Digital divide and bilateral trade in Africa

Abstract

Abstract This paper contributes to the literature on the discussed effects of digitization on bilateral trade in African countries over the period 1995-2021. We proceed in two steps. In the first, we calculate, following an absolute and bilateral approach, original digital divide indicators relating to fixed phone, cell phone, internet and broadband. In a second step, we link the calculated indicators to measures of bilateral trade (bilateral exports, bilateral imports and total bilateral trade). Given the characteristics of the data, we apply the Poisson Pseudo Maximum Likelihood (PPML) estimator. Subsequently, we deepen the empirical investigations by disaggregating the effect of the digital divide on bilateral exports and imports, estimating the effect of the bilateral digital divide (including the sign of the gap) on bilateral trade, further controlling for omission bias despite accounting for multilateral resistances, using several estimators competing with PPML, further controlling for fixed effects and multilateral resistances with Poisson Pseudo-Maximum Likelihood Regression with High Dimensional Fixed Effects, controlling for endogeneity with IV-Poisson regression, then by addressing the issue of uncertainty in the model using several Bayesian techniques. Taken together, these techniques led us to the key finding that the digital divide is a limiting factor for bilateral trade in Africa. In other words, the distance of countries from complete digital coverage or digital inequalities between countries, is an obstacle to bilateral trade in Africa. This is due to high transaction costs, complex and cumbersome administrative and commercial procedures between countries, and poor digital and physical connectivity between countries, among other things. These different situations tend to lengthen the distance between countries, reinforcing the border effect, all of which contribute to reducing the intensity of trade between countries. The various sensitivity and robustness tests corroborate this main result.