Why do developing countries tax imports so highly when the distortionary effects of tariffs are well-known and access to imported inputs can benefit economic growth (Koren and Csillag 2016, Goldberg et al. 2010)? One possibility is that, in low state capacity contexts, it is easier to enforce tariffs and hence generate revenue at the border than it is to enforce theoretically more appealing tax instruments such as a Value-Added Tax (Emran and Stiglitz 2005).
The primary objective of this project is to test this hypothesis using transactions-level domestic and trade data, to compare the extent of misreporting in Customs vs. domestic VAT. This work incorporates the implementation of a machine learning text algorithm developed to assign good codes to all business-to-business transactions (building on the methodology in Best, Hjort and Szakonyi 2017). Indeed, firms’ incentive and ability to evade taxes on a given transaction depend on the goods they trade.
The categorising of goods will further allow testing the theoretical prediction that VAT misreporting varies with the location in the production network (Hoseini 2015). The second additional benefit of coding good types in our data is that it will allow us to construct the first firm-level input-output map of a developing country’s economy. Such a map will help inform an active ongoing research agenda investigating the shape of the networks that make up countries’ economies and their consequences.
One goal is to improve the fiscal capacity of the Uganda Revenue Authority and therefore help reduce the tax compliance gap; another is to provide policymakers in Uganda (and similar countries) with a map of economic activity in their economy.