Abstract
Modern data driven Artificial Intelligence (AI), enabled by powerful machine learning models, is rapidly changing financial services, leading to a widespread diffusion of financial technologies (fintechs). While financial technologies bring important opportunities (increased financial inclusion, better transparency, lower transaction costs) they may also lead to new risks. The European Horizon 2020 FIN-TECH (2019-2021) and PERISCOPE (2021-2023) projects, have developed statistical models aimed at measuring fintech risks and at improving the security of AI applications to finance. The current widespread use of AI motivates the need to develop further statistical methods that can measure the trustworthiness of AI systems in the different application domains and in finance, in particular.
The aim of the SAFE-AI project is to propose a set of statistical metrics to assess the trustworthiness of AI applications in finance, and to formalise an AI risk management model. The metrics to be developed will fulfill the regulatory requirements recently proposed in the European AI Act, which we suggest to combine in a set of integrated S.A.F.E. statistical scores, all based on the extension of the well known Lorenz curve: from the measurement of concentration in population incomes to the measurement of concentration in machine learning predictions.
The acronym S.A.F.E. derives from the four proposed metrics: Security, referred to the resilience of the AI model output to extreme events and/or cyber attacks; Accuracy, referred to the predictive accuracy of the model; Fairness, referred to the absence of biases towards specific population groups, induced by the AI output; Explainability, referred to the capability of the model output to be understood and oversight by humans, particularly in its driving causes.
Description of research unit
● Corresponding researcher / coordinator: Emanuela Raffinetti
Assistant Professor of Statistics in the Department of Economics and Management - University of Pavia
Her research interests are mainly focused on: Explainable, Accurate, Fair and Sustainable Artificial Intelligence; Machine Learning model validation methods; assessment of operational and cyber risks; dependence analysis; sub-sampling methods; inequality measures for income distributions. She is Associate Editor of the following international scientific journals: Statistics; Frontiers in Artificial Intelligence.
Other researchers:
Associate Professor of Statistics in the Department of Economics Management - Department of Economics and Management - University of Pavia
His main research interests cover the application of econometric and network theory approaches to the study of financial stability, systemic risk and the emergence of interdependencies in economic and financial systems. Between 2016 and 2017, he worked as Big Data Analyst at OMNICOM MEDIA GROUP for the development of Marketing Analytic Models that supports marketers to create data-driven marketing plans.
● Francesca Mariani (external):
Assistant Professor of Economics in the Department of Economic and Social Sciences - Marche Polytechnic University
Her research includes analytical and computational methods for asset and option pricing; risk management; volatility forecasting. Recently, she is working on machine learning approaches for financial applications and new composite indicators. She is member of editorial board of “The Italian Journal of Economic, Demographic and Statistical Studies”.
● Gloria Polinesi (external):
Researcher in Statistics in the Department of Economic and Social Sciences - Marche Polytechnic University
Her research interests span various fields of statistics, ranging from the construction of composite indicators to the risk evaluation of portfolio models. She is the author of several articles in top journals, including the Annals of Operation Research and Computational Management Science.
● Maria Cristina Recchioni (external):
Professor of Economics Statistics in the Department of Economic and Social Sciences and director of the Department - Marche Polytechnic University
Her research interest includes: analytical and computational methods for asset and option pricing; risk management; volatility forecasting. Recently, she is working on machine learning approaches for financial applications and new composite indicators. She is co-editor in chief of “Statistical Methods and Applications”, associate editor of “European Journal of Operational Research” and “Journal of Economic Interaction and Coordination”.
Retired Full Professor (Department of Mathematics - University of Pavia)
His main research interests are focused on: mathematical and numerical methods in kinetic theory of rarefied gases; granular gases; statistical Mechanics; diffusion equations; hyperbolic systems and applications; mathematical modeling in socio-economic and related problems.
Additional research team members:
● Golnoosh Babaei (Post-doc)
● one Phd student (to be recruited)