The STAMP(sic) IT Platform: Digitalisation for Modelers


Henry J., Januario A.

4th International Conference on Intelligent and Fuzzy Systems (INFUS), Bornova, Turkey, 19 - 21 July 2022, vol.505, pp.380-390 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 505
  • Doi Number: 10.1007/978-3-031-09176-6_44
  • City: Bornova
  • Country: Turkey
  • Page Numbers: pp.380-390
  • Keywords: IT system, Modelling, Reverse stress test
  • Istanbul Technical University Affiliated: No

Abstract

Stamp(sic) IT is a data science platform providing a "model management" cooperative tool. Users can create, change and run models and analyse simulations, all in a shared digital environment. There is always a single "Stamp-ed" version of the model, as validated via an embedded governance process. Users can be "basic" who run pre-defined simulations, "advanced" who may change some of the parameters governing the pre-defined simulations, or "developers" who have full freedom to amend model and simulation designs. The original IT set-up combines notebooks to run simulations, a master and variant code repository, and a tool to compare simulation results (across varying parameters, sample, time horizon etc.). These three tasks are respectively conducted by using Jupyter, Bitbucket and MLF1ow, all integrated in one place. Screenshots document the step-by-step process involved. We finally provide a use case, by conducting a complex and computation-intensive "reverse stress-test" for a sample of euro area banks, requiring multiple model runs and specific visualisations. We specify the concept of reverse stress test for multiple institutions considered jointly. We introduce an efficient way of running simulations as an alternative to Monte Carlo simulations. Results are provided using a multiplicity of scenarios, covering a range of underlying probabilities.