Stress Testing Collateral Adequacy

Why Stress Testing Collateral Matters

A collateral estimate based on an expected unpaid obligation is a useful starting point, but it does not show how the obligation could change when experience is adverse. Claim frequency may rise, the severity mix may shift toward losses that penetrate the deductible, and reporting delays may leave more incurred claims outside the reported inventory. If the insured then fails to reimburse the insurer, the security may not cover the amount at risk.

A stress test makes these assumptions explicit and asks a practical question: under the stated adverse scenario, how much of the reimbursement obligation is not covered by current security? A stress scenario is not automatically a forecast, a probability percentile, or an expected loss. It is a conditional calculation whose value depends on the assumptions selected.

Definition: Let $O$ be the gross unpaid reimbursement obligation at the valuation date, including estimated obligations for incurred but unreported claims where appropriate. Let $q$ be the fraction of that obligation assumed to remain unpaid in the specified reimbursement-default scenario after other collections. Let $C$ be current security assumed available for this test. The scenario collateral need is $qO$, and the scenario shortfall is $\max(0,qO-C)$. An excess is $\max(0,C-qO)$.

The definition separates two questions. First, how large is the unpaid obligation under the claim and reporting assumptions? Second, what portion of it is at risk of noncollection in the defined default scenario? If the scenario assumes that all obligations remain unpaid on default, set $q=1$. If the scenario assumes some recovery outside security, $q$ is the residual unpaid fraction. Do not reduce $O$ for collections already reflected in its valuation and then subtract those same collections again through $q$.

For the examples below, deductible reimbursement per claim is modeled as $Y^L=\min(X,d)$, where $X$ is the loss and $d$ is the per-claim deductible. This assumes no separate expense reimbursement, aggregate deductible, per-claim limit, or other contract adjustment. When such terms apply, the payment calculation must reflect them before it is used in the stress test.

Build the Claim Obligation Before Applying Default

Frequency and severity assumptions determine the amount of deductible-layer payments generated by claims. For a discrete severity distribution, the expected payment per claim is the probability-weighted average of the payment at each severity. The expected obligation for a specified claim count is the count multiplied by that average. This expected amount is an input to the scenario; it should not be confused with a high-percentile aggregate loss unless the scenario explicitly specifies one.

The deductible truncates each loss payment. A claim above the deductible contributes $d$, not its full severity, to $Y^L$. Therefore, a severity stress should be applied to the loss distribution and the payment recalculated, rather than applying a severity factor mechanically to the expected payment without checking the contract layer.

Worked Example: Frequency and Severity Stress in the Deductible Layer

A hypothetical program has 40 expected incurred but unreported claims in the liability window. The baseline severity distribution is $P(X=\$3{,}000)=0.5$ and $P(X=\$8{,}000)=0.5$, with a deductible of $d=\$5{,}000$. In an adverse scenario, frequency increases by 25%, and the severity probabilities become 0.25 and 0.75, respectively. Reported open claims already have a gross unpaid reimbursement estimate of $180,000. Assume the adverse claim assumptions apply only to the unreported claims. The default scenario leaves 40% of gross obligations unpaid, and current security is $150,000. Find the stressed shortfall.

Method. Recalculate the expected deductible payment using the stressed severity distribution, apply the stressed count, add the reported open obligation, and only then apply the default fraction.

At $X=\$3{,}000$, the deductible payment is $\min(\$3{,}000,\$5{,}000)=\$3{,}000$. At $X=\$8{,}000$, it is $\min(\$8{,}000,\$5{,}000)=\$5{,}000$. Thus the stressed expected payment per claim is

E[Y^L]=0.25(\$3{,}000)+0.75(\$5{,}000)=\$750+\$3{,}750=\$4{,}500.

The stressed expected claim count is $40(1.25)=50$. The unreported-claim obligation is therefore $50(\$4{,}500)=\$225{,}000$. Add the reported open obligation, which is held fixed by the scenario:

O=\$225{,}000+\$180{,}000=\$405{,}000.

The default scenario leaves $q=0.40$ unpaid, so the amount requiring security is $0.40(\$405{,}000)=\$162{,}000$. The shortfall is $\max(0,\$162{,}000-\$150{,}000)=\$12{,}000$. Check: current security covers $150,000 of the $162,000 at-risk amount, leaving exactly $12,000 uncovered. The common mistake avoided is applying the frequency and severity stresses to the reported obligation as well, even though the scenario says those assumptions apply only to unreported claims.

Allow for Claims That Have Not Yet Been Reported

A valuation based only on reported claims can understate the obligation if claims have occurred but have not yet been reported. A simple delay model estimates the unreported count by combining the claim occurrence rate with the probability that a claim of a given age remains unreported. The model must match the scenario’s timing and reporting-delay assumptions.

For example, suppose claims occur at a constant rate of $\lambda$ claims per month during the $T$ months before valuation, and reporting delay $D$ is uniformly distributed from zero to $T$ months. A claim that occurred $a$ months ago is unreported at valuation when $D>a$, with probability $(T-a)/T$. The expected unreported count is therefore

E[N_{\mathrm{unreported}}] =\int_0^T \lambda P(D>a)\,da =\int_0^T \lambda\left(1-\frac{a}{T}\right)da =\frac{\lambda T}{2}.

The result follows because the unreported probability declines linearly from 1 for a just-occurred claim to 0 for a claim $T$ months old. The integral is the occurrence rate multiplied by the area under that probability curve. This expression applies to the stated constant-rate and uniform-delay assumptions; a different reporting pattern requires its own probability function and calculation.

Worked Example: Reporting Delay and Security Shortfall

A hypothetical portfolio has an occurrence rate of 8 claims per month during the four months before valuation. Assume each claim’s reporting delay is uniformly distributed from zero to four months. A frequency stress increases the rate by 50%. Severity is $P(X=\$3{,}000)=0.4$ and $P(X=\$9{,}000)=0.6$, and $d=\$6{,}000$. There are no other unpaid obligations in this illustration. The default scenario leaves all of these estimated obligations unpaid, and current security is $100,000. Calculate the stress amount and shortfall.

Step 1: Estimate the unreported count. Under the baseline rate, the formula gives $8(4)/2=16$ expected unreported claims. Applying the 1.50 frequency factor gives $16(1.50)=24$ stressed expected unreported claims.

Step 2: Calculate payment per claim. The deductible payments at the two severities are $\min(\$3{,}000,\$6{,}000)=\$3{,}000$ and $\min(\$9{,}000,\$6{,}000)=\$6{,}000$. Therefore,

E[Y^L]=0.4(\$3{,}000)+0.6(\$6{,}000)=\$1{,}200+\$3{,}600=\$4{,}800.

Step 3: Calculate the gross obligation. The estimated obligation is $24(\$4{,}800)=\$115{,}200$. Because the default scenario leaves all of it unpaid, $q=1$, so the scenario collateral need is $115,200.

Step 4: Compare with security. The shortfall is $\max(0,\$115{,}200-\$100{,}000)=\$15{,}200$. Check: $100,000 of security plus a $15,200 shortfall equals the $115,200 scenario need. The common mistake avoided is counting only claims reported by valuation and omitting claims whose modeled reporting delays keep them unreported.

Represent Reimbursement Default Explicitly

Default assumptions should be stated in terms of the portion of the gross obligation that is not collected in the scenario. A default probability and a default recovery assumption are different quantities. The probability describes how likely the scenario is; the unpaid fraction describes the amount at risk conditional on the scenario. For a severe collateral stress, the relevant comparison is often the conditional shortfall if the specified default occurs.

If a default probability $p$ is also supplied, a probability-weighted expected shortfall can be calculated in a simple two-state model. Suppose the shortfall is zero when default does not occur, and is $L$ when default occurs. Then expected shortfall is $pL$. This expected value is not the same as the collateral amount needed to eliminate the shortfall in the default scenario.

Worked Example: Conditional Versus Expected Shortfall

A hypothetical program has a gross unpaid obligation of $600,000. Conditional on reimbursement default, the scenario assumes that 50% of the obligation remains unpaid after other collections. Current security is $250,000. The probability assigned to this default scenario is 10%, and the shortfall is zero if default does not occur. Calculate both the conditional shortfall and the probability-weighted expected shortfall.

Conditional scenario need. The amount at risk if default occurs is $0.50(\$600{,}000)=\$300{,}000$.

Conditional shortfall. Subtract current security: $\max(0,\$300{,}000-\$250{,}000)=\$50{,}000$. Thus $50,000 is the amount by which security falls short in the stated default scenario.

Expected shortfall under the two-state assumption. The shortfall is $50,000 with probability 0.10 and zero otherwise, so $0.10(\$50{,}000)+0.90(\$0)=\$5{,}000$. Check: the expected amount is one tenth of the conditional shortfall. It does not make the $50,000 conditional gap disappear. The common mistake avoided is multiplying the obligation by the default probability and treating that expected value as the amount required to cover the default scenario.

A Disciplined Stress-Test Sequence

A clear sequence helps prevent inconsistent assumptions and double counting. State whether each stress applies to all claims or only to a defined segment, and keep the gross obligation separate from default and security calculations.

Set the valuation scope.
Identify the claims and reimbursement obligations included, the valuation date, and which amounts are already paid or collected.

Calculate stressed claim payments.
Apply the deductible to the severity assumptions, then use the specified stressed frequency. Keep reported claims fixed if the scenario says the stress affects only unreported claims.

Add reporting-delay obligations.
Estimate incurred but unreported claims from the stated occurrence and delay assumptions, and convert their count to reimbursement amounts.

Apply the default scenario.
Multiply the gross unpaid obligation by the fraction assumed unpaid conditional on default. Do not apply a default probability here unless the requested measure is probability-weighted.

Compare with current security.
Subtract security from the scenario need, flooring the shortfall at zero. State any excess separately if security is greater than the need.

The calculated result is only as meaningful as its assumptions. Frequency, severity, reporting delay, and default may not be independent in an actual portfolio. A scenario can deliberately combine adverse assumptions, but the analyst should explain that choice rather than imply that the combined case is an ordinary expected outcome. Also make clear whether $C$ is the security currently available for the obligations being tested; this tutorial treats it as fully available and makes no adjustment for instrument form, valuation haircuts, or draw restrictions.

Common Exam Traps

Applying a deductible incorrectly. For $X>d$, the payment per loss in this model is $d$, not $X$. Recalculate $Y^L$ at each severity before taking an average.
Mixing the claim stress with the default stress. First calculate the gross obligation $O$; then apply the conditional unpaid fraction $q$. Applying $q$ to some components but not others requires an explicit scenario reason.
Using a reporting delay with the wrong time window. A delay distribution and an occurrence period must use consistent time units. The formula $\lambda T/2$ is not universal; it depends on the uniform-delay and constant-rate assumptions stated above.
Treating expected shortfall as scenario protection. Multiplying a conditional gap by its probability gives an expected value under a two-state model, not the security needed to cover the full conditional gap.
Counting the same recovery twice. If an amount has already reduced gross unpaid obligations, do not also count it as a recovery when selecting $q$.

Key takeaway: Stress collateral adequacy in stages: estimate the gross unpaid obligation under explicit claim and reporting assumptions, identify the portion at risk in the defined default scenario, and compare that amount with current security. Report conditional shortfall separately from probability-weighted expected shortfall.

Check Your Understanding

For each problem, use the assumptions stated and report the requested amount in dollars.

A hypothetical deductible is $6,000. Severity is $1,000 with probability 0.3, $4,000 with probability 0.4, and $9,000 with probability 0.3. There are 90 expected claims. Find the expected total deductible payment. (Answer: $333,000)
A portfolio has 10 claims per month during the five months before valuation. Reporting delay is uniform from zero to five months. Frequency is stressed upward by 20%. If the average deductible payment is $4,000 per claim, estimate the stressed unreported-claim obligation. (Answer: $120,000)
Gross unpaid obligations are $460,000. The default scenario assumes 35% remains unpaid after other collections, and current security is $150,000. Find the conditional shortfall. (Answer: $11,000)
Conditional on default, $500,000 is unpaid and current security is $420,000. Default probability is 0.08, and shortfall is zero otherwise. Find the conditional shortfall and the probability-weighted expected shortfall. (Answer: $80,000 conditional shortfall; $6,400 expected shortfall)
Reported open obligations are $175,000. There are 24 baseline expected unreported claims; a frequency stress factor is 1.25. The stressed expected deductible payment is $3,200 per claim. The default scenario leaves 50% of total obligations unpaid, and current security is $140,000. Find the scenario shortfall or excess. (Answer: $0 shortfall; $4,500 excess)

What You'll Learn