Money laundering and the detection of bad companies: a machine learning approach for the risk assessment of opaque ownership structures

The aim of the present study is to test the relationship between corporate secrecy and financial crime on a large sample of firms, and to develop predictive models with which to improve the early detection of high-risk firms. To this end, we develop a set of secrecy indicators based on business ownership data collected on a sample of more than 2.6 million firms registered in eight European countries. To validate the indicators, we implement several machine learning models that are trained, validated, and tested against evidence of sanctions and enforcementcrimes committed by firms and/or their owners. Results show that the proposed risk indicators have a strong predictive power. Firms with (i) more complex structures, (ii) owners from high-risk jurisdictions, and (iii) links to opaque corporate vehicles are more likely to engage in illicit activities. The inclusion of macro-level information, such as geographic location and business sector, significantly improves the understanding of the phenomenon. These findings advance the empirical knowledge on corporate and financial crime, and provides anti-money laundering supervisory and investigative authorities with new risk assessment tools based on machine learning technologies.