The contrast of fraud in international trade is a crucial task of economic regulations within the European Union. The volumes involved are huge and under-reporting of traded values leads to significant depletion of the resources available for the EU and its Member States. We show two complementary approaches to fraud detection in this domain, relying on different assumptions about the fraud generation process. The first and more established path is based on robust statistical methods for linear regression and clustering. The second emerging approach is based on testing conformance of transactions digits to Benford’s law. For the latter, we also address the problem of reducing the false-positive rate and to develop corrected goodness-of-fit tests when Benford’s law does not hold.
This talk describes joint work with several people, including Lucio Barabesi, Domenico Perrotta, Marco Riani and the research staff at the Joint Research Centre of the European Commission (JRC). The motivation of this research line comes from a longstanding collaboration of the JRC with the Anti-fraud Office of the European Union (OLAF), supported by the HERCULE Programme of the European Commission