Tracking down financial statement fraud by analyzing the supplier-customer relationship network. (April 2023)

• Record Type: Journal Article
• Authors: Li, Jianping; Chang, Yanpeng; Wang, Yinghui; Zhu, Xiaoqian
• Is Part Of: Computers & industrial engineering. Volume 178(2023)
• Publication Date: 2023-04
• Subjects: Financial statement fraud -- Supplier-customer relationship -- Supply chain network -- Knowledge graph -- Graph neural network; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2023.109118
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: This study introduces supplier-customer relationships to financial fraud detection. Knowledge graph is applied to aggregate companies' complex supply-customer relationships. GNN methods are employed to analyze the supply-customer knowledge graph. Supplier-customer relationship information can significantly improve fraud detection accuracy. Abstract: The supplier-customer relationships in the supply chain reflect the transaction activities between companies, which can also imply the relationships across the financial data disclosed in companies' financial statements, thus helping to discover financial statement fraud. However, few studies can systematically analyze these complex relationships in the entire supply chain network and apply them to financial statement fraud detection studies. This paper introduces the supplier-customer relationships between companies to improve the accuracy of financial statement fraud detection. Based on the suppliers and customers data of Chinese listed companies, a supplier-customer knowledge graph is constructed to aggregate companies' complex supply-customer relationships in the whole supply chain network, and the graph neural network (GNN) methods are innovatively applied to analyze companies' financial data and their relationships in the graph to detect financial statement fraud. The empirical results indicate that the accuracy of GNN methods considering the supplier-customer relationships is significantly improved than the common machine learning methods. The AUC of the Heterogeneous Graph Transformer (HGT) method achieves 85.10%, which improves 5.19% over the result of the best-performing machine learning method. Furthermore, the results of fraud detection for different years using different periods of historical supplier-customer relationships are all improved, which shows the robustness of this study. This paper demonstrates the effectiveness of introducing supplier-customer relationships in financial statement fraud detection, providing a new perspective for regulators, investors, and researchers in future anti-fraud practices.