Calculating Diversification Credit Across Risk Categories

Why Diversification Credit Is a Calculation, Not a Label

Rating capital models may assign separate charges to categories such as premium and reserve risk, asset risk, and catastrophe risk. A simple sum treats every category charge as if all adverse outcomes occur together at their full standalone levels. A covariance-style aggregation instead uses prescribed relationships among categories to calculate a combined charge. The difference between the standalone sum and the combined amount is the diversification credit.

The calculation is useful only when the charges and correlations are defined consistently. A category charge is a model output, not automatically a standard deviation. Thus, the formula below is a covariance-style aggregation rule: it has a variance-based justification when charges behave like scaled standard deviations, but its use in a rating model depends on that model’s definitions and prescribed inputs. This distinction is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.

Definition: Let $K_i$ be the standalone capital charge for risk category $i$, and let $\rho_{ij}$ be the prescribed correlation between categories $i$ and $j$. The covariance-style aggregate charge is $$ K_{\mathrm{agg}}=\sqrt{\sum_{i=1}^{n}\sum_{j=1}^{n}\rho_{ij}K_iK_j}, $$ where $\rho_{ii}=1$ and $\rho_{ij}=\rho_{ji}$. The dollar diversification credit is $$ D=\sum_{i=1}^{n}K_i-K_{\mathrm{agg}}. $$

All charges must use the same monetary units and measurement basis. The correlation matrix must be symmetric and mathematically valid; in particular, it should be positive semidefinite. Rating models may prescribe correlations rather than estimate them from a company’s historical data. Use the values and aggregation rule specified for the model being evaluated; do not substitute a different dependence assumption merely because it seems more conservative or more realistic.

Why the Formula Has Cross Terms

The variance identity explains the structure of the calculation. Suppose category losses $L_i$ have standard deviations $\sigma_i$, and their pairwise correlations are $\rho_{ij}$. Since $\operatorname{Cov}(L_i,L_j)=\rho_{ij}\sigma_i\sigma_j$, the variance of their sum is $$ \operatorname{Var}\left(\sum_{i=1}^{n}L_i\right) =\sum_{i=1}^{n}\sigma_i^2+2\sum_{i<j}\rho_{ij}\sigma_i\sigma_j. $$ The factor of 2 appears because each pair occurs twice in the double sum: once as $(i,j)$ and once as $(j,i)$.

If each category charge is proportional to its standard deviation by the same factor, say $K_i=c\sigma_i$, then multiplying the variance identity by $c^2$ and taking the square root gives the stated aggregation formula. In practice, a rating capital charge need not be a standard deviation or share a common scaling factor with other charges. The covariance-style rule is therefore a model convention unless those additional conditions hold. It should not be presented as a universal identity for tail-risk capital.

Confirm the inputs.
List each standalone charge and the relevant pairwise correlations. Check that the charges have consistent scope, units, and time basis.

Build the quadratic sum.
Include each diagonal term $K_i^2$ and each pairwise contribution twice, or include each distinct pair once with a factor of 2.

Take the square root.
The square root of the quadratic sum is the aggregate charge, not the quadratic sum itself.

Measure the credit.
Subtract the aggregate charge from the sum of standalone charges. Divide by that standalone sum only if a percentage credit is requested.

Two-Category Aggregation

For two categories, with charges $K_1$ and $K_2$ and correlation $\rho$, the general formula reduces to

K_{\mathrm{agg}}=\sqrt{K_1^2+K_2^2+2\rho K_1K_2}.

The standalone sum is $K_1+K_2$, so the dollar credit is that sum less $K_{\mathrm{agg}}$. When $\rho=1$ and both charges are nonnegative, the formula gives $K_{\mathrm{agg}}=K_1+K_2$, hence no credit. Lower correlations generally reduce the aggregate relative to the standalone sum. A negative correlation can produce a larger credit, if such a correlation is permitted by the model and the full correlation matrix is valid.

Worked Example: Calculating Credit for Two Categories

A hypothetical insurer has a premium-and-reserve charge of $80$ thousand dollars and an asset-risk charge of $120$ thousand dollars. The rating model specifies a correlation of $0.25$ between them. Find the aggregate charge, dollar credit, and credit as a percentage of the standalone total. Report monetary amounts in thousands of dollars to two decimal places.

Method. Apply the two-category covariance-style formula, keeping the correlation term and both standalone squares.

The standalone total is $$ 80+120=200 $$ thousand dollars. The quadratic sum is $$ 80^2+120^2+2(0.25)(80)(120) =6{,}400+14{,}400+4{,}800 =25{,}600. $$

Therefore, $$ K_{\mathrm{agg}}=\sqrt{25{,}600}=160.00 $$ thousand dollars. The dollar credit is $$ D=200-160=40.00 $$ thousand dollars, and the percentage credit is $$ \frac{40}{200}\times 100\%=20.00\%. $$ Check: $160+40=200$, so aggregate plus credit equals the standalone total. The calculation avoids the mistake of treating $0.25$ as the share of capital to subtract directly; correlation enters the quadratic sum.

More Than Two Categories

With three categories, the formula can be written in a way that makes the pair contributions visible:

K_{\mathrm{agg}}^2 =K_1^2+K_2^2+K_3^2 +2\rho_{12}K_1K_2 +2\rho_{13}K_1K_3 +2\rho_{23}K_2K_3.

Each correlation applies to one pair of categories. Do not apply a single correlation to every pair unless the model specifically assigns the same value to all pairs. For larger matrices, a reliable method is to calculate the diagonal terms and each distinct pair contribution in a table, then sum them before taking the square root.

Worked Example: Aggregating Three Charges With Different Correlations

A hypothetical insurer has three charges, all in thousands of dollars: $K_1=60$ for premium and reserve risk, $K_2=80$ for asset risk, and $K_3=100$ for catastrophe risk. The model specifies $\rho_{12}=0.20$, $\rho_{13}=0.10$, and $\rho_{23}=0.40$. Calculate aggregate capital and the diversification credit as a percentage of the standalone total. Report monetary amounts to two decimals.

Method. Add the three squared charges and twice each correlation-weighted pair product. The standalone total is $$ 60+80+100=240 $$ thousand dollars.

The diagonal terms sum to $$ 60^2+80^2+100^2=3{,}600+6{,}400+10{,}000=20{,}000. $$ The three pair contributions are $$ 2(0.20)(60)(80)=1{,}920,\qquad 2(0.10)(60)(100)=1{,}200, $$ and $$ 2(0.40)(80)(100)=6{,}400. $$ Thus, the quadratic sum and aggregate charge are $$ 20{,}000+1{,}920+1{,}200+6{,}400=29{,}520, \qquad K_{\mathrm{agg}}=\sqrt{29{,}520}=171.81. $$

The dollar credit is $240-171.81=68.19$ thousand dollars, and the percentage credit is $$ \frac{68.19}{240}\times 100\%=28.41\%. $$ Check: $171.81+68.19=240.00$ thousand dollars, to the stated precision. The distinct pair correlations were used separately rather than replaced by an average correlation.

Worked Example: Inferring a Pairwise Correlation

For two hypothetical risk categories, the standalone charges are $90$ thousand dollars and $150$ thousand dollars. A model calculation reports an aggregate charge of $190$ thousand dollars under the two-category formula. Infer the correlation and calculate the dollar diversification credit. Report the correlation to four decimal places and monetary amounts to two decimals.

Method. Square the reported aggregate, substitute the known charges, and solve the formula for the unknown correlation.

Starting from $$ 190^2=90^2+150^2+2\rho(90)(150), $$ the left side is $36{,}100$, while the squared charges sum to $$ 90^2+150^2=8{,}100+22{,}500=30{,}600. $$ The cross-term coefficient is $2(90)(150)=27{,}000$. Therefore, $$ 36{,}100=30{,}600+27{,}000\rho, \qquad \rho=\frac{5{,}500}{27{,}000}=0.2037. $$ The standalone total is $90+150=240$ thousand dollars, so the credit is $240-190=50.00$ thousand dollars.

Check: substituting the unrounded correlation gives a quadratic sum of $36{,}100$, whose square root is $190$. The inferred correlation lies between $-1$ and $1$, as required for a pairwise correlation.

Interpreting the Diversification Credit

The dollar credit is an arithmetic comparison between the sum of standalone charges and the modeled aggregate charge. It is not itself an additional category of capital, and it does not imply that the insurer has actually reduced its exposure by that amount. Its size depends on the category definitions, standalone charges, and prescribed correlation inputs. If those inputs change, both the aggregate and the credit may change.

A percentage credit is $$ \frac{D}{\sum_i K_i}\times 100\%, $$ provided the standalone total is positive. State the denominator when reporting a percentage: a dollar credit and a percentage credit answer different questions. Also retain unrounded intermediate values when calculating the credit percentage; rounding the aggregate too early can change the final result.

Common Calculation Traps

Omitting the factor of 2. In a three-category calculation, each distinct pair contributes $2\rho_{ij}K_iK_j$. Counting it only once understates the quadratic sum when correlations and charges are positive.
Stopping before the square root. The double sum is the squared aggregate charge. It is not itself the capital charge.
Confusing correlation with credit. A correlation is an input to the aggregation formula, not a percentage reduction applied directly to the standalone total.
Mixing units or bases. If one charge is in dollars and another in thousands, convert them before aggregation. Likewise, charges with incompatible scope or measurement bases should not be combined without a model rule that makes them comparable.
Calling every result a diversification benefit. The difference is meaningful only relative to the model’s specified standalone and aggregate calculations. Interpret it as model output, not as a universal economic measure.

Key takeaway: Calculate the full quadratic sum using every prescribed pairwise correlation, take its square root to obtain aggregate capital, and subtract that result from the standalone total to quantify the dollar diversification credit. Interpret the output in the context of the model’s definitions and assumptions.

Check Your Understanding

For each problem, use the covariance-style formula as stated. Monetary values are in thousands of dollars unless specified otherwise.

Two categories have charges of $40$ and $70$, with correlation $0.25$. Find the aggregate charge and dollar diversification credit, rounded to two decimals. (Answer: aggregate $88.88$; credit $21.12$.)
Charges are $50$ and $90$, with correlation $0.30$. Find the aggregate charge and the credit as a percentage of the standalone total, rounding monetary amounts to two decimals and the percentage to two decimals. (Answer: aggregate $115.33$; percentage credit $17.62\%$.)
Three charges are $30$, $50$, and $70$, with pairwise correlations $\rho_{12}=0.20$, $\rho_{13}=0$, and $\rho_{23}=0.50$. Find the aggregate charge and dollar credit, to two decimals. (Answer: aggregate $111.36$; credit $38.64$.)
Two categories have charges $80$ and $120$ and correlation $0.50$. Calculate the aggregate charge and percentage credit relative to the standalone total, to two decimals. (Answer: aggregate $174.36$; percentage credit $12.82\%$.)
Two charges are $40$ and $60$, and their aggregate under the formula is $70$. Infer their correlation to four decimals, then calculate the dollar credit. (Answer: correlation $-0.0625$; credit $30.00$.)

What You'll Learn