I would like to discuss a problem with you that I see repeatedly encountered when performing complex, multifarious tasks such as bank stress tests: the lack of ability to concentrate on the "bigger picture" and the consequences this has on the analyst.
Where is the problem?
The EU-wide stress test created by the EBA is a great example of a task in which there are several different layers of tasks to perform, including:
a detailed understanding of the EBA methodology,
application of the methodology to the current organisational situation and
the implementation of all the individual steps following several already complex test scenarios.
The typical approach to implementing these complex test scenarios is to use a variety of one-off solutions such as scripts, manual data transformations in Excel, portfolio groupings, and subsequent transformations in SAS, Python, etc.
This approach requires frequent switches between high and low-level contexts as well between the programs and the test methodology itself.
Furthermore, the individual steps such as data preparation, transfers, and reading risk parameters, become very error prone and time consuming, meaning analysts have less time to spend with the results of the test.
Focus on the scenario
To help solve this problem, we created a tool to help analysts fully focus on what they need to achieve - the test scenario and the interpretation of the results- by taking care of how it is done.
The tool uses a domain-specific language (DSL) to describe the test scenario. The language is essentially standard Groovy extended by several domain-specific constructs - portfolio migrations, RWA and IFRS 9 ECL calculations, access to particular module results, access to PD and LGD repositories, execution of custom analytics code in MongoDB, and even the population of the EBA template in MS Excel.
Because we used some of these language-specific components in production, the behavior of the tool can be personalized simply by switching various features. By using our system, the analyst can focus on writing the scenario while easily calling transformation, getting results, and storing results as a simple variable.
Additionally, as we’ve already laid the framework for these smaller steps, you need only customize it too with your specific parameters.
The Conclusion
It’s rather easy to get lost in the number of required transformations, scenarios, and time shifts of risk parameter vectors. Breaking down the transformations and simplifying the operations provides the ability to focus on the big picture, which matters the most.
What is your approach to complex exercises like the upcoming 2020 EU-wide stress test?
Comments