Why Capital-Relief Comparisons Go Wrong
A reinsurance treaty or catastrophe bond may reduce an insurer’s retained loss, but a stated limit or expected payment does not by itself establish the amount of capital relief. The measurement depends on the loss horizon, the capital measure, the contract’s actual payment rules, and whether the recovery is collectible. A comparison can look numerically precise and still overstate the reduction in required capital.
The errors addressed here are especially important when reviewing capital effects across different models or reports. A defensible comparison holds the measurement basis constant, calculates the recovery under the contract, accounts for counterparty nonpayment, and compares portfolio capital before and after the transaction without reusing diversification benefits.
The comparison must use a common basis. At a minimum, check that gross and net results use the same horizon, confidence level, risk measure, portfolio perimeter, and valuation date. If one result is annual VaR and the other is a quarterly capital figure, subtracting them does not measure relief. Similarly, an expected recovery cannot simply be subtracted from VaR: the recovery may occur in states that do not determine the selected quantile.
Why Expected Recovery Is Not Capital Relief
Capital measures depend on the distribution of retained losses, not just their average. The following simple result explains an important exception and why it does not justify treating every recovery as a fixed reduction.
Let \(R=r\) in every outcome, where \(r\) is a constant, and let retained loss be \(L-r\).
For any \(x\), \(P(L-r\leq x)=P(L\leq x+r)\). The probability reaches level \(\alpha\) at the gross-loss quantile shifted down by \(r\).
Thus \(\operatorname{VaR}_{\alpha}(L-r)=\operatorname{VaR}_{\alpha}(L)-r\), so a certain fixed recovery reduces VaR by exactly \(r\).
A contractual recovery is generally not fixed: it depends on loss, attachment, exhaustion of limits, exclusions, and collectability. In that case the shift argument does not apply. The capital reduction must be calculated from the distribution of \(L-R\); it need not equal \(E[R]\), the stated limit, or the recovery in a single illustrative event.
Worked Example: Comparing Capital on the Same Horizon
Method and common mistake. Calculate gross and net VaR over the same annual horizon. The trap is to subtract a monthly capital result from an annual result.
A hypothetical insurer has an annual loss of $10 million with probability 0.02 and zero otherwise. An annual contract pays $6 million if that loss occurs. Ignore counterparty default and all other losses. Find the annual 99% gross and net VaR. Then assess a report that compares annual gross VaR with monthly net VaR, where the annual event is assigned with equal probability to one of 12 months.
Annual gross capital. The annual loss is zero with probability \(0.98\) and $10 million with probability \(0.02\). Since \(P(L\leq 0)=0.98<0.99\), but \(P(L\leq 10)=1\), annual gross \(\operatorname{VaR}_{0.99}\) is $10 million.
Annual net capital. The annual recovery is $6 million when the loss occurs, so retained loss is \(10-6=4\) million in that event and zero otherwise. The probability of zero retained loss remains \(0.98<0.99\); therefore annual net \(\operatorname{VaR}_{0.99}\) is $4 million. Annual capital relief is \(10-4=\)$6 million.
Monthly figure. The probability that the event falls in a particular month is \(0.02/12=0.0016667\). In any one month the probability of no event is \(1-0.0016667=0.9983333\), which exceeds \(0.99\). Monthly net \(\operatorname{VaR}_{0.99}\) is therefore zero. Subtracting it from annual gross VaR would suggest $10 million of “relief,” but this mixes horizons and is invalid. The like-for-like annual comparison gives $6 million.
Check. Annual net loss is exactly $6 million lower than gross loss in the event state, and both annual VaRs select that state. The resulting $6 million relief is consistent with the contract and the common annual basis.
Counterparty Risk and Contract Terms
A gross limit describes the maximum payment permitted by specified contract terms; it is not necessarily a guaranteed recovery. First apply the attachment point, coinsurance, limit, and any relevant event or aggregate conditions. Then reflect the possibility that the counterparty fails to pay, using a model appropriate to the transaction. If default is related to the insured event, assuming independence may materially understate retained loss.
Worked Example: Adjusting for Counterparty Default
Method and common mistake. Construct the net-loss distribution by outcome, including default, before finding VaR. The trap is to count the contractual recovery as certain.
In a hypothetical one-year portfolio, a covered loss is $8 million with probability 0.02 and zero otherwise. The treaty’s recovery in the loss event is $5 million if the reinsurer pays. Conditional on that loss event, the reinsurer defaults with probability 0.10 and pays nothing. Assume no other losses or recoveries. Calculate 99% gross and net VaR and the resulting capital relief.
Gross VaR. Gross loss is zero with probability \(0.98\) and $8 million with probability \(0.02\). Because \(0.98<0.99\), the 99% gross VaR is $8 million.
Outcome probabilities after transfer. The probability of a loss event with payment is \(0.02(1-0.10)=0.018\). The probability of a loss event with default is \(0.02(0.10)=0.002\). No-loss probability is \(0.98\). These probabilities sum to \(0.018+0.002+0.98=1.000\).
Net VaR. If the reinsurer pays, retained loss is \(8-5=3\) million; if it defaults, retained loss is $8 million. The cumulative probability of net loss at or below $3 million is \(0.98+0.018=0.998\), which exceeds \(0.99\). At zero, cumulative probability is only \(0.98\). Thus net 99% VaR is $3 million, and measured capital relief is \(8-3=\)$5 million.
Check. The 99% quantile falls in the payment outcome, while the default outcome occupies the uppermost 0.2% of probability. It would be wrong to conclude that default is irrelevant: it remains in the distribution, and at a higher confidence level it could determine the capital requirement.
Worked Example: A Gross Limit Is Not the Recovery
Method and common mistake. Apply the contract’s attachment, limit, and ceded share in sequence. The trap is to subtract the headline limit directly from the loss.
A hypothetical annual loss is $10 million with probability 0.02 and zero otherwise. A treaty attaches at $3 million, has a $5 million limit, and pays 80% of the covered layer. Assume the reinsurer always pays and there are no other contract restrictions. Calculate the recovery, retained loss, and 99% VaR relief.
Covered layer before coinsurance. The loss above the $3 million attachment is \(10-3=7\) million. The limit caps this amount at \(\min(7,5)=5\) million.
Actual recovery. The reinsurer pays 80% of the capped layer, so \(R=0.80(5)=4\) million. The retained event loss is \(10-4=6\) million.
Capital calculation. Gross 99% VaR is $10 million because the loss-event probability of 0.02 exceeds the upper 1% tail. Net 99% VaR is $6 million for the same reason. Measured relief is \(10-6=\)$4 million, not the $5 million gross limit.
Check. The recovery does not exceed the contract’s $5 million layer limit, and it is 80% of that layer. Subtracting the full limit would report $5 million of relief and understate retained event loss by $1 million.
Do Not Count Diversification Twice
Portfolio capital is generally not the sum of standalone business-line capital. Diversification is already reflected when a portfolio model calculates capital for the combined loss distribution. A common error is to use an existing diversification benefit as an additional deduction after comparing modeled portfolio capital before and after a transfer.
The appropriate comparison is direct: calculate total portfolio capital before the transaction and total portfolio capital after it, on the same basis, then subtract. Standalone capital changes can help explain the result, but they cannot replace the post-transaction portfolio calculation unless the capital method explicitly supports that decomposition.
Worked Example: Avoiding a Second Diversification Deduction
Method and common mistake. Compare the modeled combined-portfolio capital before and after the transfer. The trap is adding a pre-existing diversification benefit to a change in standalone capital.
A hypothetical insurer’s model gives, on the same horizon and capital basis, standalone capital of $60 million for Line A and $40 million for Line B. Combined portfolio capital is $80 million. After a risk transfer affecting Line A, standalone capital is $45 million for Line A and remains $40 million for Line B; the model gives combined portfolio capital of $70 million. Calculate actual relief and the result from the double-counting approach.
Direct portfolio comparison. Before transfer, combined capital is $80 million. After transfer, combined capital is $70 million. Measured portfolio relief is \(80-70=\)$10 million.
Standalone change. Line A’s standalone capital falls by \(60-45=15\) million, while Line B is unchanged. The pre-transfer diversification benefit is \(60+40-80=20\) million.
Incorrect calculation. Adding the $20 million pre-existing diversification benefit to the $15 million standalone reduction gives $35 million. This is not the measured relief: it treats a benefit already embedded in pre-transfer combined capital as though it were a new reduction caused by the transaction.
Check. The model’s like-for-like portfolio outputs give $10 million of relief. The $35 million alternative does not reconcile to those outputs and counts the existing diversification benefit again.
A Practical Error Check
Before accepting a capital-relief figure, trace the calculation from gross losses to net portfolio capital. The sequence below makes the key assumptions visible and helps distinguish measurement errors from genuine model differences.
Record the horizon, confidence level or other risk-measure parameter, portfolio perimeter, valuation date, and capital measure for both sides.
Apply attachment, covered-loss definitions, coinsurance, per-event and aggregate limits, and other material payment conditions to each modeled loss outcome.
Model counterparty nonpayment and any dependence between default and insured losses, or state and justify a simplifying assumption.
Apply the same capital method to the resulting net-loss distribution and the same portfolio perimeter. Do not substitute expected recovery or a headline limit for this calculation.
Calculate pre-transaction portfolio capital minus post-transaction portfolio capital. Keep premium, expenses, collateral costs, and other economic effects distinct unless the stated capital measure includes them.
Common Errors and Their Corrections
| Error | Why it fails | Correction |
|---|---|---|
| Subtracting monthly capital from annual capital | The two figures describe different loss accumulations and probability distributions. | Recalculate both on the same horizon and basis. |
| Subtracting expected recovery from VaR | An average payment does not identify the recovery in the quantile-defining outcomes. | Calculate the net-loss distribution and apply the capital measure to it. |
| Treating a stated limit as certain cash | Attachment, coinsurance, exhaustion, exclusions, and nonpayment can reduce the collectible amount. | Apply the contract and counterparty assumptions outcome by outcome. |
| Adding diversification relief twice | The combined-portfolio capital may already reflect diversification. | Compare modeled combined capital before and after, using the same method. |
These controls matter because capital relief is often used to compare risk-transfer alternatives or assess solvency and rating constraints. A misstated reduction can distort the apparent capital efficiency of a transaction even if its contract terms and price are otherwise understood. The diagnostic discipline here is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.
Check Your Understanding
For each problem, state the relevant basis and show how the stated error changes the conclusion.
- A hypothetical annual loss is $7 million with probability 0.015 and zero otherwise. An annual treaty pays $4 million if the loss occurs, with certain payment. Find gross and net annual 99% VaR and relief. A report instead compares annual gross VaR with monthly net VaR, assuming the event is assigned uniformly to one month. (Answer: Annual gross VaR = $7 million; annual net VaR = $3 million; valid annual relief = $4 million. Monthly event probability is \(0.015/12=0.00125\), so monthly net 99% VaR = $0; the cross-horizon difference of $7 million is invalid.)
- A hypothetical annual loss is $12 million with probability 0.03. A treaty attaches at $2 million, has a $6 million limit, and pays 75% of the covered layer. Assume certain payment. Find the recovery, net loss in the event, and 99% VaR relief. (Answer: Covered layer is \(\min(12-2,6)=\$6\) million; recovery is \(0.75(6)=\$4.50\) million; net event loss is $7.50 million; gross and net VaR are $12 million and $7.50 million, respectively, so relief is $4.50 million.)
- A hypothetical one-year loss is $10 million with probability 0.02. A contract pays $4 million in the event, but the counterparty defaults with conditional probability 0.20 and pays nothing on default. Find 99% gross and net VaR. (Answer: Gross VaR = $10 million. Payment probability is \(0.02(0.80)=0.016\), default-event probability is \(0.02(0.20)=0.004\), and no-loss probability is 0.98. Net loss is $6 million on payment and $10 million on default; cumulative probability through $6 million is \(0.98+0.016=0.996\), so net 99% VaR = $6 million and relief = $4 million.)
- On a consistent basis, modeled standalone capital is $50 million for Line A and $30 million for Line B; combined capital is $65 million. After a transfer, standalone capital is $42 million for A and $30 million for B; combined capital is $60 million. Calculate the true relief and show the double-counting result if the original diversification benefit is added to the standalone reduction. (Answer: True relief is \(65-60=\$5\) million. Original diversification benefit is \(50+30-65=\$15\) million, and standalone reduction is \(50-42=\$8\) million. Adding them gives $23 million, which double counts and is not the modeled portfolio relief.)