MSCA Individual Research Projects
Strengthening European financial service providers through applicable reinforcement learning
- Host institution: University of Twente.
- Starting month: M3.
- Duration: 36 months.
- Pillar 1: Introduction to AI for financial applications (WWU, 4 ECTs), Work Package 2
Objectives
Reinforcement Learning (RL) has become popular for automating uncertain decision-making in complex environments. Deep reinforcement learning can make impressive algorithmic decisions in closed environments, but real-world applications in open environments are harder. This project examines how RL can advance digital finance.
Expected Results
The project will address several RL implementation issues in digital finance. Utility-based RL deliverables will improve financial decision-making by developing multi-criteria analysis, extreme scenarios, and risk management methods. RL in decision-support will be optimised for explainability, regulatory compliance, model abstractions, and human judgement. We will also examine technological challenges like Digital Twin environments, machine learning pipelines, and digital finance ecosystem integration.
Planned Secondments
- CAR, Altin Kadareja (CEO), M6, 18 months, applied research on Fintech innovations with Deep learning
- ECB, Lukasz Kubicki, M27, 4 months, training on EU principles, supervision policies and research
Modelling green credit scores for a network of retail and business clients
- Host institution: University of Twente.
- Starting month: M6.
- Duration: 36 months.
- Pillar 1: Sustainable finance (UNA, 4 ECTs), Work Package 5
Objectives
Some markets use green credit scores to assess SME credit risk in sustainable and circular economies. Simultaneously, network customers' default likelihood has been studied. This study develops and deploys green credit score models that account for customers' networks. We show the impact and give financial institutions methods to improve credit risk assessment and access.
Expected Results
Green credit score models will be developed and implemented. These models inform SMEs about their carbon footprint, their main risks in a low-carbon economy, and how to mitigate them. SMEs leading on sustainability could gain easier access to capital by demonstrating positive relationships between creditworthiness and sustainability, creating a fairer credit risk assessment that explicitly factors in sustainability metrics and encouraging low-carbon measures.
Planned Secondments
- SWE, Prof. Dr. Tadas Gudaitis, M12, 18 months, ESG and credit score modelling
- ECB, Lukasz Kubicki, M33, 4 months, exposure to globally leading central bank, research training on EU principles, supervision
Industry standard for blockchain
- Host institution: University of Twente, The Netherlands (UTW)
- Starting month: M9
- Duration: 36 months
- Pillar 1: Introduction to Blockchain applications in finance (HUB, 4 ECTs), Work Package 4
Objectives
Cryptocurrencies and other digital assets have proliferated in recent years, elevating blockchain technology. Decentralised finance requires it. Blockchain is well-established in digital finance, but it lacks maturity and scalability. This project leverages the extensive and diverse DIGITAL network, which includes financial industry leaders, to set a blockchain industry standard.
Expected Results
The project will map and categorise blockchain technology literature and practice to create an industry standard. DIGITAL includes the ECB, multinational banks, and leading digital finance tech companies. This combination lets us analyse blockchain and find commonalities and desires that could lead to a standard. We will work on several blockchain use cases, inspired by industrial partners, to ensure that the results do not stall at a normative framework level but trickle down to concrete and relevant demonstrations of the standard implemented in digital finance.
Planned Secondments
- DEL, Alberto Ferrario, M18, 18 months, business modelling, research on use-cases and prototypes
- Fraunhofer Institute, Prof. Dr. Ralf Korn, M12, 4 months, applied industry-research, exposure to world-leading research centre and infrastructure
A recommender system to re-orient investments towards more sustainable technologies
- Host institution: University of Twente, The Netherlands (UTW)
- Starting month: M12
- Duration: 36 months
- Pillar 1: Sustainable finance (UNA, 4 ECTs), Work Package 5
- Work Packages: WP5, WP6, WP7, WP8
Objectives
Recommender systems are well-known information filtering systems that suggest items most relevant to a user. To our knowledge, there are none that suggest investments in sustainable technologies and businesses. This project will develop and deploy a recommender system to help financial institutions and their clients invest in sustainable technologies.
Expected Results
The project informs user groups about investment sustainability. Sustainability KPI mapping and evaluation are project deliverables. The recommender system's explainability is crucial. Thus, the recommendations will be tailored to multiple user classes with appropriate explanations and interpretations. The system recommends sustainable investments, monitors portfolio performance, and dynamically updates financial and sustainable KPIs.
Planned Secondments
- RAI, Dr. Stefan Theußl, M24, 18 months, research exposure in a global business environment,
- ARC, Prof. Dr. Ioannis Emiris, M18, 4 months, applied industry-research, exposure to world-leading research centre and infrastructure
Fraud detection in financial networks
- Host institution: WU Vienna University of Economics and Business, Austria (WWU)
- Starting month: M3
- Duration: 36 months
- Pillar 1: Introduction to Blockchain applications in finance (HUB, 4 ECTs), Work Package 4
- Work Packages: WP4, WP6, WP7, WP8
Objectives
Detecting fraud is currently one of the most important topics in Finance. However, it is also one of the most complex, given that fraudsters typically represent and generate a highly dynamic system, requiring that the boundaries and objectives of any system designed to detect and reduce fraud be constantly adapted to new extrinsic structures. This enables the definition of not only a static fraud detection system, but also a dynamic AI learning system, particularly in relation to network analysis.
Expected Results
On a meta-level, a set of Machine Learning and Artificial Intelligence models will be defined to enable a research-based approach that can be applied directly in financial institutions. The models are defined in such a way that the outcomes of the learning process within the institutions can be used to define and design new algorithms from a scientific standpoint. The work on network algorithms during the process of designing Machine Learning environments, will result in the publication of seminal papers.
Planned Secondments
- RAI, Dr. Stefan Theußl, M6, 18 months, research exposure in a global business environment
- ECB, Dr. Lukasz Kubicki, M27, 4 months, exposure to globally leading central bank, research training on EU principles, supervision
Collaborative learning across data silos
- Host institution: WU Vienna University of Economics and Business, Austria (WWU)
- Starting month: M9
- Duration: 36 months
- Pillar 1: Foundation of data science (BBU, 4 ECTs), Work Package 1
Objectives
Connecting several dozen different data pipeline components and integrating an excessive number of APIs to leverage siloed data is a significant barrier to the comprehensive implementation of AI-based systems in finance. Currently, very little research is devoted to addressing all of the challenges associated with training, testing, and deploying cutting-edge ML and DL methods while leveraging siloed data. We will concentrate on the data challenges that finance service providers face by proposing solutions to streamline data collection, resolve data quality issues, and structure data to support downstream processes.
Expected Results
APIs for integrating Machine Learning and Deep Learning algorithms into FinTech processes necessitate careful abstraction of the specified input and output, which is the responsibility of the researchers to simplify and aggregate the complexity. This project produced a large number of API definitions that are closely related to research papers in the fields of theory of Artificial Intelligence and Machine Learning, as well as theory of Finance applications in various sub-fields such as security and compliance. The API specification itself should not only be integrated into financial institutions' business processes, but should also provide fruitful input for new research papers that are of interest to readers and users of all involved fields of research.
Planned Secondments
- SWE, Prof. Dr. Tadas Gudaitis, M18, 18 months, research on prototype implementations, applied research
- FRA, Prof. Dr. Ralf Korn, M12, 4 months, applied industry-research, exposure to world-leading research centre and infrastructure
Risk index for cryptos
- Host institution: Humboldt University Berlin, Germany (HUB)
- Starting month: M3
- Duration: 36 months
- Pillar 1: Introduction to Blockchain applications in finance (HUB, 4 ECTs), Work Package 4
Objectives
The cryptocurrency market is exceptional: it is volatile, with a constantly shifting market structure. As cryptocurrencies evolve into a class of investable assets, the need for an index product arises. We investigate the dependencies of tail risk events within cryptocurrencies, which entails identifying coins with high or low joint tail event risks. Based on this, we intend to develop a risk index for cryptocurrencies to measure joint tail events, which will be an important tool for communicating risks to the public and regulators.
Expected Results
Develop a risk index to understand, measure and forecast upcoming risk flows from all cryptocurrency market participants and risk drivers. The possibility of pinpointing co-tress components in a dynamic network context makes the tool versatile and flexible for Digital Finance. The index will provide a thorough understanding of cryptocurrencies, and measure the dependencies and spillover effects in tail risk events within cryptocurrencies. It helps investors to manage risks and support decision-making.
Planned Secondments
- ROY, Dr. Michael Althof, M6, 18 months, research in innovation-driven business, use-case implementation
- FRA, Prof. Dr. Ralf Korn, M27, 4 months, applied industry-research, exposure to world-leading research centre and infrastructure
Detecting anomalies and dependence structures in high dimensional, high frequency financial data
- Host institution: Humboldt University Berlin, Germany (HUB)
- Starting month: M9
- Duration: 36 months
- Pillar 1: Foundation of data science (BBU, 4 ECTs), Work Package 1
Objectives
Herding, a well-known financial anomaly, is thought to cause high volatility, volatile prices, and low liquidity (Bikhchandani and Sharma, 2000). Greed and herd behaviour caused the seventeenth-century tulip mania, the 1995–2000 Internet bubble, and the 2015 Chinese stock market crash. This project studies high-dimensional sentiment networks and herd behaviour on the stock market. To better fit investor sentiment, the project will calibrate the option pricing model, Stochastic Volatility and Correlated Jump (SVCJ).
Expected Results
The project will detect anomalies like herd behaviour and dependence structures in high-dimensional, high-frequency financial data. We plan to create a tail event-driven network that graphs or matrices the interconnections of a large panel to understand sentiment network mechanics. That will inform our herd behaviour detection and option pricing model calibration. 1) Publications in prestigious journals available via public repositories, 2) Presentations at prestigious conferences, and 3) Knowledge exchange
Planned Secondments
- INT, Prof. Dr. Marco Bianchetti, M18, 18 months, Analyse quantitative modelling & technology and risk
- ARC, Prof. Dr. Ioannis Emiris, M12, 4 months, applied industry-research, exposure to world-leading research centre and infrastructure
Audience-dependent explanations
- Host institution: University of Naples Federico II, Italy (UNA)
- Starting month: M3
- Duration: 36 months
- Pillar 1: The need for eXplainable AI: methods and applications in finance (BFH, 4 ECTs), Work Package 3
Objectives
To address the issue of explainability in complex models, the literature has proposed an ever-expanding list of post-hoc explainability methods that can be used to gain some understanding of the inner workings of complex models. However, explaining the inner workings of algorithms and their interpretation is entirely dependent on the target audience. The existing literature fails to match the growing number of explainable AI (XAI) methods with the varying explainability requirements of stakeholders. To promote the widespread adoption of AI-based systems in finance, additional research is required to map the requirements of explainable systems across the various stakeholders in the finance industry
Expected Results
Finance decision-makers and AI model builders don't understand XAI's capabilities or ESG's impact on society and economy. This project promotes dialogue and knowledge transfer between those camps. It facilitates AI, Sustainable Finance, and ESG Technology innovation and collaboration. The following channels will disseminate expected results: 1) technical reviews, newspapers, and magazines, 2) public events (workshops for results presentation), and 3) knowledge exchange with stakeholders and project partners.
Planned Secondments
- SWE, Prof. Dr. Tadas Gudaitis, M6, 18 months, policies for asset, sustainable fund management
- FRA, Prof. Dr. Ralf Korn, M27, 4 months, improve know-how transfer by using and implementing advanced financial models
Experimenting with Green AI to reduce processing time and contributes to creating a low-carbon economy
- Host institution: University of Naples Federico II, Italy (UNA)
- Starting month: M9
- Duration: 36 months
- Pillar 1: Sustainable finance (UNA, 4 ECTs), Work Package 5
Objectives
Green AI supports the use of resources more efficiently and conserves them for future generations. Multiple applications have been presented in different areas, however, there are no studies exploring the impact that the use of green AI concepts can have in the Financial industry. This research objective focuses on experimenting with green AI concepts in multiple applications in finance, analysing economical and practical impact of its deployment in industry. It facilitates the exchange of innovative ideas and cooperation opportunities in the field of Environmental, Social, and Governance (ESG), Sustainable Finance, and ESG Technology.
Expected Results
The project aims at providing reports about pricing and risk management of green financial instruments across all asset classes, with a focus on new products development, model validation, model risk management, funding and counterparty risk, fair and prudent valuation, applications. It aims at focusing on financial inclusion and inequality. The WP will also have a strong focus on discussion and disseminating of the main results with the aim of spreading the culture of green AI and creating a table for the discussion of new proposals and rules. 1) publications in open access journals, 2) presentations at prestigious conferences and 3) knowledge exchange with stakeholders and project partners 4) General outreach (Media, Open Science Day).
Planned Secondments
- INT, Prof. Dr. Marco Bianchetti, M18, 18 months, Study ESG-related decision making, related risk management
- ARC, Prof. Dr. Ioannis Emiris, M12, 4 months, Develop models to industrialise evidence-based, data-driven sustainable finance
Applications of Agent-based Models (ABM) to analyse finance growth in a sustainable manner over a long-term period
- Host institution: Kaunas University of Technology, Lithuania (KUT)
- Starting month: M6
- Duration: 36 months
- Pillar 1: Sustainable finance (UNA, 4 ECTs), Work Package 5
Objectives
Agent-based systems are computer models that simulate the behaviours and interactions of autonomous agents, either as individuals or in groups, in order to gain a deeper understanding of how a system behaves and what factors influence its outcomes. In agent-based modelling, a system is represented as a collection of autonomous decision-making units, or agents (ABM). Each agent evaluates its own situation and makes decisions according to a set of rules. Agents are capable of a variety of appropriate behaviours for the system they represent. ABM has been utilised in numerous financial investigations. The literature contains few ABM studies that model economies and markets while assuming the industry's adoption of sustainable finance
Expected Results
This study aims to use agent-based models to simulate different market scenarios in which industry agents take sustainable actions. Long-term financial growth will be analysed, and the findings will aid in the development and modification of industry policies and strategies. A public repository containing a library of the developed agent-based models is another anticipated outcome. The WP will place a strong emphasis on disseminating and the anticipated outcomes. Several channels, including peer-reviewed articles in high-impact journals, research talks at national and international conferences, and use case presentations at industry workshops, will be utilised to accomplish this objective.
Planned Secondments
- DEL, Alberto Ferrario, M12, 18 months, analyse finance growth in an applied research setting
- ARC, Prof. Dr. Ioannis Emiris, M33, 4 months, applied industry-research, exposure to world-leading research centre and infrastructure
Developing industry-ready automated trading systems to conduct EcoFin analysis using deep learning algorithms
- Host institution: Bucharest University of Economic Studies, Romania (ASE)
- Starting month: M12
- Duration: 36 months
- Pillar 1: Introduction to AI for financial applications (WWU, 4 ECTs), Work Package 2
Objectives
This IRP will focus on addressing the challenges associated with automated trading systems in the direction of industry-ready platforms, i.e. minimising the risks of mechanical failures, improving the explainability of the underlying AI/ML models used in automated trading systems to better address performance-related issues, and also addressing ESG/CSR and ethical issues. This area will contribute significantly to Green Finance, thereby addressing the European Green Deal.
Expected Results
The project's outcomes will provide financial institutions with new automated trading tools. The primary anticipated outcome of the project is the design of new trading algorithm solutions for mitigating the risks of mechanical failures, enhancing the explainability of the underlying AI/ML models used in automated trading systems to better address performance-related issues, as well as ESG/CSR and ethical concerns. The anticipated results will be disseminated through the following channels: 1) Publications in prestigious journals made widely accessible through public repositories; 2) Presentations at prestigious conferences; and 3) Knowledge exchange with project partners.
Planned Secondments
- ROY, Dr. Michael Althof, M24, 18 months, research on crypto assets for prototype and user acceptance
- ARC, Prof. Dr. Ioannis Emiris, M18, 4 months, applied industry-research, exposure to world-leading research centre and infrastructure
Predicting financial trends using text mining and NLP
- Host institution: The Babes-Bolyai University, Romania (BBU)
- Starting month: M6
- Duration: 36 months
- Pillar 1: Introduction to Blockchain applications in finance (HUB, 4 ECTs), Work Package 1
Objectives
This ESR's primary objective is to improve the use of AI-based natural language processing (NLP) solutions in order to predict credit risk and fiscal fraudulent behaviour based on speech text from audit reports, social media, and other sources. Predicting non-compliance based on free-text responses from survey respondents' perceptions. Constructing attitudinal indices based on free text and incorporating them into behavioural models, along with other qualitative or quantitative factors, in order to predict the likelihood of system fraud or the level of risk associated with accreditation.
Expected Results
Constructing large databases that provide both qualitative and quantitative data for use in the development of AI algorithms for both public and private entities (prediction of tax fraud) (banks, FinTechs offering credit services, etc.). Using text mining and NLP, evaluate the viability of various models that could predict the risk of fraudulent behaviour in the financial sector. Utilisation of these models in both the public sector (public policy formulation) and the private sector (help banks and FinTechs in credit scoring).
Planned Secondments
- RAI, Dr. Stefan Theußl, M12, 18 months, research exposure in a global business environment, trend modelling
- ECB, Dr. Lukasz Kubicki, M33, 4 months, exposure to globally leading central bank, research training on EU principles, supervision
Challenges and opportunities for the uptaking of technological development by industry
- Host institution: Cardo S.R.L, Italy (CAR)
- Starting month: M6
- Duration: 36 months
- Pillar 1: Introduction to AI for financial applications (WWU, 4 ECTs), Work Package 2
Objectives
The building blocks of any institutional investor's loan portfolio are cash flows. Using public and proprietary data, the ESR will conduct research and develop a machine learning tool capable of performing grouped time series forecasting on a private debt portfolio spanning multiple geographies, sectors, and whose features can also be grouped at other levels, such as loan amount and interest rate. In our innovation-driven industry, we analyse the obstacles and opportunities associated with adopting technological advances.
Expected Results
The project's outcomes will contribute to the expanding body of knowledge concerning the applications of cutting-edge machine learning and artificial intelligence techniques to traditional financial problems. Specifically, the first phase of the project will concentrate on missing value imputation for loan payment time series, while the second phase will adopt a more general predictive approach, that of grouped time series forecasting, possibly incorporating the first step. The anticipated outcome will be three research/conference papers describing the data analysis, modelling approaches, and experimental results.
Planned Secondments
- HUB, Prof. Dr. Wolfgang Härdle, M12, 18 months, research on Fintech innovations and implementations
- ARC, Prof. Dr. Ioannis Emiris, M33, 4 months, applied industry-research, exposure to world-leading research centre and infrastructure
Deep Generation of Financial Time Series
- Host institution: Cardo S.R.L, Italy (CAR)
- Starting month: M12
- Duration: 36 months
- Pillar 1: Introduction to Blockchain applications in finance (HUB, 4 ECTs), Work Package 1
Objectives
Macroeconomics factors such as central banks’ interest rates, inflation, unemployment rate, house price indices, to name a few, are of foremost importance in Financial Markets. The aim of this project is to benchmark various methods from classical statistical learning and modern machine learning in order to predict their point value in the future. As a second step the student will be using the above predictions to forecast future market scenarios in a what-if fashion.
Expected Results
The project's outcomes will contribute to the expanding body of knowledge concerning the applications of cutting-edge machine learning and artificial intelligence techniques to traditional financial problems. We will apply recent findings from the ML literature on time series forecasting in the first step. In the second phase of the project, the ESR will be able to conduct research in the field of causal inference in finance, which also appears to be an extremely promising area of study. The anticipated outcome will be three research/conference papers describing the data analysis, modelling approaches, and experimental results.
Planned Secondments
- WWU, Prof. Dr. Kurt Hornik, M24, 18 months, theoretical modelling and mathematics for deep learning
- Fraunhofer Institute, Prof. Dr. Ralf Korn, M18, 4 months, applied industry-research, exposure to world-leading research centre and infrastructure
Investigating the utility of classical XAI methods in financial time series
- Host institution: Bern Business School, Switzerland (BFH)
- Starting month: M6
- Duration: 36 months
- Pillar 1: Introduction to AI for financial applications (WWU, 4 ECTs), Work Package 2
Objectives
The introduction of complex ML and DL methods for financial time series forecasts potentially enables higher predictive accuracy but this comes at the cost of higher complexity and thus lower interpretability. For cross sectional data classical XAI approaches can lead to valuable insights about the models’ inner workings, but these techniques generally cannot cope well with longitudinal data (time series) in the presence of dependence structure and non-stationarity. The literature currently does not offer any XAI approach that is specifically developed for financial time series. Further research is needed on developing explainability methods that can be applied to complex models like deep learning methods (DL) which preserve and exploit the natural time ordering of the data.
Expected Results
Within this IRP, we will propose a set of novel explainability functions that are specifically tailored for financial time series. We envision a framework for XAI in finance that addresses the shortcomings of existing methods. Namely, under existing, perturbation-based XAI methods, if features are correlated, the artificial coalitions created will lie outside of the multivariate joint distribution of the data. Furthermore, generating artificial data points through random replacement disregards the time sequence hence producing unrealistic values for the feature of interest. In addition to the novel, finance-tailored methodology for obtaining explanations, the project will also aim to produce industry-ready deployments of the novel XAI techniques developed.
Planned Secondments
- INT, Prof. Dr. Marco Bianchetti, M12, 18 months, quantitative modelling & technology and risk management
- FRA, Prof. Dr. Ralf Korn, M18, 4 months. Research needs to be validated with industry to achieve the envisioned impact
Fair Algorithmic Design and Portfolio Optimization under Sustainability Concerns
- Host institution: Bern Business School, Switzerland (BFH)
- Starting month: M6
- Duration: 36 months
- Pillar 1: The need for eXplainable AI: methods and applications in finance (BFH, 4 ECTs), Work Package 3
Objectives
The surge in interest in algorithmic fairness and sustainability is present in numerous fields of study, including finance and portfolio management in particular. This project's objective is to create new portfolio optimization models that address some of the difficulties associated with incorporating fairness and sustainability into investment management. The objective of the project is to increase understanding of the source and methods for eliminating algorithmic bias in finance in order to generate sustainable outcomes. The project will equip financial institutions with new sustainable and equitable algorithmic solutions to increase customer trust.
Expected Results
The primary anticipated outcome of the project is the development of new algorithmic solutions for multiple areas of finance, such as sustainable portfolio management. The project will equip financial institutions with new tools to comply with EU sustainability regulations. The subsequent anticipated outcome is the publication of a library containing all of the designed algorithms in a public repository. A significant emphasis will be placed on the dissemination of the anticipated results, which will be accomplished through the following channels: at least one publication in prestigious open-access journals and at least three presentations at prestigious conferences and open events. The final outcome of the project will be a comprehensive exchange of knowledge with project partners.
Planned Secondments
- ROY, Dr. Michael Althof, M24, 18 months, for training in portfolio optimization of ETFs
- FRA, Prof. Dr. Ralf Korn, M18, 4 months, for training in portfolio optimization in the presence of sustainability scenarios