It has been stated that the best way to create the reverse stress test for certain types of risks is as a part of the whole stress testing framework. As such, it is implied that the stress testing framework is a singular entity and should be treated accordingly. However, more often than not, the clients approach us with demands to improve their current stress test framework via adding/modifying reverse stress tests for one or more specific types of risk. The approach Finalyse uses to do this is fairly general, so that other companies may find a general insight in our approach too.
This step obviously entails familiarising ourselves with what exactly it is that we are being asked to do. Being crystal clear about this is very important – even companies working on their own stress tests internally should not underestimate this step. It also, and perhaps most importantly, consists of the analysis of available input: stress testing framework documentation, stress testing process, market data (historical series), AS-IS calculations and results, related IT infrastructure as well as recommendations from the auditors. As these will be the key components, it is very important that they are stated explicitly.
Enhancing Current risk framework
The second step is to assess, verify and to the extent possible, improve the existing stress test framework, which can encompass the following:
- Checking the relevance of the scenarios currently in use
- Time horizon of the stress
- Checking for further evidence explaining the validity of stress scenarios
- Analysis of the relevance of the stress tests in light of the structure and composition of the portfolio
- Review of the interactions between the different risk factors
- Review of the stress testing policy
For this end, Finalyse has designed and built a pragmatic tool using a historical time series of market data to compute statistical KPIs that are relevant for the stress testing purposes, and to evaluate the plausibility of the scenarios currently in use.
Develop Reverse Stress Test Framework
The third step is to develop a reverse stress testing framework for the scope of the particular risk factor in question.
The building blocks of our reverse stress testing framework would be the following ones:
Bank failure target
In the case of reverse stress testing, we start with the definition of the level of bank failure needed to assess the stress test scenarios. This does not specifically mean insolvency, but would rather relate to a level of P/L loss or capital loss that cannot be easily corrected by prospective management actions. In practice, we will look at the same key output metrics used in the usual stress testing framework and define the appropriate level of magnitude in joint collaboration with Risk and Business teams.
Tail Risk Analysis
A pragmatic way to perform this analysis is to look at the loss distribution simulated for VaR and to isolate the main risk factors responsible for losses in the neighbourhood of the level previously identified. In a historical VaR approach, the space of risk factor realisation may be too narrow because only risk factors combinations effectively observed in the past will show up. As a complementary approach, P/L loss could be simulated in a sensitivity-based Monte-Carlo setting, assuming a relatively simple fat-tail multivariate distribution that replicates the volatility and correlation structure displayed by historical series of risk factor returns. Once most influential risk factors are identified in the tail region under focus, the dimension of the problem can be further narrowed by grouping similar simulations using clustering methods, each cluster leading to the definition of a possible scenario.
The technological choices for developing the tool supporting these calculations depend on the outcome of the requirements analysis phase.
Identifying and running the unviable scenarios
Scenarios derived from the tail risk analysis should then be calibrated more precisely – using a sensitivity approach - to produce a sufficient level of loss. We would typically analyse the relevance of those scenarios through discussions with risk or business experts, and compare them with historical data and similar shocks used in other stress testing exercises (e.g. EBA stress testing). The likelihood of the scenarios would also be assessed.
Full revaluation and quantification of impacts
The following step would be to add these scenarios in the stress test framework and validate the losses calculated under full-revaluation.
The last but not necessarily least step is to provide a consistent documentation of the stress testing framework, in line with market best practices. For many of our clients, the existing documentation of previous models tends to be fragmented and aged. In most cases, it is necessary to build up a structured document outlining the stress testing approach of the company as such, consistently with market best practices and highlighting the links to regulatory requirements. The updated stress test features and reverse market stress tests approach would fall logically in that structured document.
This approach, we believe, has the best chance to add a new reverse stress test for a specific type (or types – or indeed all of the reverse stress testing) to the existing stress testing. Note that it must be done with a full understanding of the current stress test framework, which will need, to some extent, to be reworked or reassessed.