Expected Credit Loss And Security Loads

Why Expected Credit Loss Matters

In a large-deductible arrangement, an insurer may pay covered claims and later seek reimbursement from the policyholder. The amount owed is not necessarily the amount ultimately collected: the counterparty may default, collateral may realize less than its stated value, and the remaining unsecured balance may be only partly recovered. A pricing indication that treats every reimbursement as certain can therefore understate expected cost.

Expected credit loss (ECL) quantifies the average shortfall attributable to counterparty default over a stated horizon. It can be expressed as a dollar amount, a rate on exposure at default, or a component of an indicated premium. The calculation depends on keeping the default event, collateral recovery, and unsecured recovery assumptions distinct.

Definition: Let $EAD$ be the exposure at default, $PD$ the probability of default over the measurement horizon, and $LGD$ the dollar loss conditional on default. Under fixed $EAD$ and $PD$, expected credit loss is $$ ECL = PD \times E[LGD \mid \text{default}]. $$ The conditional expectation allows collateral and unsecured recovery to vary across outcomes after default. Unless stated otherwise, amounts are measured at the same valuation date, with no discounting for the time between default and recovery.

Modeling Collateral and Unsecured Recovery

Suppose collateral has face amount $V$ and a haircut $h$, where $0 \leq h \leq 1$. The haircut-adjusted collateral value is $C=V(1-h)$. If the collateral is assumed fully realizable at this adjusted value in default, the amount credited against the exposure is capped at $EAD$. The residual unsecured balance is the exposure not covered by that recovery.

Let $R$ be the fraction of the residual unsecured balance recovered conditional on default. With full realization of haircut-adjusted collateral, the conditional loss is

LGD = \max(EAD-\min(EAD,C),0)(1-R).

If $C \leq EAD$, this simplifies to $LGD=(EAD-C)(1-R)$. The formula assumes collateral proceeds are applied first to the exposure and that $R$ applies only to the remaining unsecured balance. It does not apply unchanged if collateral and unsecured recoveries overlap, recovery costs have not been allowed for, or the collateral is not available to satisfy this obligation.

If the adjusted collateral itself may not be fully realized, let $K$ be the fraction of $C$ recovered in default. The collateral recovery is then $\min(EAD,KC)$, and the residual unsecured balance becomes $\max(EAD-\min(EAD,KC),0)$. The conditional loss is that balance multiplied by $1-R$. When $K$ or $R$ is uncertain, calculate conditional loss in each outcome and probability-weight those losses conditional on default.

Why the Expected-Loss Formula Holds

The loss is zero if default does not occur. If default occurs, the loss is $LGD$, which may vary with the recovery outcome. Taking the unconditional expectation therefore gives the probability of default multiplied by the expected loss conditional on default:

E[\text{credit loss}] = P(\text{default})E[LGD\mid\text{default}] + P(\text{no default})\times 0 = PD \times E[LGD\mid\text{default}].

This derivation also explains an important modeling point: recovery scenarios must be weighted by their probabilities conditional on default. A recovery percentage is not itself a default probability, and multiplying by $PD$ does not replace the need to calculate the conditional loss correctly.

Set the horizon and exposure.
Use the $EAD$ and $PD$ that refer to the same horizon and reimbursement obligation.

Estimate recoveries conditional on default.
Apply any collateral haircut and realization assumption, then calculate the residual balance and unsecured recovery.

Average conditional losses.
For multiple outcomes, multiply each outcome's loss by its probability given default and sum.

Apply the probability of default.
Multiply the conditional expected loss by $PD$ to obtain ECL.

Choose a load basis.
Divide ECL by the stated exposure or premium basis, or add the dollar ECL to a base indication if that treatment is appropriate.

Worked Calculations

Worked Example: Probability-Weighted Unsecured Recovery

A hypothetical reimbursement exposure has $EAD=\$200{,}000$, $PD=4\%$, and effective collateral of $\$80{,}000$. Conditional on default, the unsecured recovery fraction is 25% with probability 40% or 75% with probability 60%. Calculate ECL and the load rate on EAD.

Method and assumptions. Collateral is fully realized at its stated effective value and applied first. The recovery scenarios and probabilities are conditional on default. There are no additional recovery costs.

Step 1: Calculate the unsecured balance. It is $\$200{,}000-\$80{,}000=\$120{,}000$.

Step 2: Find the expected unsecured recovery fraction. The probability-weighted fraction is $0.40(0.25)+0.60(0.75)=0.10+0.45=0.55$. Thus the expected loss fraction on the unsecured balance is $1-0.55=0.45$, and conditional expected loss is $\$120{,}000(0.45)=\$54{,}000$.

Step 3: Apply the probability of default. $ECL=0.04(\$54{,}000)=\$2{,}160$.

Step 4: Calculate the EAD load. $\$2{,}160/\$200{,}000=0.0108=1.0800\%$.

Check and trap avoided. The conditional loss is based on the 45% unrecovered share, not the 55% recovered share. The resulting ECL is less than the default probability times the full unsecured balance: $\$2{,}160 < 0.04(\$120{,}000)=\$4{,}800$, as expected when unsecured recovery is positive.

Worked Example: Minimum Face Collateral for an ECL Target

A hypothetical account has $EAD=\$500{,}000$, $PD=2\%$, and unsecured recovery of 50% conditional on default. Collateral has a 10% haircut. Find the minimum collateral face amount, to the nearest cent, that makes ECL no more than $\$2{,}000$, and verify the rounded amount.

Method and assumptions. Collateral is fully realized at its haircut-adjusted value and applied before unsecured recovery. The face amount is not greater than needed to cover the exposure.

Step 1: Express ECL in terms of effective collateral. If $C$ is the effective collateral value, then $ECL=0.02(\$500{,}000-C)(1-0.50)$. The target requires $0.01(\$500{,}000-C)\leq \$2{,}000$.

Step 2: Solve the inequality. Dividing by 0.01 gives $\$500{,}000-C\leq \$200{,}000$, so $C\geq \$300{,}000$. Since $C=0.90V$, the face amount must satisfy $V\geq \$300{,}000/0.90=\$333{,}333.333\ldots$. Rounding upward to cents gives $\$333{,}333.34$.

Step 3: Verify the rounded amount. Effective collateral is $0.90(\$333{,}333.34)=\$300{,}000.006$. The unsecured balance is $\$500{,}000-\$300{,}000.006=\$199{,}999.994$. Conditional loss is $\$199{,}999.994(0.50)=\$99{,}999.997$, so ECL is $0.02(\$99{,}999.997)=\$1{,}999.99994$, which is no more than $\$2{,}000$.

Check and trap avoided. One cent less face collateral gives effective collateral of $\$299{,}999.997$, ECL of $0.02(0.50)(\$200{,}000.003)=\$2{,}000.00003$, just above the target. The 50% unsecured recovery must be applied to the balance after collateral; omitting it would produce a different threshold.

Worked Example: Collateral Realization Scenarios and a Premium Load

A hypothetical reimbursement has $EAD=\$100{,}000$, $PD=3\%$, and haircut-adjusted collateral of $\$60{,}000$. Conditional on default, 100% of that value is realized with probability 70%, while 80% is realized with probability 30%. The unsecured recovery fraction is 40% in either scenario. If the base indicated premium is $\$12{,}000$, calculate ECL, the load rate on EAD, and the load as a percentage of base premium.

Method and assumptions. The realization scenarios are conditional on default. Recovered collateral is applied first; the 40% unsecured recovery applies to the balance left after collateral recovery.

Step 1: Calculate the loss in each default scenario. With full realization, collateral recovery is $\$60{,}000$, leaving $\$40{,}000$ unsecured. The loss is $\$40{,}000(1-0.40)=\$24{,}000$. With 80% realization, collateral recovery is $0.80(\$60{,}000)=\$48{,}000$, leaving $\$52{,}000$ unsecured. The loss is $\$52{,}000(0.60)=\$31{,}200$.

Step 2: Average conditional losses. Conditional expected loss is $0.70(\$24{,}000)+0.30(\$31{,}200)=\$16{,}800+\$9{,}360=\$26{,}160$.

Step 3: Apply $PD$ and calculate the two rates. ECL is $0.03(\$26{,}160)=\$784.80$. The EAD load is $\$784.80/\$100{,}000=0.7848\%$. The load relative to base premium is $\$784.80/\$12{,}000=0.0654=6.54\%$.

Check and trap avoided. If full collateral realization were incorrectly assumed in every default, ECL would be $0.03(\$24{,}000)=\$720$, understating this scenario-based result by $\$64.80$. The premium load and EAD load use different denominators; neither percentage should be reported without naming its basis.

Translating ECL Into a Pricing or Security Load

An ECL amount can be translated into a rate only after identifying the denominator. A load on exposure is $ECL/EAD$. A load as a percentage of base premium is $ECL/P_0$, where $P_0$ is that premium. If the pricing method adds expected dollar loss to a base indication, the adjusted indication is $P_0+ECL$; if the method instead applies a multiplicative load to premium, its factor is $1+ECL/P_0$. These are equivalent dollar additions under the stated assumptions, but the reported percentage depends on the chosen base.

The load should represent expected credit loss not already included elsewhere in the indication. Adding it twice, or applying an exposure-based rate to a premium denominator without conversion, misstates the indicated amount. If EAD changes with the claim estimate, collateral, or reimbursement limit, recalculate ECL using that exposure rather than carrying forward a rate from an inconsistent base.

What ECL Does—and Does Not—Measure

ECL is a probability-weighted average, not a worst-case loss, a collateral target by itself, or a measure of the funds needed to withstand a severe default scenario. A security requirement may be set above ECL to address uncertainty, liquidity, or a specified stress tolerance. Conversely, posted collateral does not eliminate ECL if it is subject to a haircut, uncertain realization, or residual unsecured exposure.

For this calculation, specify whether the haircut already reflects expected realization costs and whether the unsecured recovery estimate is conditional on default. Keep those assumptions consistent across scenarios. Do not multiply an already probability-weighted recovery estimate by a second probability, and do not apply an unsecured recovery fraction to the full exposure when collateral has already reduced the balance at risk.

Key takeaway: Calculate each default outcome's net shortfall after collateral and unsecured recovery, average those shortfalls conditional on default, and then multiply by $PD$. State whether the resulting load is measured against EAD or premium, and keep it distinct from a separate collateral or stress-security requirement.

Check Your Understanding

For each hypothetical reimbursement, calculate the requested expected-loss amount or load. Recovery assumptions are conditional on default.

An account has $EAD=\$150{,}000$, $PD=3\%$, effective collateral of $\$45{,}000$, and a 60% unsecured recovery fraction. Find ECL and the load rate on EAD. (Answer: ECL $\$1{,}260$; EAD load $0.8400\%$)
An account has $EAD=\$320{,}000$, $PD=1.5\%$, collateral face amount $\$100{,}000$, a 20% haircut, and 50% unsecured recovery. A base premium is $\$60{,}000$. Find ECL and ECL as a percentage of base premium. (Answer: ECL $\$1{,}800$; premium load $3.00\%$)
An account has $EAD=\$250{,}000$, $PD=2.4\%$, and effective collateral of $\$90{,}000$. Conditional unsecured recovery is 25% with probability 20% and 75% with probability 80%. Find ECL and the EAD load. (Answer: ECL $\$1{,}344$; EAD load $0.5376\%$)
An account has $EAD=\$400{,}000$, $PD=2\%$, a 12% collateral haircut, and 40% unsecured recovery. Find the minimum collateral face amount, rounded upward to cents, that makes ECL no more than $\$2{,}500$. (Answer: $\$217{,}803.04$)

What You'll Learn