A Statistical Approache to Fraud Detection in External Trade

Abstract

Although detection of fraud in trade normally requires extensive on the spot investigations, it is very often the case that early stage investigative steps are based on specific trade data analyses. This is based on the fact that import/export fraud mechanisms always affect, to a certain extend, the patterns of related trade data series. In this paper, based on the availability of such import/export trade data series, we give an outline of a statistical approach for fraud modeling with respect to a specific fraud scheme. The basic idea is to separate the available trade data series into a historical data part and a present data part, and then use the former part to predict the latter part and to compute the corresponding forecast errors. Then, by means of a clustering method, trade flows with similar patterns are grouped together, and outliers, as well as clusters matching the fraud scheme pattern under consideration, are identified