Market Risk Stress Testing Beyond VaR

By Jonas Osman Abdelghafour.

Value-at-Risk (VaR) is a summary of the loss distribution under normal market conditions. It is useful for limits, capital, and daily risk reporting, but by construction it does not describe the tail beyond its confidence level and it does not accommodate structural breaks. Stress testing exists precisely to cover what VaR cannot see. The 2019 revisions to the Basel market risk framework (FRTB) and the ECB's guide to internal models both reinforce this: expected shortfall and stressed calibrations improve the tail measurement, but scenario-based stress testing remains a first-order control.

What stress testing adds

A market-risk stress framework typically layers three elements. Sensitivity analysis isolates the impact of a single risk factor move (parallel curve shifts, equity index moves, credit spread widening). Historical scenarios replay actual episodes (October 1987, September 2008, March 2020, September 2022). Hypothetical scenarios construct forward-looking narratives that combine factor moves the historical record has not yet produced but that plausible transmission mechanisms could produce.

The value is in the third layer. Historical replays are useful but by definition backward-looking. Well-constructed hypothetical scenarios reflect current portfolio exposures, current market microstructure, and current tail risks.

Scenario construction discipline

A defensible scenario carries a narrative, calibrated factor shocks, an assumed correlation structure, and a stated horizon. Correlation assumptions typically break down under stress — historical correlations tighten and diversification benefits shrink — and this should be reflected explicitly rather than left implicit in a covariance matrix.

Factor shocks should be justified against a specific reference point: a historical episode of comparable severity, a regulatory reference scenario (EBA / ECB stress tests, NGFS climate scenarios), or a portfolio-specific tail analytic such as extreme value theory applied to the underlying factor. Calibrating to "a 3-sigma move" without saying which distribution and which sample is not a justification.

Reverse stress testing

Reverse stress tests invert the ordinary exercise. They start from a defined failure outcome — for a trading book, that might be a loss that breaches a capital or liquidity threshold — and identify the combinations of market moves and behavioural responses that would cause it. The exercise typically surfaces vulnerabilities the forward stress set misses: correlated risks across desks, hidden basis exposures, and behavioural assumptions (such as client hedging demand or funding availability) that would break under stress.

The output is not a probability estimate. It is a targeted list of vulnerabilities and mitigation actions, prioritised by plausibility.

Governance and use

Stress testing outputs should be tied to specific decisions: limits, hedging actions, capital planning, and — under ORSA or ICAAP — to strategic choices. A stress test that produces a large number and no action is a governance failure. Escalation triggers should be defined: for example, a scenario loss exceeding a stated fraction of Tier 1 capital triggers a management response and a board notification.

For insurers, the market-risk stress set feeds into the broader ORSA scenario design; for banks, into ICAAP and recovery planning. Consistency between the market-risk scenarios used for capital and those used for treasury/liquidity is important — inconsistent scenario families produce confused management responses.

Common failure modes

Three failure patterns recur. Scenario libraries that grow indefinitely without periodic pruning become unmanageable and are then treated as a run-and-file exercise. Correlation assumptions frozen at pre-crisis levels understate concentration risk under stress. Reverse stress tests written as narrative essays with no linkage to actual portfolio exposures produce no actionable output.

Limitations

Stress testing depends on scenario imagination. It cannot capture events the designers cannot conceive of, and it can create a false sense of security if the scenario set is treated as exhaustive. Complementing structured scenarios with periodic ad-hoc thematic stresses — geopolitical fragmentation, sovereign redenomination, sudden dollar liquidity withdrawal — reduces but does not eliminate this risk. See sovereign risk and geopolitical fragmentation for one such theme.

Conclusion

VaR and expected shortfall answer questions about normal-market risk. Stress testing — sensitivity, historical, hypothetical, and reverse — answers questions about the events that most threaten the business. Both are needed, and only the second is a genuine test of resilience. Related notes: ORSA scenario design and board use, liquidity stress testing under LCR and NSFR.

Written by Jonas Osman Abdelghafour, actuary and financial risk manager. Background and contact details are on the about page.