Financial companies must assess investments according to environmental and social criteria (ESG = Environmental, Social & Governance). The data on this topic is still rather sparse, resulting in a high degree of uncertainty, which is exacerbated by a lack of market standards which could provide improved alignment. Furthermore, the legal challenges to cooperation are significant.
In this use case, Deloitte combines and standardises various sustainability assessments for particular groups of banks and asset managers (e.g. funds) on the basis of EuroDaT. The use of modern methods such as distributed learning and computing should allow one to analyse business-critical data in such a way that no raw data can be reconstructed and that, post-factum, it is not known which companies were assessed by which bank.
The basic idea: The financial institutions share their own assessments and from these an ESG ranking of the companies is calculated. If sufficient financial institutions participate, this leads to a strong market representation, which in turn has a high value for the involved companies.
Individual ESG assessments of financial institutions involve both sensitive business secrets and information that could influence the market and thus cannot be published easily.
Using mathematical algorithms and intelligent technical implementation together with the data trustee EuroDaT achieves the following:
- The internal models and ESG assessments of the financial institutions remain permanently and unconditionally confidential.
- Even a successful attack on communication links or by a participant does not completely undermine confidentiality.
- The trustee does not store or analyse any of the provided data.
As a consequence, none of the provided information can be reconstructed post-factum. Only the overall result, i.e. the ranking of companies according to ESG criteria is published. However, this also implies a very low probability of detecting fraud. In order to discourage fraudsters, the algorithms are designed to be robust; a single fraudster is unable to substantially influence the overall result and is therefore deterred by the poor cost-to-benefit ratio.
Technical safety measures as well as legal assessments will secure the entire project, enabling financial institutions to participate securely.
The techniques developed in this project are also transferable to other cases, such as data aggregation.
Keywords: Data Sovereignty, Differential Privacy, Federated Learning, Plausible Deniability, Multi-Party-Computation, Homomorphic Encryption
ZEVEDI, Saarland University, DFKI