Primary tabs

University of Pavia

yes
yes, however a further risk/opportunity should be included.
There is need to address the informational" risks posed by Fintech activities to their customers, which are amplified by the great interconnectdness of the Fintech ecosystem. Due to their "disintermediated" nature, and in the light of a possibly aggressive marketing activity, Fintechs may assess credit scoring and/or credit risk of a counterparty in a way that does not accurately estimate the underlying probability of default. This bias may be amplified by the systemic risk that arise among Fintech borrowers, highly intereconnected with each other so that, when a counterparty fails, there may be a strong contagion effect on other, neighbouring, counterparties, that may further bias the estimate of the probability of default. To tackle this issue, and allow the devlopment of the innovative services provided by fintechs, we suggest to use "network" based models to estimate default which, using transactional/networked data naturally available for fintechs, improve the estimation of credit risk, both at the idiiosyncretic and at the systemic level. These models belong to the so-called financial network models, employed so far to measure systemic risks and contagion. Our R&D work, documented by research published in relevant finance scientific journal,s shows how it can effectively improve predictive perfomance, on real case studies that concern European Fintechs."
Credit institutions will likely evolve, to a fintech" style model, either directly or through partnerships. This means that informational risks, as described in the previous comment, amplify and become more systematically important. The same for cyber operational risks."
yes, I would suggest that the suggested training and workshop activity for supervisors indicated could be extended to include fintech researchers from the academia
yes
yes
yeses
yes
yes
yes
yes
Yes, although I would add that machine learning and AI algorithms should be explicitly evaluated, looking at what actual statistical advantage they bring, for example in terms of improved predictive accuracy. Supervisors cannot establish the coding" correctness of an algorithm, however, they can establish its "statistical" ("learning") performance"
yes
yes
Paolo Giudici
++393316421199