Reverse Sensitivity Analysis for Risk Modelling

We consider the problem where a modeller conducts sensitivity analysis of model consisting random input factors, corresponding output interest, and baseline probability measure. The seeks to understand how (the distribution factors as well output) changes under stress on output’s distribution. Specifically, for variable, we derive unique stressed that is closest in Wasserstein distance satisfies stress. further model, including inputs, which can be calculated numerically efficient way from set Monte Carlo samples implemented R package SWIM CRAN. proposed reverse framework model-free allows stresses such (a) mean variance, (b) any distortion risk measure Value-at-Risk Expected-Shortfall, (c) expected utility type constraints, thus making suitable models.