A BAYESIAN APPROACH TO STRESS TESTING AND SCENARIO ANALYSIS

Riccardo Rebonato

I present a new approach to stress testing that combines the elicitation of subjective (marginal or conditional) probabilities of events with the specification of a simple causal structure among them. By so doing, stress events are placed in an approximate but coherent probabilistic framework. The approach only requires the risk manager to provide ‘simple’ and cognitively resonant input probabilities. The techniques of linear programming and Bayesian nets then ensure the consistency of the subjective inputs and facilitate the derivation of the desired joint probabilities.