Artificial intelligence is expected to become a fundamental part of any business operations, and it is difficult to envisage an industry that will remain unaffected. The financial sector revealed to be interesting because of the several opportunities coming from artificial intelligence applications, especially those related to machine learning methodologies. The machine learning domain of artificial intelligence is devoted to the development of methods for emulating human learning by avoiding explicit programming of computers and instead allowing them to continuously learn from data, which is a critical component of this subject. Industry and academic research are currently examining all possible applications in banking and finance: it turns out that the range of tasks that can be performed more efficiently with these new technologies is extremely broad, encompassing nearly all operations, from customer experience enhancement to more effective management and compliance. The aim of this paper is to identify the main determinants of a bank default employing different machine learning algorithms, so trying to understand the weight of different variables in terms of impact on it. Machine learning algorithms have never been used to study bank default. In particular, an element of originality of our study refers to thefact that among the algorithms that will be employed there is a graph neural network (GNN) that has never been employed in a financial context. A further element of originality deals with the inclusion of competition among the possible default determinants. The dataset consists of all the banks, active or dead, in the Euro Area in the period 2018-2020 for which it was possible to obtain from Bank Focus (Bureau Van Dijk) the values of all the accounting variables needed in the models. The results are useful both from a micro and macro-economic point of view: as for the first, they could help the bank managers to optimise business strategies focusing on the business areas that reveal to have a strong impact on the default probability; for the latter, they may be useful to give financial authorities some indications about the variables that should be carefully monitored, for example through the stress tests, to assess the stability of the entire banking system.
Lagasio, V., Pampurini, F., Grazia Quaranta, A., Pezzola, A., Predicting Bank Default with Machine Learning, 39th EBES CONFERENCE - ROME. PROGRAM AND ABSTRACT BOOK, EBES Publications, Instambul 2022: 91-92 [https://hdl.handle.net/10807/218566]
Predicting Bank Default with Machine Learning
Pampurini, Francesca;
2022
Abstract
Artificial intelligence is expected to become a fundamental part of any business operations, and it is difficult to envisage an industry that will remain unaffected. The financial sector revealed to be interesting because of the several opportunities coming from artificial intelligence applications, especially those related to machine learning methodologies. The machine learning domain of artificial intelligence is devoted to the development of methods for emulating human learning by avoiding explicit programming of computers and instead allowing them to continuously learn from data, which is a critical component of this subject. Industry and academic research are currently examining all possible applications in banking and finance: it turns out that the range of tasks that can be performed more efficiently with these new technologies is extremely broad, encompassing nearly all operations, from customer experience enhancement to more effective management and compliance. The aim of this paper is to identify the main determinants of a bank default employing different machine learning algorithms, so trying to understand the weight of different variables in terms of impact on it. Machine learning algorithms have never been used to study bank default. In particular, an element of originality of our study refers to thefact that among the algorithms that will be employed there is a graph neural network (GNN) that has never been employed in a financial context. A further element of originality deals with the inclusion of competition among the possible default determinants. The dataset consists of all the banks, active or dead, in the Euro Area in the period 2018-2020 for which it was possible to obtain from Bank Focus (Bureau Van Dijk) the values of all the accounting variables needed in the models. The results are useful both from a micro and macro-economic point of view: as for the first, they could help the bank managers to optimise business strategies focusing on the business areas that reveal to have a strong impact on the default probability; for the latter, they may be useful to give financial authorities some indications about the variables that should be carefully monitored, for example through the stress tests, to assess the stability of the entire banking system.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.