$60{,}000=\$30{,}000$. After the load it is \$39,124.14. The calculation avoids treating correlated line losses as independent.

Common mistake avoided. Adding the two standard deviations gives \$7,000, which is not the portfolio standard deviation. The covariance calculation gives approximately \$6,082.76. Adding standard deviations would be appropriate only in a different dependence situation, such as perfectly positively correlated deviations with the stated marginal standard deviations.

Choosing and Reporting the Load

A selected multiple should be reported alongside the risk measure and the premium base used to express the load. For example, “$k=1.2$ times the standard deviation of annual underwriting results” is more informative than reporting only a dollar amount. If the amount is shown as a rate provision, state whether it is divided by the original premium or by another stated base; the denominator affects the percentage.

The selected multiple has a straightforward effect: increasing it increases the contingency amount linearly. For the same modeled risk, doubling $k$ doubles the load. It does not alter the expected losses, their variance, or the model’s assumptions. A load should therefore not be presented as though it removes volatility, guarantees a profitable period, or compensates automatically for model error.

State what varies. Identify whether the standard deviation is for losses alone or for the full underwriting result.
Use the right aggregate variance. Account for claim frequency, severity, and dependencies where the model requires them.
Keep profit distinct. Expected underwriting profit and a variability-based contingency load are different provisions, even if both affect the premium.
Show units and basis. Report the load in dollars for the modeled period and identify the premium denominator for a percentage.

Key takeaway: A variability-based contingency load is $k\sigma_R$. The standard deviation measures modeled spread; the selected multiple scales the added cushion. Neither the multiple nor the load, by itself, guarantees protection against every adverse outcome.

Check Your Understanding

For each problem, use the stated model and report dollar loads to the nearest cent when needed.

A hypothetical book has 50 independent policies, each with claim probability $0.04$, and a fixed \$2,000 payment per claim. What is the load at $k=1$? Assume a one-period model with interest rate $0\%$; mortality is not applicable. (Answer: \$2,771.28)
A hypothetical portfolio has $N\sim\operatorname{Poisson}(25)$ claims, with independent severity equally likely to be \$500 or \$1,500. What is the contingency load at $k=0.5$? Assume interest rate $0\%$ and no mortality. (Answer: \$2,795.08)
Two independent hypothetical lines have loss standard deviations of \$9,000 and \$12,000. What is the total-loss standard deviation and the load at $k=1.1$? (Answer: \$15,000 standard deviation; \$16,500 load)
Two hypothetical aggregate loss components have standard deviations of \$8,000 and \$6,000, with correlation $-0.25$. Calculate the total standard deviation and a load at $k=0.75$. (Answer: \$8,717.80 standard deviation; \$6,538.35 load)
A hypothetical book has an underwriting-result standard deviation of \$20,000, and the selected multiple is $0.6$. If the original premium is \$250,000, calculate the load and its percentage of premium. (Answer: \$12,000; 4.80%)

What You'll Learn