Systematic and Systemic Stress Tests
Research priority: current stress test procedures consider only a handful of scenarios and do not quantify the plausibility of scenarios. They also mostly neglect second-round effects arising from the response of financial institutions to shocks. Current procedures can therefore lead to an illusion of safety, to hasty reactions to alarming stress test results, and to underestimation of risks.
In this project, we aim to develop second-generation stress tests that are both systemic and systematic. These systematically analyze a variety of sufficiently plausible scenarios to identify the worst case for the financial system. These stress test procedures also model the systemic risk that can arise from distress sales and networks of debt and ownership linkages among financial institutions when institutions are affected by the responses of other institutions to stress events. Based on results in various areas of the academic literature, we aim to develop a robust model using available data and implement a software prototype for systematic and systemic stress testing.
Content of the research project:
We propose to develop the core elements of a modern stress test procedure:
- The automated evaluation of a large number of scenarios reduces the risk of overlooking dangerous but plausible scenarios. All current stress test procedures use only a handful of scenarios. There are no procedures to control whether other, equally plausible scenarios might not be more dangerous.
- Quantifying the plausibility of scenarios prevents consideration of highly implausible scenarios. Because most stress test procedures currently used by government agencies avoid quantifying the plausibility of scenarios, they could trigger premature responses to alarming stress test results in implausible scenarios.
- The use of distribution scenarios enables consistent treatment of the banking and trading books.
- The modeling of distress sales allows realistic quantification of the consequences of the response of financial institutions to stress scenarios. The well-known fact that the most important mechanisms in crises propagate through distress sales is ignored in current stress test procedures.
- Modeling contagion effects in networks allows realistic quantification of the consequences of financial institution failures. Current stress test procedures ignore the contractual linkages between financial institutions that arise from interbank debt and property rights.
Applied research methods:
We briefly describe the methods used to develop the five core elements mentioned above:
- The automated evaluation of a large number of scenarios is a challenge for modeling, data provision, and data processing. We are trying to develop models that are simple enough to allow rapid, automated evaluation of scenarios. This requirement must be balanced with the requirements for accuracy and level of detail.
- The plausibility of point scenarios (realizations of the joint risk factor distribution) is quantified by their Mahalanobis distance, as first proposed by Studer and then developed by Breuer and Krenn. The plausibility of distribution scenarios is quantified by their relative entropy or other f-divergence, as proposed by Breuer and Csiszar.
- The determination of the worst case distribution scenarios is implemented as in Breuer and Csiszar. The determination of the worst case scenarios for the combination of banking and trading books still needs to be solved.
- Emergency sales, their price effects, and the consequences for financial institutions are modeled (Cont and Schaaning (2016)). This requires data on both the balance sheets of all major financial institutions in the system and the market depth of the major asset markets in which these financial institutions operate.
- Patch effects in networks are modeled as in Noe and Eisenberg and implemented (Elsinger and Summer).
|project name||Systematic and Systemic Stress Tests|
|Program||Jubilee Fund of the Oesterreichische Nationalbank|
|Project Index Number||17671|
|Project Duration||September 2018 - July 2021|
|Project Budget||141,000 EUR|
Dr. Martin Gächter
Financial Market Authority Liechtenstein
Dr Martin Summer
Mag Claus Puhr
Prof. Dr. Branko Urosevic
University of Belgrade
Dr Eric Schaanning
European Central Bank