Predicting Trade Mispricing: A Gaussian Multivariate Anomaly Detection Model

Abstract

Objective - This paper predicts a measurement indicator for the trade mispricing channel and its effectiveness in identifying IFFs. Methodology – A model gaussian multivariate anomaly detection algorithm, for classifying between legal and illegal transactions that are suspicious in terms of misreporting was developed. The method is a machine learning technique and uses data from South Africa, Botswana, the USA, and China over a period from 2000 to 2019, to learn whether there are any intriguing differences in the model performance based on these countries and the effect of other factors. Imports and Exports are used as features of the model while the net flow derived from these features is used as the third feature of the model. Imports and exports data are sourced from IMF’s Direction of Trade Statistics database. Annual tariffs data and corruption data come from the WDI database and Transparency International’s Corruption Perception Index, respectively. Data for ‘accounting and auditing standards’ comes from the world economic forum. Findings - The result showed that while the model may be effective in detecting IFFs due to mispricing, other factors may however contribute to irregularities of trading data that is flagged as IFFs. This in addition to accounting for total quantum, also provides details empowering governments with the information to stimulate and drive the desire to curb IFFs from its different sources and channels. Novelty - This study contributes to the debate on trade mispricing by proving a baseline measurement to help detect and track IFFs. Type of Paper: Empirical JEL Classification: F17, Q02 Keywords: Gaussian Multivariate Anomaly Detection; GMAD; Illicit Financial Flow; IFF., Trade Mispricing; Reference to this paper should be made as follows: Opeyemi, O.I; Mendon, D; Lenhle, D. (2022). Predicting Trade Mispricing: A Gaussian Multivariate Anomaly Detection Model, J. Bus. Econ. Review, 7(1), 61–74. https://doi.org/10.35609/jber.2022.7.1(2)