Why the Choice of Capital Method Matters
A regulatory capital method is not just a calculation. It determines which features of an insurer's risk profile affect the capital requirement, how much information the insurer must supply, and how easily an outside reader can understand or compare the result. A method that is straightforward to apply may be less responsive to a particular insurer's risks; a more tailored method may require substantially more data, controls, and validation.
This tutorial compares three broad approaches: factor-based methods, formula-based methods, and internal models. The labels are not used identically in every regulatory system. Here, a factor-based method means a relatively direct charge applied to a measured exposure; a formula-based method means a prescribed structure combining defined risk components and rules; and an internal model means a company-developed model used to estimate capital under regulatory requirements. These are useful distinctions for comparison, not universal legal definitions.
The methods should be compared on a consistent basis. Before interpreting different capital outputs, check that they refer to the same valuation date, risk scope, gross or net basis, capital definition, confidence level, and time horizon. A difference in results may reflect a different method, but it may instead reflect a different basis of measurement.
What Each Approach Captures
| Dimension | Factor-based | Formula-based | Internal model |
|---|---|---|---|
| Basic structure | Exposure multiplied by prescribed factor | Prescribed modules, calculations, and aggregation rules | Company-developed representation of risks and capital |
| Risk sensitivity | Usually responsive to the exposure measure and factor | Responsive to the inputs and distinctions recognized by the formula | Potentially responsive to company-specific exposures and modeled relationships |
| Data demand | Often relatively limited | Moderate; depends on the required inputs and modules | Often high, including data for calibration, testing, and ongoing monitoring |
| Comparability | Often easier to compare when definitions and factors match | Often comparable under common prescribed rules | May be harder to compare because models and assumptions can differ |
| Important limitation | A factor can miss risk differences not reflected in the exposure measure | A prescribed structure may not fit every company's risk profile | Detailed modeling does not eliminate estimation, implementation, or governance risk |
These are tendencies, not guarantees. A well-designed formula can distinguish risks more carefully than a crude factor, while a poorly specified internal model can be less reliable than a transparent prescribed calculation. Model complexity is not evidence by itself that a capital estimate is accurate.
Worked Example: Sensitivity of a Factor Charge
Worked Example: Sensitivity of a Factor Charge
A hypothetical insurer has $12 million of annual premium exposure subject to a prescribed factor of 8% and $30 million of reserve exposure subject to a factor of 3%. The charges are additive in this simplified example. Calculate the total charge, then determine its change if annual premium exposure rises to $15 million while reserve exposure is unchanged. Explain what a change in loss mix alone would do if neither exposure measure changed.
Method. Apply each factor to its stated exposure, then add the charges. This example assumes the factor definitions and exposures are applicable as stated and that no additional adjustment or aggregation rule applies.
Premium charge before the change:
Reserve charge before the change:
The initial total is \(960{,}000+900{,}000=\$1{,}860{,}000\). After premium exposure rises, the premium charge is \(15{,}000{,}000 \times 0.08=\$1{,}200{,}000\). The reserve charge remains \(\$900{,}000\), so the new total is \(1{,}200{,}000+900{,}000=\$2{,}100{,}000\). The increase is \(\$2{,}100{,}000-\$1{,}860{,}000=\$240{,}000\).
If the loss mix changes but both measured exposures remain unchanged, this factor calculation remains \(\$1{,}860{,}000\). The result follows because the calculation uses only the stated exposures and factors. The method avoids representing any risk feature that is not reflected in those inputs.
Check. The increase also equals the premium exposure increase, \(\$3{,}000{,}000\), multiplied by 8%: \(3{,}000{,}000 \times 0.08=\$240{,}000\). A common mistake is to apply 8% to the new total of premium and reserves; that incorrectly uses one factor for two differently rated exposure bases.
How the Approaches Differ in Practice
A factor-based method is easiest to interpret when the exposure measure is well defined and the factor is appropriate for the business being measured. It can be operationally efficient and relatively easy to reproduce. Its central limitation is structural: if two portfolios have the same measured exposure but meaningfully different risk characteristics, a single factor may give them the same charge. The result may be insensitive to changes that matter but are absent from the formula's inputs.
A formula-based method can recognize more distinctions. It may have separate components for different risk categories, prescribed stress calculations, and specified rules for combining results. This structure can improve consistency across companies while allowing more risk detail than a single exposure factor. It remains prescribed: a company does not ordinarily replace the rule simply because its own analysis suggests a different result. The formula's transparency and comparability depend on common definitions, implementation, and disclosure of its inputs.
An internal model can be designed to reflect company-specific exposures, business practices, and risk relationships. It may therefore be more sensitive to a change that a standard factor or formula does not distinguish. That potential advantage comes with demands: suitable data, credible calibration, technical implementation, controls, documentation, validation, and ongoing review. Internal-model output can also be sensitive to assumptions and model choices. A tailored result is not automatically more accurate or more comparable.
For this comparison, risk sensitivity means how a capital result responds to a change in exposure or risk characteristics. Comparability means how readily results can be compared across companies on a genuinely consistent basis. Neither term is a claim that one approach is universally superior. The appropriate evaluation depends on the purpose of the capital measure and the risks the method is intended to capture.
Worked Example: Comparing Capital Results on a Common Basis
Worked Example: Comparing Capital Results on a Common Basis
A hypothetical insurer calculates one-year capital on a consistent net-loss basis using three methods. A simple factor method produces a total of $2.00 million. A prescribed formula has three risk-module charges of $1.20 million, $0.80 million, and $0.60 million, with prescribed pairwise correlations of 0.25, 0.10, and 0.20, respectively. For this example, its aggregation rule is the square root of the sum of squared charges plus twice each prescribed correlation times the corresponding product of charges. An internal model uses the following discrete annual loss distribution: $0.8 million with probability 0.900, $2.0 million with probability 0.070, $4.0 million with probability 0.025, and $8.0 million with probability 0.005. Its capital measure is the smallest loss whose cumulative probability is at least 99.5%. Compare the three results.
Method. Calculate each result under its stated rule. The example stipulates a common horizon and net-loss basis, but the methods still use different structures and assumptions; the outputs should not be treated as interchangeable without checking their definitions.
For the prescribed formula, the sum of squared module charges, in millions of dollars squared, is:
The three correlation terms are:
Thus the total under the prescribed formula is:
For the internal model, cumulative probability is 0.900 at $0.8 million, 0.970 at $2.0 million, 0.995 at $4.0 million, and 1.000 at $8.0 million. Therefore, the smallest loss whose cumulative probability reaches at least 0.995 is $4.00 million.
The three results are $2.00 million, approximately $1.804 million, and $4.00 million, respectively. The difference does not by itself establish which result is best. The factor result reflects its prescribed exposure factors; the formula result reflects its modules and aggregation rule; the internal-model result reflects its specified loss distribution and percentile definition.
Check. The formula's radicand is \(2.44+0.480+0.144+0.192=3.256\), and \(\sqrt{3.256}\approx1.8044\). A common percentile trap is to select $8 million because it is the largest loss. The 99.5th percentile is instead the smallest loss at which cumulative probability reaches 0.995, which is $4 million here.
Why Factor Charges Respond Linearly
The basic factor rule also explains exactly what such a method can and cannot respond to. Let \(E\) be a nonnegative exposure measured in the units required by the method, and let \(f\) be its prescribed factor. If the charge is \(C=fE\), then for two exposure values \(E_1\) and \(E_2\):
Thus, with a constant factor, the charge changes in direct proportion to the measured exposure. If exposure doubles, the charge doubles. But if the exposure is unchanged and risk characteristics absent from the factor calculation change, this equation produces no change in the charge. That is not an arithmetic error; it is a limitation of the information represented by the method. The result may not apply if the prescribed factor varies by category, contains thresholds, or includes other adjustments.
Worked Example: A Change in Risk Mix
Worked Example: A Change in Risk Mix
A hypothetical insurer has $10 million of exposure in class A and $2 million in class B. A simple factor approach assigns 4% to class A and 10% to class B. The insurer transfers $1 million of exposure from A to B, leaving total exposure unchanged. Under a separate prescribed formula, the calculated total is $0.72 million before the transfer and $0.80 million afterward. An internal model, using the same stated capital basis, reports $0.68 million before and $0.92 million afterward. Calculate the factor charges before and after, compare the changes, and interpret the results without treating any approach as a guaranteed benchmark.
Method. Apply each factor to its class exposure before and after the transfer. For the other two methods, calculate the changes from the stated outputs. All figures are hypothetical; the formula and model outputs are supplied rather than derived from unspecified rules or assumptions.
Before the transfer, the factor charges are \(10{,}000{,}000 \times 0.04=\$400{,}000\) for class A and \(2{,}000{,}000 \times 0.10=\$200{,}000\) for class B. The total is \(\$600{,}000\).
After the transfer, class A exposure is \(\$9{,}000{,}000\), giving a charge of \(9{,}000{,}000 \times 0.04=\$360{,}000\). Class B exposure is \(\$3{,}000{,}000\), giving a charge of \(3{,}000{,}000 \times 0.10=\$300{,}000\). The new factor total is \(\$660{,}000\), an increase of \(\$60{,}000\).
The prescribed-formula result rises from \(\$720{,}000\) to \(\$800{,}000\), an increase of \(\$80{,}000\). The internal-model result rises from \(\$680{,}000\) to \(\$920{,}000\), an increase of \(\$240{,}000\). All three respond to the transfer in this hypothetical, but by different amounts. Their differences could reflect different risk distinctions, assumptions, or calibrations; the figures alone do not reveal which explanation is correct.
Check. Total exposure is \(\$12\) million both before and after. A flat 5% factor applied to that total would remain \(\$600{,}000\), missing the class-mix change. The stated class-specific factors instead increase the charge by \(\$60{,}000\). A common mistake is to say that a factor method must be insensitive to mix: it is insensitive only to changes not reflected in its factor categories or exposure definitions.
Data, Comparability, and Model Risk
The data burden generally increases as the calculation becomes more tailored, though the precise burden depends on the method. A factor approach may need reliable exposure measures and correct classification. A formula may additionally require consistent inputs for several prescribed components. An internal model may need historical and scenario data, explicit assumptions, controlled code and processes, evidence that inputs are suitable for their intended use, and monitoring for changes in performance or business conditions.
Comparability is strongest when methods use common definitions, assumptions, and rules. A prescribed factor or formula can support this consistency, but only if companies apply the same requirements correctly. Internal models can reflect different risk profiles, but comparisons require attention to company-specific modeling choices, data, calibration, and scope. Two numbers described as capital for the same horizon may still not have the same meaning.
Model risk is the possibility that a model's specification, inputs, implementation, or use causes an unsuitable result. It affects all three approaches: factors may be poorly matched to exposures, prescribed formulas may encode limitations or assumptions, and internal models may contain estimation or implementation errors. More detail can create additional opportunities for errors as well as additional ways to represent risk. Validation and governance are therefore essential complements to calculation, not optional evidence that a sophisticated number is correct.
Check Your Understanding
For each problem, identify the applicable method and keep the stated measurement basis in view.
- A hypothetical insurer has $18 million of exposure subject to a constant 5% factor. Calculate the charge. If exposure increases by 10% with the factor unchanged, calculate the new charge and increase. (Answer: Initial charge $900,000; new charge $990,000; increase $90,000.)
- A simple factor calculation applies 4% to $8 million of one exposure and 7% to $3 million of another, with additive charges. Calculate the total. (Answer: \(0.04(8{,}000{,}000)+0.07(3{,}000{,}000)=\$530{,}000\).)
- A prescribed formula has two module charges of $0.90 million and $1.20 million and a prescribed correlation of 0.30. Under the square-root aggregation rule for two modules, calculate the result to the nearest thousand dollars. (Answer: \(\sqrt{0.90^2+1.20^2+2(0.30)(0.90)(1.20)}=\sqrt{2.898}\approx\$1.702\) million.)
- An internal model's annual loss distribution has losses of $0.5 million with probability 0.80, $1.5 million with probability 0.12, $3 million with probability 0.06, and $6 million with probability 0.02. Under the definition of capital as the smallest loss whose cumulative probability is at least 0.95, find capital. (Answer: $3 million, since cumulative probability is 0.92 at $1.5 million and 0.98 at $3 million.)
- A company has limited historical data for a new coverage but has reliable exposure records and a prescribed factor. State one reason the factor may be practical and one caution in interpreting its result. (Answer: It may be practical because it uses available exposure data and is relatively straightforward to apply; its result may not reflect risk differences the exposure measure and factor do not capture.)