APPLICATION OF MACHINE LEARNING FOR CLUSTER GROUPING OF PEER-TO-PEER COMPANIES IN INDUSTRY CLASSIFICATION

A data-based grouping system has been developed using machine learning tools to identify market perceptions and, in turn, group companies into clusters at various levels of detail. A permanent measure of similarity between companies has also been developed, which is used to group companies into clusters and form hedged portfolios. Groups of companies united in the same clusters had strong homogeneous risks and returns, while different clusters of companies demonstrated diverse and changing risk parameters. When analyzing clusters, it was found that companies grouped by the proposed method have a higher correlation of profitability outside the sample, but less stability and interpretability compared to companies grouped according to the standard industry classification. An interactive visual classification system has been developed based on machine learning to identify clusters and similar companies.