Specifying and Communicating the Semantics of Stylized Facts

Martin, Jan Niklas (2025) Specifying and Communicating the Semantics of Stylized Facts. Masters thesis, Institute for Visual and Analytic Computing, University of Rostock.

Text Masterarbeit_Jan_Niklas_Martin.pdf Restricted to Registered users only Download (2MB)

Abstract

The usage of simulation systems has become more prominent over the last decades [BA13]. An essential aspect of model development is model validation, which checks the representativeness of a model. A model that has not been thoroughly tested for correct behavior can produce incorrect results. In the worst case, this can lead to incorrect conclusions in the real world. Model validation is therefore of enormous relevance in order to establish trust in simulation models. Typically, model validation is carried out using real data. In some cases, this is not possible for various reasons, e.g. because it involves investigating a scenario that has not yet occurred and therefore cannot be empirically recorded, or because collecting sufficient real data would be too costly or time-consuming. Accordingly, there is a need for model validation methodologies that can be performed without real data. An established approach is the use of stylized facts, which summarize domain-specific and empirically determined knowledge. A language was developed in [WZA+23] that enables the formalization of stylized facts. The language has not been named, we will simply refer to it as SF-DSL. A separate model checker was also developed that can evaluate established stylized facts. However, the meaning of expressions in stylized facts can change in each use case. For example, a ’strong correlation’ can be interpreted differently in different research fields. Therefore, the language lacks the ability to flexibly adapt the semantics of the expressions it contains. To solve this problem, we extend the language in this work by a semantic parameterization, which allows an adaptation of the expressions contained in the language. This parameterization allows changing numerical values such as thresholds used but also the logical evaluation of certain operators. In the absence of transparent communication during model validation, there is a risk of eroding public and scientific trust in the respective simulation model. To mitigate this risk, we generate assurance cases during validation, to ensure clarity and simplicity in the validation process. The assurance cases argue for the evaluation results of the stylized facts in the context of the semantic parameterization used. This serves the transparent communication of model validation. Especially because assurance cases are intuitive to read and do not require a great deal of prior technical knowledge. Another benefit of this approach is that it facilitates the modeling itself by enabling the modeller to recognize precisely why certain stylized facts are not fulfilled by the model. Thus, the extension of SF-DSL increases the application possibilities, since domain-specific conditions can be recorded in the semantic parameterization, and improves the transparency of the model validation through the automatically generated assurance cases.