Systematic scenario selection: stress testing and the nature of uncertainty

TLDR

The paper proposes a method for selecting multidimensional shock scenarios in financial stress testing and suggests extensions for non‑monotonic loss functions and univariate shocks. The method samples points of arbitrary severity from a plausible joint probability distribution and performs a grid search of sparse, well‑distributed scenarios, positioning itself between traditional and reverse stress testing. The approach reduces blind spots in stress testing and the authors provide tested, commented Matlab source code.

Abstract

We present a technique for selecting multidimensional shock scenarios for use in financial stress testing. The methodology systematically enforces internal consistency among the shock dimensions by sampling points of arbitrary severity from a plausible joint probability distribution. The approach involves a grid search of sparse, well distributed, stress-test scenarios, which we regard as a middle ground between traditional stress testing and reverse stress testing. Choosing scenarios in this way reduces the danger of 'blind spots' in stress testing. We suggest extensions to address the issues of non-monotonic loss functions and univariate shocks. We provide tested and commented source code in Matlab®.