Core Path · L6.2.11

Scenario generation and model-based stress tests

14 minScenario generationVAR(1)Conditional scenariosReverse stress testMahalanobis distance

By the end you will be able to

  1. 1Build a real-world scenario generator (a VAR(1) on curve components and FX with resampled shocks), check its stability and compare its 12-month distribution with history.
  2. 2Turn a single shock into a full scenario with the conditional normal and t distributions, and read the loss distribution around it.
  3. 3Find the most likely move that causes a given loss by minimising the Mahalanobis distance, linear and under full revaluation.
  4. 4Judge a reverse stress scenario's plausibility honestly: in 25 dimensions, along the loss direction, and against real days.

Why it matters

What is the most likely way for the sleeve to lose £5m in a day? Search all 25 factors for the smallest move that does it, and the answer is plain: US Treasury yields up about 23–29bp across the curve, euro yields up about 18bp, a Mahalanobis distance of 6.02. Along the direction that hurts the book that is a six-sigma day. Yet 74 of the last 500 real days were farther from average than 6.02. Both statements are true. This lesson is about building scenarios like this one, and about saying honestly how plausible they are.