Stress Testing

A stress test begins with identification of the risk factors that actually drive portfolio value: the level and shape of the yield curve, credit spreads by rating bucket, equity index levels, implied volatility, currency rates, commodity prices, funding costs and haircuts. Each factor is assigned a shock magnitude, and every position is then revalued under the shocked factor set rather than scaled by a single sensitivity measure. The difference between stressed value and current value is the stress loss, which can be expressed in currency terms, as a percentage of net asset value, or as an impact on a regulatory capital ratio.

The critical modelling choice is how factors move together. A univariate sensitivity test moves one factor and holds the rest fixed; a full stress scenario imposes an internally consistent joint move, because in crises spreads widen while equities fall, correlations converge toward one and liquidity evaporates simultaneously. Full revaluation matters most for options and structured credit, where convexity means a linear approximation badly understates loss at large shock sizes.

Historical stress tests replay an actual episode, applying the observed factor moves of the 1987 crash, the 1998 sovereign and hedge fund crisis, 2008 or the March 2020 liquidity squeeze to today's holdings. Their advantage is internal consistency, since the co-movements are real rather than assumed; their weakness is that the next crisis will not be a copy of the last, and the current portfolio may have no meaningful exposure to the factors that dominated the old event.

Hypothetical stress tests specify a forward-looking narrative, such as a sovereign default, a disorderly rate spike or a geopolitical shock, and translate it into factor moves. Reverse stress testing inverts the exercise: rather than asking what a given shock costs, it starts from an unacceptable outcome, for example breaching a covenant, exhausting capital or failing a margin call, and works backwards to identify which combinations of factor moves would produce it. Reverse tests are particularly effective at surfacing concentrations and hidden dependencies that a standard scenario library never touches.

Stress testing is embedded in supervision. Banks project revenues, provisions and risk-weighted assets over a multi-quarter horizon under supervisory scenarios, and the resulting minimum capital ratio constrains dividends and buybacks. Insurers run analogous exercises for underwriting and market shocks under their own solvency regimes. Central counterparties stress default fund adequacy against the simultaneous failure of their largest clearing members, and money market and open-end funds stress redemption pressure against the saleability of their assets.

On the buy side, stress results feed position limits, collateral and margin planning, and liquidity buffers. A fund with leveraged or derivative exposure cares less about the mark-to-market loss itself than about the cash the loss consumes: a stressed variation margin call arriving while credit lines tighten is what converts a paper loss into forced liquidation. Stress testing is therefore as much a funding and liquidity discipline as a valuation exercise, which is precisely why it complements rather than duplicates statistical loss measures.

Stress losses carry no probability attached. That is the source of both their usefulness and their misuse: a stress number cannot be aggregated across scenarios, compared like-for-like with a distributional risk measure, or read as an expected loss. The correct interpretation is conditional, namely the loss the portfolio would sustain if this specific configuration of shocks occurred, and the decision it supports is whether the institution could survive that outcome and remain solvent, funded and compliant.

Candidates should be able to distinguish sensitivity analysis, scenario analysis and stress testing, explain why full revaluation is needed for non-linear instruments, identify the assumptions most likely to fail under stress, and recognise reverse stress testing as a distinct methodology. The recurring analytical errors are assuming correlations hold at their historical averages, treating liquidation values as observable mid-market prices, and ignoring second-round effects such as counterparty failure, rating downgrades and margin spirals that amplify the initial shock.