How Capital Factors And Model Assumptions Shape Results

Why a Capital Result Depends on Its Inputs

A factor-based capital result is not a stand-alone property of an insurer. It is conditional on the risk categories included, how exposures are defined, which factors are applied, and how the resulting charges are aggregated. Two calculations can use the same arithmetic and still produce different capital results because one of those inputs differs.

This tutorial traces those dependencies in a compact hypothetical model. All monetary amounts in the examples are in thousands of dollars ($000). The factors and exposures are illustrative, not data or factors from a particular rating agency. The aim is to explain how assumptions flow through a calculation, not to reproduce any published rating model or determine a rating.

Definition: For risk category $i$, let $E_i$ be the exposure measured on the model's specified basis and let $q_i$ be the factor applied to that exposure. The category charge is $K_i=q_iE_i$. An exposure basis includes the definition, measurement date, scope, and units of the quantity to which the factor applies. A factor is meaningful only together with that basis and the model's calibration assumptions.

For example, a premium factor might apply to a specified measure of premium, while a reserve factor might apply to a defined reserve balance. The factor is not an independent estimate of capital: it translates a particular exposure measure into a charge under a particular model. Before comparing results, establish whether the categories, exposure definitions, factor calibrations, and included adjustments are consistent.

A Compact Illustrative Calculation

Assume a hypothetical model calculates charges for premium risk, reserve risk, and asset risk, then aggregates them using a correlation formula. The model inputs are:

Premium exposure: $E_P=80{,}000$, with factor $q_P=8\%$.
Reserve exposure: $E_R=120{,}000$, with factor $q_R=6\%$.
Asset exposure: $E_A=50{,}000$, with factor $q_A=4\%$.

The model uses correlations $\rho_{PR}=0.25$, $\rho_{PA}=0.10$, and $\rho_{RA}=0.15$. In this example, the correlation matrix is assumed to be valid, and the model's aggregation rule is

C=\sqrt{K_P^2+K_R^2+K_A^2+ 2\rho_{PR}K_PK_R+ 2\rho_{PA}K_PK_A+ 2\rho_{RA}K_RK_A}.

This is an illustrative model specification, not a universal rule for every rating-capital model. It applies here because the hypothetical model defines its charges and aggregation in this way. In a real application, use the stated model's own definitions and aggregation rules.

Worked Example: Establishing a Baseline Before Changing Assumptions

Calculate the three category charges and the aggregate capital for the hypothetical insurer. Identify the calculation method and check the result.

Method: calculate each factor charge first, then apply the specified aggregation rule. The premium charge is $K_P=0.08(80{,}000)=6{,}400$. The reserve charge is $K_R=0.06(120{,}000)=7{,}200$. The asset charge is $K_A=0.04(50{,}000)=2{,}000$. All three charges are in $000.

Substitute those charges and the three correlations into the model:

\begin{aligned} C^2={}&6{,}400^2+7{,}200^2+2{,}000^2\\ &+2(0.25)(6{,}400)(7{,}200) +2(0.10)(6{,}400)(2{,}000)\\ &+2(0.15)(7{,}200)(2{,}000)\\ ={}&40{,}960{,}000+51{,}840{,}000+4{,}000{,}000\\ &+23{,}040{,}000+2{,}560{,}000+4{,}320{,}000\\ ={}&126{,}720{,}000. \end{aligned}

Therefore $C=\sqrt{126{,}720{,}000}=11{,}256.998\ldots$ $000, or $11,257,000 to the nearest thousand dollars. As a check, the aggregate exceeds the largest individual charge, $7,200,000, but is less than the sum of the three charges, $15,600,000, which is reasonable for these positive correlations and this aggregation formula.

Common mistake avoided: adding the three charges gives $15,600,000, but that is not the specified aggregate. The model requires the squared charges and correlation terms to be combined before taking the square root.

Why the Aggregation Formula Has These Terms

The correlation terms matter because a model that allows for dependence does not generally treat category charges as if they occur independently. One way to see the structure is to consider a model in which each charge $K_i$ is treated as a scale parameter analogous to a standard deviation and the aggregation formula is based on the variance of a weighted sum. For two components, the variance identity is $\operatorname{Var}(X+Y)=\operatorname{Var}(X)+\operatorname{Var}(Y)+2\operatorname{Cov}(X,Y)$. If the component scales are $K_i$ and $K_j$, and their correlation is $\rho_{ij}$, the covariance term is represented as $\rho_{ij}K_iK_j$. Adding all component variances and pairwise covariance terms gives the expression under the square root.

That reasoning explains why a correlation term is multiplied by two in the expanded expression: each unordered pair contributes the covariance term to the variance of the sum. It does not establish that every rating model's charges are actual standard deviations or that every model uses this aggregation. Those are model assumptions, which must be checked rather than inferred from the formula.

Key takeaway: A change in aggregate capital can originate in a change to an exposure, a factor, or an aggregation assumption. Keep those sources separate when explaining a result.

Factor Calibration and Exposure Definition

A factor change and an exposure change are different model changes, even though both change $K_i=q_iE_i$. A factor change may represent a different calibration or stress severity. An exposure change may instead reflect a different business perimeter, data date, accounting definition, or gross-versus-net basis. If both change together, the resulting charge alone does not reveal which change caused the difference.

Worked Example: Isolating the Effect of a Factor Calibration

Suppose the premium exposure remains $80,000,000, but the hypothetical premium factor is revised from 8% to 9%. The reserve and asset charges and all correlations remain at their baseline values. Calculate the revised aggregate and compare it with the baseline.

Method: hold all inputs fixed except the factor being tested. The revised premium charge is $0.09(80{,}000)=7{,}200$ $000. The other charges remain $7{,}200$ and $2{,}000$ $000.

The revised squared aggregate is

\begin{aligned} C_{\mathrm{new}}^2={}&7{,}200^2+7{,}200^2+2{,}000^2\\ &+2(0.25)(7{,}200)(7{,}200) +2(0.10)(7{,}200)(2{,}000)\\ &+2(0.15)(7{,}200)(2{,}000)\\ ={}&51{,}840{,}000+51{,}840{,}000+4{,}000{,}000\\ &+25{,}920{,}000+2{,}880{,}000+4{,}320{,}000\\ ={}&140{,}800{,}000. \end{aligned}

Thus $C_{\mathrm{new}}=\sqrt{140{,}800{,}000}=11{,}865.918\ldots$ $000, or $11,866,000 to the nearest thousand dollars. Relative to the baseline of $11,257,000, the increase is about $609,000. The factor rises by one percentage point, but aggregate capital does not rise by one percentage point: the factor acts on one exposure, and the changed charge then enters the aggregation formula.

Check: the premium charge rises by $800,000, from $6,400,000 to $7,200,000. The aggregate rises by less than that amount, which is consistent with the specified formula. The comparison is valid because the exposure basis, other charges, and correlations were held fixed.

The local effect of a factor change can also be described analytically. Write the aggregate as $C=\sqrt{Q}$, where $Q$ is the expression inside the square root. For a particular category $i$, differentiating $C^2=Q$ with respect to $K_i$, while holding the other charges and correlations fixed, gives

2C\frac{\partial C}{\partial K_i} = 2K_i+2\sum_{j\ne i}\rho_{ij}K_j, \qquad \frac{\partial C}{\partial q_i} = E_i\frac{K_i+\sum_{j\ne i}\rho_{ij}K_j}{C}.

The second equality uses $K_i=q_iE_i$ and holds when $C>0$ and the exposure $E_i$ is fixed. It is a local sensitivity: for a finite factor change, recompute the aggregate for an exact result. For the baseline premium category, the charge-weighted numerator is $6{,}400+0.25(7{,}200)+0.10(2{,}000)=8{,}400$ $000. Thus $\partial C/\partial K_P=8{,}400/11{,}256.998\ldots\approx0.7462$. The formula helps explain why a change in one category's charge need not pass through one-for-one to aggregate capital.

Data and Perimeter Assumptions

Exposure data have to match the factor's intended definition. A reserve total might change because a data extract includes an additional reserve component, uses a different valuation date, or applies a different gross or net convention. It is not enough to say that the factor stayed constant: the resulting charge is comparable only if the exposure has the same scope and meaning.

Worked Example: Revising the Reserve Exposure After a Data Reconciliation

Assume a data reconciliation changes the reserve exposure from $120,000,000 to $132,000,000 because a reserve component previously excluded from the hypothetical model's exposure definition is now included. The reserve factor remains 6%; the other charges and all correlations remain at baseline. Calculate the revised reserve charge and aggregate, and compare them with baseline.

Method: apply the unchanged factor to the corrected exposure, then reaggregate. The revised reserve charge is $0.06(132{,}000)=7{,}920$ $000, an increase of $7{,}920-7{,}200=720$ $000.

\begin{aligned} C_{\mathrm{new}}^2={}&6{,}400^2+7{,}920^2+2{,}000^2\\ &+2(0.25)(6{,}400)(7{,}920) +2(0.10)(6{,}400)(2{,}000)\\ &+2(0.15)(7{,}920)(2{,}000)\\ ={}&40{,}960{,}000+62{,}726{,}400+4{,}000{,}000\\ &+25{,}344{,}000+2{,}560{,}000+4{,}752{,}000\\ ={}&140{,}342{,}400. \end{aligned}

Therefore $C_{\mathrm{new}}=\sqrt{140{,}342{,}400}=11{,}846.620\ldots$ $000, or $11,847,000 to the nearest thousand dollars. Compared with the baseline, the increase is about $590,000. The reserve charge rose by $720,000, while the aggregate rose by less. The result reflects both the corrected exposure and the model's aggregation rule.

Check: the factor remains 6%, and $6\%$ of the revised $132,000,000 exposure is $7,920,000. If the extra reserve component were outside the factor's intended definition, including it would not be a valid correction; the exposure scope must be settled before calculating.

Aggregation Assumptions Also Matter

Even when exposures and factors do not change, the model's dependence assumptions can change the result. The following hypothetical comparison alters only one correlation. It is not a new estimate of the insurer's exposures; it shows the effect of a different aggregation assumption under the same charges.

Worked Example: Changing One Correlation Assumption

Use the baseline charges of $6,400,000, $7,200,000, and $2,000,000. Suppose the assumed premium-reserve correlation changes from 0.25 to 0.50, with the other correlations unchanged. Calculate the revised aggregate and its change from baseline.

Method: isolate the changed term. Only the premium-reserve term changes, by $2(0.50-0.25)(6{,}400)(7{,}200)=23{,}040{,}000$ in squared $000 units. The baseline squared aggregate was $126{,}720{,}000$, so the revised squared aggregate is $126{,}720{,}000+23{,}040{,}000=149{,}760{,}000$.

The revised aggregate is $\sqrt{149{,}760{,}000}=12{,}237.646\ldots$ $000, or $12,238,000 to the nearest thousand dollars. The increase from the baseline is about $981,000. The increase is not calculated by multiplying the old aggregate by the correlation change; the correlation changes a cross term inside the square root.

Check: the revised squared aggregate exceeds the baseline squared aggregate by exactly the additional correlation term, $23,040,000 in squared $000 units. The revised correlation matrix remains valid for these illustrative values.

Exam Traps and Model Comparison

A useful explanation of a capital movement identifies the input that changed and traces its effect through the calculation. Avoid presenting a changed result as evidence that the insurer's underlying risk necessarily changed: a new data definition or calibration can change modelled capital even if the underlying portfolio is otherwise unchanged.

Applying a factor to the wrong base. A factor calibrated for one exposure definition cannot automatically be applied to a differently defined amount. Confirm scope, timing, and gross or net treatment first.
Changing several assumptions without attribution. If exposure, factor, and correlation all change, calculate controlled comparisons that change one input at a time before describing the drivers.
Dropping the correlation terms. In a model that specifies correlation aggregation, adding charges or taking the square root of only the sum of squares does not implement the model.
Using a local sensitivity as an exact finite-change result. A derivative describes a local rate of change. Recalculate the full formula when a factor or exposure changes materially.
Confusing modelled capital with a rating conclusion. A capital calculation is one input to a broader assessment. Its interpretation depends on the relevant model and assessment framework.

Key takeaway: A defensible capital comparison records the exposure basis, factor, calibration context, aggregation rule, and data scope for each result. If one changes, show where it enters the calculation and distinguish that model effect from a change in underlying risk.

Check Your Understanding

For each hypothetical problem, use the stated model assumptions and keep monetary amounts in $000 unless indicated otherwise.

A model uses a premium exposure of 45,000 at a 6% factor, a reserve exposure of 70,000 at a 4% factor, and an asset exposure of 30,000 at a 3% factor. Calculate each category charge. (Answer: 2,700; 2,800; and 900 $000, respectively.)
Using charges $K_1=2{,}700$, $K_2=2{,}800$, and $K_3=900$, a model uses correlations $\rho_{12}=0.20$, $\rho_{13}=0.10$, and $\rho_{23}=0.30$ in the stated square-root aggregation formula. Calculate the aggregate to the nearest $000. (Answer: $\sqrt{20{,}962{,}000}=4{,}578.4\ldots$, or $4,578,000.)
A factor rises from 5% to 6% on an unchanged exposure of 40,000. Calculate the old and new category charges and their difference. (Answer: 2,000 and 2,400 $000; the charge increases by 400 $000.)
A reserve exposure is corrected from 60,000 to 66,000 at an unchanged factor of 4%. Aggregate the reserve charge with a second charge of 1,800 using correlation 0.25 and the two-risk formula $C=\sqrt{K_1^2+K_2^2+2\rho K_1K_2}$. Give the revised aggregate to the nearest $000. (Answer: the revised reserve charge is 2,640; $C=\sqrt{12{,}585{,}600}=3{,}547.6\ldots$, or $3,548,000.)
For baseline charges of 6,400, 7,200, and 2,000, the premium-reserve correlation increases from 0.25 to 0.50 while all other inputs stay fixed. Find the revised aggregate and its increase from baseline, to the nearest $000. (Answer: revised aggregate $12,238,000; increase approximately $981,000.)

What You'll Learn