Empirical Specifications and Data

food exports to the United States increased at the same pace, but from a much lower base (US$350 million in 2000, less than a quarter of the value of exports to the EU).

Similarly, ECOWAS countries’ exports of chemicals to the EU increased from US$100 million in 2000 to US$810 million in 2013 (figure 3.3, panel c), and their exports of plastics to the EU increased from US$110 million in 2000 to US$910 million in 2011 (figure 3.3, panel d), posting average annual growth rates of 17 percent and 22 percent, respectively.

ECOWAS exports of vehicles and automobile parts to the EU increased from US$350 million in 2000 to US$1.5 billion in 2007, before plunging to less than US$150 million by 2012 (figure 3.3, panel d). However, ECOWAS exports of apparel to the EU consistently decreased, from US$210 million in 2000 to US$67 million in 2015 (an average annual negative growth rate of 6 percent) (figure 3.3, panel f).

Nonmineral exports to the US, by sector. By contrast, ECOWAS exports of these categories of goods to the United States have been very low, with less-marked trends except for food and plastic products (figure 3.3, panels b and d). ECOWAS exports of food to the United States increased from US$350 million in 2000 (less than one-fourth the exports to the EU) to US$1.2 billion in 2015 (one-fifth the exports to the EU), posting an average annual growth rate of 9 percent. ECOWAS exports of plastics to the United States increased from US$50 million in 2000 to a maximum of US$360 million in 2011, before declining and stabilizing at around US$100 million in 2015.

The rest of this chapter assesses whether these differences (in volume and composition) in ECOWAS exports to the EU and the US result from differences in the design and implementation of the AGOA and EBA. The chapter also makes inferences about the potential impacts on ECOWAS countries of redesigning these two preferential trade agreements.

Empirical Specifications and Data

Following Santos Silva and Tenreyro (2006), we use the Poisson pseudomaximum likelihood estimation approach to account for the heteroscedasticity of bilateral trade flows as well as zero trade flows. The basic equation to be estimated for each year t is

Export lnX AGOAijpt ijt ijpt kt k , ,∑ α β ( ) = + ijpt +

∈ k nTT , { } EBA FE FEt ijpt i j ijptγ ε + + +

, (3.1)

where Xijt is a vector of gravity variables; nT and T are nontextile and textile products, respectively; AGOAnT,ijpt takes the value 1 only for year t when the AGOA is in effect in country i and covers nontextile product p exported to the United States; AGOAT,ijpt takes the value 1 only for year t when the AGOA is in effect in country i and covers textile product

p exported to the United States; EBAijpt takes the value 1 only for year t when the EBA is in effect in country i and covers product p exported to any of the 28 EU member countries; FEi is the full set of reporter fixed effects; and FEj is the full set of partner fixed effects. In the empirical assessment, we also interact the AGOA and EBA variables with a dummy variable specifying West African countries, to single out the impact of the two preferential agreements on this subregion.

Given the difference in country eligibility between the AGOA and EBA— with the EBA covering only LDCs and the AGOA covering any country that is approved by the United States—it is important to be able to assess any differential treatment of countries by both preferential agreements. We focus on West African countries and compare the trade impact estimated for all the AGOA and EBA beneficiaries with that estimated only for West African countries. The final equation used for the empirical assessment of the trade impacts of the AGOA and EBA is the following:

_ Export lnX West Africa AGOAijpt ijt ijpt kt k , ,∑ α β ( ) = + × × ijpt West Africa EBA FE FE

∈ k nTT , { } ijpt i j ijpt _ ,t γ ε+ × × + + + (3.2)

where West_Africa is a dummy variable taking the value 1 only if country i is a West African country. This formulation allows us to focus on ECOWAS countries and assess the differentiated trade impacts of the AGOA and EBA on them.

We use the approach suggested by Anderson and van Wincoop (2003) to account for multilateral trade resistance, focusing on all the bilateral flows between Sub-Saharan African countries, non-African LDCs, and the EU and the US to estimate the model.3 We therefore have 91 partner countries for each of the 92 reporter countries.

Because AGOA product eligibility is granted at a fairly disaggregated (Harmonized System [HS] 6-digit) level, we use disaggregated export flows to assess the trade impacts of the AGOA and the EBA. To reduce the size of the data set, we use the 4-digit (1,241 products) instead of the 6-digit (more than 5,000 products) disaggregation level. To ensure the “squareness” of the data set for product coverage, we complete it as needed with zero trade flows for any 4-digit product exported at least once by any of the reporter countries to any of the partner countries during 2001–15.

Finally, to deal with missing trade flows, we use the Database for International Trade Analysis (BACI), a unique data set of harmonized trade flows.4 The data set was initially constructed by Gaulier and Zignago (2010), using United Nations Comtrade data, and is regularly updated by the Centre for Prospective Studies and International Information (CEPII). CEPII also provides a full set of traditional gravity variables (such as bilateral distance, contiguity, common language, and common colonizer), which was first used in Head, Mayer, and Ries (2010).