Systematic Stress Testing

We started developing a framework which allows to perform financial stress tests in a systematic manner. Currently, such stress test evaluate a few (maybe isolated) expert scenarios. However, given that there exist various risk parameters, like macroeconomic indizes and changes in ratings, the question is whether these few stress test provide the whole picture. Performing stress tests in a systematic manner, means evaluation of many, thousands or millions, scenarios and looking at the distribution of the results to find extreme but plausible stress scenarios. This requires a) a model generating scenarios, b) defining performance criteria, c) a method to determine plausibility, and d) optimization methods for finding the worst case. The framework will use the DEF to perform the evaluation of the single scenarios.

In the current version we use a normal distribution for the scenario generating model and availavle time series to estimate the parameters of the normal distribution. Moreover, a single criteria is used to indicate the performance for a single scenario. The plausibility and the search for the worst case have still to be developed, but the ideas are outlined in A systematic approach to multi-period stress testing of portfolio credit risk and Systematic stress tests with entropic plausibility constraints. Other open questions are concerned with other scenario generating models and applying multi-objective approaches if more than one criteria is considered.