Drivers Of Export Performance In Africa: Evidence From Machine Learning Approach

Abstract

There has been discussion on what factors contribute to Africa's export performance for many years. The factors that are most important for Africa’s export performance are still unclear, nonetheless, given that prior research has focused on the preferential selection of covariates within the framework of many potential drivers of export performance. The fundamental issue with these contributions is that, depending on certain model assumptions and specifications, even shaky variables can be considered significant. To deal with this and properly inform policy, the study conducted a systematic review on drivers of export performance from 1980 to 2021. The study then employed four machine learning regularization algorithms approaches, namely standard lasso, minimum BIC lasso, square root lasso and the elasticnet, to analyze trends in a dataset with 87 covariates and determine the main factors influencing Africa's export. The findings indicated that, 7 factors (public private partnership investment, net trade, service traded, stock traded, domestic credit, inflation, gasoline price) were the actual drivers of export performance in Africa. The inferential estimates were obtained by applying double-selection Lasso, partialing-out lasso instrumental variable regression, and partialing-out lasso linear regression. Policy recommendations are also provided to inform policy appropriately.