Classifying companies that have registered headquarters and subsidiaries in OFCs using unsupervised learning techniques

Alexandra Georgiana SIMA
Gheorghe HURDUZEU
Stefan Alexandru IONESCU

Keywords

cluster analysis, OFC, k-means, financial indicators

Abstract

We proposed an improved methodology for estimating, prioritizing and evaluating the effects of the transfer of activities, assets and profits to offshore financial centers (OFCs) and shadow banking systems. We collected financial information for 2019 for approximately 16,000 companies, the raw data being used to build a set of financial rates that covered four main areas: liquidity, solvency (risk), activity and profitability. The data set comprises three broad categories of companies: companies with registered headquarters in an OFC, companies that have headquarters registered in classic jurisdictions but having subsidiaries in OFCs, companies registered in non-OFC jurisdictions and which have no connection with OFCs. We initially used hierarchical cluster analysis, an amalgamation method that starts from a number of clusters equal to the number of companies considered. We continued by using partitioning algorithms, superior in terms of performance. The k-means method provided the best results. We noticed a good separation of companies that do not carry out activities in OFCs, but the solvency indicators offer a better separation within the groups of companies that carry out activities in OFCs.