Aggregating Risk Charges With Covariance Formulas

Why Aggregate Charges With a Covariance Formula?

A regulatory capital calculation may first produce charges for distinct risk categories and then combine them using a prescribed covariance formula. The aggregate is generally not the arithmetic sum: the formula recognizes specified relationships among the categories. The calculation is mechanical only after the inputs are identified correctly. A charge, a correlation, and a separate formula adjustment are different things and must remain distinct throughout the work.

The method below uses hypothetical charges and prescribed correlations. It is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus, but it does not assert that one formula or one set of correlations applies in every regulatory regime. In practice, follow the applicable rule’s definitions, scope, and order of operations.

Definition: Let $K_i$ be the nonnegative capital charge for risk category $i$, and let $\rho_{ij}$ be the prescribed correlation parameter for categories $i$ and $j$. For $n$ categories, the covariance aggregate is $$ K_{\mathrm{agg}}=\sqrt{\sum_{i=1}^{n}K_i^2+2\sum_{i<j}\rho_{ij}K_iK_j}. $$ The $K_i$ values and the $\rho_{ij}$ values are inputs to this calculation. Any additional charge, cap, floor, scaling factor, tax adjustment, or other prescribed modification is applied separately, in the order specified by the applicable formula.

Inputs, Scope, and Assumptions

Before calculating, establish which category charges belong in the aggregation and which relationships the rule prescribes. A charge might already incorporate specified exposure measures or risk factors. Do not recompute it or include an omitted category unless the instructions require that. Likewise, do not substitute an empirical correlation estimate for a prescribed parameter.

The formula is a prescribed aggregation rule. It does not, by itself, say that each charge is an observed standard deviation of a random loss. If a model does interpret the charges as standard deviations, the usual mathematical interpretation requires centered category losses with those standard deviations and the stated correlations. A regulatory formula may instead use the same algebra as a convention without asserting that probabilistic model.

For a valid correlation matrix, each $\rho_{ij}$ lies between $-1$ and $1$, the diagonal entries are $1$, and the matrix is positive semidefinite. These conditions ensure that the corresponding quadratic form is nonnegative for every vector of charges. Prescribed matrices should be used as supplied; if a problem asks you to assess a proposed matrix, checking that every pairwise correlation is in range is necessary but not sufficient for validity.

The charges must use a common unit before aggregation. If one category is stated in dollars and another in thousands of dollars, convert one of them first. Since the formula squares charges and then takes a square root, the final aggregate has the same monetary unit as the input charges.

How the Covariance Terms Enter

The diagonal contributions are the squared charges, $K_i^2$. Each distinct pair contributes one cross term, $2\rho_{ij}K_iK_j$. The factor of two is needed because, when a variance of a sum is expanded, both ordered covariance terms occur: the covariance of category $i$ with $j$, and the covariance of $j$ with $i$. The formula sums over $i<j$ so each pair is listed once, then multiplies that sum by two.

\operatorname{Var}\left(\sum_{i=1}^{n}X_i\right) =\sum_{i=1}^{n}\operatorname{Var}(X_i) +2\sum_{i<j}\operatorname{Cov}(X_i,X_j)

For a short justification of the covariance form, suppose, just for this interpretation, that centered random variables $Z_i$ each have variance $1$, and set $X_i=K_iZ_i$. Then $\operatorname{Var}(X_i)=K_i^2$ and $\operatorname{Cov}(X_i,X_j)=K_iK_j\rho_{ij}$. Substituting these into the variance-of-a-sum identity gives the expression under the square root. The square root is the standard deviation of the sum in that model. When applying a regulatory formula, however, use the prescribed aggregation as written; do not infer that the underlying charges are actual standard deviations unless the model says so.

List the charges.
Record each included category charge with its unit. Convert all amounts to a common unit.

List the pairwise parameters.
For every distinct pair, record the prescribed $\rho_{ij}$. Do not omit a pair merely because its correlation is zero.

Calculate diagonal and cross terms.
Square every charge, calculate each $2\rho_{ij}K_iK_j$, and show the terms separately.

Take the square root and check.
Add the terms, verify the radicand is nonnegative, and take the nonnegative square root. Round only as instructed.

Apply separate adjustments.
Apply any stated addition, subtraction, cap, floor, or multiplier after the covariance subtotal only if that is the specified order.

Worked Calculations

Worked Example: Two Charges and One Prescribed Correlation

A hypothetical insurer has two category charges of $12$ thousand dollars and $16$ thousand dollars. The prescribed correlation is $0.25$. Calculate the covariance aggregate in thousands of dollars, to two decimal places. No separate formula adjustment applies.

Method. Use the two-category formula, which has one diagonal term for each charge and one doubled cross term. This is a deterministic prescribed-charge calculation: no interest rate, mortality assumption, or loss distribution is needed or assumed. The amounts and correlation are the complete inputs.

K_{\mathrm{agg}} =\sqrt{12^2+16^2+2(0.25)(12)(16)}

Step 1: Square the charges. $12^2=144$, and $16^2=256$, so the diagonal total is $144+256=400$.

Step 2: Calculate the cross term. The charge product is $12(16)=192$. The doubled correlation-weighted term is $2(0.25)(192)=0.5(192)=96$.

Step 3: Add and take the square root. The radicand is $400+96=496$. Thus $K_{\mathrm{agg}}=\sqrt{496}=22.271057\ldots$ thousand dollars, which rounds to $\$22.27$ thousand.

Check and common mistake avoided. The result is greater than either individual charge and less than their sum, $28$, as expected for this positive, less-than-one correlation. Omitting the factor of two would give $\sqrt{448}$, which is not the prescribed calculation. The correct doubled cross term is $96$.

Worked Example: Three Categories and a Separate Add-On

A hypothetical regulatory formula combines three charges, in thousands of dollars: property $10$, liability $20$, and credit $15$. Its prescribed correlations are $\rho_{\text{property, liability}}=0.20$, $\rho_{\text{property, credit}}=0.10$, and $\rho_{\text{liability, credit}}=0.40$. After aggregation, the rule adds a separate $3.50$ thousand dollar charge. Find the final amount to two decimal places.

Method. First compute the covariance aggregate using all three distinct pairs. Then add the stated $3.50$ thousand dollar adjustment. As in the prior example, this formula calculation has no interest rate, mortality assumption, or probability distribution input; the category charges and prescribed correlations are given.

Step 1: Sum the squared charges.

$$ 10^2+20^2+15^2=100+400+225=725 $$

Step 2: Calculate every doubled cross term. Property and liability contribute $2(0.20)(10)(20)=80$. Property and credit contribute $2(0.10)(10)(15)=30$. Liability and credit contribute $2(0.40)(20)(15)=240$. The total cross-term contribution is $80+30+240=350$.

Step 3: Calculate the covariance subtotal. The radicand is $725+350=1{,}075$, so the aggregate is $\sqrt{1{,}075}=32.787192\ldots$ thousand dollars.

Step 4: Apply the separate add-on. Add $3.50$: $32.787192\ldots+3.50=36.287192\ldots$ thousand dollars. Rounded to two decimal places, the final amount is $\$36.29$ thousand.

Check and common mistake avoided. The add-on is applied to the square-root result, not inserted inside the radicand. Adding $3.50$ to the radicand would change the formula and would not represent the stated adjustment.

Worked Example: Four Charges, Common Correlation, and Ordered Adjustments

A hypothetical formula has four category charges, in thousands of dollars: $8$, $12$, $10$, and $6$. It prescribes a correlation of $0.30$ for every distinct pair. After finding the covariance aggregate, the formula applies a $5\%$ multiplier and then adds $1.20$ thousand dollars. Calculate the final amount to two decimal places.

Method. There are four diagonal terms and six distinct pairs. Calculate the diagonal sum and the sum of the six charge products, use the common correlation for each cross term, and then apply the adjustments in their stated order. No interest rate, mortality assumption, or distributional parameters are part of this prescribed calculation.

Step 1: Calculate the diagonal sum.

$$ 8^2+12^2+10^2+6^2=64+144+100+36=344 $$

Step 2: List and sum all six pair products. The products are $8(12)=96$, $8(10)=80$, $8(6)=48$, $12(10)=120$, $12(6)=72$, and $10(6)=60$. Their sum is $96+80+48+120+72+60=476$.

Step 3: Calculate the cross-term total and aggregate. Each product is multiplied by $2(0.30)=0.60$, so the total cross contribution is $0.60(476)=285.60$. The radicand is $344+285.60=629.60$, and the covariance aggregate is $\sqrt{629.60}=25.091831\ldots$ thousand dollars.

Step 4: Apply the multiplier, then the fixed add-on. The $5\%$ multiplier gives $1.05(25.091831\ldots)=26.346423\ldots$. Adding $1.20$ gives $27.546423\ldots$ thousand dollars. The final amount, rounded to two decimal places, is $\$27.55$ thousand.

Check and common mistake avoided. Four categories produce six pairs, not four or twelve, when each pair is counted once before doubling. The final amount is not obtained by multiplying each category charge by $1.05$ before aggregation unless the rule specifically directs that order.

Interpretation and Exam Traps

A positive correlation increases the radicand relative to the sum of squared charges, while a negative correlation decreases it. A zero correlation removes that pair’s cross contribution, but it does not remove either category’s squared charge. The aggregate is not automatically the sum of charges, nor is it automatically bounded by the largest charge: its value depends on the full set of charges and prescribed parameters.

Keep a calculation table or equivalent written layout so each pair is visible. For $n$ categories, there are $n(n-1)/2$ distinct pairs. For example, four categories have $4(3)/2=6$ pairs. If all charges are nonnegative and all correlations are nonnegative, the radicand cannot be less than the diagonal sum; that is a useful arithmetic check. If prescribed correlations include negative values, do not use that check without accounting for their effects.

Forgetting the factor of two. The formula written as a sum over $i<j$ requires $2\rho_{ij}K_iK_j$ for each pair.
Omitting a pair. With three or more categories, every distinct pair contributes, even if its prescribed correlation is zero.
Mixing units. Convert all charges to the same unit before squaring or multiplying them.
Combining formula stages. Keep a covariance subtotal separate from any explicitly stated add-on, cap, floor, or multiplier, and follow the specified order.
Over-interpreting the result. A prescribed correlation formula is not evidence by itself that the charges are empirical standard deviations or that the formula captures every dependence feature.

For a final reasonableness check, verify that every pair appears once in the written pair list, each cross term uses the correct two charges and parameter, the radicand is nonnegative, and the result has the same unit as the charges. Then check whether the separate adjustments were applied exactly as stated.

Check Your Understanding

For each problem, retain the stated units and round the final amount to two decimal places when needed. Treat all inputs as hypothetical prescribed formula values; no interest, mortality, or loss-distribution assumptions are required.

Two category charges are $9$ and $12$ thousand dollars, with prescribed correlation $0.50$. Find their covariance aggregate. (Answer: $\$18.25$ thousand)
Charges of $5$ and $12$ thousand dollars have prescribed correlation $0$. A separate $0.75$ thousand dollar add-on is applied after aggregation. Find the final amount. (Answer: $\$13.75$ thousand)
Three charges are $10$, $15$, and $20$ thousand dollars. The correlations for pairs $(1,2)$, $(1,3)$, and $(2,3)$ are $0.20$, $0$, and $0.10$, respectively. Find the covariance aggregate. (Answer: $\$29.07$ thousand)
Four charges are $3$, $4$, $6$, and $8$ thousand dollars, and every distinct pair has prescribed correlation $0.20$. A separate $0.50$ thousand dollar add-on follows aggregation. Find the final amount. (Answer: $\$14.22$ thousand)
Charges of $7$ and $24$ thousand dollars have prescribed correlation $0.25$. The rule multiplies the covariance aggregate by $1.04$, then adds $2.00$ thousand dollars. Find the final amount. (Answer: $\$29.69$ thousand)

What You'll Learn