Scenario analysis & stress testing

Definition

Scenario analysis revalues the portfolio under specified adverse states — price shocks, volatility spikes, correlation convergence, liquidity gaps — and reports the P&L each scenario produces. Where statistical risk measures summarize the past’s distribution, scenarios ask a different question: “what does THIS book lose if THAT happens?” It is the platform’s required risk lens for undefined-risk structures and the source of the R figure their sizing uses.

How it works / structure

• Scenario grid (engine-executable): underlying moves (e.g. ±5/10/20%), volatility shifts (IV ±25/50%, crushes and spikes), combined states (down-and-vol-up — the equity stress pairing), time steps (today vs next expiry), and rate moves where relevant; options books revalue with full repricing, not delta approximation — gamma and vega are why (port-portfolio-greeks, greek-gamma, greek-vega).
• Historical replays: named episodes (1987-scale day, 2008 autumn, 2020 March, 2018 vol event) applied to today’s book — anchors the grid in realized physics.
• Correlation stress: pairwise correlations pushed toward crisis levels (risk-correlation-exposure) — the scenario layer where “hedged” books confess.
• Outputs: worst-scenario loss vs the drawdown budget (risk-max-drawdown-budget), the R input for undefined-risk sizing (risk-fixed-fractional), and the specific scenario that dominates — WHICH state hurts is as informative as how much.

When it applies

Mandatory for undefined-risk options structures (naked shorts, ratios, jade lizard put sides — their “R” is a scenario number); books with optionality where linear measures mislead; pre-event checks (what does the book lose if the event gaps this?); and portfolio ratification (the budget is tested against the grid, not against history’s average).

Risk profile & failure modes

• Imagination bounds: scenarios cover what was specified; the realized crisis is routinely a combination not on the grid — grids constrain, they do not enumerate reality (the regulatory literature’s own stated limit).
• Severity anchoring: grids calibrated to recent memory understate; the pre-2008 grids famously lacked 2008.
• False precision: scenario P&L inherits model assumptions (repricing models, liquidity assumptions — fills at scenario prices may not exist, ms-liquidity).
• Checklist decay: scenarios run but not acted on are ritual; the platform wires worst-scenario loss into sizing so the output binds.

Evidence & limits

The methodology canon is regulatory: the Federal Reserve’s DFAST scenario design and the Basel Committee’s stress-testing principles document both the practice and its known limits (scenario coverage, model risk). For trading books the technique is standard practice; its value is structural (bounding known unknowns), and its blind spot — the unspecified scenario — is permanent and acknowledged.

Falsifiable-thesis examples

Illustrations only, not signals:

• “This book’s worst grid-scenario loss (down 10% + IV +50%) stays within 60% of the drawdown budget this quarter” — falsified by the weekly grid runs.
• “In the next realized 5%+ index down-day, the book’s actual loss will not exceed 1.5× the corresponding grid scenario’s estimate” — falsified by the realized comparison.

Cross-references

• Inputs it stresses: port-portfolio-greeks, greek-gamma, greek-vega, risk-correlation-exposure
• Outputs it feeds: risk-max-drawdown-budget, risk-fixed-fractional (undefined-risk R)
• Structures that require it: strategy-strangle (short), strategy-ratio-spread, strategy-jade-lizard
• Liquidity honesty: ms-liquidity, regime-liquidity