Why the Model Boundary Matters
An economic-capital result is meaningful only in relation to a clearly specified perimeter. The perimeter determines whose losses count, which activities and risks are represented, and which events can generate modeled losses. Two models can use the same capital measure and calibration yet produce different results because they cover different exposures.
The boundary is not automatically the insurer’s legal-entity boundary, nor does it necessarily match the scope of operational control. An insurer may retain losses arising from an outsourced activity, owe payment under a guarantee, or have material exposure to an affiliate. Conversely, an affiliate’s loss does not automatically belong in a model of the insurer on a standalone basis. The decision depends on the model’s intended perspective and the insurer’s actual economic, contractual, or financial exposure.
A perimeter should be stated in terms that can be applied consistently. For example, “the insurer’s material risks” is not enough by itself: the model owner must identify which entities and activities are covered, what kinds of loss-generating events are in scope, and how relevant obligations to or from other parties are treated. The boundary should also remain consistent with the model’s stated purpose and materiality decisions.
Four Dimensions of the Perimeter
A useful boundary description addresses four connected dimensions. These decisions define the model’s scope; they do not, by themselves, specify the exposure units or detailed model components.
| Dimension | Boundary question | Typical boundary case |
|---|---|---|
| Entities | Which legal entities or parts of the organization are represented? | A subsidiary included in a group model but not in a standalone insurer model |
| Portfolios and activities | Which insurance portfolios, investments, services, or other activities are in scope? | A legacy portfolio still producing claim obligations |
| Risks | Which sources of possible loss or recovery are represented? | Losses associated with a service provider’s failure |
| Events | Which event types can produce an in-scope loss or recovery? | A guarantee call, litigation outcome, or operational disruption |
The same exposure can be inside one model boundary and outside another. A group model may cover multiple subsidiaries because it measures risk from a group perspective. A standalone model may cover only one insurer, while still including that insurer’s contractual obligations to other group members. The model’s perspective must therefore be stated before boundary cases are decided.
A Practical Test for Boundary Decisions
For each potentially relevant exposure, ask whether it can create a material loss or recovery for the entity or group represented by the model. Then identify the mechanism: ownership, a contract, a guarantee, an indemnity, a service dependency, or another economic connection. A risk being managed by another party does not, by itself, remove the insurer’s exposure. Likewise, common ownership does not, by itself, make every affiliate loss an insurer loss.
State whether the model represents a standalone insurer, a group, or another defined set of entities. Apply boundary decisions from that perspective.
Determine whether a loss can reach the modeled entity through its own activity, a legal obligation, a financial connection, or a retained responsibility.
Specify the event or condition that produces the payment, loss, or recovery. For contingent obligations, distinguish the trigger from the possible payment amount.
Make sure each economic loss is represented once, and that an exposure has not been omitted merely because another party operates or manages the activity.
Record inclusions, exclusions, materiality judgments, and any limitations so that users can understand what the resulting capital estimate does and does not cover.
Contingent Obligations and the Model Perspective
A contingent obligation can create exposure even if no payment is currently due. The relevant questions are whether the modeled entity could be required to pay, what event triggers payment, and what amount could be payable under the obligation. For example, if an insurer guarantees a subsidiary’s debt, a standalone model of the insurer may need to represent the insurer’s guarantee payment when the specified trigger occurs. The subsidiary’s underlying losses are not automatically the insurer’s losses; the insurer’s exposure arises through the guarantee.
For a simple contingent payment, let \(I\) equal 1 when the trigger occurs and 0 otherwise, and let \(G\) be the payment due if the trigger occurs. The insurer’s loss from the obligation is \(IG\). This follows directly from the two cases: if the trigger does not occur, \(I=0\) and the payment is zero; if it occurs, \(I=1\) and the payment is \(G\). More detailed terms, such as caps or conditions, change the payment amount or the trigger definition, but not the need to represent the obligation from the perspective of the entity that owes it.
At a group level, an obligation between two included group entities can appear as a loss to one entity and a receipt to another. Whether those entries cancel depends on the model’s perspective and the economic transaction represented. An internal transfer may cancel in a group total if the group has no corresponding loss to an outside party. But cancellation should not erase a genuine external obligation, such as a group guarantee owed to an external creditor. A standalone insurer model and a group model may therefore treat the same contract differently.
Worked Example: A Guarantee in Standalone and Group Models
A hypothetical insurer has issued a guarantee to an external lender covering up to $7 million of a subsidiary’s debt. Under the contract, a qualifying subsidiary default requires the insurer to pay the amount owed, up to the cap. A standalone model of the insurer and a consolidated model of the parent group are both being designed. Decide whether the guarantee belongs in each boundary, and state what loss each model should represent.
Step 1: Establish the perspectives. The standalone model represents the insurer. The consolidated model represents the parent group, including both the insurer and the subsidiary.
Step 2: Identify the insurer’s obligation. The insurer has a contractual obligation to the external lender. If the qualifying default occurs and the full $7 million cap is payable, the insurer can incur a $7 million loss. The guarantee is therefore within the standalone insurer’s exposure perimeter, subject to the contract’s trigger and payment terms.
Step 3: Apply the group perspective. The guarantee is also relevant to the group’s relationship with the external lender. In a group model, the subsidiary’s debt and the guarantee cannot be treated as two independent external losses if they describe the same amount owed to the lender. The model should represent the group’s external economic loss once, using a consistent treatment of the debt and guarantee.
Step 4: State the conclusion. Include the guarantee in the standalone insurer model because the insurer owes payment under the contract. Include the external debt exposure in the group model, but do not add a second copy of the same external loss merely because the insurer also guarantees it.
Method and trap avoided. Decide inclusion from the modeled entity’s actual obligation, then check the group total for duplicate representation. A common mistake is to exclude the guarantee from the insurer model because the subsidiary’s debt is the underlying exposure, or to count both the debt and guarantee as separate full group losses.
Externally Managed Activities and Retained Exposure
Outsourcing transfers an activity, but it does not necessarily transfer every loss that activity can cause. An insurer may remain responsible for policyholder obligations, incur costs when a service provider fails, or retain a contractual right to recover only part of a loss. In those cases, the insurer may retain exposure even though it does not operate the activity itself.
The boundary should distinguish the service provider’s own losses from losses that can reach the insurer. If a vendor suffers a loss that creates no obligation, payment, or material economic consequence for the insurer, that vendor loss may sit outside the insurer’s perimeter. If a vendor disruption causes the insurer to incur costs or breach its obligations, the insurer’s resulting exposure may be in scope. A recovery from the vendor should be represented only to the extent the insurer has a relevant right to recover and the model’s assumptions support that treatment.
Worked Example: A Retained Loss From an Outsourced Service
A hypothetical insurer outsources a claims-data service. For a specified measurement period, assume a data breach occurs with probability 0.0200. If it occurs, the insurer incurs a $6 million loss from its own response costs and obligations. A contract provides a $1 million reimbursement if the provider pays; assume payment occurs with probability 0.8000 conditional on a breach. If the reimbursement is not paid, the insurer receives nothing. There are no other losses in this example. Determine the insurer’s exposure perimeter and calculate the 99% VaR of its net loss. Assume these are the complete, mutually exclusive outcomes and use \(\operatorname{VaR}_{0.99}(X)=\inf\{x:P(X\leq x)\geq0.99\}\).
Step 1: Decide what belongs in the insurer’s boundary. The provider operates the service, but the insurer incurs its own $6 million loss if the breach occurs. That retained loss is within the insurer’s perimeter. The possible reimbursement is also relevant because it changes the insurer’s net loss; it is not a reason to remove the gross exposure from consideration.
Step 2: Calculate the unconditional probabilities. The probability of no breach is \(1-0.0200=0.9800\). The probability of a breach with reimbursement is \(0.0200\times0.8000=0.0160\). The probability of a breach without reimbursement is \(0.0200\times(1-0.8000)=0.0040\). The three probabilities sum to \(0.9800+0.0160+0.0040=1.0000\).
Step 3: Calculate net loss in each outcome. With no breach, loss is $0. With a breach and reimbursement, net loss is \($6\text{ million}-$1\text{ million}=$5\text{ million}\). With a breach and no reimbursement, net loss is $6 million.
Step 4: Find the quantile. The cumulative probability at $0 is 0.9800. At $5 million it is \(0.9800+0.0160=0.9960\), which is at least 0.9900. Therefore the 99% VaR is $5 million.
Check. The distribution has probability 0.9800 at $0, 0.0160 at $5 million, and 0.0040 at $6 million; these probabilities total 1.0000. The $5 million quantile is correct because the cumulative probability first reaches 0.9900 at that loss amount.
Method and trap avoided. Build the insurer’s loss distribution from its retained loss and the stated recovery outcomes. Do not exclude the exposure just because a vendor manages the service, and do not subtract the $1 million reimbursement in every breach outcome when it is collected only with probability 0.8000.
Worked Boundary Decisions and Capital Results
A boundary decision can change the modeled loss distribution and therefore the capital result. The calculation must preserve the stated outcome probabilities and distinguish losses that occur in separate states from losses that occur together. The following example uses the quantile definition stated explicitly in the problem; it is a boundary illustration, not a general lesson on selecting a capital measure.
Worked Example: Including a Contingent Affiliate Guarantee
A hypothetical insurer is modeling its standalone exposure at a 99.5% confidence level. The mutually exclusive outcomes for the measurement period are: ordinary claims loss of $4 million with probability 0.9400; a direct catastrophe loss of $10 million with probability 0.0500; and a guarantee call on an affiliate’s external borrowing, producing total insurer loss of $13 million with probability 0.0100. The guarantee is legally owed by the insurer. Calculate the 99.5% VaR with the guarantee included, then calculate it if the guarantee is wrongly omitted. Use \(\operatorname{VaR}_{0.995}(X)=\inf\{x:P(X\leq x)\geq0.995\}\).
Step 1: Confirm the boundary. Because the insurer is legally required to pay when the guarantee is called, the guarantee exposure belongs in the standalone insurer model. The three outcomes are stated to be mutually exclusive and already give total losses, so their loss amounts are not added to one another.
Step 2: Calculate VaR with the guarantee included. The cumulative probability at $4 million is 0.9400. At $10 million it is \(0.9400+0.0500=0.9900\). At $13 million it is \(0.9900+0.0100=1.0000\). Since 0.9900 is below 0.9950 and 1.0000 is at least 0.9950, the 99.5% VaR is $13 million.
Step 3: Recalculate after omitting the guarantee. In the guarantee-call state, the omitted model records only the ordinary $4 million loss. Its distribution is therefore $4 million with probability \(0.9400+0.0100=0.9500\), and $10 million with probability 0.0500. The cumulative probability at $4 million is 0.9500; at $10 million it is 1.0000. The 99.5% VaR under this incorrect boundary is $10 million.
Check. With the guarantee included, the cumulative probability at $10 million is only 0.9900, below the 0.9950 target. With it omitted, the guarantee-call state has been incorrectly reduced to $4 million, causing the cumulative probability to reach 1.0000 at $10 million. The difference arises from the changed boundary, not from adding the separate guarantee and catastrophe losses together.
Method and trap avoided. Sort the loss outcomes and accumulate their probabilities, preserving the mutually exclusive states. A common error is to add losses from separate states as though they happened together, or to claim that $10 million is the included 99.5% VaR even though its cumulative probability is only 0.9900.
Common Boundary Errors
- Equating the perimeter with operational control. A provider’s management of an activity does not remove losses the insurer retains through its own obligations or economic exposure.
- Including every affiliate loss in a standalone model. Common ownership alone does not establish that an affiliate’s loss is the insurer’s loss. Identify the connection that transfers the exposure.
- Ignoring a contingent obligation until payment is due. A guarantee or indemnity may create exposure before a claim is made because the triggering event can occur within the modeled scope.
- Counting one economic loss twice. A group model should not treat an external debt and a guarantee of that same debt as two separate full losses without a reason that distinguishes the exposures.
- Leaving exclusions unexplained. A documented exclusion makes the result’s limitations visible and supports consistent decisions when the model is reviewed or updated.
Check Your Understanding
For each problem, state the relevant boundary decision before calculating any requested amount.
- A hypothetical insurer owns an affiliate but has no guarantee, service contract, or other obligation connected to the affiliate’s $3 million loss. Should that loss automatically be included in the insurer’s standalone model? (Answer: No; ownership alone does not make the affiliate’s loss an insurer loss.)
- An insurer outsources policy administration. A specified vendor outage occurs with probability 0.0100 and creates a $2 million loss for the insurer; no other losses are in scope. Does the outage-related exposure belong in the insurer’s perimeter, and what is the insurer’s loss if the outage occurs? (Answer: Yes; $2 million.)
- A guarantee pays $5 million if a specified event occurs with probability 0.0200, and pays nothing otherwise. What is the expected payment under these assumptions? (Answer: \(0.0200\times$5\text{ million}=$100{,}000\).)
- A hypothetical insurer has losses of $2 million with probability 0.9700 and $8 million with probability 0.0300. A legally owed guarantee adds a further $4 million only in the $8 million state. What is the 99% VaR when the guarantee is included? (Answer: $12 million; cumulative probability at $2 million is 0.9700 and at $12 million is 1.0000.)
- A group model includes a subsidiary and its insurer parent. The subsidiary owes $6 million to an external lender, and the parent guarantees that same debt; assume the group’s total obligation to the lender is $6 million, not $12 million. What amount should the group model represent for this single external debt obligation? (Answer: $6 million once, rather than double-counting the debt and guarantee.)