Global value chains' position and value capture: Firm evidence in agri-food industry

Value creation forms the basis for the construction of global value chains (GVCs) and has received significant scholarly attention, yet the issue of value capture or power distribution along supply chains, "within" industries, is still unresolved. A recent framework of property rights (Antràs and Chor, 2013; Alfaro et al., 2019) highlights how final firms exert power over their suppliers to optimally organize their sequential production process. In such an environment, how can suppliers act strategically to counterbalance the power of the final firm? We contribute, theoretically and empirically, to a better understanding of the extent to which the division of surplus in the agri-food sector is affected by suppliers' positioning in GVCs. We argue that: (1) further upstream position of exports and further downstream position of imports, and consequently specialization of the production process along agri-food GVCs increase the bargaining power of suppliers; (2) these effects are more pronounced in more upstream position of suppliers production process in GVCs; (3) suppliers that specialize in the most downstream stages increase their bargaining power by importing further upstream and exporting further downstream, and thus performing more stages in GVCs; and (4) the mechanism is that the effects observed on surplus, whether in terms of upstream or downstream position of the production process, are mainly due to the upgrading of the product mix. Using the matched French Customs-AMADEUS 2002-2017 data, we build on the bilateral stochastic frontier model to measure the two-sided division of surplus of upstream/midstream suppliers and their export destination markets. We link this dataset to the U.S. inputoutput table converted to the NACE Rev.2 level, which identifies agri-food industries at a very detailed level, and compute upstreamness indicators for each industry and firms' exports and imports, following recent approaches in the literature. Our hypotheses (1), (2) and (4) are strongly supported, while hypothesis (3) is only weakly supported.