SupTech workshop on text mining with machine learning
The fourth SupTech workshop at the Czech National Bank covered text mining with machine learning.
SupTech workshop on credit risk analysis in P2P lending
The second SupTech workshop within the FIN-TECH project presented carefully curated use cases on advanced scoring models in peer-to-peer lending. The case studies embodied the blend of standard big data analytics techniques and network analysis. Participants acknowledged that presented use cases might be implemented in their work.
The second day of the workshop was comprised of industry presentations by three Czech fintech firms: Twisto, Roger and Bulletproof AI. Michal Kročil, Chief Risk Officer at Twisto, showed how his company achieves the balance between great user experience and accurate risk analysis in payment services. Adam Šoukal, co-founder and CEO of Roger a.s., talked about the company’s products in online invoice trading and how electronic data interchange is achieved in financial technologies. Martin Řehák, founder and CEO of Bulletproof AI, showed how different standard techniques in big data analytics might result in biases and financial losses. Workshop participants discussed what exactly should be regulated in FinTech activities including the discrimination of clients due to unavailability of data, personal responsibility of model builders (similar to financial institutions), technical robustness of machine learning models to fraud and the overall accuracy of used models. Representatives of the participating firms also pointed out that the quality of data collection and preparation, the complexness of data and model features might be a particular way for FinTech regulation. The discussion resulted in an interesting suggestion on creating specific protocols, that might be beneficial for firms and regulator to test models on specific data samples (created by the regulator with known and potential misspecifications and outliers), and generating stress testing scenarios based on sectoral development and different hypotheses (not only a macroeconomic stress scenario, provided by the regulator).