Abstract

Financial data, related to companies listed on the Tunisian Stock Exchange, were collected and analyzed according to the methodology applied in machine learning on over two different time periods. A particular interest was focused on the periods before and during the Covid-19 crisis. The results obtained in this paper show, on the one hand, that an empirical diversification based on unsupervised learning algorithms is possible and on the other hand, a good coherence with the corporates financial state in Tunisia. This paper shows, for instance, that the k-means algorithm makes it possible to segment companies according to several criteria and to discover the aberrant behavior of certain companies with an abnormal financial situation. These results were confirmed by other outlier detection algorithms.

Keywords: Unsupervised learning, Stock market, Finance, CAPM (Capital Asset Pricing Model), Machine learning, Asset management