Counterparty Credit Adjustments to Program Value

Why Counterparty Credit Changes Program Value

A reinsurance contract can promise a valuable recovery while leaving the cedent exposed to the possibility that the reinsurer cannot pay. Program value therefore depends not only on the contractual recovery, premium, and expenses, but also on the amount expected to be collected and the variability of that collection. A larger promised recovery is not automatically a better economic outcome if it is concentrated with a weak counterparty or is poorly collateralized.

This tutorial focuses on adjusting reinsurance benefits for counterparty credit risk. It distinguishes the expected credit loss from a separate charge for the uncertainty of credit loss. The latter is a decision convention, not a universal pricing rule: its level should be stated whenever it is used. The analysis assumes a specified horizon and, unless noted otherwise, treats promised recoveries, default probabilities, collateral, and recovery rates as known for that horizon.

Definition: Let $R$ be the promised recovery over the analysis horizon, $q$ the counterparty's probability of default, $c$ the fraction of $R$ protected by collateral that is available in default, and $r$ the fraction of the remaining unsecured amount recovered after default. Under this simplified model, the credit loss is $L=R(1-c)(1-r)D$, where $D=1$ if the counterparty defaults and $D=0$ otherwise. The expected collectible recovery is $R-E[L]$.

The model treats collateral and post-default recovery as distinct. Collateral protects the fraction $c$ first; the recovery rate $r$ applies only to the amount left unsecured. This convention assumes collateral is legally enforceable, available when needed, and valued without a haircut or delay. If collateral has a haircut, incurs costs, or is exposed to the same event that causes the counterparty default, its effective protection must be reduced or modeled explicitly. The formula also assumes a single default event over the horizon; it does not by itself model changing credit quality, timing of payments, or disputes.

Expected Credit Loss and Collectible Recovery

Let $s=R(1-c)(1-r)$ be the loss conditional on default. Since $D$ is a Bernoulli random variable with $P(D=1)=q$, the loss is $s$ with probability $q$ and zero with probability $1-q$. Therefore:

$$ E[L]=qs=qR(1-c)(1-r) $$

The expected collectible recovery is the promised amount less this expected credit loss:

E[\text{collectible recovery}] =R-qR(1-c)(1-r) =R\left[1-q(1-c)(1-r)\right].

The justification is direct: with probability $1-q$, the cedent collects $R$; with probability $q$, it collects $R-s$. Thus the expected collection is $(1-q)R+q(R-s)=R-qs$. This expected value is not a guarantee for any individual year. It is an average under the stated probability model.

Worked Example: Adjusting a Collateralized Recovery

A hypothetical treaty promises a recovery of $1,000 thousand. The reinsurer's default probability over the treaty horizon is $q=0.0200$. Collateral is expected to cover $c=0.2500$ of the promised recovery, and the cedent expects to recover $r=0.4000$ of the remaining unsecured amount in default. The premium is $925 thousand and expenses are $8 thousand. Calculate the expected collectible recovery and the expected net program value. Assume no discounting and no other benefits or costs.

Method and assumptions. Use the one-period Bernoulli default model. The recovery, collateral fraction, and recovery rate are fixed; there is no interest adjustment, and collateral is fully available as specified. The expected net program value is expected collectible recovery less premium and expenses.

Step 1: Find the unsecured fraction. Collateral covers 0.2500 of the recovery, so the remaining fraction is $1-0.2500=0.7500$.

Step 2: Find the loss fraction conditional on default. The fraction not recovered from the unsecured amount is $1-0.4000=0.6000$. The conditional credit-loss fraction of the promised recovery is $0.7500(0.6000)=0.4500$. Thus the loss in default is $1{,}000(0.4500)=\$450$ thousand.

Step 3: Calculate expected credit loss. The expected loss is $0.0200(450)=\$9$ thousand.

Step 4: Calculate the expected collectible recovery and net value. The expected collection is $1{,}000-9=\$991$ thousand. After premium and expenses, expected net value is $991-925-8=\$58$ thousand.

Check. The probability-weighted collection is $0.9800(1{,}000)+0.0200(550)=980+11=\$991$ thousand, agreeing with the expected-loss calculation. The common mistake avoided is applying the 40% recovery rate to the entire promised recovery before recognizing that collateral protects 25% of it.

Comparing Program Alternatives

For a consistent comparison, define each alternative's value on the same basis. If $B_j$ is the expected program benefit before counterparty credit loss for alternative $j$, $P_j$ its premium, and $C_j$ its other costs, then expected value is:

$$ V_j=B_j-P_j-C_j-E[L_j]. $$

Here, $B_j$ must exclude the credit loss already represented by $E[L_j]$. If the starting benefit is already credit-adjusted, subtracting expected credit loss again would double count it. When program alternatives offer the same promised recovery, the premium difference and credit adjustment can still lead to different rankings.

Worked Example: A Cheaper Counterparty Versus Better Credit Protection

Two hypothetical alternatives each promise $800 thousand of recovery over the same horizon and have no other expenses. Alternative A has a premium of $610 thousand, default probability $q_A=0.0300$, no collateral ($c_A=0$), and a 50% recovery on the unsecured amount ($r_A=0.5000$). Alternative B has a premium of $620 thousand, default probability $q_B=0.0200$, collateral covering half the recovery ($c_B=0.5000$), and a 40% recovery on the unsecured amount ($r_B=0.4000$). With no discounting, compare expected net values.

Method and assumptions. Apply $E[L]=qR(1-c)(1-r)$ separately to each alternative. The promised recovery and horizon are identical, and there are no expenses other than the stated premiums.

Alternative A. The conditional loss is $800(1-0)(1-0.5000)=800(1)(0.5000)=\$400$ thousand. Expected credit loss is $0.0300(400)=\$12$ thousand. Expected collectible recovery is $800-12=\$788$ thousand, so expected net value is $788-610=\$178$ thousand.

Alternative B. The conditional loss is $800(1-0.5000)(1-0.4000)=800(0.5000)(0.6000)=\$240$ thousand. Expected credit loss is $0.0200(240)=\$4.8$ thousand. Expected collectible recovery is $800-4.8=\$795.2$ thousand, so expected net value is $795.2-620=\$175.2$ thousand.

Comparison and check. Alternative A's expected net value is $178-175.2=\$2.8$ thousand higher. B's credit adjustment improves expected collection by $795.2-788=\$7.2$ thousand, but its premium is $620-610=\$10$ thousand higher. The premium difference exceeds the expected credit benefit by $10-7.2=\$2.8$ thousand, matching the value comparison. The mistake to avoid is selecting B solely because it has lower expected credit loss without also accounting for its higher premium.

Concentration and the Variability of Credit Loss

Expected credit loss alone does not describe the uncertainty of collection. In the single-counterparty model, $L=sD$, so $E[L]=qs$ and:

\operatorname{Var}(L) =E[L^2]-E[L]^2 =qs^2-(qs)^2 =s^2q(1-q).

The variance is zero when default is certain or impossible, and is largest for a fixed severity when the probability is near one-half. For multiple counterparties, total credit loss is the sum of their losses. Its expectation is the sum of expected losses whether or not defaults are independent, but its variance depends on dependence:

\operatorname{Var}\left(\sum_j L_j\right) =\sum_j\operatorname{Var}(L_j) +2\sum_{j<k}\operatorname{Cov}(L_j,L_k).

If defaults are independent and the loss severity for each counterparty is fixed, the covariance terms are zero. Splitting exposure among independent counterparties can then reduce variance while leaving total expected credit loss unchanged. If defaults are positively dependent, the covariance terms increase aggregate variance and reduce the diversification benefit. Independence should not be assumed just because the reinsurers have different names; shared ownership, geography, collateral banks, or exposure to the same shock can create dependence.

Worked Example: Applying a Credit-Loss Risk Charge

A hypothetical program has an expected value of $230 thousand after premiums, expenses, and expected credit losses have already been reflected. Its remaining portfolio credit-loss uncertainty consists of two independent counterparties. Each has default probability $q=0.2000$ and a conditional credit-loss severity of $87.5 thousand. Management applies an additional charge equal to 50% of the standard deviation of aggregate credit loss. Calculate the risk charge and value after that charge.

Method and assumptions. Each counterparty's loss is $L_j=87.5D_j$, where $D_j$ is Bernoulli with probability 0.2000. The two default indicators are independent. The stated charge is a management convention, applied after expected credit losses have already been included in the $230 thousand value.

Step 1: Calculate the variance for one counterparty. For each, $\operatorname{Var}(L_j)=87.5^2(0.2000)(0.8000)$. Since $87.5^2=7{,}656.25$ and $0.2000(0.8000)=0.1600$, the variance is $7{,}656.25(0.1600)=1{,}225$ thousand-squared.

Step 2: Aggregate the variances. Independence makes the covariance zero, so total variance is $1{,}225+1{,}225=2{,}450$ thousand-squared. The standard deviation is $\sqrt{2{,}450}=49.497474\ldots$ thousand.

Step 3: Calculate the charge and adjusted value. The charge is $0.50\sqrt{2{,}450}=24.748737\ldots$ thousand, or $24.7487 thousand before final rounding. The value after the charge is $230-24.748737\ldots=205.251263\ldots$ thousand, or $205.25 thousand to the nearest $0.01 thousand.

Check. Each counterparty's expected loss is $0.2000(87.5)=\$17.5$ thousand, so expected aggregate credit loss is $35 thousand. That expected loss is already in the stated $230 thousand starting value and is not deducted again. The common mistake avoided is rounding the standard deviation too early: using the unrounded value gives a charge of approximately $24.7487 thousand and a value of approximately $205.2513 thousand.

Worked Example: Comparing One Counterparty With Two Independent Counterparties

A hypothetical $600 thousand recovery is placed either with one counterparty or equally with two independent counterparties. Each counterparty has default probability $q=0.0400$, and 50% of the amount exposed to that counterparty is lost if it defaults. There is no collateral. Assume the expected program value after premiums, expenses, and expected credit losses is $230 thousand under either arrangement. Management charges 40% of the standard deviation of credit loss. Compare the risk charges and values after charge.

Method and assumptions. Use Bernoulli default indicators, with independent defaults in the split arrangement. The risk charge is $0.40\operatorname{SD}(L)$; the starting expected value is already credit-adjusted, so the charge is the only further deduction.

One counterparty. The loss severity in default is $600(0.5000)=\$300$ thousand. Its variance is $300^2(0.0400)(0.9600)=90{,}000(0.0384)=3{,}456$ thousand-squared. The standard deviation is $\sqrt{3{,}456}=58.7878\ldots$ thousand. The charge is $0.40(58.7878\ldots)=\$23.5151\ldots$ thousand, which rounds to $23.52 thousand. Value after charge is $230-23.5151\ldots=206.4849\ldots$, or $206.48 thousand.

Two counterparties. Each receives $600/2=\$300$ thousand, so each conditional loss severity is $300(0.5000)=\$150$ thousand. Each variance is $150^2(0.0400)(0.9600)=22{,}500(0.0384)=864$ thousand-squared. Independence gives aggregate variance $864+864=1{,}728$ thousand-squared. The standard deviation is $\sqrt{1{,}728}=41.5692\ldots$ thousand. The charge is $0.40(41.5692\ldots)=\$16.6277\ldots$ thousand, which rounds to $16.63 thousand. Value after charge is $230-16.6277\ldots=213.3723\ldots$, or $213.37 thousand.

Comparison and check. Splitting reduces the charge by $23.5151\ldots-16.6277\ldots=6.8874\ldots$, or $6.89 thousand to the nearest cent. Expected credit loss is unchanged: with one counterparty it is $600(0.5000)(0.0400)=\$12$ thousand; with two it is $2(300)(0.5000)(0.0400)=\$12$ thousand. The split improves value here only under the stated independent-default and charge assumptions. The common mistake avoided is treating split exposure as if it reduced expected loss; it reduces variance in this model, not the expected loss.

Practical Interpretation and Exam Traps

The expected-value adjustment and the risk charge answer different questions. Expected credit loss measures the average shortfall under the model. A standard-deviation charge or another risk measure reflects uncertainty around that expected shortfall. Such a charge must be applied consistently: do not deduct it if the starting value already includes it, and do not present a chosen multiple as a universal market or regulatory requirement.

A useful comparison sequence is to calculate, for each alternative, the conditional loss in default, expected credit loss, expected collectible recovery, and expected net value. Add a separate risk charge only if the stated decision framework calls for one. This sequence makes it easier to identify whether a value difference comes from the premium, expected credit loss, or concentration of exposure.

Applying recovery to the wrong amount. If collateral protects fraction $c$, apply $r$ to the remaining fraction $1-c$, not to the full promised recovery.
Confusing expected loss with loss in default. The loss in default is $s$; expected loss is $qs$. Multiplying by $q$ twice understates the loss.
Assuming diversification without checking dependence. Variances add only without covariance terms when losses are independent or uncorrelated. Positive dependence can materially reduce the benefit of splitting.
Double counting credit adjustments. Confirm whether the starting program value already includes expected credit loss before subtracting it.
Overstating collateral protection. Collateral is not automatically equivalent to cash if it can be delayed, disputed, impaired, or unavailable in the same stress that causes default.

These calculations support a credit-aware comparison of reinsurance arrangements, but they remain conditional on the assumptions supplied. A more detailed model may need counterparty-specific default probabilities, uncertain recoveries, collateral haircuts, timing, and dependence between underwriting losses and counterparty default. The central discipline is to distinguish contractual recovery from expected collection, then distinguish expected collection from the additional compensation required for uncertainty.

Check Your Understanding

For each problem, use the stated horizon and assumptions; report money in thousands unless otherwise indicated.

A treaty promises $500 thousand. Default probability is $0.0200$, collateral covers $0.2000$ of the recovery, and recovery on the unsecured amount is $0.5000$. Find expected credit loss and expected collectible recovery. (Answer: Expected credit loss is $4 thousand; expected collectible recovery is $496 thousand.)
A hypothetical $1,200 thousand recovery has default probability $0.0250$, collateral fraction $0.4000$, and recovery rate $0.2500$ on the unsecured amount. Find the conditional loss in default and expected credit loss. (Answer: Conditional loss is $540 thousand; expected credit loss is $13.5 thousand.)
A $400 thousand recovery is split equally between two independent counterparties. Each has default probability $0.0500$, no collateral, and a 60% loss fraction in default. Calculate total expected credit loss and aggregate standard deviation of credit loss. (Answer: Expected credit loss is $12 thousand; aggregate standard deviation is $\sqrt{1{,}368}=36.9865\ldots$, or $36.99 thousand.)
Two independent counterparties each support a $250 thousand recovery. Each has default probability $0.0400$, no collateral, and a 50% loss fraction in default. Management applies a 30% charge to aggregate credit-loss standard deviation. Find the expected credit loss, the charge, and the expected-value reduction including expected credit loss and charge. (Answer: Expected credit loss is $10 thousand; standard deviation is $\sqrt{1{,}200}=34.6410\ldots$ thousand; charge is $10.39 thousand; total reduction is $20.39 thousand.)

What You'll Learn