Identifying tax evasion in shell companies and fraudulent credits of Brazil’s state value-added tax (ICMS)

Abstract

Companies that issue tax documents to defraud the tax authorities with the transfer of credits of Brazil’s state value-added tax (ICMS) without the movement of goods cause financial losses to the government and, therefore, to society as a whole. Several initiatives to combat tax fraud have successfully used data analysis and Machine Learning techniques. This work sought to investigate the use of these techniques in identifying a specific practice of tax fraud, practiced by shell companies, formed exclusively to issue non-due ICMS credits, the tax on operations related to the circulation of goods, and the provision of interstate, intercity, and communication services. Based on document analysis and consultation with auditors and specialists, typologies and variables relevant to identifying tax evasion events carried out by shell companies were identified. Around these variables, data from the Finance Department of the Federal District were collected and prepared. With this data, it was possible to explore the use of predictive models based on Machine Learning capable of pointing out potentially fraudulent behavior. The good results obtained by these models demonstrate their potential as part of systematic monitoring and fiscal audits by tax authorities.

Keywords:
ICMS; machine learning; shell companies; tax evasion