Netting and Guarantees in Credit Loss Models

Why Netting and Guarantees Matter

A counterparty may owe an insurer money on some contracts while the insurer owes that same counterparty money on others. If the contracts can be legally closed out and combined, the amount at risk may be substantially smaller than the sum of the positive contract values. A guarantee can further transfer part of the resulting exposure to another obligor. Neither technique makes credit risk disappear: netting depends on enforceable rights, and a guarantee replaces some counterparty risk with guarantor risk.

The key modeling task is to follow the contractual sequence carefully. First determine which transactions can legally be netted. Then calculate the positive amount owed after closeout and any recognized collateral adjustment. Finally, identify which part is supported by a guarantee and model the guarantor’s ability to pay. This tutorial is aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.

Definition: A closeout netting set is a group of transactions for which a legally enforceable agreement permits their values to be combined into one net closeout amount following a specified event, such as counterparty default. If $V_j$ is the value of transaction $j$ from the reporting institution’s perspective, positive when the counterparty owes the institution, the net amount before collateral is $\sum_j V_j$. The credit exposure is its positive part, $\max(\sum_j V_j,0)$.

A negative net amount means the institution owes the counterparty at closeout; it is not a positive credit exposure to the institution. A netting agreement applies only to the transactions and parties within its legally effective scope. It does not automatically permit obligations with an affiliate, a different legal entity, or a transaction outside the agreement to offset a receivable.

For a stylized calculation that recognizes collateral held, let $C$ be eligible collateral available to reduce the net amount at closeout, measured in the same currency and at the same valuation date. The residual exposure for netting set $k$ is

E_k=\max\left(\sum_{j\in k}V_j-C_k,0\right).

This formula assumes the specified collateral can be applied to the closeout amount and that its value has already been adjusted appropriately for the model. Actual documentation and collateral terms determine what can be recognized. The examples below use no collateral so that the effect of netting and guarantees is clear.

From Net Exposure to Credit Loss

Exposure is an amount at risk, not itself a loss. A loss model also needs a default event and a recovery assumption. Write $p_C=P(D_C)$ for the probability that the counterparty defaults over the modeled horizon, and $R_C$ for the recovery rate on an unguaranteed claim against that counterparty. In the simple model, recovery is a fraction of the claim and the loss given default is $1-R_C$. If exposure $E$ is fully unguaranteed, expected loss is

$$ E[L]=p_C E(1-R_C). $$

This expression assumes the exposure and recovery amount are fixed for the calculation horizon, and that the default probability and recovery assumption refer to the same horizon and claim definition. More detailed models can make exposure, default, and recovery jointly variable; the simple expression should not be used as if those quantities were independent when they are not.

Definition: A guarantee is a contractual promise by a guarantor to pay some specified amount if the primary obligor fails to meet its obligation. For a single exposure $E$, let $H$ be the portion covered by the guarantee and $U=E-H$ the unguaranteed portion. A cap and a coverage fraction can be represented by $H=\min(E,\alpha E,K)$, where $\alpha$ is the stated fraction and $K$ is the guarantee cap. The exact definition of covered claims, timing, and conditions must follow the guarantee terms.

For the calculation that follows, assume the counterparty defaults, the unguaranteed part recovers at rate $R_C$, and the guarantee pays the covered amount in full if the guarantor remains solvent. If the guarantor also defaults, the covered claim recovers at rate $R_G$. Let $q=P(D_G\mid D_C)$, the probability of guarantor default conditional on counterparty default. Then the expected loss conditional on counterparty default is

E[L\mid D_C]=U(1-R_C)+qH(1-R_G).

Multiplying by the counterparty’s default probability gives the unconditional expected loss in this model:

E[L]=p_C\left[U(1-R_C)+qH(1-R_G)\right].

The conditional probability $q$ matters. Substituting an unconditional guarantor default probability assumes, without justification, that guarantor and counterparty default events are independent. If they are more likely to fail together, the guarantee provides less protection than an independence assumption would suggest.

Worked Applications

Worked Example: Comparing Gross and Netted Exposure

A hypothetical insurer has four transactions with one counterparty, all in millions of dollars. Their closeout values from the insurer’s perspective are $+1.20$, $-0.70$, $+0.90$, and $-0.20$. Assume all four are in one legally enforceable netting set, there is no collateral, the counterparty’s default probability over the horizon is $0.030$, and its recovery rate is $0.40$. Calculate the expected loss with netting, and compare it with the result if the transactions cannot be netted.

Method. First find the positive net closeout amount. Then apply the loss-given-default assumption to that exposure and multiply by the default probability.

The signed net value is

1.20-0.70+0.90-0.20=1.20\text{ million dollars}.

Because this is positive, the netted exposure is $E=1.20$ million dollars. The loss-given-default rate is

$$ 1-R_C=1-0.40=0.60. $$

The loss conditional on counterparty default is $1.20(0.60)=0.72$ million dollars. Thus expected loss is

0.030(0.72)=0.0216\text{ million dollars}=\$21{,}600.

Without enforceable netting, the two positive transaction values are separate claims. Their sum is $1.20+0.90=2.10$ million dollars. Under the same recovery and default assumptions, the conditional loss is $2.10(0.60)=1.26$ million dollars, and expected loss is $0.030(1.26)=0.0378$ million dollars, or $\$37{,}800$. The check is that netting reduces the modeled exposure from $\$2.10$ million to $\$1.20$ million, so expected loss falls by the same proportion under these fixed assumptions. The common mistake is to add only the positive values even though the contracts are legally netted; that ignores the $\$0.90$ million of offsetting obligations.

Worked Example: Applying a Guarantee Cap and Guarantor Risk

A hypothetical counterparty has a netted exposure of $\$800{,}000$. A guarantee covers 50% of the exposure, subject to a cap of $\$300{,}000$. The counterparty recovery rate on unguaranteed claims is $0.35$. Conditional on counterparty default, the guarantor defaults with probability $0.10$; if it defaults, its recovery rate on the guaranteed claim is $0.20$. If the guarantor does not default, it pays the covered claim in full. The counterparty default probability is $0.020$. Find expected loss.

Method. Calculate the guarantee amount as the smaller of the percentage-based amount and the cap, then apply the two recovery assumptions to the appropriate portions.

The percentage-based guarantee amount is $0.50(\$800{,}000)=\$400{,}000$. The cap reduces the covered amount to

H=\min(\$400{,}000,\$300{,}000)=\$300{,}000.

The unguaranteed amount is $U=\$800{,}000-\$300{,}000=\$500{,}000$. Its loss conditional on counterparty default is $\$500{,}000(1-0.35)=\$325{,}000$. On the guaranteed portion, a loss occurs in this model only if the guarantor also defaults. Conditional on that event, the loss is $\$300{,}000(1-0.20)=\$240{,}000$; conditional on counterparty default, its expected contribution is $0.10(\$240{,}000)=\$24{,}000$.

Total expected loss conditional on counterparty default is $\$325{,}000+\$24{,}000=\$349{,}000$. Therefore

E[L]=0.020(\$349{,}000)=\$6{,}980.

Without the guarantee, the conditional loss would be $\$800{,}000(0.65)=\$520{,}000$, giving expected loss of $0.020(\$520{,}000)=\$10{,}400$. The guarantee lowers modeled expected loss by $\$3{,}420$, but does not eliminate it. The common mistake is to treat all $\$400{,}000$ of the percentage-based amount as covered; the contractual cap limits coverage to $\$300{,}000$.

Why the Netting Formula Uses a Positive Part

The positive-part operation follows directly from the direction of the closeout payment. If the signed closeout balance is positive, the counterparty owes the institution that amount, which can be lost if the counterparty defaults. If the balance is zero or negative, there is no amount owed to the institution: a negative balance means the institution owes money instead. Therefore the credit exposure is the larger of the signed balance and zero, rather than the absolute value of the balance.

Combine only eligible transaction values.
Use the signed values for contracts within the same legally enforceable closeout netting set.

Determine the amount owed to the institution.
Take the positive part of the net closeout amount, after any collateral treatment justified by the model and contract.

Allocate covered and uncovered claims.
Apply the guarantee’s covered percentage, cap, and scope to the exposure; do not apply it to unrelated obligations.

Model default and recovery.
Apply counterparty recovery to the unguaranteed claim and guarantor default and recovery assumptions to the guaranteed claim.

Legal Scope and Residual Risk

A calculation can be mathematically correct and still understate risk if it assumes rights the institution does not have. Netting is not simply a convenient portfolio aggregation rule. It requires the relevant contracts to be covered by an arrangement whose closeout provisions are enforceable under the applicable circumstances. If enforceability is uncertain, a prudent model may need to show a separate scenario with less or no netting rather than treating the modeled offset as certain.

Guarantees also have a defined scope. A guarantee may have a cap, cover only specified transactions, exclude particular claims, or require that procedural conditions be met. A guarantee from a financially weak or highly correlated guarantor may provide limited economic protection even if its stated amount is large. In particular, a guarantor whose credit quality deteriorates when the counterparty deteriorates can create substantial wrong-way risk.

The following comparison keeps the distinct effects visible:

Feature	What it changes in the model	Risk that remains
Closeout netting	Combines signed transaction values within an eligible netting set	Legal enforceability, scope, and any positive net claim still matter
Guarantee	Transfers a specified part of a claim to the guarantor	Coverage limits, guarantor default, recovery, and dependence remain
Both together	Guarantee applies to the exposure remaining after the modeled closeout calculation, subject to its terms	Incorrect netting scope or guarantee allocation can understate the claim at risk

A particularly important exam trap is netting across legally separate sets. Suppose one set has a positive net closeout amount while another has a negative one. Combining the two signed totals is valid only if the relevant agreements permit cross-set netting. Otherwise the negative amount cannot be used to reduce the positive claim. The guarantee should likewise be assigned only to the claim it actually covers.

Worked Example: Separate Netting Sets and a Limited Guarantee

A hypothetical insurer has two groups of transactions with one counterparty, with values in thousands of dollars. Set A contains $+900$ and $-600$, and those transactions can be legally netted. Set B contains $+500$ and $-200$, but the two transactions are not covered by enforceable netting rights, so the positive claim cannot be offset by the negative amount. A guarantee covers up to $\$250{,}000$ of Set A’s exposure and does not cover Set B. The counterparty default probability is $0.040$, and its recovery rate is $0.40$. Conditional on counterparty default, the guarantor default probability is $0.15$, with recovery rate $0.30$. Calculate expected loss.

Method. Calculate exposure separately for each legal set, then apply the guarantee only to the eligible Set A claim.

Set A’s net amount is $900-600=300$ thousand dollars, so its exposure is $\$300{,}000$. In Set B, the transactions cannot be netted, so the positive claim produces $\$500{,}000$ of exposure; the $\$200{,}000$ negative value does not reduce it. Total modeled exposure is $\$300{,}000+\$500{,}000=\$800{,}000$.

The guarantee covers $H=\$250{,}000$ of Set A. The remaining unguaranteed exposure is $\$800{,}000-\$250{,}000=\$550{,}000$. Its loss conditional on counterparty default is $\$550{,}000(1-0.40)=\$330{,}000$. If the guarantor defaults, the loss on the covered claim is $\$250{,}000(1-0.30)=\$175{,}000$. Its expected contribution conditional on counterparty default is $0.15(\$175{,}000)=\$26{,}250$.

The conditional expected loss is $\$330{,}000+\$26{,}250=\$356{,}250$, so expected loss is

E[L]=0.040(\$356{,}250)=\$14{,}250.

A mistaken calculation might net every transaction across both sets: $900-600+500-200=\$600{,}000$. That would understate exposure by $\$200{,}000$, because it uses Set B’s negative value to offset a positive claim despite the stated lack of enforceable netting. The correction is to calculate each legally distinct set according to its own rights, as above.

Common Modeling Checks

Do not confuse gross and net values. Gross positive transaction values are appropriate only when the applicable contracts cannot be netted; signed values belong in the calculation for an enforceable netting set.
Do not treat a negative balance as negative exposure. The positive-part operation sets exposure to zero when the institution owes at closeout.
Do not apply a guarantee before determining its scope. A cap or transaction restriction can leave a substantial uncovered amount.
Do not assume the guarantor is independent of the counterparty. Use a conditional default probability, or otherwise model their joint default behavior.
Do not interpret reduced expected loss as eliminated risk. Legal uncertainty, default dependence, and recovery uncertainty can remain material.

These calculations are deliberately structured so that exposure reduction and loss reduction can be audited separately. First verify the legal set and its closeout balance. Then verify collateral and guarantee allocation. Finally check that every probability and recovery assumption is applied to the correct claim and event.

Check Your Understanding

For each problem, calculate using the stated legal scope and default assumptions.

A legally enforceable netting set has transaction values of $+\$420{,}000$, $-\$170{,}000$, $-\$80{,}000$, and $+\$260{,}000$. There is no collateral. The counterparty default probability is $0.025$, and its recovery rate is $0.45$. What is expected loss? (Answer: $\$5{,}912.50$)
A counterparty has exposure of $\$1{,}200{,}000$. A guarantee covers 40% of exposure, capped at $\$350{,}000$. The counterparty recovery rate on uncovered claims is $0.40$; the guarantor’s recovery rate is $0.50$, and its default probability conditional on counterparty default is $0.08$. The counterparty default probability is $0.03$. What is expected loss? (Answer: $\$15{,}720$)
Set A has transaction values $+\$700{,}000$ and $-\$450{,}000$, with enforceable netting. Set B has values $+\$400{,}000$ and $-\$150{,}000$, but its transactions cannot be netted. A guarantee covers $\$200{,}000$ of Set A only. The counterparty recovery rate is $0.30$; the guarantor recovery rate is $0.40$, and its default probability conditional on counterparty default is $0.20$. The counterparty default probability is $0.05$. What is expected loss? (Answer: $\$16{,}950$)
A netting set has exposure of $\$600{,}000$, of which $\$300{,}000$ is guaranteed. The counterparty recovery rate is $0.50$, the guarantor recovery rate is $0.20$, and the guarantor default probability conditional on counterparty default is $0.25$. If the counterparty default probability is $0.02$, what is expected loss under the stated guarantee model? (Answer: $\$4{,}200$)
Two transaction groups have signed values, in thousands of dollars, of $+800,-500$ in Group A and $+350,-250$ in Group B. Netting is enforceable within each group but not across the groups. There is no guarantee or collateral; counterparty recovery is $0.25$, and default probability is $0.04$. Calculate total exposure and expected loss. (Answer: exposure $\$400{,}000$; expected loss $\$12{,}000$)

What You'll Learn