Successful STSM application

Transparent Network and Graphical Models for FinTech Analysis

Luciana Dalla Valle, Associate Professor of Data Science and Statistics, University of Plymouth, UK.

STSM title: Transparent Network and Graphical Models for FinTech Analysis. Host institution: University of Pavia, Department of Economics and Management, Italy. Supervisor at the University of Pavia, Italy: Prof. Claudia Tarantola STSM project: Recently, regulators are facing the pressing need for the availability of transparent quantitative tools for a correct appraisal of financial risks. However, several AI approaches currently adopted for financial data are “black-box” models and non-transparent. Graphical models are probabilistic models expressing the conditional dependence structure between random variables and allowing a transparent and immediate interpretation of results. The project will develop interpretable, explainable and transparent graphical and network models for the analysis of financial products, in particular cryptocurrencies, integrating asset prices and textual information gathered from social media platforms.

Robustified Markowitz approach for cryptocurrencies

Alla Petukhina, Lecturer and researcher, University of Applied Sciences for Engineering and Economics – HTW Berlin, Germany.

STSM title: Robustified Markowitz approach for cryptocurrencies. Host institution: Bucharest University of Economic Studies – ASE Bucuresti, Bucharest, Romania. Supervisor at ASE Bucuresti: Prof. Dr. Daniel Traian Pele. STSM project: The proposed research collaborative project is an extension of the previous research project, Klochkov et al. (2021) We compare performance of robustified Markowitz portfolios with 7 other well-known risk-based allocation approaches for 88 assets, including 42 cryptocurrencies. The robustified approach stabilizes weights of assets along with improvement of diversification benefits in comparison with other benchmarks.

Towards Ai-empowered sustainability and transparency of Ai financial products

Galena Pisoni, Innovation and Entrepreneurship (I&E) Coordinator, University Cote d’Azur, France

STSM title: Towards Ai-empowered sustainability and transparency of Ai financial products Host institution: Norwegian University of Science and Technology, Trondheim Host at Norwegian University of Science and Technology: Prof. Dr. Rita Pimentel STSM project: The aim of this STSM was to initiate new research collaboration between our respective home departments on two lines of importance to the COST action, namely, i) on financial intelligence and the application of artificial intelligence (AI) to financial context especially in relation to tracing sustainability metrics and compliance with ESG goals for companies, thought big data, and in the implementation of AI-based solutions for assisting in complex business processes for this aim, and ii) transparency of Ai financial products, that is how decisions made by AI systems can be made more understandable for users, thus such systems be more accepted by final users.

How taxpayers perceive fiscal system? – an AI approach

Ioana Florina Coita, PhD Lecturer, University from Oradea, Faculty of Economics, Romania.

STSM title: How taxpayers perceive fiscal system? – an AI approach Host institution: Zeppelin University (ZU), Germany; Supervisor at ZU, Germany: Prof. Dr. Florentina PARASCHIV, Chair of Finance Department; STSM project: Purpose of the project was to identify the literature gap and structure the problem statement regarding taxpayers’ perception of fiscal system and its impact upon voluntary compliance. We aimed at evaluating several AI and econometrical models that would fit our data and objectives needed to identify the factors, internal and external, that impact taxpayers’ behaviour in relation to tax system. Main results referred to building research hypothesis and objectives:” Identifying patterns of the taxpayers that trust in State” and” Which are the features of tax system that taxpayers are most sensitive about when it comes to voluntary compliance?”. We used an online survey for collecting behavioural data which we explored using LASSO method compared to other ML algorithms (logistic regression, decision tree, support vector machine, naïve bayes). Interesting results were revealed by applying natural language processing along with machine learning tools in order explore sentiment according to people who trust and those who do not, based on their gender differences.

Fintech and Artificial Intelligence in Finance – Towards a transparent financial industry (FinAI)

Ahmad Amine Loutfi, PhD Fellow, Norwegian University of Science and Technology (NTNU), Norway.

STSM title: Fintech and Artificial Intelligence in Finance – Towards a transparent financial industry (FinAI). Host institution: Institute of Wealth & Asset Management, Zurich University of Applied Sciences (ZHAW), Switzerland. Supervisor at ZHAW, Switzerland: Prof. Dr. Peter Schwendner, Director of Institute. Supervisor at NTNU, Norway: Prof. Dr. Per Bjarte Solibakke. STSM project: In this project, we propose to study the extent to which alternative data (News/reports) can predict electricity spot prices. This will allow us to assess the backtesting of relevant investment strategies. The results of this project will also be used to extend an ongoing research project where we study electricity spot price prediction based only on conventional data. We aim to augment the conventional data set with a new feature which reflects the alternative data, and then run the newly augmented dataset through the same neural network model and then compute the new loss function results in order to assess the models’ performance with and without alternative data.

Genetic Programming for the Fraudulent Activity Detection: Performance and Transparenct Perspectives

Stjepan Picek, assistant professor, Delft University of Technology, The Netherlands.

STSM title: Genetic Programming for the Fraudulent Activity Detection: Performance and Transparenct Perspectives. Host institution: University of Zagreb, Faculty of Electrical Engineering and Computing (FER), Croatia. Supervisor at FER, Croatia: Prof. Dr. Domagoj Jakobovic, Full professor Start date: 2021-01-11 End date: 2021-01-26 STSM project: In this project, we will explore available datasets for fraudulent behavior classification and evaluate their common characteristics. Based on it, we will start a series of experiments with techniques from the machine learning domain and genetic programming to compare their performance. Finally, we will investigate what are the transparency considerations when using genetic programming and what kind of interpretability/explainability one can hope to achieve with such techniques.

Contagion dynamics in high frequency – modeling shock impacts in cryptocurrency markets

Danial Florian Saef, PhD Researcher, Humboldt University Berlin, Germany. Member of International Research Training Group 1792 “High Dimensional Nonstationary Time Series”.

STSM title: Contagion dynamics in high frequency – modeling shock impacts in cryptocurrency markets Host institution: University College London, United Kingdom. Supervisor at Humboldt University Berlin: Prof. Dr. Wolfgang Karl Härdle Supervisor at UCL: Prof. Dr. Tomaso Aste. [t.aste@ucl.ac.uk] STSM project: Shocks in financial markets often lead to severe contagion effects, especially in strongly correlated and highly capitalised markets these effects can be observed almost immediately. A large branch of literature has already focused on the phenomenon of financial contagion. However, the recent availability of high frequency data gives new possibilities to investigate it. A gap in the literature exists regarding the effects of financial contagion in correlated assets w.r.t. high frequency data. The newly emerging cryptocurrency market is highly capitalised, yet it differs from traditional markets due to non-stop trading, lower volume and high correlation among currencies. These properties cause larger volatility and make cryptocurrencies more vulnerable to contagion effects. We analyse a diverse tick cryptocurrency tick dataset and test for jumps following Lee & Mykland (2012), and then aim to model cryptocurrencies as a network of interconnected assets. We want to show that patterns in contagion dynamics exist and that they can be used to predict how future shocks evolve. Eventually, this model aims at helping to understand the dynamics of this new and largely unregulated asset class better. The non-parametric nature makes it also adaptable to applications in traditional assets, thus our goal is to provide a general framework for modeling the dynamics in correlated financial time series with jumps in high frequency.

A Data-driven Evolutionary Case-based Reasoning Approach for Financial Risk Detection

Wei Li, PhD Fellow, Norwegian University of Science and Technology, Norway. Member of Centre for Banking and Finance.

STSM title: A Data-driven Evolutionary Case-based Reasoning Approach for Financial Risk Detection. Host institution: Humboldt University Berlin, Germany. Supervisor at Norwegian University of Science and Technology: Prof. Dr. Florentina Paraschiv. Supervisor at Humboldt University Berlin: Prof. Dr. Wolfgang Karl Härdle. STSM project: In this mobility, the main targets are to improve my current paper about financial risk prediction applying machine learning method, and build up common interests in machine learning application in finance and make a collaboration for new papers.

Transparency in regulatory Benchmarking

Jasone Ramírez-Ayerbe, PhD student, Universidad de Sevilla, Spain.

STSM title: Transparency in regulatory Benchmarking. Host institution: Copenhagen Business School, Denmark. Supervisor at CBS, Denmark: Prof. Dr. Dolores Romero Morales. Supervisor at Universidad de Sevilla, Spain: Prof. Dr. Emilio Carrizosa. STSM project: In this STSM, we will investigate novel Mathematical Optimization formulations to construct counterfactual explanations in regulatory Benchmarking, i.e., a set of actions that can be taken by an instance such that the model at hand would have assigned a higher efficiency to the firm. This project aims at the development of innovative Mixed Integer Programming formulations to enhance transparency in regulatory Benchmarking and to apply it to the Benchmarking of Electricity Distribution System Operators.

Efficient algorithmic and computational tools for Bayesian inference of systemic risk interlinkages

Chalkis Apostolos, PhD student, University of Athens (NKUA), Greece.

STSM title: Efficient algorithmic and computational tools for Bayesian inference of systemic risk interlinkages. Host institution: Inria Research Center, Paris, France. Supervisor at NKUA, Greece: Prof. Dr. Ioannis Emiris. Supervisor at Inria Research Center, Paris, France: Dr. Elias Tsigaridas. STSM project: In this STSM, we will study the systematic risk interlinkages between European and international banks. In particular, we will study the significant variation in the cross-section of systemic risk measures of large banks during the recent financial crisis in a broad sample of countries. To address this problem, this STSM builts upon a novel computational and algorithmic framework to incorporate several Bayesian models to inference on the covariance matrix of multivariate distributions.