Define severity through stated factor changes, starting conditions, joint relationships, and time paths. Evaluate portfolio losses separately from those inputs. Use a documented ladder of conditional shocks and sensitivity checks, without attaching probabilities or regulatory pass thresholds that the exercise has not established.
Separate input severity from portfolio damage
Input severity describes the disturbance: how far equities fall, how much yields change, or how long financing remains unavailable. Portfolio damage describes the consequences of that disturbance for the actual holdings. An enormous move in a factor with little exposure can produce a small loss, while a moderate move in a concentrated or leveraged exposure can dominate the entire result.
Keep these concepts in different report columns. Otherwise, the scenario that happens to produce the largest loss can become labeled the most severe by definition, concealing gaps elsewhere. A portfolio with little rate exposure does not make a large rate shock economically mild. It simply responds differently from a portfolio whose valuation depends strongly on the same part of the curve.
State the starting point and measurement convention
The same endpoint can imply different shocks from different starting values. A spread ending at 300 basis points represents a 200-basis-point widening from 100, but only a 50-basis-point widening from 250. Relative changes answer another question entirely. Record starting level, stressed level, absolute change, and relative change where useful, without treating these alternative descriptions as interchangeable severity measures.
Historical percentiles can provide context only when the underlying sample, horizon, and transformation are stated. A daily equity return percentile cannot be compared directly with a quarterly change in unemployment. The Federal Reserve's 2025 scenario publication illustrates explicit paths and starting conditions; its published assumptions are exercise-specific. They are not universal severity thresholds for investment portfolios, nor are they probability estimates.
Distinguish a sudden shock from persistent pressure
An instantaneous price change is suitable for examining immediate mark-to-market sensitivity. A prolonged scenario may also alter coupons received, refinancing terms, option expiries, and liquidity demands. Two paths with the same final price can create different interim losses and cash requirements. Specify whether severity refers to the endpoint, the worst point along the path, or the cumulative burden over the horizon.
For a static linear exposure, scaling shocks can provide a convenient first comparison. That convenience weakens when contractual features activate or positions evolve. An option can cross a strike, a callable bond can change effective duration, and a collateral threshold can trigger a payment. Recompute those mechanisms at each severity level rather than multiplying a small-shock loss by an arbitrary factor.
Build a transparent hypothetical ladder
Consider a synthetic USD 1,000,000 portfolio with USD 600,000 in equities, USD 300,000 in bonds, and USD 100,000 in cash. Give the bonds modified duration of four and ignore convexity, credit changes, income, and currency effects. Level 1 assumes equities fall 10% and bond yields rise 50 basis points. The duration approximation gives a 2% bond loss and a USD 66,000 portfolio loss.
Level 2 uses an 18% equity decline and a 100-basis-point yield rise, producing USD 108,000 plus USD 12,000 of losses, or 12% of the portfolio. Level 3 uses 28% and 150 basis points, producing USD 168,000 plus USD 18,000, or 18.6%. These are deliberately constructed sensitivity levels, not historical observations, calibrated probabilities, or evidence that duration remains accurate at the largest shock.
The table separates assumed market moves from calculated portfolio effects. Cash is fixed at par solely for this instantaneous illustration. Real cash arrangements can carry counterparty, currency, or access risk, and a longer horizon introduces interest and payment flows. Those dimensions would require additional assumptions rather than being captured by changing the severity label attached to this simple calculation.
| Synthetic level | Equity return | Yield rise | Approximate bond return | Portfolio loss USD |
|---|---|---|---|---|
| 1 | -10% | 50 bp | -2% | 66,000 |
| 2 | -18% | 100 bp | -4% | 120,000 |
| 3 | -28% | 150 bp | -6% | 186,000 |
Challenge the joint scenario, not just each extreme
Combining every factor's worst observed move can create a scenario with no coherent timing or economic mechanism. It may still serve as a deliberately harsh sensitivity envelope, but should be labeled accordingly. A joint scenario needs a reason why rates, spreads, currencies, and equities move together. Internal consistency matters even when the purpose is to explore unusually adverse conditions.
The Basel Committee's 2018 principles discuss severe yet plausible scenarios and material risks. For portfolio education, plausibility is an argument to document, not a certification stamp. Describe the originating disturbance, transmission channels, and conditions that could weaken the proposed combination. Keep a more extreme boundary case available when useful, but do not disguise it as an estimated forecast or observed episode.
Keep sensitivity ranges separate from likelihood
A range of losses across severity levels shows how the model responds to different assumptions. It is not automatically a confidence band. A confidence band needs a statistical model and an interpretation of repeated sampling or predictive uncertainty. A scenario ladder built by judgment has neither merely because it contains several rows and one row has been named central.
Do not average scenario losses unless the weighting scheme has an explicit analytical purpose. Equal weights are not neutral probabilities when scenarios were selected to challenge vulnerabilities. Report individual results, their principal drivers, and any changes in position rankings. A threshold crossing deserves attention, but its meaning depends on who set the threshold and whether it concerns value, liquidity, or another constraint.
Document calibration and revisit it deliberately
Retain the rationale for every shock size, including historical reference windows, model estimates, or clearly labeled judgment. Review whether changed portfolio exposures make the scenario library incomplete. Avoid quietly weakening a shock after an uncomfortable result or strengthening it until a predetermined conclusion appears. Versioned assumptions allow later readers to distinguish a genuine portfolio change from a revised test.
End with the conditional response curve and its limits. Explain where approximation error could be material, which exposures have incomplete coverage, and whether losses accelerate under harsher inputs. That makes severity useful for comparison and model challenge. It does not convert a stress test into an investment recommendation, a quantified event probability, or an assurance that losses cannot exceed the final level.
Review checkpoints
- Shock magnitude and portfolio loss measure different things.
- Severity includes starting conditions, duration, and joint relationships.
- A sensitivity ladder is not a probability distribution or confidence interval.
Source notes
Continue the analysis
Historical Versus Hypothetical Stress Scenarios
Choose between historical factor replays, constructed scenarios, and transparent hybrids without confusing them with portfolio backtests.
Reverse Stress Testing a Portfolio
Start with a defined loss or liquidity threshold and investigate the combinations of shocks that could reach it.
Designing Coherent Portfolio Stress Scenarios
Connect an economic narrative to consistent factor shocks, horizons, and valuation channels without counting the same effect twice.
Where duration and convexity stop explaining the loss
Audit rate-shock units, duration conventions, curve shape, embedded options, and the gap between approximations and bond revaluation.
First edition, 16 October 2026. Hypothetical values are teaching inputs, not observed portfolio or provider results.