The World Trade Web: using network analysis and machine learning as tools for public policy decision-making

Abstract

The World Trade Web (WTW) contains a wealth of information that upon rigorous analysis can aid governments in public policy decision-making. In my attempt to provide this valuable input, this dissertation uses two main methods: weighted network analysis and machine learning. First, the topology of the WTW is explored, described, and analyzed. Secondly, the relationship between countries’ trade network characteristics and their income is modeled. Lastly, deep learning is used to predict trade interactions between countries using quantitative, dyadic binary, and categorical variables. Insightful remarks are obtained: countries with higher PCGDP tend to associate with more neighbors that are themselves weaker, reciprocate fewer of their trade links, and trade more strongly with countries that are themselves stronger, and have a higher export to GDP Ratio. The improved trade forecasting model obtained can result in better GDP forecasts, which can aid with the optimization of tariffs, quotas, and subsidies.