Artifical intelligence for credit risk model, or how do machine learning algorithms compare to traditional models?

Rajka, László and Pollák, Zoltán (2024) Artifical intelligence for credit risk model, or how do machine learning algorithms compare to traditional models? ECONOMY AND FINANCE: ENGLISH-LANGUAGE EDITION OF GAZDASÁG ÉS PÉNZÜGY, 11 (3). pp. 232-257. ISSN 2415-9379

Preview

Text 232-257RajkaPollakE.pdf - Published Version Download (1MB) | Preview

Abstract

A new generation of credit risk management models has surfaced as a result of the technology revolution marked with artificial intelligence, which in short is a term for models based on machine learning. Expert systems represented the past in the development of credit risk models over some decades, while traditional statistical models, e.g., logistic regression are the present and machine learning methods are expected to be the future. The objective of this study is to describe and empirically analyse the classification algorithm XGBoost, one of the most promising examples of the latter machine learning models to reveal the degree of increase in efficiency machine learning algorithms can achieve compared to the traditional modelling methods currently regarded to be industrial best prac- tice. In our study, both Artificial Neural Network (ANN) and XGBoost, models relying on artificial intelligence, have surpassed logistic regression in terms of efficiency of classification. Although machine learning methods have an excel- lent capability of prediction, the interpretation of decision-making models they offer is quite cumbersome compared to their traditional peers, which is a disad- vantage. Because of the “black box nature” of machine learning methods based on artificial intelligence, banks are currently limited regarding their application. Therefore, the authors propose the current rules and guidelines corresponding to the traditional models should be reviewed so as to give way to banks for the ap- plication of machine learning models and, as a result, to improve the efficiency of their credit risk management.

Actions (login required)

Edit Item