DETECTING CROSS-BORDER TRANSACTION PATTERNS USING MACHINE LEARNING: THE CASE OF INDONESIA

Abstract

We explore the intercompany cross-border trade involving Indonesian entities between 2018 and 2022 to identify to assess the risk of undervaluing in the export/import of goods with machine learning algorithms. Utilizing both micro and macro-level data, this study uncovers unique intercompany cross-border patterns, exposing a higher risk of undervaluing in export/import activities between Indonesian entities and their counterparts overseas. Our findings indicate the following: 1) Foreign investment-based companies are more inclined to engage in cross-border trade compared to domestic-based companies, posing higher risks, 2) Intercompany cross-border trade aligns with the extended gravity model, 3) Certain countries exhibit similar patterns in affiliation transactions, each pattern potentially associated with specific risk profiles and 4) Higher risk is evident in the number of trade transactions with low-tax jurisdictions. These findings offer valuable insights for identifying the potential risk of undervaluing trade. Consequently, this can benefit Customs by using the findings as risk indicators to improve its cross-border management, promote trade fairness, and optimize state revenue.