Why Stress Severity Needs Calibration
A stress scenario is useful only if its severity has a defensible basis. A scenario that is too mild may fail to reveal a material vulnerability; one built by combining every worst-case assumption may be so extreme that it does not represent a plausible adverse condition. Calibration is the disciplined choice of scenario magnitudes, supported by evidence and judgment, for a defined risk, horizon, and exposure.
The earlier material on selecting stress tests and constructing adverse solvency scenarios establishes how to identify a risk and describe a scenario. Here the focus is narrower: how severe should the scenario be, and what makes that severity plausible? Evidence may include the insurer’s own experience, external events or studies, expert judgment, and forward-looking changes in the environment or the insurer’s portfolio. No single source automatically determines the answer.
“Plausible” does not mean “most likely,” and it does not require assigning a precise probability. Historical occurrence is informative, but a scenario can be plausible even if the exact combination has not occurred before. Conversely, a number taken from a historical event is not automatically appropriate: exposure, prices, coverage, business mix, and conditions may have changed. Calibration therefore asks both what evidence supports the magnitude and how relevant that evidence is to the current insurer.
A Practical Evidence Framework
Start by fixing the measurement basis. Specify the assessment horizon, the variable being stressed, the units, and the exposure to which the stress applies. For example, “a 15% increase in annual claims” is incomplete unless it is clear which claims, relative to which baseline, over what period, and whether the figure is gross or net of reinsurance.
Then assess the evidence sources. Internal experience can be highly relevant to the insurer’s portfolio, but a short or changing history may be unrepresentative. External evidence can extend the record, but differences in geography, business mix, policy terms, and vulnerability need consideration. Expert judgment helps interpret gaps and emerging conditions, but its rationale and range should be recorded rather than presented as an observed fact. Forward-looking considerations include changes in exposure, concentration, technology, legal conditions, and the broader risk environment.
| Evidence source | Useful contribution | Calibration question |
|---|---|---|
| Internal historical experience | Shows outcomes under the insurer’s own portfolio and practices | Have exposure, coverage, prices, or operations changed? |
| External evidence | Broadens the set of observed events and relevant benchmarks | How comparable are the event and the insurer’s current exposure? |
| Expert judgment | Interprets sparse data, emerging risks, and unobserved conditions | What rationale, assumptions, and uncertainty support the judgment? |
| Forward-looking considerations | Reflects conditions that historical data may not capture | What has changed, and how does that change affect severity? |
A useful calibration record identifies the selected severity, the evidence considered, adjustments made, material limitations, and why the chosen case is plausible. It can also record a lower or higher sensitivity and an extreme boundary case, provided these are not confused with the selected plausible stress. This makes the judgment reviewable without implying a precision that the evidence does not support.
Scaling Historical Experience
When past losses are informative, a common first step is to restate them to a current exposure and price level. For a simple proportional model, assume the historical loss is directly proportional to exposure and to the relevant claim-cost index. Assume also that coverage, loss definition, and the relationship between exposure and loss severity remain unchanged. Under those assumptions, a historical loss \(L_h\) can be scaled as follows:
Here \(E_h\) and \(E_c\) are historical and current exposures, and \(I_h\) and \(I_c\) are the corresponding price or claim-cost indices. The exposure ratio adjusts the amount of risk represented; the index ratio adjusts the cost level. This is a scaling approximation, not a general loss model.
Specify the event, coverage basis, currency, time period, and whether the loss is gross or net of reinsurance.
Check whether historical exposures, terms, and operating conditions are sufficiently comparable to the current situation.
Show each exposure or price-level adjustment separately, and explain any additional judgmental adjustment.
Decide whether the result supports a plausible severe stress, a sensitivity, or an extreme boundary case, and document why.
Worked Example: Restating a Historical Property Loss
A hypothetical insurer recorded a covered property loss of $80 million in a past wind event. At that time, the relevant insured exposure was $4.0 billion and the claim-cost index was 100. Current comparable exposure is $5.0 billion and the index is 110. Assume losses scale proportionally with exposure and the index; assume the coverage definition and portfolio mix are unchanged. The loss measure is gross of reinsurance. Estimate the restated event loss and assess its use as a stress anchor.
Method. Apply the exposure ratio and claim-cost index ratio separately:
The restated loss is:
As a check, the exposure adjustment alone raises the loss from $80 million to $100 million; the 10% cost-level adjustment then raises it to $110 million. The restated amount is a defensible historical anchor under the stated assumptions, but it is not automatically the selected stress. The insurer must still assess whether the event remains relevant and whether current conditions justify a different severity. The common mistake is to add 25% and 10% to get a 35% increase: the correct multiplicative adjustment is \(1.25\times1.10=1.375\), or a 37.5% total increase.
Using External Evidence and Judgment
External evidence is most useful when the comparison is explicit. A benchmark should not be copied without asking whether the insurer has more or less exposure, different vulnerability, or a different cost basis. Where an adjustment can be quantified and justified, show it. Where it cannot be quantified reliably, record the qualitative difference and explain how it affects the choice.
Expert judgment should complement evidence rather than disguise its absence. A clear judgment record states who supplied the input, what assumptions were considered, the range of views where relevant, and why the final selection was chosen. It is often more informative to show a supported range than to present a single number as if it were known precisely.
Worked Example: Adjusting an External Benchmark
A hypothetical external study reports a $160 million loss for an event affecting $8.0 billion of comparable exposure. The benchmark is at a claim-cost index of 100; the insurer’s current index is 105 and its current comparable exposure is $6.0 billion. An engineering review supports a vulnerability factor of 1.15 for the insurer relative to the benchmark. Assume the loss scales proportionally with exposure and cost index, and that the vulnerability factor applies multiplicatively. The review gives a reasonable factor range of 1.00 to 1.20.
Method. Adjust the benchmark for scale, price level, and relative vulnerability:
First apply the exposure factor: \(\$160\text{ million}\times0.75=\$120\text{ million}\). Then apply the cost factor: \(\$120\text{ million}\times1.05=\$126\text{ million}\). Finally apply the central vulnerability factor:
The same calculation at the bottom and top of the stated vulnerability range gives \(\$126\text{ million}\times1.00=\$126\text{ million}\) and \(\$126\text{ million}\times1.20=\$151.2\text{ million}\). A selection near $145 million is therefore traceable to the external benchmark and the stated engineering judgment. It is not an objectively exact estimate: the range shows the effect of uncertainty in the judgmental factor. The common mistake is to apply the vulnerability factor to the original $160 million and overlook both the current exposure scale and the cost-level change.
Plausible Severe Stress Versus Extreme Boundary Case
A plausible severe scenario should be challenging but coherent with the evidence and assessment horizon. An extreme boundary case asks a different question: what would happen under a deliberately adverse limit? For example, a plausible scenario may use a severe market movement supported by historical and external evidence, while a boundary case may test a substantially larger movement selected to explore a limit. The boundary case should not be described as equally likely, or necessarily plausible on the same basis.
A frequent calibration error is to combine the worst observed value for every risk, even when those observations came from different periods and conditions. The result may be a useful boundary case, but it is not automatically a coherent plausible scenario. For a plausible combined scenario, consider whether the chosen conditions can occur over the same horizon and whether there is a defensible narrative linking them. Do not infer that simultaneous marginal extremes are supported merely because each occurred separately.
Worked Example: Choosing a Market and Claims Stress
A hypothetical insurer is setting a one-year stress. Its evidence review identifies the following severe-but-supported ranges: equity-market declines of 18% to 25%, spread widening of 140 to 200 basis points, and claims-cost increases of 7% to 12%. Internal experience and external market evidence support those ranges; experts judge that a common adverse economic environment could affect all three risks during the year. For a separate boundary case, the committee is considering equity declines of 40%, spread widening of 350 basis points, and claims-cost increases of 25%. Select and translate a plausible severe stress using equity decline of 22%, spread widening of 180 basis points, and claims-cost increase of 10%. Current exposures are $400 million in equities, $300 million in spread-sensitive assets with modified duration 4, and $100 million of annual claims.
Method. First check that each selected stress lies within its evidence-supported range: 22% lies between 18% and 25%; 180 basis points lies between 140 and 200; and 10% lies between 7% and 12%. The stated common adverse environment supplies a coherent scenario rationale, rather than treating three unrelated maxima as one event.
The equity loss under the selected stress is \(0.22\times\$400\text{ million}=\$88\text{ million}\). For spread-sensitive assets, use the first-order duration approximation \(\text{percentage price change}\approx-\text{duration}\times\text{yield change}\). A widening of 180 basis points is \(0.018\), so the estimated price decrease is \(4\times0.018=0.072\), or 7.2%. The corresponding first-order value decrease is \(0.072\times\$300\text{ million}=\$21.6\text{ million}\). This approximation assumes a small yield change and ignores convexity and other valuation effects; it is not an exact repricing. The claims-cost increase is \(0.10\times\$100\text{ million}=\$10\text{ million}\).
These are separate translations of the selected stress magnitudes, not a complete solvency-impact calculation. The proposed boundary values—40%, 350 basis points, and 25%—are well outside the stated plausible ranges, so they should be labelled and justified as a separate extreme boundary case, not silently substituted for the plausible stress. The common mistake is to select 25%, 200 basis points, and 12% simply because each is the upper endpoint. Upper endpoints can be useful for a severe scenario, but their joint use still needs a coherence rationale; it is not established by the marginal ranges alone.
Calibration Checks and Common Traps
Before finalizing a severity, check that the scenario is measurable, internally consistent, and relevant to the insurer’s current position. The following questions help expose weak calibration:
- Is the basis clear? Record the horizon, baseline, units, coverage, and gross or net treatment.
- Are adjustments justified? Avoid scaling historical experience for exposure while ignoring material changes in terms, mix, or vulnerability.
- Is the evidence comparable? Explain material differences between an external benchmark and the insurer’s own risk.
- Is judgment visible? Distinguish observed data from selected assumptions and describe the rationale for judgmental adjustments.
- Are scenario labels accurate? Keep plausible severe stresses, sensitivities, and boundary cases distinct.
A stress calibration does not become reliable merely because it has many decimal places. Precision should reflect the quality of evidence. A rounded selection supported by a transparent range and rationale is often more defensible than an apparently exact value built on weak comparability. Clear calibration supports an ORSA assessment aligned with the CAS Exam 9 (Risk Management for Actuaries) syllabus.
Check Your Understanding
For each problem, state the calculation or reasoning that supports your answer; distinguish a supported stress from a boundary case where relevant.
- A hypothetical insurer had a covered loss of $48 million at exposure of $3.0 billion and index 100. Current exposure is $3.6 billion and the index is 110. Assuming proportional scaling and unchanged coverage, what is the restated loss? (Answer: $63.36 million)
- An external benchmark is $90 million at exposure $5.0 billion and index 100. Current exposure is $4.0 billion and index 108. Assume proportional scaling and no additional vulnerability adjustment. What is the adjusted benchmark? (Answer: $77.76 million)
- A severe-but-supported claims-cost range is 6% to 11%, while an expert proposes 14% based on a documented emerging condition. What should the calibration record do before treating 14% as a plausible stress? (Answer: Explain the emerging-condition evidence, the reason it extends beyond the observed range, key assumptions and uncertainty, and whether 14% is being selected as a plausible stress or a separate boundary case; the range alone does not validate it.)
- For a one-year scenario, internal and external evidence support equity declines of 15% to 22% and claims increases of 5% to 9%. The proposed scenario uses 20% and 8%, with a documented common economic narrative. Is it within the stated evidence ranges, and what additional caution applies? (Answer: Yes, both magnitudes are within range; the joint scenario still needs a coherent common-horizon rationale, which the narrative should support.)
- A portfolio has $250 million of spread-sensitive assets with modified duration 3. A stress widens spreads by 120 basis points. Using a first-order duration approximation, estimate the value decrease and state one limitation. (Answer: \(3\times0.012=3.6\%\), so the estimated decrease is $9.00 million; the approximation ignores convexity and is not exact repricing.)