Stress Testing a Quant Portfolio

Stress testing reprices a book under scenarios the sample never produced, which is exactly what a historical covariance matrix cannot do.

Stress testing a quant portfolio means repricing the book under scenarios the historical sample used to estimate risk never actually produced, because a covariance matrix estimated from a calm market cannot tell a risk manager what happens once correlations move together in a broad selloff. Value at risk and expected shortfall figures, covered in risk management, VaR, CVaR and expected shortfall, are quantiles drawn from whatever distribution the historical or parametric model assumes, and neither approach is built to describe a combination of shocks the sample window did not contain.

The most direct stress test applies the actual price moves from a past crisis, the 2008 credit crisis, the March 2020 liquidity shock, or the 2022 rate shock, to the portfolio's current positions, regardless of when those positions were actually opened. This answers a concrete question: what would today's book have lost if a known historical episode happened again tomorrow, holding position sizes fixed at today's level. The value of the exercise depends on the book's current composition resembling, in risk character if not in exact names, the kind of exposure that suffered in the original episode, so a desk usually runs several historical scenarios rather than relying on just one.

A second approach shocks specific risk factors directly rather than replaying an entire historical period: rates up 100 basis points, implied volatility up 10 points, credit spreads wider by a set amount, applied through the factor loadings described in factor models and how desks use them to see how the book's P&L responds. This lets a risk manager test combinations that have not occurred historically but are plausible given current macro conditions, and it isolates the effect of a single factor rather than mixing several shocks together as a historical replay necessarily does. A desk running an options book typically extends this to include a shock to implied volatility skew specifically…

A risk team typically maintains a standing library of scenarios rather than designing a fresh set for every review, covering a mix of historical replays, hypothetical shocks calibrated to current volatility levels, and reverse stress cases specific to the book's largest exposures. The library is updated as market conditions change and as the book's composition shifts, since a scenario calibrated to a portfolio's exposures from a year earlier may no longer stress the positions that actually matter today.

Results are usually reported alongside the book's current VaR and expected shortfall figures rather than in isolation, giving the risk committee both a probabilistic view of everyday risk and a deterministic view of what a specific, named event would do to the book. The two figures answering different questions is the point: VaR describes a likely range of outcomes under current conditions, while a stress scenario describes a specific, plausible tail event that a quantile estimate alone would not necessarily flag as material.