Dealing with Dependence in Risk Simulations

Abstract

Difficulties arise in choosing a suitable sampling scheme for risk simulations when variables are not judged to be independent of each other. This paper shows that, by considering the unconditional distributions of the variables as transformations of the normal distribution, a model can be constructed to overcome these difficulties. This model takes account of the extent of the dependence between the variables while at the same time keeping to a minimum the total number of individual probability assessments which must be made by management.

The model is illustrated with an example and in addition the way in which a similar model can be developed to describe the probabilistic growth of a variable in a risk simulation is discussed.