Correlation Stress in a Historical VAR Setting

This paper describes a reasonably straightforward methodology for assessing the impact of different correlation stress assumptions in a historical VaR setting. This is accomplished by first introducing continuous empirical distribution functions for the marginal risk factor distributions, and using this to define standardized normal drivers for each. Then assuming a normal copula structure, a mixing of these standardized drivers is introduced in order to arrive at the target correlation structure. Finally the mixed drivers are re-converted into values for the risk factors using the inverse marginal distribution functions, from which a stressed correlation VaR calculation can be performed