Worst-Case Scenarios as a Stress Testing Tool for Risk Models

Abstract:     
We propose a new approach to stress testing risk models of financial institutions. In this approach, a scenario is fully defined by the frequency of scenario occurrence and lower bound of anticipated loss severity. All available scenarios are ordered by their frequency and severity to identify worst-case scenarios and only worst-case scenarios augment the loss distribution in a risk model. By doing that, we ensure that while the information in all scenarios is considered, only those that negatively affect the tail of the loss distribution are taken into account for the purpose of stress testing the risk model. The proposed approach has several advantages: (i) it has a built-in feature which ensures that a stressed risk model cannot produce a risk estimate that is lower than the one derived from the historical data based model; (ii) it does not require assumptions on scenario loss distributions, thereby simplifying the scenario generation process; and (iii) the approach can be applied to various types of risk, such as market or operational risk.

Number of Pages in PDF File: 23

working papers series