Network Analysis of Bilateral Trade Data Under Asymmetry

Trade statistics is a vivid example of bilateral data with asymmetry. Exporter and importer report their own versions of a flow between them, that frequently differ in dozens of times. In order to construct a network of trade relations we need to choose only one value for each weighted edge. We propose new methodology that aims to deal with the problem of mirror statistics. Our approach includes outliers detection step, the analysis of National Compilation and Reporting Practices survey and the construction of coherence metric of trade. We apply the proposed approach to real global trade data and compare several network statistics with corresponding statistics of networks that are constructed on export and import data. The key advantage of our methodology is that it does not depend on commodity selection and can be applied to trade networks on various levels.