Identifying suspicious groups of affiliated-transaction-based tax evasion in big data

Abstract Affiliated-transaction-based tax evasion (ATTE) is a new strategy in tax evasion that is carried out via legal-like transactions between a group of companies that have heterogeneous, complex and covert interactive relationships to evade taxes. Existing studies cannot effectively detect ATTE behaviors since (i) they perform well only for determining the abnormal financial status of individuals and ineffectively address the interactive relationships among companies, (ii) they aim at detecting ATTE from the perspective of structural characteristics, which leads to a poor false-positive rate, and (iii) few of them perform well in most sectors of companies. Effectively detecting suspicious groups according to both structural characteristics of ATTE groups and business characteristics of ATTE means (BC-ATTEM) remains an open issue. In this paper, we propose an affiliated-parties interest-related network (APIRN) for modeling affiliated parties, interest-related relationships, and their properties for identifying ATTE. Then, we identify the behavioral patterns of ATTE via topological pattern abstraction from APIRN and theoretical inference of BC-ATTEM. Based on the above, we further propose a hybrid method, namely, 3TI, for identifying ATTE suspicious groups via three steps: tax rate differential detection, topological pattern matching and tax burden abnormality identification. Experimental tests that are based on two years of real-world tax data from a province in China demonstrate that 3TI can identify ATTE suspicious groups with higher accuracy and better generality than existing works. Moreover, we identify various interesting implications and provide useful guidance for ATTE inspection based on an analysis of our experimental results.