ESSAYS ON GLOBAL VALUE CHAINS AND AGRI-FOOD SYSTEMS TRANSFORMATION By Mohammed Beroud A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Agricultural, Food, and Resource Economics – Doctor of Philosophy Economics – Dual Major 2024 ABSTRACT The renewed attention to food trade can be attributed to recent disruptions in global agri- food supply chains. Therefore, research into the impact of globalization on domestic agri-food systems is of special interest not only to policymakers but also to researchers. In fact, the transformation of agri-food systems serves as a major catalyst for structural change—a crucial process of economic development. In this dissertation, I examine three key aspects of the transformation of agri-food systems in response to participation in global agri-food value chains, hereafter referred to as GVCs. Chapter 1 explores the impact that GVCs have on the distribution of value-added among the various segments of domestic agri-food value chains, as measured by food dollar expenditures. Chapter 2 delves into the effects of participation in GVCs on the distribution of labor income within domestic agri-food value chains, with a particular focus on the decline in farm labor’s share of income as a key feature of the transformation in agri-food systems. Finally, Chapter 3 addresses consumer issues at the downstream end of the value chains, investigating the impact of GVC participation on the long-term pass-through of international food price spikes to domestic consumer markets. The first essay, titled “How Does Participation in Global Value Chains Impact Agri-food Systems Transformation?”, examines the relationship between participation in global value chains and the transformation of agri-food systems, as measured by the contribution of each segment in the value chain to food dollar expenditures. The study analyzes a panel dataset of 61 countries and three distinct types of agri-food value chains, representing 90% of the global economy for the period 2005-2015. The empirical results suggest that higher participation in global agri-food production networks is associated with a decrease in the farm's share of total value-added. More interestingly, this study shows that globalization causes a change in the distribution of value-added among the so-called “hidden middle” industries, such as manufacturing, transportation, and retail. This finding is important as it challenges the conventional wisdom about expanding post-farm industries in the agri-food systems transformation narrative. In the second essay, titled “Global Agri-food Value Chains and Farm Labor's Share of Income”, I study the impact of participation in global agri-food value chains on the farm labor's share of income. Specifically, I examine the distribution of income between farm labor and other agricultural production factors such as capital as well as between farm and post-farm labor. Using nationally representative longitudinal data for the period 2005-2015, I find that the decline in the farm labor’s share of agricultural income is exclusively attributable to stronger backward linkages (connections a country has with its foreign suppliers of agri-food inputs). Conversely, the decline in the farm labor’s share of total labor income within the entire agri-food value chain is more associated with higher forward linkages (connections of a country to its foreign downstream partners). Surprisingly, the empirical results suggest that the effects on farm labor are spilling over from the global food and beverage value chain. This finding contrasts with the prevailing idea that attributes the distributional effects of globalization on farm labor directly to participation in the global agricultural value chain. The third essay, titled “Buffer or Conduit? Global Agri-food Value Chains and Food Price Transmission”, explores how participation in global agri-food value chains affects food price transmission. Using a two-step regression analysis, I first estimate country-specific long-run pass- throughs, which measure the extent to which changes in international food prices lead to changes in domestic consumer food prices. This study estimates the long-run pass-throughs for 173 countries and four episodes of international price spikes between 2000 and 2022. I then examine how participation in GVCs impacts the long-run pass-through estimates. I find that higher two- sided or mixed participation (more intermediate positioning in GVCs) decreases food price transmission, while pure backward and forward GVC related-trade (participation in later and earlier stages of production, respectively) increase the transmission of international food price spikes. This study challenges the dominant view that higher real integration into GVCs transmits global food price spikes to domestic markets. Specifically, trade policies should focus more on promoting intermediate positioning in GVCs to stabilize domestic food prices. This work is dedicated to my mother, my brother, and my fiancée Amina, whose unwavering support and faith in me have illuminated my path. iv ACKNOWLEDGEMENTS I extend my deepest gratitude to Professor Titus Awokuse for his unwavering patience and invaluable assistance throughout my journey at Michigan State University. His guidance has been a cornerstone of my academic development, and for that, I am profoundly thankful. I am equally grateful to the rest of my guidance committee: Professor Christian Ahlin, Professor Saweda Liverpool-Tasie, Professor Thomas Reardon, and Professor Jeffrey Wooldridge. Their invaluable input and constructive criticism have greatly contributed to enhancing the quality and depth of my research. My appreciation extends to all the professors I had the privilege of learning from in both the Agricultural, Food, & Resource Economics and Economics departments. Your teachings have had a profound impact on my intellectual growth and academic pursuits. Your dedication to imparting knowledge and fostering understanding has indelibly marked my academic journey. I am eternally grateful to Professor Rachid Doukkali, my former mentor back in the engineering school, for his belief in my potential and his support in initiating my doctoral studies at Michigan State University. His encouragement and guidance have been pivotal in reaching this milestone in my academic career. I am sincerely grateful to Mohammed VI Polytechnic University for funding my thesis, which has been essential in pursuing my research endeavors and achieving my academic goals. To all mentioned and unmentioned, who have contributed to my journey in any way, your support has been a guiding light towards this achievement. Thank you. v TABLE OF CONTENTS CHAPTER 1: HOW DOES PARTICIPATION IN GLOBAL VALUE CHAINS IMPACT AGRI- FOOD SYSTEMS TRANSFORMATION? ................................................................................... 1 REFERENCES .......................................................................................................................... 20 APPENDIX: TABLES AND FIGURES................................................................................... 24 CHAPTER 2: GLOBAL AGRI-FOOD VALUE CHAINS AND FARM LABOR'S SHARE OF INCOME ....................................................................................................................................... 37 REFERENCES .......................................................................................................................... 50 APPENDIX A: TABLES AND FIGURES ............................................................................... 53 APPENDIX B: ENDOGENEITY ISSUES............................................................................... 66 CHAPTER 3: BUFFER OR CONDUIT? GLOBAL AGRI-FOOD VALUE CHAINS AND FOOD PRICE TRANSMISSION ............................................................................................................. 68 REFERENCES .......................................................................................................................... 82 APPENDIX A: TABLES AND FIGURES ............................................................................... 85 APPENDIX B: ROBUSTNESS CHECK ................................................................................. 89 vi CHAPTER 1: HOW DOES PARTICIPATION IN GLOBAL VALUE CHAINS IMPACT AGRI-FOOD SYSTEMS TRANSFORMATION? 1 1.1 Introduction Globalization and its implications for the strength of interlinkages and interdependencies among countries matter for the quality of domestic agri-food value chains (AVCs). Recent developments in global health, politics, and markets emphasize the importance of global agri-food value chains (hereafter GVCs) for AVCs. Relevant examples include the disruptions in global commerce and food supply chains caused by COVID-19 and the Russo-Ukrainian war. These global events further highlight the need for a better understanding of how different aspects of globalization impact domestic AVCs across countries and along the various segments of the food supply chain. Reardon (2015) and Barrett et al. (2022) outline the three stages of agri-food systems transformation (traditional, transitional, and modern) and document how agri-food systems in low- income countries have been experiencing a rapid transformation in recent decades. They note that a hallmark of moving from the traditional to the modern stage of transformation is the decline in the farm share of consumer food dollar expenditure as an economy increases its participation in GVCs. The distributional effect of globalization is a topic of interest to both researchers and policy makers because the transformation of agri-food systems serves as a major catalyst for structural change—an essential driver of economic growth (Timmer 2009; Barrett et al. 2022). Despite several descriptive studies on the relationship between globalization and agri-food systems transformation, very little empirical analyses exist on this important phenomenon, as pointed out by Feyaerts, Van den Broeck and Maertens (2020) and Barrett et al. (2022). In this study, I examine the linkages between GVCs and AVCs with particular focus on its effects on the distribution of food dollar expenditures as a country engages more with the rest of the world. How does participation in GVCs across countries of different income levels impact the distribution of food dollar expenditures within AVCs? I hypothesize that more participation in GVCs would reduce the farm share of total value-added due to increased agricultural productivity, global knowledge spillover, foreign direct investments, and access to foreign markets. Consequently, the downstream segments of the AVC would expand with an increasingly dominant presence of supermarkets and fast-food chains. The contrasting hypothesis posits that participation in GVCs is likely to have more impact on the transportation and storage industries due to long- distance shipments and the perishable nature of agricultural and food products. Moreover, I expect that countries supplying raw materials and intermediate inputs may experience distinct effects 2 compared to countries involved in the final stage of production and distribution. These distinctions can be attributed to differences in market dynamics, power relations, and value capture along the global value chains. This study contributes to the literature in three ways. First, I present novel empirical evidence on the impact of participation in GVCs on agri-food systems transformation, contrasting the predominant focus on the impact of international trade on various segments of the AVC in the existing literature (see Gaigné and Gouel (2022) for a review). Traditional trade statistics, which mainly use gross imports and exports, attribute the total value of an imported product to the exporting country. A global production network, however, spans at least two countries, with each participating country contributing value at different production stages. This study exploits recent advancements that have introduced decomposition methodologies designed to quantify a country's contribution to the final value of food products, uncovering nuances that aggregate trade data might overlook (Koopman, Wang and Wei 2014; Borin and Mancini 2019)1. Such methodologies make it possible to analyze positioning of countries in the GVC by decomposing the GVC participation measure into forward and backward linkages2. Second, this paper contributes to a growing literature on the impacts of participation in GVCs on various economic outcomes such as agricultural productivity, food security, and food prices (Lim 2021; Lim and Kim 2022; Montalbano and Nenci 2022; Dalheimer, Bellemare and Lim 2023). In this study, I propose a value chain perspective to examine the impact of globalization on the agri-food systems. The general idea suggests that variations in the extent of participation in GVCs impact the distribution of value-added between the different segments of the AVC, thus driving the transformation of the agri-food system. Third, the existing literature mainly focuses on the farm segment of the value chain, neglecting the role of the intermediate industries—also known as the “hidden middle”—that are crucial to the functioning of the system (Reardon 2015; Yi et al. 2021). For this, this paper also examines the midstream industries such as food manufacturing, transportation and storage, and wholesale and retail trade. 1 Giunta, Montalbano and Nenci (2022) show that the available inter-country input–output data can be used to compensate for the scarcity of firm-level data for evidence-based GVC analyses. 2 Backward linkages refer to the import content of a country’s exports or the value of raw materials that a country imports from upstream sectors in other countries. This contrasts with forward linkages, which refer to the exports of output by downstream sectors to the importing country to be later re-exported. Therefore, backward linkages are equivalent to "upstream" participation and forward linkages are equivalent to "downstream" participation in GVCs. 3 I preview three key findings from this study. First, the empirical findings suggest that higher participation in GVCs reduces the farm and manufacturing shares of total value-added and that these effects are exclusively attributed to backward linkages. Second, the retail and food services sectors expand due to higher GVC participation, reflecting a shift towards service-oriented agrarian economies and bypassing the traditional manufacturing phase in structural change, as supported by Lim (2021). Third, poorer countries in the sample experience a greater marginal effect on the value-added distribution—in accordance with the “catch-up effect” principle in the theory of convergence. The findings align with the idea that farms get a smaller share of the value- added due to globalization, but also uncover surprising results that some midstream industries are actually shrinking rather than expanding. This paper challenges the prevailing narrative within the literature on agri-food systems transformation, which suggests that globalization invariably leads to the expansion of post-farm industries. The rest of this chapter is organized as follows. Section 1.2 provides a brief overview of studies on globalization and agri-food systems transformation. Section 1.3 presents the measurement methods and data used in this study. Section 1.4 discusses the analytical framework and some methodological issues. Section 1.5 presents the empirical findings and Section 1.6 contains the concluding remarks and implications. 1.2 Literature Review Trade is the most studied aspect of globalization. Research on the impact of global economic integration through trade on various components of the agri-food system often relies on empirical analyses and anecdotal evidence, as trade theory provides inconclusive predictions regarding this causal relationship. This section delves into the extensive body of research examining the multidimensional impacts of trade in agricultural and food products. Early research on agricultural trade focused on the welfare gains from liberalization. Studies such as Reimer and Li (2010) have examined the crop sector and found that the median country experiences small gains from trade, amounting to only 0.6% of GDP. In the case of Japan, Bernhofen and Brown (2005) estimate the gains from agricultural liberalization in the 1850s at around 9% of GDP. The review conducted by Costinot and Rodríguez-Clare (2014), and later confirmed by Farrokhi and Pellegrina (2023), suggests that these gains are much larger than previously estimated, particularly when models allow for multiple crops instead of one aggregate sector. Beyond agricultural trade itself, the agricultural sector also gains from trade in agricultural 4 inputs (e.g., fertilizers, machinery, planting materials, etc.) to enhance its productivity (McArthur and McCord 2017; Porteous 2020; Farrokhi and Pellegrina 2023). Another gain from international trade comes from the fact that agricultural production is more stable globally compared to individual national levels. In this regard, international trade redistributes food among countries and mitigates the risk of fluctuations in national food production due to the inter and intra-annual variability in climatic conditions and other natural hazards such as disasters and droughts (Chapoto and Jayne 2009; Burgess and Donaldson 2010; Ferguson and Gars 2020). Also, the redistributive effects of trade stabilize food prices at the country level, which is another benefit for food security (Jacks, O’rourke and Williamson 2011; Gouel and Jean 2015). Welfare gains from globalization are conditional on trade cost reductions. The seminal work of Donaldson (2018) provides insights into how railroad access significantly boosts district level agricultural real income in India. However, this positive effect diminishes by half when accounting for the district's predicted trade share, highlighting the role of trade as a key mechanism for railroads' economic impact. These trade costs are particularly higher in Africa due to its poor transport infrastructure (Porteous 2019), which greatly reduces the positive impacts of trade policy reforms (Atkin and Donaldson 2015). In sum, these studies underscore how trade costs reductions are crucial for the gains from trade in agriculture. However, trade costs reductions may result in unequal distributional effects across regions and between farmers and non-farmers (Porteous 2020; Sotelo 2020). The specifics of such distributional effects are understudied in the case of agri-food value chains (Gaigné and Gouel 2022). Moving beyond research on the agricultural sector, a thin body of literature presents empirical evidence for other segments of the value chain. The role of intermediaries in agri-food trade is of special interest because they account for a large share of consumers’ food expenditures. For example, post-farmgate industries represent at least 73% of the value chain in the global food system according to the data presented in Yi et al. (2021). Ahn, Khandelwal and Wei (2011) suggest that intermediaries with large business portfolios can leverage economies of scope to distribute multiple agri-food products efficiently. Consequently, intermediaries can enter and operate in foreign markets at lower costs compared to individual agricultural producers (Gaigné, Latouche and Turolla 2018). These efficiency gains have positive impacts on aggregate productivity in the food industry, mainly due to a shift in market share towards more efficient firms and the exit of less efficient ones (Ruan and Gopinath 2008; Olper, Pacca and Curzi 2014). 5 Amid recent global value chains disruptions, a burgeoning stream of literature explores the impacts of integration of agri-food sectors into GVCs on various economic outcomes. Studies by Lim (2021), Montalbano and Nenci (2022), and Dalheimer et al. (2023) provide insights into how participation in global agri-food production networks affects agricultural productivity, food prices, and structural change. The work of Lim (2021) examines how GVC participation impacts structural transformation. The author suggests that modern agrarian economies are bypassing the manufacturing sector due to higher participation in GVCs. Specifically, more participation in GVCs leads to increases in GDP and employment shares in the agriculture and services sectors, but these shares decrease in the manufacturing sector. Given the important role of agri-food systems transformation in driving structural change, the impact of globalization on agri-food systems warrants further exploration. While several studies have investigated the various impacts of GVCs, more research is needed to document the impact of globalization on the distribution of welfare gains along the entire AVC, as advocated by Gaigné and Gouel (2022). Previous studies have conceptually discussed that globalization is a key driver of agri-food systems transformation (Feyaerts et al. 2020; Barrett et al. 2022). Other than globalization, Yi et al. (2021) represents the sole study within the literature that studied the determinants of agri-food systems transformation measured by the farm share. This study identified three meta-drivers of farm share of consumers’ food dollar expenditure: agricultural productivity, urbanization, and economic growth. While these are important drivers of farm share investigated in the literature, the role of globalization has not been explicitly considered in an empirical study. 1.3 Measurement Methods and Data This section introduces the methods employed to measure participation in GVCs and the average payment from each food dollar expenditure that producers receive in each segment of the AVC. I also present the set of covariate variables used in the empirical analysis of the casual relationship between globalization and agri-food systems transformation. 1.3.1 Participation in GVCs Interest in trade statistics and supply and demand linkages between countries has increased in response to the growing global fragmentation of production processes. However, there is no agreement on how to measure participation in global value chains, as methodologies propose various decompositions of gross exports. Earlier papers face some challenges regarding how to 6 account for items that are recorded several times in a given gross trade flow due to the back-and- forth shipments that occur in a cross-national production process, how to disaggregate gross exports by commodity or sector, and how to treat direct and indirect linkages between countries (Koopman et al. 2014). To overcome these challenges, this paper adopts the framework of Borin and Mancini (2019) (see Figure 1.1 for an illustration of the decomposition method). In the context of global value chains, gross exports can be decomposed to retrieve the value of imported intermediate inputs in a country’s exports and the value of a country's exports that are later re-exported by a trading partner. The first step of the decomposition consists of obtaining the GVC related exports by excluding the domestic value-added of country 𝑖 absorbed directly by its trading partner 𝑗 without any further re-exportations (i.e., Directly Absorbed Value-Added in exports (𝐃𝐀𝐕𝐀𝐗𝑖𝑗)) from its gross exports to 𝑗 (𝐄𝑖𝑗) for sector 𝑠: 𝐆𝐕𝐂𝐗𝑖𝑗 𝑠 = 𝐄𝑖𝑗 𝑠 𝑠 − 𝐃𝐀𝐕𝐀𝐗𝑖𝑗 𝑠 For the two country case, GVC-participation is simply computed as the share of 𝐆𝐕𝐂𝐗𝑖𝑗 in total exports is given by: 𝐆𝐕𝐂𝑖𝑗 𝑠 = 𝑠 𝐆𝐕𝐂𝐗𝑖𝑗 𝑠 𝐄𝑖𝑗 . The GVC share is easily computed as 𝐆𝐕𝐂𝑖 𝑠 = 𝑠 ∑ 𝐆𝐕𝐂𝐗𝑖𝑗 𝑗≠𝑖 𝑠 ∑ 𝐄𝑖𝑗 𝑗≠𝑖 , The overall GVC participation indicator can be decomposed into backward and forward components: where and 𝐆𝐕𝐂𝑖𝑗 𝑠 = 𝐩𝐆𝐕𝐂Backward𝑖𝑗 𝑠 𝑠 + 𝐩𝐆𝐕𝐂Forward𝑖𝑗 𝐆𝐕𝐂Forward𝑖𝑗 𝑠 = 𝐃𝐕𝐀𝑖𝑗 𝑠 𝑠 − 𝐃𝐀𝐕𝐀𝐗𝑖𝑗 𝑠 𝐄𝑖𝑗 𝑠 𝐅𝐕𝐀𝑖𝑗 𝑠 𝐄𝑖𝑗 𝑠 is the ratio of the difference between value of exports of sector 𝑠 in country 𝐆𝐕𝐂Backward𝑖𝑗 𝑠 = 𝐆𝐕𝐂Forward𝑖𝑗 𝑖 to country 𝑗 that is created by domestic production factors and contributes to gross domestic 𝑠 is the ratio of the value of exports of 𝑠 . 𝐆𝐕𝐂Backward𝑖𝑗 product (DVA) and DAVAX, over 𝐄𝑖𝑗 sector 𝑠 in country 𝑖 to country 𝑗 that originates from imported inputs. Backward linkages refer to the connections a country or sector has with its suppliers. In the context of GVCs, these linkages 7   measure the value-added by foreign suppliers in the production process of a country's exports. Essentially, backward linkages capture the imported inputs used in the production of goods and services for export. High backward linkage in a country's exports indicates significant dependence on foreign inputs, which is crucial for policymakers and businesses to understand the sourcing patterns and the vulnerability of export sectors to global supply chain disruptions. Forward linkages, on the other hand, represent the extent to which a country or sector's output is used as an input by other countries in their production processes. In GVC terms, forward linkages measure the value-added by a country to the exports of other countries. It indicates the contribution of a country's exports to the value-added in the production of downstream products abroad. A strong forward linkage suggests that a country is an important supplier in global production chains, providing intermediate goods that are essential for the manufacture of final food products in other countries. For country 𝑖’s participation in global networks for sectors 𝑠 = {Agriculture, Fishing, Food & Beverage}, the overall GVC-related indicator is written as follows: 𝐆𝐕𝐂𝑖 = 𝑗≠𝑖 𝑠 ∑ ∑ 𝐆𝐕𝐂𝐗𝑖𝑗 𝑠 𝑠 ∑ ∑ 𝐄𝑖𝑗 𝑗≠𝑖 𝑠 = 𝑠 ∑ ∑ 𝐃𝐕𝐀𝑖𝑗 𝑗≠𝑖 𝑠 𝑠 − ∑ ∑ 𝐃𝐀𝐕𝐀𝐗𝑖𝑗 𝑠 𝑗≠𝑖 𝑠 ∑ ∑ 𝐄𝑖𝑗 𝑗≠𝑖 𝑠 + 𝑠 ∑ ∑ 𝐅𝐕𝐀𝑖𝑗 𝑗≠𝑖 𝑠 𝑠 ∑ ∑ 𝐄𝑖𝑗 𝑗≠𝑖 𝑠 . This study uses the Inter-Country Input-Output (ICIO) tables from the UNCTAD-Eora Global Value Chain Database to measure country-level GVC participation for 61 countries from 2005 to 2015. Descriptive statistics in Table 1.1 present the GVC share by sector and type of linkage. The 'Agriculture' sector has a GVC share just above the mean of 0.377, while the 'Fishing' sector aligns with it, and the 'Food and Beverage' sector is slightly below. The data shows noteworthy variations in GVC shares across sectors and countries, as can be seen from the standard deviation. Figure 1.2 depicts the levels of participation in GVCs across countries. Interestingly, Luxembourg has the highest total GVC share and backward participation indicators, while Romania leads in forward participation. Conversely, Kazakhstan, Indonesia, and Hong Kong have the lowest GVC indicators. In terms of regional analysis, Europe appears to have the highest values in all three categories. On the other hand, Asia and Africa register the lowest values of participation in GVCs. To draw a comparison, some high-income countries such as the United States record moderate scores across all three sectors and indicators. 8   1.3.2 Food Dollar Expenditures The Economic Research Service (Canning 2011) pioneered the use of Input-Output (IO) analysis to measure the average food dollar expenditures and the industry share of those expenditures in the value chain (see Figure 1.3 for an illustration). The main contribution compared to the old ERS Marketing Bill estimation involves netting out industry-to-industry direct and indirect payments that do not contribute to the final market value of the food dollar. A detailed mathematical derivation can be found in Canning (2011) and an extension of the computational method to multiple countries is presented in Yi et al. (2021). The resulting expression of interest from the standard computational method is the contribution of industry 𝑖 to the food dollar expenditure, denoted by 𝑠𝑖 and illustrated as follows: 𝑠𝑖 = 𝑣𝑖 𝑦𝑖 . In this expression, 𝑣𝑖 is obtained from two key calculations. First, we compute the net gross industry output by netting out industry-to-industry sales and food imports. Then, we multiply it by a square diagonal matrix, where elements represent the value-added coefficients for all industries, as well as those of their subcontracting industries. The resulting matrix is divided industry-wise by the total food dollar expenditure, 𝑦𝑖. In sum, the ratio measures the average payment from each food dollar expenditure that producers in industry 𝑖 receive for their raw food dollar commodities, in a given country and year. Supply chain industries are clustered into major industry groups based on their contributions to the different stages of agri-food production and services. In this study, I consider five industry groups as a simple aggregation of industries that comprise the agri-food value chain and produce similar products, as defined by FAO3. Figure 1.4 depicts the industry groups along the agri-food value chain. The flowchart starts with "Primary Food Production" representing the farm level of food production, then moves to the post-farm activities which include "Manufacture of food," followed by "Transportation & Storage," and finally "Wholesale & Retail Trade" along with "Food Services." Each of these segments represent a set of agribusinesses, and the arrows between the boxes suggest the sequential relationship between these production stages. In a major advance, Yi et al. (2021) develop a standardized method to estimate the distribution of consumer food expenditures among the different segments of the value chain for a 3 Details on the industry groups can be found in the following FAOSTAT webpage: “Value shares by industry and primary factors (2022)”. Webpage is updated on November 28, 2022 and can be accessed through: www.fao.org/faostat/en/#data/GFDI. 9 set of countries. In dealing with data constraints and accounting conventions in IO tables, Yi et al. (2021) complement the ‘Food At Home’ (FAH) and ‘Food Away From Home’ (FAFH) type of value chains with the ‘Food & Tobacco At Home’ (FTAH) value chain. Identical calculations are performed for the three types of value chains. This paper uses the consumer food dollar expenditure series to quantify each industry's contribution to the total agri-food value-added using the OECD international database for Input- Output tables. Table 1.2 provides a breakdown of industry shares for the different types of value chains. In the FAH value chain, the data represents consumers preparing meals at home and purchasing directly from retail stores. This direct transaction with retailers explains why the average share for ‘Accommodation and Food Service’ activities is as low as 0.004. In contrast, the ‘Wholesale and Retail Trade’ segment experiences a higher mean share of 0.460. The variability in these shares is evidenced by standard deviations ranging between 0.003 to 0.058. The FTAH value chain is analogous to the FAH chain, with the primary distinction being its inclusion of tobacco. Despite this difference, the two chains exhibit similar patterns. However, in the FAAFH value chain, the ‘Accommodation and Food Service’ segment reaches a peak value of 0.682 due to consumers opting to dine away from home. 1.3.3 Covariate Variables This study accounts for several control variables in the empirical analysis of the relationship between GVC participation and the five industry shares, as measured by the food dollar expenditure series4. The inclusion of these covariates is based on their conceptual linkages to understanding how globalization impacts agri-food systems transformation. First, the level of national income is a significant factor in the transformation of agri-food systems for several reasons. Higher income levels can lead to increased demand for non-staple, higher-quality, and more processed food products, according to Bennett's Law (Bennett 1954). This demand stimulates the post-farm segments of the agri-food value chain, such as food retail and food service activities (Behrman and Deolalikar 1987; Bouis and Haddad 1992; Pingali 2007; Jensen and Miller 2008). Furthermore, higher wages at the household level increase the opportunity cost of time, which in turn impacts the demand for food away from home (Senauer, Sahn, and Alderman 1986; Ma et al. 2006). 4 Given the small sample size, this study follows Yi et al. (2021) in opting for a more compact specification, focusing only on the meta-drivers of the agri-food systems transformation. 10 Second, urbanization is directly linked to changes in agri-food systems. The theory of urbanization and migration highlights the role of transnational migration and rural-to-urban migration in shaping the spatial distribution of economic activities and food demand (Jensen and Miller 2008). As more people move to urban areas, the demand for food changes in terms of quantity, quality, and diversity, leading to the growth of post-farmgate segments of the value chain, driven by long-distance transportation, cold storage, and processing (Reardon and Timmer 2014; Tschirley et al. 2015). Third, agricultural productivity is crucial for the transformation of agri-food systems. The development of new crop varieties, such as dwarf wheat and semi-dwarf rice, and improved agricultural management practices, has been instrumental in shaping the structure of agri-food systems worldwide (Feder, Just, and Zilberman 1985). Precision agriculture, biotechnology, and genetics are among the innovations that have been the most significant contributors to internal changes within the agri-food systems in recent decades (Whitaker and Kolavalli 2006; Pardey et al. 2016; Aggarwal et al. 2018). Empirically, I include three time-varying country-level covariates in the analysis: income, measured as GDP per capita (PPP) in $US; urbanization, represented as the percentage of the total population; and agricultural productivity, defined by the ratio of Gross Production Value (constant 2004-2006 million US$) to Agricultural Land (1000 ha). Data for these variables are sourced from the World Bank's data catalogue and FAOSTAT. Table 1.3 provides descriptive statistics for control variables. Urbanization, measured as a percentage of the total population, has a mean of 73.47% with a standard deviation (SD) of 15.12, ranging from 19.17% to 100%. GDP per capita (PPP) averages at $31,947 with a SD of $17,685, and ranges from $1,969 to $97,864. Agricultural Productivity, measured in 1000 $US/ha, has a mean of 3.253 with a substantial SD of 7.938, and ranges from 0.0242 to 77.63. 1.4 Analytical Framework and Empirical Issues The empirical model is written in a simple panel regression framework: ′𝑋𝑖𝑡 + 𝛽3𝑡𝑦𝑝𝑒𝑖𝑡𝑚 + 𝜏𝑡 + 𝜖𝑖𝑡𝑚 s𝑖𝑡𝑚 = 𝛼0 + 𝛽𝜊ln (GVC)𝑖𝑡 + 𝛽1 where subscripts 𝑖, 𝑡, and 𝑚 denote country identifier, year, and type of the value chain, respectively. I run separate regressions for the five industry shares, s𝑖𝑡𝑚, measured as the average payment from each food dollar expenditure that producers in each industry receive for their raw food dollar commodities, in a given country and year. GVC𝑖𝑡 is participation in GVCs, which 11 measures the extent of cross-border value addition through backward and forward linkages in global agri-food production networks. 𝑋𝑖𝑡 is a vector of the meta-drivers of agri-food systems transformation and 𝑡𝑦𝑝𝑒𝑖𝑡𝑚 is a categorical variable that assigns values as follows: 1 for FAH, 2 for FTAH, and 3 for FAFH, with the FAH series serving as the baseline. Finally, 𝛿𝑖 and 𝜏𝑡 represents the country and year fixed effects, respectively. 𝜖𝑖𝑡𝑚 is an error term with mean zero. Year fixed effects are included to eliminate the correlation between the error term and the treatment variable that arises from constant factors across countries in a given year, such as the global financial crisis in 2008-2009. I exclude country fixed effects for several reasons. First, country fixed effects consume degrees of freedom, which might be a concern given the limited number of observations per country in the dataset. In the case where certain countries are underrepresented, including country fixed effects can lead to unreliable estimates. Second, this study’s interest is in examining global trends rather than country-specific effects. That is, the interactions between countries are an important aspect of the study, similar to Yi et al. (2021). Including country fixed effects might not capture these cross-country dynamics. To ensure that each country contributes equally to the overall estimate, despite the differences in the number of observations available for each type of the value chain, this study uses the following weighting scheme: i) if a country has only one estimate, we assign a weight of 1; ii) if a country has two estimates, each one is assigned a weight of 1/2; and iii) in the case of the three estimates, each one gets a weight of 1/3. In this manner, in each case, the total weight for each country sums up to 1. A linear modeling framework such as that in Yi et al. (2021) is not appropriate for continuous fractional response variables as it may generate fitted values outside the (0,1) range. Therefore, this paper employs the Panel Quasi-Maximum Likelihood (PQMLE) to estimate the fractional regression model (Papke and Wooldridge 1996). This approach allows capturing the true average partial effects that may diminish with increasing levels of participation in GVCs. I also include time averages of the covariates and year dummies to avoid large spurious estimates. 1.5 Results and Discussion 1.5.1 Baseline Estimation Table 1.4 reports the baseline estimation results where each panel represents a segment of the value chain for different specifications and estimators. The findings suggest that participation in GVCs has a significant and negative effect on the farm’s share of total value-added. Specifically, 12 a 10% increase in participation in GVCs reduces the farm share in the agri-food value chain by 0.003 share point (Panel A). These findings are in line with the agri-food systems transformation narrative and Yi et al. (2021), at least with respect to the declining share of value-added at the farm level. This could imply that other stages in the value chain, such as processing or distribution, capture a larger share of the value as a result of increased globalization. In the post-farm segments, however, participation in GVCs has a positive impact on food services, transportation, and storage industries of the value chain, but with varying slopes (see Panels B to E). Moreover, the estimate on manufacture of food, beverages, and tobacco products is not statistically significant at the 10% level. More interestingly, participation in GVCs has a negative and significant effect on the share of wholesale and retail trade sectors. This suggests that globalization causes a shift of resources within the so-called “hidden middle” industries of the value chain. Contrary to the common narrative that post-farm industries expand with increased globalization, this study finds that certain sectors within the "hidden middle" contract due to higher participation in GVCs. A possible interpretation of these results is that participation in GVCs involves long- distance shipment and storage leading to an increase in the share of the corresponding segments in the value chain. However, increased participation in GVCs may promote vertical integration and favor large international retailers, reducing the contribution of domestic wholesalers and retailers in the agri-food value chain—a mechanism that warrants further investigation in future research. Previous research partially explains the empirical findings presented in this section by considering that the intermediate industries enhance the efficiency and aggregate productivity of the AVC. These improvements occur through increased economies of scope and the reallocation of market share towards more efficient firms (Ruan and Gopinath 2008; Ahn, Khandelwal and Wei 2011; Olper, Pacca and Curzi 2014; Gaigné, Latouche and Turolla 2018). 1.5.2 Endogeneity Issues In this section, I employ an instrumental variable (IV) estimation approach to treat potential endogeneity issues in the previous analysis. Endogeneity can arise from omitted variable bias, measurement error, simultaneity, or reverse causality, leading to biased and inconsistent estimates. First, there may be unobserved factors that influence both the industry share in total value added and participation in GVCs such as technological advancements or industry-specific policies that could impact both variables. Second, food dollar series or participation in GVCs might be 13 difficult to measure accurately due to difficulties in tracing value addition across segments of the AVC or national borders. Third, there might be a bidirectional relationship between the industry share in the total value added and participation in GVCs. Industries with high added value might be more likely to participate in GVCs, but at the same time, participation in GVCs might allow industries to specialize and increase their share of value added. Industries might also adjust their strategies based on their performance in GVCs, leading to feedback loops where past participation influences future value-added shares and vice versa. Fourth, industries may not randomly decide to participate in GVCs. Instead, they might self-select into GVCs based on their expected benefits, which are often correlated with unobservable factors such as politics, national strategic priorities, or industry competitiveness. To overcome these issues, this paper uses the weighted sum of a country's geographical distances with all its trading partners as an IV, ∑ 𝑤𝑖𝑗𝑡𝑑𝑖𝑗 𝑗≠𝑖 , where the weights are bilateral import flows between country 𝑖 and 𝑗 in year 𝑡, and 𝑑𝑖𝑗 denotes distance between 𝑖 and 𝑗. Data to construct the IV comes from Conte, Cotterlaz and Mayer (2022). This instrument is likely to be relevant, as it is reasonable to assume that countries closer to their trading partners and with substantial trade flows are more likely to form a production network, mainly due to reduced transportation costs and enhanced trade opportunities—directly follows from the gravity model theory. Over time, countries might experience shifts in their primary trading partners, driven by geopolitical, technological, or resource-based factors. These shifts in trade patterns inherently validate the dynamic nature of the IV, as the changing bilateral import flows between countries over time reflect evolving production networks. Moreover, this instrument is also likely to satisfy the exogeneity assumption because the weighted geographical distance is determined by factors that are exogenous to the structure of the AVC. In examining the estimation Table 1.5, I find different estimates regarding the impact of GVC participation on the industry shares across the middle segments of the value chain. Differences between IV and baseline estimates suggest that the baseline model might be affected by endogeneity issues. IV estimates are more reliable because the F-statistics for the first stage regression (Table 1.6) are far above the commonly used rule-of-thumb threshold of 10, which suggests that the instrument is strong. Given the significant coefficients of the instrument and the high F-statistics, it can be concluded that the instrument is relevant in the first stage regression. The coefficient for residuals in the fractional probit regression is significant for all panels except 14 Panel A, which indicates that GVC participation is correlated with the error term in baseline regression. The findings reveal a consistent negative association between GVC participation and the shares of agriculture and manufacturing sectors, with the latter diverging significantly from the baseline estimate. Notably, a 10% increase in GVC participation decreases the share of the manufacturing in the AVC by 0.01 share point. Conversely, retail and services sectors exhibit positive growth from GVC participation of 0.004 and 0.008 respectively. The transportation & storage sector, although experiencing a slight negative impact, shows a less significant effect in comparison. These results suggest a realignment within economies towards sectors that either directly benefit from globalization or adapt to shifting comparative advantages, reflecting broader trends of economic transformation influenced by GVC participation. The significant negative impact of GVC participation on the manufacturing sector can be attributed to increases in output. However, more output may not proportionately increase the sector's share in the economy if there is even faster growth in other sectors, particularly services, that benefit more directly from globalization and technological change. This finding is in total agreement with Lim (2021), which finds that economies bypass the manufacturing phase to develop services directly due to higher integration into GVCs—a trend that diverges from the conventional trajectory of structural change. 1.5.3 Heterogeneity by Income In this section, I further estimate the average marginal effects of participation in GVCs on the outcome variables at different percentiles of income (or GDP per capita). The use of PQMLE enables leveraging the inherent non-linearities of the econometric model to avoid partitioning the dataset into subgroups of countries. This approach preserves the entire dataset and statistical power of the analysis. The empirical results from Table 1.7 suggest that the effect of participation in GVCs on various segments of the value chain differs across income percentiles. First, the magnitude of the effect in the farm segment tends to decrease as income rises, with the most significant effect observed at the lowest percentile (see column 1)—in accordance with the “catch-up effect” principle in the theory of convergence. In lower-income countries, agriculture often constitutes a larger share of the economy and employs a significant portion of the labor force in relatively low-productivity activities. As countries integrate into GVCs, there may be a reallocation of resources away from traditional agriculture towards more industrial and service-oriented sectors, which offer higher value addition and productivity gains. In higher- 15 income countries, agriculture tends to be more technologically advanced and capable of integrating into GVCs without substantial displacement of traditional activities, thus lessening the negative impact. Second, the manufacturing sector in column (2) shows a consistent negative association with GVC participation across all income levels, with the adverse effects tapering off in lower- income countries. In the manufacturing sector, GVC participation might lead to a shift towards more advanced manufacturing processes that are less labor-intensive but more capital and technology-intensive. Lower-income countries, with their comparative advantage in labor- intensive manufacturing, may see a reduction in the share of manufacturing as they lose ground to more cost-effective producers abroad. In contrast, higher-income countries can maintain or even enhance their manufacturing sector's value through specialization in high-tech, high-value-added manufacturing niches within GVCs. Third, the impact in the transportation and storage segment appears to increase with rising income, peaking at the highest percentile (column 3). The growing negative impact with income could reflect the evolving nature of logistics and transportation demands in higher-income economies. As these countries move up the value chain, their internal transportation and storage needs may become more sophisticated but also more efficient, reducing the overall share of this sector in the economy. Additionally, higher-income countries might outsource part of these services internationally, relying on global logistics networks rather than domestic infrastructure, which could explain the increased negative impact. Fourth, contrary to the trends observed in agriculture and manufacturing, the retail sector in column 4 benefits from higher GVC participation across all income percentiles, with a positive and statistically significant effect that remains relatively stable across income levels. This suggests that retail sectors universally gain from integration into GVCs, regardless of the country's income level. The positive impact across income levels may be attributed to the expansion of consumer markets and the diversification of consumer goods facilitated by GVCs. Retail sectors benefit from the increased availability and variety of imported goods, as well as from improved supply chain efficiencies. In both lower and higher-income countries, retail can thrive by catering to the growing demand for both basic and luxury imported goods, making it a sector that consistently benefits from global trade integration. Finally, the services sector in the last column also sees a universally positive impact from 16 GVC participation, with the magnitude of the effect slightly decreasing as income rises. The universally positive but decreasing impact as income rises could be due to the increasing importance of services in the global economy and the ease with which these services can be traded across borders thanks to digital technologies. Lower and middle-income countries may initially experience substantial gains as they tap into international markets for services, ranging from customer support to software development. However, as countries become wealthier, the services sector becomes more saturated, and the marginal gains from additional GVC participation may diminish. Furthermore, higher-income countries might already have a well-developed services sector that faces stiffer global competition, slightly reducing the benefits of further GVC integration. In sum, while the agriculture and manufacturing sectors generally face negative impacts that lessen with higher income levels, the retail and services sectors experience benefits from GVC participation across the board. The transportation and storage sector's negative impact grows with income, albeit it remains a relatively minor effect compared to the other sectors. Lower-income countries often specialize in the upstream activities of the GVC, whereas wealthier nations, which focus on transportation, storage, and food manufacturing, might not experience greater impacts on farming but would be more sensitive to changes in post-farmgate segments of the value chain. Lower-income countries may also lack good transportation and storage infrastructure, thus the relative impact of globalization on these segments might be less considerable due to smaller scales of operations and potentially lower reliance on advanced logistics systems. This could be an explanation of why countries experience small gains from trade in agricultural and food production on average (Bernhofen and Brown 2005; Reimer and Li 2010; Costinot and Rodríguez-Clare 2014), as discussed in Section 1.2 of this Chapter. 1.5.4 Heterogeneity by Region Table 1.8 shows the average partial effects of participation in GVCs on the industry shares across different regions. Each coefficient represents the change in the industry share associated with participation in GVCs for Asia, Europe, and Americas5. In Asia, participation in GVCs is associated with a significant decrease in the share of manufacturing, transportation & storage, and retail in the total value-added. Conversely, it significantly increases the services sector share. The agriculture sector sees a decrease, though not statistically significant. For Europe, GVC 5 Africa was excluded from the analysis due to a lack of sufficient data. 17 participation significantly reduces the share of the manufacturing and transportation & storage sectors, while significantly increasing the services sector share. The effects on the agriculture and retail sectors are positive but not statistically significant. In the Americas, GVC participation significantly increases the shares of all segments except for food services6. In sum, the table suggests that the impact of GVC participation varies significantly across regions and sectors. In Asia and Europe, there's a shift away from traditional sectors like agriculture, manufacturing, and transportation towards services. In contrast, the Americas show an overall positive impact across most sectors, except for a notable decrease in the food services sector. 1.5.5 Heterogeneity by Type of Linkage This section examines how positioning in GVCs affects the distribution of value-added across industries, by breaking down total GVC participation into forward and backward linkages, as follows: s𝑖𝑡𝑚 = 𝛼0 + 𝛽𝜊𝐹ln (GVCForward)𝑖𝑡 + 𝛽𝜊𝐵ln (GVCBackward)𝑖𝑡 + 𝛽1 ′𝑋𝑖𝑡 + 𝛽3𝑡𝑦𝑝𝑒𝑖𝑡𝑚 + 𝜏𝑡 + 𝜖𝑖𝑡𝑚 Four key findings emerge from table 1.9. The consistent impact of backward linkages across sectors underscores the role of downstream positioning in global agri-food production networks in transforming domestic agri-food systems. Higher participation in GVCs, particularly through importing intermediate goods and using them for further re-exportation, may displace domestic agricultural production. This can occur as countries import more cost-effective or higher- quality agricultural inputs or products, impacting domestic agriculture. For the manufacturing sector, more GVC integration may lead to domestic agri-businesses potentially favoring more specialized, higher-value-added foreign manufacturing activities at the expense of more basic domestic production processes. Similarly, the demand for transportation and storage may shift, reflecting changes in the nature and volume of trade flows, possibly emphasizing more efficient, integrated logistics solutions that do not proportionally benefit all domestic providers. Conversely, the positive impact from backward linkages on retail suggests that importing intermediate goods can enhance retail offerings, diversifying and elevating the quality of goods available to consumers. This reflects the benefits of downstream positioning in GVCs, where 6 When specifically including dummies for the US and Canada, the significant effects are more mixed and generally weaker. 18 access to a global supply chain enhances retail sector competitiveness and variety. Also, the services sector thrives from backward integration due to higher demand for food services related to changes in diets and lifestyles. As income increases, Bennett's Law suggests a shift in dietary composition towards higher-value foods, with individuals and families spending less time on food preparation at home. Consequently, there is a marked increase in consumption of food away from home, fueling the growth of food services such as fast-food chains. 1.6 Concluding Remarks Globalization changes the structure of agri-food systems in many ways. However, scant empirical studies exist on the causal relationship between participation in GVCs and agri-food systems transformation—an engine of structural change. The existing studies focus on the farm segment, which may be misleading or at best incomplete. This study offers a value chain perspective to examine the effects of increased participation in global agri-food networks on domestic value chains. I constructed a panel dataset from 61 countries and three distinct types of agri-food value chains derived from the food dollar expenditure series, representing 90% of the global economy. To estimate the impact of participation in GVCs, I employed the PQMLE to estimate the fractional regression model. This study also conducted an analysis of heterogeneity by income and type of linkage and addressed some endogeneity issues related to the relationship between participation in GVCs and agri-food systems transformation. Three key findings emerged from this study. First, participation in GVCs (particularly backward participation) has a significant and negative effect on the farm and manufacturing shares of total value-added. Second, the retail and food services sectors expand due to higher GVC participation, similar to the findings in Lim (2021). Third, poorer countries in the sample experience a greater marginal effect on the value-added distribution—in accordance with the “catch-up effect” principle in the theory of convergence. The empirical findings in this paper align with the idea that farms get a smaller share of the value-added due to globalization, but also uncover surprising results that some midstream industries are actually shrinking rather than expanding. This paper challenges the prevailing idea within the literature on agri-food systems transformation, which posits that globalization invariably leads to the expansion of post-farmgate industries. Contrary to this widespread narrative, the evidence presented herein demonstrates a more varied impact of globalization on these sectors. 19 REFERENCES Aggarwal, S., B. Giera, D. Jeong, J. Robinson, and A. Spearot. 2022. “Market access, trade costs, and technology adoption: Evidence from northern Tanzania.” Review of Economics and Statistics, 1-45. Ahn, J. Bin, A.K. Khandelwal, and S.J. Wei. 2011. “The role of intermediaries in facilitating trade.” Journal of International Economics 84(1):73–85. Atkin, D., and D. Donaldson. 2015. “Who’s Getting Globalized? The Size and Implications of Intra-national Trade Costs.” Workin Paper N°21439. Barrett, C.B., T. Reardon, J. Swinnen, and D. Zilberman. 2022. “Agri-food Value Chain Revolutions in Low- and Middle-Income Countries.” Journal of Economic Literature 60(4):1316–1377. Behrman, J. R., and A. B. Deolalikar. 1987. “Will developing country nutrition improve with income? A case study for rural South India.” Journal of political Economy, 95(3), 492-507. Bennett, M. K. 1954. “The world's food.” Vol. 78, No. 1, p. 83. LWW. Bernhofen, D.M., and J.C. Brown. 2005. “An Empirical Assessment of the Comparative Advantage Gains from Trade: Evidence from Japan.” American Economic Review 95(1):208–225. Borin, A., and M. Mancini. 2019. “Measuring What Matters in Global Value Chains and Value- Added Trade.” World Bank Policy Research Working Paper N°8804. Bouis, H. E., and L. J. Haddad. 1992. “Are estimates of calorie-income fxelasticities too high?: A recalibration of the plausible range.” Journal of Development Economics, 39(2), 333-364. Burgess, R., and D. Donaldson. 2010. “Can openness mitigate the effects of weather shocks? Evidence from India’s famine era.” American Economic Review 100(2):449–453. Canning, P. 2011. “A Revised and Expanded Food Dollar Series: A Better Understanding of Our Food Costs.” N°1477-2017-4009. Chapoto, A., and T.S. Jayne. 2009. “The impacts of trade barriers and market interventions on maize price predictability: Evidence from Eastern and Southern Africa.” Working Paper. Conte, M., P. Cotterlaz, and T. Mayer. 2022. “The CEPII gravity database.” Working Paper N°2022-05. Costinot, A., and A. Rodríguez-Clare. 2014. “Trade Theory with Numbers: Quantifying the Consequences of Globalization.” Handbook of International Economics 4:197–261. Dalheimer, B., M.F. Bellemare, and S. Lim. 2023. “Global Agricultural Value Chains and Food Prices.” Working Paper. 20 Donaldson, D. 2018. “Railroads of the Raj: Estimating the Impact of Transportation Infrastructure.” American Economic Review 108(4–5):899–934. Farrokhi, F., and H.S. Pellegrina. 2023. “Trade, Technology, and Agricultural Productivity.” Journal of Political Economy 131(9):2509–2555. Feder, G., R. E. Just, and D. Zilberman. 1985. “Adoption of agricultural innovations in developing countries: A survey.” Economic Development and Cultural Change, 33(2), 255-298. Ferguson, S.M., and J. Gars. 2020. “Measuring the impact of agricultural production shocks on international trade flows.” European Review of Agricultural Economics 47(3):1094–1132. Feyaerts, H., G. Van den Broeck, and M. Maertens. 2020. “Global and local food value chains in Africa: A review.” Agricultural Economics 51(1):143–157. Gaigné, C., and C. Gouel. 2022. “Trade in agricultural and food products.” Handbook of Agricultural Economics 6:4845–4931. Gaigné, C., K. Latouche, and S. Turolla. 2018. “Vertical Ownership and Export Performance: Firm-Level Evidence from the Food Industry.” American Journal of Agricultural Economics 100(1):46–72. Giunta, A., P. Montalbano, and S. Nenci. 2022. “Consistency of micro- and macro-level data on global value chains: Evidence from selected European countries.” International Economics 171:130–142. Gouel, C., and S. Jean. 2015. “Optimal Food Price Stabilization in a Small Open Developing Country.” The World Bank Economic Review 29(1):72–101. Jacks, D.S., K.H. O’rourke, and J.G. Williamson. 2011. “Commodity Price Volatility and World Market Integration since 1700.” The Review of Economics and Statistics 93(3):800–813. Jensen, R. T., and N. H. Miller. 2008. “Giffen behavior and subsistence consumption.” American economic review, 98(4), 1553-1577. Koopman, R., Z. Wang, and S.-J. Wei. 2014. “Tracing Value-Added and Double Counting in Gross Exports.” American Economic Review 104(2):459–494. Lim, S. 2021. “Global Agricultural Value Chains and Structural Transformation.” Working Paper. Lim, S., and S.W. Kim. 2022. “Global agricultural value chains and employment growth.” Journal of Agricultural and Applied Economics, 1(4):402–418. Ma, H., J. Huang, F. Fuller, and S. Rozelle. 2006. “Getting rich and eating out: consumption of food away from home in urban China.” Canadian Journal of Agricultural Economics/ Revue Canadienne d'Agroeconomie, 54(1), 101-119. McArthur, J.W., and G.C. McCord. 2017. “Fertilizing growth: Agricultural inputs and their effects 21 in economic development.” Journal of Development Economics 127:133–152. Montalbano, P., and S. Nenci. 2022. “Does global value chain participation and positioning in the agriculture and food sectors affect economic performance? A global assessment.” Food Policy 108:102235. Olper, A., L. Pacca, and D. Curzi. 2014. “Trade, import competition and productivity growth in the food industry.” Food Policy 49(P1):71–83. Papke, L.E., and J.M. Wooldridge. 1996. “Econometric methods for fractional response variables with an application to 401(k) plan participation rates.” Journal of Applied Econometrics 11(6):619–632. Pardey, P. G., C. Chan-Kang, S. P. Dehmer, and J. M. Beddow. 2016. “Agricultural R&D is on the move.” Nature, 537(7620), 301-303. Pingali, P. 2007. “Westernization of Asian diets and the transformation of food systems: Implications for research and policy.” Food policy, 32(3), 281-298. Porteous, O. 2019. “High Trade Costs and Their Consequences: An Estimated Dynamic Model of African Agricultural Storage and Trade.” American Economic Journal: Applied Economics 11(4):327–66. Porteous, O. 2020. “Trade and agricultural technology adoption: Evidence from Africa.” Journal of Development Economics 144:102440. Reardon, T., and C. P. Timmer. 2014. “Transformation of the Agrifood Industry in Developing Countries.” The Oxford Handbook of Food, Politics, and Society, Ronald J. Herring (ed.) Reardon, T. 2015. “The hidden middle: the quiet revolution in the midstream of agrifood value chains in developing countries.” Oxford Review of Economic Policy 31(1):45–63. Reimer, J.J., and M. Li. 2010. “Trade Costs and the Gains from Trade in Crop Agriculture.” American Journal of Agricultural Economics 92(4):1024–1039. Ruan, J., and M. Gopinath. 2008. “Global productivity distribution and trade liberalisation: evidence from processed food industries.” European Review of Agricultural Economics 35(4):439–460. Senauer, B., D. Sahn, and H. Alderman. 1986. “The effect of the value of time on food consumption patterns in developing countries: evidence from Sri Lanka.” American Journal of Agricultural Economics, 68(4), 920-927. Sotelo, S. 2020. “Domestic Trade Frictions and Agriculture.” Journal of Political Economy 128(7):2690–2738. Tschirley, D., T. Reardon, M. Dolislager, and J. Snyder. 2015. “The rise of a middle class in East transformation.” Journal of and Southern Africa: Implications for food system 22 International Development, 27(5), 628-646. Timmer, C.P., 2009. “A world without agriculture: The structural transformation in historical perspective (p. 96).” Washington, DC: Aei Press. Whitaker, M., and S. Kolavalli. 2006. “Floriculture in Kenya.” Technology, Adaptation, and Exports, 335. Yi, J., E.M. Meemken, V. Mazariegos-Anastassiou, J. Liu, E. Kim, M.I. Gómez, P. Canning, and C.B. Barrett. 2021. “Post-farmgate food value chains make up most of consumer food expenditures globally.” Nature Food 2021 2:6 2(6):417–425. 23 Table 1.1 Descriptive statistics of GVC shares APPENDIX: TABLES AND FIGURES Total GVC share GVC share Backward participation Forward participation Agriculture GVC share Backward participation Forward participation Fishing GVC share Backward participation Forward participation Food and Beverage GVC share Backward participation Forward participation N 865 865 865 865 865 865 865 865 865 865 865 865 mean 0.377 0.231 0.145 0.403 0.169 0.234 0.377 0.228 0.148 0.367 0.260 0.107 s.d. min max 0.120 0.127 0.048 0.124 0.096 0.070 0.193 0.191 0.082 0.123 0.133 0.038 0.117 0.048 0.045 0.097 0.026 0.069 0.097 0.026 0.000 0.154 0.063 0.039 0.855 0.769 0.291 0.719 0.636 0.385 1.080 1.109 0.446 0.873 0.810 0.257 24 Total Gross Exports Domestic Content Foreign Content Domestic value-added Double counted Foreign value-added Double Counted Value-added Exports Reflection Directly absorbed Indirectly absorbed Figure 1.1 Value-added decomposition of total exports based on Borin and Mancini (2019) and Koopman et al. (2014) Notes: this schematic representation provides a detailed breakdown of aggregate exports into several components. First, the total export of a country is decomposed into the part of exports that includes value added by other countries (‘Foreign Content’), which is also subdivided into the value-added contributed by foreign countries in the production of the exported goods the foreign value-added that is counted multiple times. Second, the total export of a country is also decomposed into the portion of exports that originates from the exporting country itself, which is further broken down into ‘Domestic value-added’ (the value-added by the exporting country in the production of exported goods) and ‘Domestic Double Counted’ (the portion of the domestic value-added that is counted more than once in the export value due to re-imports of intermediate goods). The ‘Value-added Exports’ captures the direct and indirect value-added in exports, excluding any double counted portions. It's further divided into the part of value-added exports directly used by the importing countries and the part of value-added exports that undergoes further processing before being used. ‘Reflection’ captures the portion of domestic value-added that is ultimately absorbed abroad but reflected into the domestic economy. 25 Figure 1.2 Average GVC participation by type of linkage between 2005 and 2015 26 Farm segment Post-farm segments (e.g., Transportation, retail, …) Figure 1.3 Illustration of the food dollar expenditure along the agri-food value chain based on Canning (2011) Notes: this illustration depicts the answer to the question: “For what our food dollar pay?” The standard method uses Input-Output analysis to measure average food dollar expenditures and the industry share of those expenditures in the value chain. 27 Agriculture Manufacture m r a F m r a f - t s o P Transportation & Storage Retail Services Figure 1.4 Industry groups aggregation along the agri-food value chain Notes: The figure is a flowchart depicting the industry groups along the agri-food value chain. The chart starts with "Primary Food Production" representing the farm level of food production, then moves to the post-farm activities which include "Manufacture of food," followed by "Transportation & Storage," and finally "Wholesale & Retail Trade" along with "Food Services." Each of these segments is contained within its own box, and the arrows between the boxes suggest the sequential relationship between these production stages. 28 Table 1.2 Summary statistics of industry shares by type of value chains Food At Home Agriculture Manufacture Transportation & Storage Retail Services Food & Tobacco at Home Agriculture Manufacture Transportation & Storage Retail Services Food & Accommodation Away From Home Agriculture Manufacture Transportation & Storage Retail Services N mean s.d. min max 34 0.216 0.058 0.139 0.343 34 0.249 0.034 0.180 0.324 34 0.069 0.020 0.037 0.106 34 0.460 0.049 0.348 0.558 34 0.004 0.003 0.000 0.014 160 0.188 0.058 0.008 0.445 160 0.235 0.042 0.098 0.355 160 0.092 0.109 0.011 0.513 160 0.480 0.137 0.103 0.840 160 0.004 .003 0.000 0.014 671 0.064 0.048 0.005 0.333 671 0.105 0.038 0.028 0.338 671 0.027 0.009 0.004 0.068 671 0.122 0.043 0.036 0.430 671 0.682 0.081 0.418 0.872 Notes: Tags for industry groups are as follows. ‘Agriculture’ indicates Agriculture, Forestry and Fishing; ‘Manufacture’ for Manufacture of food, beverages, and tobacco products; Transportation and storage; ‘Retail’ refers to Wholesale and retail trade; and ‘Services’ indicates Accommodation and food service activities. 29 Table 1.3 Summary statistics of control variables Variables N mean s.d. min Urbanization (% of total population) 865 73.47 15.12 19.17 max 100 GDP per capita (PPP) in $US 865 31,947 17,685 1,969 97,864 Agricultural Productivity (1000 $US/ha) 865 3.253 7.938 0.0242 77.63 30 Table 1.4 Baseline estimation results Estimation Model Panel A: Agriculture Panel B: Manufacture OLS (1) (2) (3) (4) PQMLE -0.027*** (0.006) -0.027*** (0.006) Coeff. -0.198*** (0.045) APE -0.028*** (0.006) Coeff. -0.202*** (0.045) APE -0.029*** (0.006) 0.006 (0.005) 0.005 (0.005) 0.035 (0.027) 0.007 (0.005) 0.033 (0.028) 0.006 (0.005) Panel C: Transportation & Storage 0.014*** (0.005) 0.014*** (0.005) 0.169*** (0.056) 0.013*** (0.004) 0.174*** (0.056) 0.013*** (0.004) Panel D: Retail -0.013* (0.007) -0.014* (0.007) -0.053* (0.031) -0.012* (0.007) -0.054* (0.031) -0.012* (0.007) Panel E: Services 0.021* (0.011) YES Type of value chain YES Year FE NO Time averages YES Constant Observations 865 Notes: Robust standard errors in parentheses. OLS is the Ordinary Least Squares and APE is the Average Partial Effects.*** p<0.01, ** p<0.05, * p<0.1. The outcome variable for each panel represents the share of each segment in the AVC. The independent variables are in logarithmic terms and are omitted for expositional purposes. Tags for industry groups are as follows. ‘Agriculture’ indicates Agriculture, Forestry and Fishing; ‘Manufacture’ for Manufacture of food, beverages, and tobacco products; Transportation and storage; ‘Retail’ refers to Wholesale and retail trade; and ‘Services’ indicates Accommodation and food service activities. 0.066** (0.034) YES YES YES YES 865 0.021** (0.011) YES YES YES YES 865 0.019* (0.010) YES NO YES YES 865 0.021* (0.011) YES NO NO YES 865 0.063* (0.034) YES NO YES YES 865 31 Table 1.5 Instrumental Variable estimation Estimation Model Panel A: Agriculture Panel B: Manufacture Panel C: Transportation & storage Panel D: Retail 2SLS (1) (2) -0.042*** (0.012) -0.042*** (0.013) PQMLE (4) Coeff. -0.241*** (0.092) (5) APE -0.034*** (0.013) -0.093*** (0.021) -0.093*** (0.021) -0.464*** (0.100) -0.092*** (0.020) -0.007 (0.005) -0.007 (0.005) -0.117** (0.057) -0.009** (0.004) 0.040** (0.018) 0.038** (0.017) 0.197** (0.085) 0.044** (0.019) Panel E: Services 0.103*** (0.029) YES Type of value chain NO Year FE NO Time averages YES Constant Observations 865 Notes: Bootstrap standard errors in parentheses. APE is the Average Partial Effects.*** p<0.01, ** p<0.05, * p<0.1. The outcome variable in each panel is the share of each segment in the AVC. The independent variables are in logarithmic terms and are omitted for expositional purposes. Tags for industry groups are as follows. ‘Agriculture’ indicates Agriculture, Forestry and Fishing; ‘Manufacture’ for Manufacture of food, beverages, and tobacco products; Transportation and storage; ‘Retail’ refers to Wholesale and retail trade; and ‘Services’ indicates Accommodation and food service activities. 0.259*** (0.088) YES YES YES YES 865 0.103*** (0.029) YES YES NO YES 865 0.081*** (0.027) YES YES YES YES 865 32 (2) -0.551*** (0.039) -0.152 (0.160) -0.138 (0.502) -0.044 (0.168) 0.180*** (0.037) 0.162*** (0.035) 2.689*** (0.390) F(19, 845) = 51.48 YES YES 865 0.492 Table 1.6 First stage regression of the IV estimation Weighted distance Income Urbanization Agricultural Productivity Type of the value chain FTAH FAAFH Constant F-statistics Year FE Time averages Observations R-squared Notes: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 2.718*** (0.392) F(16, 848) = 59.80 YES NO 865 0.487 (1) -0.549*** (0.040) 0.098*** (0.016) -0.008 (0.048) 0.087*** (0.009) 0.188*** (0.037) 0.160*** (0.035) 33 Table 1.7 Heterogeneity by income level (1) (2) Agriculture Manufacture Transportation (3) Percentiles 5% 7101 $US 25% 19652 $US 50% 29047 $US 75% 42657 $US 95% 63419 $US -0.055** (0.024) -0.037** (0.015) -0.122*** (0.029) -0.098*** (0.019) -0.032*** (0.012) -0.089*** (0.017) -0.026** (0.011) -0.022** (0.009) -0.080*** (0.016) -0.072*** (0.017) & Storage -0.006 (0.004) -0.008** (0.004) -0.009** (0.005) -0.010* (0.005) -0.012* (0.007) (4) Retail (5) Services 0.044** (0.018) 0.088*** (0.027) 0.044*** (0.017) 0.082*** (0.026) 0.045** (0.0175) 0.079*** (0.025) 0.045** (0.0181) 0.076*** (0.024) 0.045** (0.0189) 0.072*** (0.024) Notes: Coefficients are the Average Partial Effects. With Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Legends: (1) Agriculture, Forestry and Fishing; (2) Manufacture of food, beverages, and tobacco products; (3) Transportation and storage ; (4) Wholesale and retail trade; and (5) Accommodation and food service activities. 34 Table 1.8 Average partial effects of participation in GVCs on agri-food systems transformation by regions Dependent Variables Asia Europe Americas Agriculture Manufacture Transportation Retail Services (1) -0.042 (0.028) 0.083 (0.111) 0.331*** (0.054) (2) -0.129*** (0.023) -0.258** (0.127) 0.322*** (0.081) & storage (3) -0.023*** (0.008) -0.393** (0.193) 0.087** (0.039) (4) -0.107*** (0.024) 0.101 (0.336) 0.132** (0.053) (5) 0.269*** (0.042) 0.617** (0.246) -0.749*** (0.121) Notes: Robust standard errors are in parentheses and significance levels are denoted by: *** p<0.01, ** p<0.05, and * p<0.1. The model incorporates the type of value chain as a key explanatory variable, alongside year fixed effects and time averages of control variables. The regions include a list of countries as follows: (1) Asia: Australia, Brunei Darussalam, Cambodia, China (Hong Kong SAR), India, Indonesia, Israel, Japan, Kazakhstan, Malaysia, Philippines, Republic of Korea, Russian Federation, Saudi Arabia, Singapore, Thailand; (2) Europe: Austria, Belgium, Bulgaria, Croatia, Cyprus, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands (Kingdom of the), Norway, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden, Switzerland, Türkiye, United Kingdom of Great Britain and Northern Ireland; and (3) Americas: Argentina, Brazil, Canada, Chile, Colombia, Costa Rica, Mexico, Peru, United States of America. Tags for industry groups are as follows. ‘Agriculture’ indicates Agriculture, Forestry and Fishing; ‘Manufacture’ for Manufacture of food, beverages, and tobacco products; Transportation and storage; ‘Retail’ refers to Wholesale and retail trade; and ‘Services’ indicates Accommodation and food service activities. 35 Table 1.9 Heterogeneity by type of linkages Model Panel A: Agriculture Panel B: Manufacture Panel C: Transportation & Storage Panel D: Retail Panel E: Services Forward Linkages APE Coeff. -0.007 -0.047 (0.00513) (0.036) Backward Linkages APE Coeff. -0.028*** -0.197*** (0.005) (0.033) -0.011 (0.020) -0.036 (0.029) 0.053* (0.027) 0.014 (0.027) -0.002 (0.004) -0.222*** (0.037) -0.044*** (0.007) -0.003 (0.002) 0.012* (0.006) 0.004 (0.008) -0.084** (0.034) -0.006** (0.003) 0.071** (0.035) 0.016** (0.008) 0.192*** (0.029) 0.059*** (0.009) Covariates Type of value chain Year dummies Time averages Constant Observations Notes: Robust standard errors in parentheses. APE is the Average Partial Effects. *** p<0.01, ** p<0.05, * p<0.1. Outcome variable is the share of each segment in the DVC. Independent variables are in logarithmic terms. Tags for industry groups are as follows. ‘Agriculture’ indicates Agriculture, Forestry and Fishing; ‘Manufacture’ for Manufacture of food, beverages, and tobacco products; Transportation and storage; ‘Retail’ refers to Wholesale and retail trade; and ‘Services’ indicates Accommodation and food service activities. YES YES YES YES YES 865 YES YES YES YES YES 865 YES YES YES YES YES 865 YES YES YES YES YES 865 36 CHAPTER 2: GLOBAL AGRI-FOOD VALUE CHAINS AND FARM LABOR'S SHARE OF INCOME 37 2.1 Introduction The first chapter examined one feature of the agri-food systems transformation, specifically the distribution of value added along the value chain. This chapter shifts focus to labor—an essential component to the functioning of the system. In fact, the agri-food systems transformation serves as a major catalyst for structural change, which is a crucial process of economic development (Barrett et al. 2022)7. A key feature of this transformation is the decline in the farm labor's share of income. Over the decade from 2005 to 2015, the farm labor’s share of agricultural income declined globally by an average of 7.79 percent. During the same period, the farm labor's share of total labor income in the agri-food value chain also decreased globally by an average of 12.02 percent (FAO 2022). The widespread decline in farm labor’s share of income across most countries suggests a common mechanism driving this unequal distribution of income in agri-food value chains, potentially attributable to globalization. Does more participation in global agri-food value chains impact the farm labor’s share of income? There are at least two strands of research that examine the impact of different dimensions of globalization on labor income. First, studies focusing on the microeconomic aspects offer valuable insights into the impact of agri-food trade on employment, wages, and income distribution. At the farm level, contract farming along with the emergence of export value chains has led to considerable employment growth in agriculture (Maertens and Swinnen 2009; Neven et al. 2009; Mano et al. 2011; Colen, Maertens, and Swinnen 2012; Meemken and Bellemare 2020)8. Moreover, improved labor standards and codes of conduct have been associated with enhanced workers' well-being (Barrientos, Dolan, and Tallontire 2003; Meemken et al. 2019), including increased employment duration and higher wages (Colen, Maertens, and Swinnen 2012). The post- harvest segments of the value chain also experience new employment opportunities arising in labor-intensive agri-food activities (Lagakos 2016; Dolislager et al. 2020). In terms of income distribution, employment in modern agri-food value chains has increased particularly for the poor 7 The rapid transformation of agri-food systems has positive impacts on rural poverty alleviation (Maertens and Swinnen 2009; Liverpool-Tasie, Adjognon, and Reardon 2016; Van den Broeck, Swinnen, and Maertens 2017) and food security (Chege, Andersson, and Qaim 2015; Bellemare and Novak 2017). 8 At the same time, modern agricultural value chains benefited greatly from automation (e.g., field machinery, irrigation systems, greenhouse automation, etc) in the last century, mainly due to labor-saving technologies (Edan et al. 2009). 38 (Van den Broeck, Swinnen, and Maertens 2017; Ogutu, Ochieng, and Qaim 2020)9. Second, macroeconomic studies have explored the role of trade openness and exports in impacting labor shares across a range of countries with different income levels (Böckerman and Maliranta 2012; Leblebicioğlu and Weinberger 2021; Panon 2022). These studies find mixed evidence on the effect of trade on labor shares, including possible increases in labor shares due to improved capital equipment quality and lower costs. The existing research has provided valuable insights into the impact of agri-food trade on employment, wages, and income distribution. Yet, the narrow focus on specific segments of the value chain offers an incomplete understanding of how globalization causes changes in the distribution of income within the entire agri-food value chain. In addition, while the effects of aggregate trade on labor shares are country- and commodity-specific, whether trade increases or decreases the fam labor’s share of income remains an empirical question. To address these gaps, I study the impacts of participation in Global Agri-food Value Chains (hereafter GVCs) on the distribution of income between farm labor and other agricultural factors such as capital as well as between farm and post-farm labor. The main research hypothesis posits that participation in global agri-food production networks would have a negative impact on the farm labor’s share of income due to skill and sector-biased differentials. In other words, globalization may exacerbate the wage gap between high and low-skilled labor or between farm and post-farm laborers due to differences in economic competitiveness, thus reducing the overall farm labor’s share of income. This study contributes to the literature in two ways. First, I provide novel empirical evidence on the impact of participation in GVCs on the distribution of income in domestic agri- food value chains based on nationally representative longitudinal data, representing 90% of the global economy. The existing research mainly focuses on the farm segment, often overlooking the critical role of the "hidden middle" intermediate industries in the reallocation of income between the different production activities along the value chain (Reardon 2015; Dolislager et al. 2020; Christiaensen, Rutledge, and Taylor 2021). In-depth analysis of how participation in GVCs impact farm labor compensation relative to the compensation received by workers in post-farm segments, this study enriches the debate on labor income reallocation along the value chain. 9 The distribution of income is important because in extreme instances income inequality may lead to food riots and social unrest (Gana 2012; Bellemare 2015; Bush and Martiniello 2017). These concerns escalated amidst the COVID- 19 pandemic (Ahmed Pissarides and Stiglitz 2020; Clark, d’Ambrosio, and Lepinteur 2021; Deaton 2021). 39 Second, the literature predominantly focuses on how international trade impacts the economy-wide labor share of income (Karabarbounis and Neiman 2014; Rognlie 2016; Fan 2019; Leblebicioğlu and Weinberger 2021; Panon 2022). However, aggregate trade statistics do not capture the granularity of global trade such as the length of the production network and the directionality of supply chain linkages (whether backward or forward linkages). The production network is complex since a global value chain involves at least two countries in the production of a final consumer good, with each country adding value at different stages of the production process (Gereffi and Fernandez-Stark 2016). In this paper, I employ a more detailed decomposition of trade flows to measure participation in GVCs, following the methodology proposed by Borin and Mancini (2019). Furthermore, I differentiate between backward and forward participation and their distinct impacts on the farm labor’s share of income. Two main findings emerge from this paper. First, I find that the decline in the farm labor’s share of agricultural income is exclusively attributable to backward linkages in the GVC. Conversely, the decline in the farm labor’s share of total labor income within the entire agri-food value chain is more associated with forward linkages, especially for poorer countries. Second, the empirical findings in this study indicate that the effects on farm labor are spilling over from the global food and beverage value chain. This contrasts with the conventional wisdom that attributes the distributional effects of globalization on farm labor directly to participation in the global agricultural value chain. Policy interventions aiming at smoothing the unequal distribution of income may focus on positioning in the global food and beverage chain rather than targeting agricultural trade per se. The remainder of the chapter is organized as follows. The next section proposes a theoretical framework on the effects of participation in GVCs on the labor share of income. Section 2.3 explores data, measurement, and some stylized facts, after which I present the empirical framework. Subsequent sections contain the empirical findings and concluding remarks. 2.2 Theoretical Framework In this section, I present a simple model following the standard theory of Kohli (1990), Harrigan (1996), and Harrigan and Balaban (1999). The model discusses the general equilibrium relationship between factor prices, factor supplies, and final goods prices. The standard model assumes vertical integration within the value chain, which implies that intermediate inputs are netted out in the analysis. This paper attempts, however, to isolate the effect of international trade 40 in intermediate inputs (or tradeable) on labor compensation. Consider a small open economy characterized by fixed aggregate factor supplies, constant returns to scale and competitive market clearing. Dixit and Norman (1980) show that the solution to the classical maximization problem of total profit yields the revenue function—also called the GDP function—, as a relation of technology levels, prices, and production inputs: 𝑔 = 𝑟(𝜹𝒑, 𝝅) (1) where 𝜹 is a vector of parameters representing Hicks-neutral technological progress in each industry relative to a base period, 𝒑 denotes prices of 𝑁 final products, and 𝝅 denotes the production inputs, referring to both factor supplies and intermediate inputs. Following a long- standing tradition in trade theory, this paper assumes that a country's revenue function can be represented by a translog function, written as: 𝑙𝑛 𝑟(𝜹𝒑, 𝝅) = 𝑎00 + ∑  𝑁 𝑖=1  𝑎0𝑖 𝑙𝑛 𝛿𝑖𝑝𝑖 + 1 2 ∑  𝑁 𝑖=1  ∑  𝑁 𝑗=1  𝑎𝑖𝑗 𝑙𝑛 𝛿𝑖𝑝𝑖 ⋅ 𝑙𝑛 𝛿𝑗𝑝𝑗 + ∑  𝑀 𝑘=1  𝑏0𝑘 𝑙𝑛 𝜋𝑘 + 1 2 ∑ 𝑘=1  ∑  𝑀  𝑀 𝑙=1  𝑏𝑘𝑙 𝑙𝑛 𝜋𝑘 ⋅ 𝑙𝑛 𝜋𝑙 (2) + ∑  𝑁 𝑖=1  ∑  𝑀 𝑘=1  𝑐𝑖𝑘 𝑙𝑛 𝛿𝑖𝑝𝑖 ⋅ 𝑙𝑛 𝜋𝑘 where the summations over goods 𝑖 and 𝑗 run from 1 to 𝑁, and the summations over production inputs 𝑘 and 𝑙 run from 1 to 𝑀. Harrigan and Balaban (1999) impose the following two restrictions: i. ii. Symmetry of cross effects: 𝑎𝑖𝑗 = 𝑎𝑗𝑖 and 𝑏𝑘𝑙 = 𝑏𝑙𝑘 for all 𝑖, 𝑗, 𝑘, and 𝑙. Homogeneity of degree one in endowments and prices: ∑  𝑁 ∑  𝑁 𝑖=1  𝑎0𝑖 𝑖=1  𝑎𝑖𝑗 = ∑ = ∑  𝑀 𝑘=1  𝑏0𝑘 = 1,  𝑁 𝑗=1  𝑎𝑖𝑗 = ∑ 𝑘=1  𝑏𝑘𝑙 = ∑  𝑀  𝑀 𝑙=1  𝑏𝑘𝑙 = ∑ 𝑘=1  𝑐𝑖𝑘 = ∑  𝑁  𝑀 𝑖=1  𝑐𝑖𝑘 = 0 (3) Computing the derivative of equation (2) with respect to 𝑙𝑛 (𝜋𝐿) yields 𝑠𝐿 = 𝑏0𝐿 + ∑  𝑀 𝑙=2 𝑏𝐿𝑙 𝑙𝑛 ( 𝜋𝑙 𝜋1 ) + ∑  𝑁 𝑖=2 𝑐𝑖𝐿 𝑙𝑛 ( 𝛿𝑖𝑝𝑖 𝛿1𝑝1 ) (4) where the labor share of total income is computed as follows: 𝑠𝐿 ≡ 𝜕 𝑙𝑛(𝑟(𝜹𝒑, 𝝅)) 𝜕 𝑙𝑛(𝜋𝐿) = 𝜕𝑟(𝜹𝒑, 𝝅) 𝜕𝜋𝐿 𝜋𝐿 𝑟(𝜹𝒑, 𝝅) = 𝑤𝜋𝐿 𝑟(𝜹𝒑, 𝝅) (5) with 𝑤 being the compensation received by workers. In words, Eq. (4) states that the share of labor in total income is a log-linear function of relative production inputs, technology levels, and prices. 2.3 Measurement Methods and Data This section presents data and computational methods used in this paper to compute the 41 share of income for farm labor and participation in global agri-food production networks. 2.3.1 Farm Labor’s Share of Income Recall from Chapter 1, the average payment from each food dollar expenditure that producers in industry 𝑖 receive for their raw food dollar commodities can be written as follows: 𝑠𝑖 = 𝑣𝑖 𝑦𝑖 (6) The computation of factor return composition is straightforward. By rewriting Eq. (6) with the addition of a second subscript 𝑝 to define the primary production factors, we get the share of total value added that a primary production factor receives in each industry, 𝑠𝑖,𝑝: 𝑠𝑖,𝑝 = 𝑣𝑖,𝑝 𝑦𝑖 (7) This formula serves as an empirical measure of 𝑠𝐿 in Eq. (5). This study focuses on labor (𝑙) compensation across farm (𝑓) and post-farm segments of the value chain (see Figure 2.1 for an illustration). Using Eq. (7), we get the value of farm labor compensation—composed of wages, salaries, and benefits—relative to the compensation of all primary production factors involved in the agricultural sector such as capital (called farm labor’s share of agricultural income in this paper): 𝑠𝑓,𝑙 ≡ 𝑣𝑓,𝑙 ∑ 𝑣𝑓,𝑝 𝑝 (8) and the value of farm labor compensation relative to the total labor compensation across all segments of the agri-food value chain (or farm labor's share of total labor income): 𝑠̃𝑓,𝑙 ≡ 𝑣𝑓,𝑙 ∑ 𝑣𝑖,𝑙 𝑖 (9) This study constructs a dataset to compute formulas (8) and (9) using the FAOSTAT platform, which provides data for three types of agri-food value chains: Food At Home (FAT), Food and Tobacco at Home (FTAT), and Food and Accommodation Away from Home (FAAFH). This data is distinguished by its industry-wise and factor-wise disaggregation and is obtained using the OECD international database for Input-Output tables. Table 2.1 provides descriptive statistics of both measures of the farm labor’s share of income categorized by type of value chain. In a detailed exploration of how the farm labor’s share of income changes between 2005 and 2015, Figure 2.2 depicts a graphical representation of the national level variations of farm labor’s share of income. Overall, the data exhibits a noticeable 42 dispersion compared to the mean, indicative of sufficient variability across countries to run a regression analysis. Understanding these differences using cross-country analysis may not only provide insight into unique national income distribution within the agri-food value chain, but also the broader forces shaping labor markets on a global scale. 2.3.2 Participation in GVCs The measurement method of participation in GVCs is similar to that in Section 1.3.1 of Chapter 1. As previously discussed in that section, this study adopts the framework of Borin and Mancini (2019), which consists of decomposing the total gross export into various components to quantify the import content of a country’s exports or the value of raw materials that a country imports from upstream sectors in other countries (backward linkages). Also, the method makes it possible to quantify the exports of output by downstream sectors to the importing country to be later re-exported (forward linkages). Illustration of the method and descriptive data are presented in Figure 1.1 and Table 1.1 of Chapter 1, respectively. 2.3.3 Control Variables The empirical examination of the income distributional effects of globalization in agri-food value chains relies on a suite of control variables. Table 2.2 below succinctly summarizes these variables, contrasts the theoretical and empirical measures, and includes conceptual comments regarding their effects. Drawing upon this table, higher income levels lead to increased demand for processed foods, prompting a labor reallocation towards sectors producing these goods and away from agriculture, resulting in a reduction of the farm labor share of income. Urbanization further exacerbates this trend by relocating labor and resources towards urban-centered industries. Conversely, labor productivity and education represent positive forces, suggesting that increases in agricultural labor productivity and a more educated workforce can lead to higher compensation for farm labor due to gains in efficiency and the ability to earn higher wages. However, the transition towards more capital-intensive farming practices, as indicated by investments in net capital stocks and access to electricity, generally reduces the income share of labor by substituting human labor with machines (e.g., field machinery, irrigation systems, greenhouse automation, etc). The impact of market prices on the farm labor share of income is mixed, potentially allowing for higher compensation to farm labor if farm-gate prices increase, though this is contingent upon the compensations received by other production factors. The crop mix also plays 43 a critical role; labor-intensive crops such as fruits and vegetables may increase the labor share of income compared to grain crops, which are more likely to be mechanized. Finally, investments in research and development can lead to agricultural innovations that may reduce the labor share of income by introducing more efficient agricultural practices and technologies, further pushing agriculture towards mechanization and high productivity that do not necessarily benefit farm labor directly. These variables were selected based on existing literature and the availability of data from various sources, including UNCTAD-Eora, World Development Indicators, and FAOSTAT. I also predict the expected direction of the impact, forming an empirical hypothesis for the subsequent section. For additional details, Table 2.3 offers descriptive statistics of the variables presented in the summary table. 2.4 Empirical Framework The coefficients of Eq. (4) can be estimated econometrically with annual panel data on output, farm labor’s share of income, product prices, endowments, intermediate inputs, and a measure of Hicks-neutral technological progress10. Panel data makes it convenient to control for technical change other than the Hicks-neutral technological progress using country and year fixed effects. To estimate the theoretical equation, I assume that the actual labor share deviates from Eq. (4) by an error term with mean zero, denoted by 𝜖. The empirical model is written as follows: s𝑖𝑡𝑚 = 𝛼0 + 𝛽𝜊ln (GVC)𝑖𝑡 + 𝛽1 ′𝑋𝑖𝑡 + 𝛽2𝑡𝑦𝑝𝑒𝑖𝑡𝑚 + 𝛿𝑖 + 𝜏𝑡 + 𝜖𝑖𝑡𝑚 (10) The subscripts 𝑖, 𝑡, and 𝑚 denote country identifier, year, and type of the value chain, respectively. I run separate regressions for the two measures of the farm labor’s share of income, as discussed in Section 3.1. GVC𝑖𝑡 is the measure of participation in global agri-food production networks. 𝑋𝑖𝑡 is a vector of the control variables I summarized in Table 1 and 𝑡𝑦𝑝𝑒𝑖𝑡𝑚 is a categorical variable that assigns values as follows: 1 for FAH, 2 for FTAH, and 3 for FAFH, with the FAH series serving as the baseline. 𝛿𝑖 and 𝜏𝑡 represents the country and year fixed effects, respectively. The modeling framework is similar to that is Chapter 1, particularly regarding the use of Panel Quasi-Maximum Likelihood (PQMLE) to estimate the fractional response model in (10). The goal in this study is to estimate 𝛽𝜊 to show the effect of participation in GVCs on the 10 Empirical proxies for the theoretical variables under discussion are summarized in table 2.2 of section 3.3.3. 44 farm labor’s share of income by testing the null hypothesis 𝐻0 ∶ 𝛽𝜊 = 0 versus the alternative hypothesis 𝐻𝑎: 𝛽𝜊 ≠ 0. To establish a causal statement, it is necessary to control for endogeneity of GVC participation. The estimate of 𝛽𝜊 may be biased if, for example, the income distribution within the agri-food value chains causes changes in the extent to which a country participates in GVCs. However, reverse causality is unlikely in this case for three reasons. First, there is no plausible theory in agricultural economics that predicts how changes in the distribution of income could lead to changes in participation in GVCs, this weakens the argument for bi-directional causality. In fact, while cheap labor could be a factor conditioning participation in global production networks, the specific distribution of income along the value chain is less likely to impact a country’s participation in GVCs. Second, participation in GVCs often precedes changes in income distribution because it's less plausible that changes in income distribution would influence past participations in GVCs. Chapter 1 supports this argument since I showed that higher participation in GVCs often precedes changes in value distribution across industries primarily because entering GVCs catalyzes agri-food systems transformation. Third, changes in the distribution of income are unlikely to cause changes in the same period's measure of participation in GVCs, which largely mitigates the risk of simultaneity bias (see Appendix B for a robustness check). 2.5 Results and Discussion In this section, I delve into the core findings of this study, beginning with a presentation of the baseline estimation results. I then study heterogeneity by type of linkage, national income and global agri-food sector. 2.5.1 Baseline Estimation Results Table 2.4 presents the baseline estimation results with the dependent variable being the farm labor’s share of agricultural income. Across all models, the coefficient for the GVC indicator is negative and statistically significant, indicating that a 10% increase in GVC participation decreases the farm labor’s share of agricultural income by 0.003 point. This negative relationship remains consistent across all models, with only slight variations in the magnitude. In terms of the farm labor's share of total labor income, the coefficient for the GVC indicator is not statistically significant at the 1% level, as can be seen from table 2.5. The lack of significance could be due to the opposing effects of upstream and downstream linkages, which might be masking the overall effect. I will delve into this decomposition in the subsequent section. 45 A possible explanation for the negative relationship between participation in GVCs and the farm labor’s share of income could be that as a country participates more in global agri-food value chains, it may cause a shift of resources from labor-intensive activities to more capital-intensive activities. This could be due to technology transfer, outsourcing, or specialization in crops where the country has a comparative advantage. The rest of the variables in both table 2.4 and 2.5 show varying effects across models, indicating that these covariates might also impact the farm labor’s share of income. The FAAFH estimates are consistently less than those of the FAH series. This result emphasizes the greater role of food services and facility rental costs in domestic agri-food value chains. In some cases, I find no evidence of differences between FTAH and FAH series. The inclusion of tobacco in the FAH category doesn't result in a significant difference between these two types of the value chain. 2.5.2 Does the Type of Linkage Matter? In this section, I analyze the distinct effects of backward and forward participation on the farm labor’s shares of income. I perform a regression analysis similar to the baseline model but break down the total GVC participation indicator into backward and forward components. The model is re-written as follows: s𝑖𝑡𝑚 = 𝛼0 + 𝛽𝜊𝑈ln (Forward)𝑖𝑡 + 𝛽𝜊𝐷ln (Backward)𝑖𝑡 + 𝛽1 ′𝑋𝑖𝑡 + 𝛽2𝑡𝑦𝑝𝑒𝑖𝑡𝑚 + 𝛿𝑖 + 𝜏𝑡 + 𝜖𝑖𝑡𝑚 (11) From table 2.6, I find that increased backward participation decreases the farm labor's share of agricultural income. Higher backward linkages, characterized by increased imports of raw products for additional manufacturing before export, can lead to a reduction in the procurement of domestic agricultural products by local agri-food businesses, especially if the imported inputs offer advantages over domestic alternatives in terms of cost, quality, or adaptability for the production processes. These dynamics may reallocate labor income from farm to post-farm segments or from labor to agricultural capital, thus reducing the farm labor's share of income. Conversely, the decline in the farm labor's share of total labor income is more associated with forward participation. However, the opposing effects of upstream and downstream linkages cancel out in the overall impact of participation in GVCs, as noted in the previous section. Increased upstream participation boosts the demand for a country's raw agri-food products that might benefit agricultural capital to the detriment of labor. Also, there would be an increase in demand for activities related to basic processing and packaging, or international transportation 46 of these products. As demand for labor in post-farm segments increases, this could lead to a rise in post-farm labor compensation. Consequently, the distribution of income changes in favor of the post-farm workers. 2.5.3 Heterogeneity by Income Level In this section, I estimate the average marginal effects of participation in GVCs on the outcome variables across different income percentiles, measured by the GDP per capita. Similar to Chapter 1, the use of PQMLE enables leveraging the inherent non-linearities of the econometric model to avoid partitioning the dataset into subgroups of countries. This approach preserves the entire dataset and statistical power of the analysis. Table 2.7 presents the estimation results for both measures of the farm labor’s share of income. Interestingly, the impact of participation in global agri-food production networks on the farm labor’s share of income varies with national income levels. Columns (1), (3), and (5) contain coefficients that are statistically significant at the 1% level, consistent with the previous findings. Within each column, the observed effects diminish as income (in terms of percentile levels) rises. Consequently, poorer countries experience a greater reduction in the farm labor’s share of income compared to their wealthier counterparts. This is likely because poorer countries often exhibit greater potential for rapid transformation of their agri- food systems—known as the catch-up effect in the economic literature. In essence, as these countries participate in global agri-food production networks, the relative inefficiencies or underutilized resources are addressed more intensively, leading to more changes in the distribution of income within the agri-food value chain compared to richer economies. 2.5.4 Heterogeneity by Region Table 2.8 presents an analysis of how participation in GVCs impacts the share of farm labor in agricultural income and the share of farm labor in total labor income across three regions: Asia, Europe, and the Americas11. The analysis is segmented into three panels (A, B, and C), each representing one of the regions. The estimation in each panel includes the other control variables, year fixed effects, and the type of value chain as part of the baseline model. I opted for the linear model in my analysis due to the failure of the non-linear model to converge, which can likely be attributed to insufficient observations and variability within regional clusters. However, it is noteworthy that the linear and non-linear estimates are close enough in the baseline estimations, indicating that the linear model 11 Africa was excluded from the analysis due to a lack of sufficient data. 47 provides a reasonable approximation of the relationships being studied. The linear approach remains a robust and reliable method for analyzing the available data within the constraints of our dataset. The table suggests that the impact of GVC participation and the direction of this participation (backward or forward) varies significantly across regions. In Asia (Panel A), GVC participation tends to decrease the share of farm labor in agricultural income but increases it in total labor income. In Europe (Panel B), forward participation benefits the share of farm labor in total labor income, while in the Americas (Panel C), both GVC participation and backward participation strongly favor an increase in the share of farm labor in agricultural income. These differences could be due to varying levels of industrialization, the nature of agricultural practices, and the structure of economies in these regions. 2.5.5 Sectoral Analysis of Participation in GVCs In this section, I re-estimate the baseline model for each global agri-food sector. The results are shown in tables 2.9 and 2.10. For the ‘Agriculture’ and ‘Fishing’ global value chains, the estimates are not statistically significant, suggesting that the effect is not statistically different from zero. However, a 10% increase in the ‘Food & Beverage’ global value chain participation (e.g., importing fruit concentrate to produce bottled juice products that are then exported to their final destinations) correlates with a statistically significant decline of approximately 0.001 share point in the farm labor's share of total labor income and about 0.005 share point in the farm labor's share of agricultural income. Unlike raw agricultural products or fish, which might be traded with minimal processing, products in the 'Food & Beverage' sector often undergo substantial value addition before reaching the final consumer. As countries participate more in the global processed food market, agribusinesses will hire more labor, especially high-skilled labor, to face foreign competition in this market. Consequently, this results in a diminished share of income for farm labor, controlling for factors such as technological change, agricultural labor productivity and capital intensity. Previous research tends to assume a somewhat direct relationship between trade in a particular sector and labor in that sector. The findings indicate that the effects of globalization may be spilling over from the global ‘Food & Beverage’ value chain, rather than the global agricultural value chain per se. I suggest that the processed global food market might play a more critical role in determining domestic farm labor income than the global agricultural market. 48 2.6 Concluding Remarks and Policy Recommendations The declining farm labor’s share of income underscores the ongoing transformation of agri- food systems. In most countries, the proportion of income allocated to farm labor has been on a downward trend, suggesting that globalization may be responsible for this shift. Yet, studies that only focus on income distribution within specific segments of the value chain might overlook a large part of the story on how globalization causes changes in the distribution of income within the entire agri-food value chain. In this chapter, I examined the distribution of income between farm labor and other agricultural factors such as capital, and between farm and post-farm labor. I have empirically estimated the causal relationship between participation in global agri- food production networks and the share of income distributed to farm labor. The main result of this chapter suggests that the decline in the farm labor’s share of agricultural income is exclusively attributable to backward linkages in GVCs. In contrast, the decline in the farm labor’s share of total labor income within the entire agri-food value chain is more associated with forward linkages. Moreover, the empirical results indicated that such an impact is higher in poorer countries compared to wealthier ones. Finally, the sectoral analysis showed that unlike raw agricultural products or fish, higher participation in global ‘Food & Beverage’ value chains leads to a statistically significant decline in the farm labor’s share of income. There are several policy implications of these findings. Understanding what determines the distribution of income in agri-food value chains can help design targeted interventions, subsidies, or income support programs. While targeted aid to farm labor may smooth the unequal distribution of income along the value chain, this study indicate that policies may focus on the country’s positioning into the global food and beverage chain rather than targeting agricultural trade per se. Moreover, the empirical evidence presented in this chapter can lead to better-informed consumers who might favor products from more equitable chains, in light of the growing consumer concerns about fair trade and equitable income distribution. 49 REFERENCES Ahmed, F., N. E. Ahmed, C. Pissarides, and J. Stiglitz. 2020. “Why inequality could spread COVID-19.” The Lancet Public Health, 5(5), e240. Arellano, M., and S. Bond. 1991. “Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations.” The Review of Economic Studies, 58(2), 277- 297. Arellano, M., and O. Bover. 1995. “Another look at the instrumental variable estimation of error- components models.” Journal of Econometrics, 68(1), 29-51. Barrett, C. B., T. Reardon, J. Swinnen, and D. Zilberman. 2022. “Agri-food value chain revolutions in low-and middle-income countries.” Journal of Economic Literature, 60(4), 1316-1377. Barrientos, S., C. Dolan, and A. Tallontire. 2003. “A gendered value chain approach to codes of conduct in African horticulture.” World development, 31(9), 1511-1526. Bellemare, M. F. 2015. “Rising food prices, food price volatility, and social unrest.” American Journal of Agricultural Economics, 97(1), 1-21. Bellemare, M. F., and L. Novak. 2017. “Contract farming and food security.” American Journal of Agricultural Economics, 99(2), 357-378. Blundell, R., and S. Bond. 1998. “Initial conditions and moment restrictions in dynamic panel data models.” Journal of Econometrics, 87(1), 115-143. Böckerman, P., and M. Maliranta. 2012. “Globalization, creative destruction, and labour share change: evidence on the determinants and mechanisms from longitudinal plant-level data.” Oxford Economic Papers, 64(2), 259-280. Borin, A., and M. Mancini. 2019. “Measuring what matters in global value chains and value-added trade.” World Bank Policy Research Working Paper N°8804. Bush, R., and G. Martiniello. 2017. “Food riots and protest: Agrarian modernizations and structural crises.” World Development, 91, 193-207. Canning, P. 2011. “A revised and expanded food dollar series: a better understanding of our food costs.” N°1477-2017-4009. Chege, C. G., C. I. Andersson, and M. Qaim. 2015. “Impacts of supermarkets on farm household nutrition in Kenya.” World Development, 72, 394-407. Christiaensen, L., Z. Rutledge, and J. E. Taylor. 2021. “The future of work in agri-food.” Food Policy, 99, 101963. Clark, A. E., C. d’Ambrosio, and A. Lepinteur. 2021. “The fall in income inequality during COVID-19 in four European countries.” The Journal of Economic Inequality, 19, 489-507. 50 Colen, L., M. Maertens, and J. Swinnen. 2012. “Private standards, trade and poverty: GlobalGAP and horticultural employment in Senegal.” The World Economy, 35(8), 1073-1088. Deaton, A. 2021. “Covid-19 and global income inequality” (No. w28392). National Bureau of Economic Research. Dixit, A., and V. Norman. 1980. “Theory of international trade: A dual, general equilibrium approach.” Cambridge University Press. Dolislager, M., T. Reardon, A. Arslan, L. Fox, S. Liverpool-Tasie, C. Sauer, and D. L. Tschirley. 2020. “Youth and adult agrifood system employment in developing regions: Rural (peri- urban to hinterland) vs. urban.” The Journal of Development Studies, 57(4), 571-593. Edan, Y., S. Han, and N. Kondo 2009. “Automation in agriculture.” Springer Handbook of Automation 1095-1128. Fan, J. 2019. “Internal geography, labor mobility, and the distributional impacts of trade.” American Economic Journal: Macroeconomics, 11(3), 252-288. FAO. 2022. “Value shares by industry and primary factors.” Last Update November 28, 2022. Retrieved from https://www.fao.org/faostat/en/#data/GFDI Gana, A. 2012. “The rural and agricultural roots of the Tunisian revolution: When food security matters.” The International Journal of Sociology of Agriculture and Food, 19(2), 201-213. Gereffi, G., and K. Fernandez-Stark. 2016. “Global value chain analysis: a primer.” Center on Globalization, Governance & Competitiveness (CGGC). Duke University, North Carolina, USA. Karabarbounis, L., and B. Neiman. 2014. “The global decline of the labor share.” The Quarterly journal of economics, 129(1), 61-103. Kohli, U. 1990. “Price and quantity elasticities in US foreign trade.” Economics Letters, 33(3), 277-281. Lagakos, D. 2016. “Explaining cross-country productivity differences in retail trade.” Journal of Political Economy, 124(2), 579-620. Leblebicioğlu, A., and A. Weinberger. 2021. “Openness and factor shares: Is globalization always bad for labor?.” Journal of International Economics, 128, 103406. Liverpool-Tasie, L., S. Adjognon, and T. Reardon. 2016. “Transformation of the food system in Nigeria and female participation in the Non-Farm Economy (NFE).” Working Paper. Maertens, M., and J. F. Swinnen. 2009. “Trade, standards, and poverty: Evidence from Senegal.” World development, 37(1), 161-178. Mano, Y., T. Yamano, A. Suzuki, and T. Matsumoto. 2011. “Local and personal networks in 51 employment and the development of labor markets: Evidence from the cut flower industry in Ethiopia.” World Development, 39(10), 1760-1770. Meemken, E. M., and M. F. Bellemare. 2020. “Smallholder farmers and contract farming in developing countries.” Proceedings of the National Academy of Sciences, 117(1), 259-264. Meemken, E. M., J. Sellare, C. N. Kouame, and M. Qaim. 2019. “Effects of Fairtrade on the livelihoods of poor rural workers.” Nature Sustainability, 2(7), 635-642. Neven, D., M. M. Odera, T. Reardon, and H. Wang. 2009. “Kenyan supermarkets, emerging middle-class horticultural farmers, and employment impacts on the rural poor.” World development, 37(11), 1802-1811. Ogutu, S. O., D. O. Ochieng, and M. Qaim. 2020. “Supermarket contracts and smallholder farmers: Implications for income and multidimensional poverty.” Food Policy, 95, 101940. Papke, L. E., and J. M. Wooldridge. 1996. “Econometric methods for fractional response variables to 401 (k) plan participation rates.” Journal of Applied with an application Econometrics, 11(6), 619-632. Panon, L. 2022. “Labor share, foreign demand and superstar exporters.” Journal of International Economics, 139, 103678. Reardon, T. 2015. “The hidden middle: the quiet revolution in the midstream of agrifood value chains in developing countries.” Oxford Review of Economic Policy, 31(1), 45-63. Rognlie, M. 2016. “Deciphering the fall and rise in the net capital share: accumulation or scarcity?.” Brookings Papers on Economic Activity, 2015(1), 1-69. Van den Broeck, G., J. Swinnen, and M. Maertens. 2017. “Global value chains, large-scale farming, and poverty: Long-term effects in Senegal.” Food policy, 66, 97-107. Yi, J., E.M. Meemken, V. Mazariegos-Anastassiou, J. Liu, E. Kim, M.I. Gómez, P. Canning, and C. B. Barrett. 2021. “Post-farmgate food value chains make up most of consumer food expenditures globally.” Nature Food, 2(6), 417-425. 52 APPENDIX A: TABLES AND FIGURES Farm segment Post-farm segments (e.g., Transportation, retail, …) Figure 2.1 Illustration of the food dollar expenditures along the agri-food value chain and the labor share of these expenditures Notes: this illustration depicts the answer to the question: “For what our food dollar pay?” The standard method uses Input-Output analysis to measure average food dollar expenditures and the industry share of those expenditures in the value chain. The payments received by each segment of the value chain are then distributed to the production factor (e.g., labor and capital). 53 Table 2.1 Summary statistics of farm labor’s share of income N mean Standard Deviation min max Food At Home Farm labor's share of agricultural income 34 0.217 Farm labor's share of total labor income 34 0.142 Food & Tobacco at Home Farm labor's share of agricultural income 160 0.207 Farm labor's share of total labor income 160 0.127 Food & Accommodation Away from Home 0.057 0.055 0.064 0.057 0.116 0.309 0.064 0.298 0.056 0.377 0.009 0.401 Farm labor's share of agricultural income 671 0.189 Farm labor's share of total labor income 671 0.041 0.071 0.048 0.010 0.379 0.0001 0.304 Farm Labor’s Share of Income Farm labor's share of agricultural income 865 0.193 Farm labor's share of total labor income 865 0.061 0.069 0.062 0.010 0.379 0.0002 0.401 54 Farm Labor's Share of Total Labor Income Farm Labor's Share of Agricultural Income Figure 2.2 Percent change of the farm labor’s share of income between 2005 and 2015 55 Comment Predicted sign Empirical measure Source Table 2.2 Summary of the control variables Theoretical measure Globalization I provided a theoretical discussion on the effects of globalization on the farm labor’s share of income in section 2. Income High income boosts demand for processed food. This demand spurs labor shifts toward the sectors producing such goods. Urbanization Rapid urbanization moves labor and resources from farming to post-farm industries. Labor Productivity If labor becomes more productive, this could mean higher farm labor compensation due to gains in efficiency. Capital Intensity If farming becomes more capital- intensive, the share of income going to labor may decrease. - - - - + + - - GVC participation (%) UNCTAD-Eora Backward participation (%) UNCTAD-Eora Forward participation (%) UNCTAD-Eora GDP per capita (PPP) World Development Indicators Urbanization (% total population) World Development Indicators Agriculture value-added per worker (constant 2015 US$) FAOSTAT Net Capital Stocks (Agriculture, Forestry and Fishing) Value million US$, 2015 prices FAOSTAT Access to electricity (% of population) World Development Indicators 56 Table 2.2 (cont’d) Market Prices If farm-gate prices are high, there might be more to distribute amongst workers, but this also depends on other input costs. Education Crop mix Investments in Research and Development More educated and skilled workforce might be able to receive higher wages. Some agricultural products are more labor-intensive than others. For instance, fruit orchards or vegetable farms may require more manual labor compared to grain crops, which are often highly mechanized. Thus, the specific product mix in a region can affect farm labor's income share. Investments in agricultural R&D can lead to innovations that influence productivity, crop variety, pest resistance, etc., all of which can impact labor's share of income. + + - + + - Producer Price Index (2014- 2016 = 100) in agriculture FAOSTAT Government expenditure on education, total (% of GDP) World Development Indicators Share of Cereals Production in total gross production Value FAOSTAT Share of Fruits Production in total gross production Value FAOSTAT Share of Vegetables Production in total gross production Value FAOSTAT Research and development expenditure (% of GDP) World Development Indicators 57 Table 2.3 Summary statistics of control variables N mean s.d min max Urbanization (% total population) 865 73.47 15.12 19.17 100 Access to electricity (% of population) 856 98.01 7.835 20.50 100 GDP per capita (PPP) in $US 865 31,947 17,685 1,969 97,864 Government expenditure on education (% of GDP) 765 13.57 3.864 4.646 31.37 Agriculture value-added per worker (constant 2015 US$) 865 28,695 26,222 880.6 114,697 Net Capital Stocks (Primary) in million US$, 2015 prices 865 67,762 124,598 187.8 640,034 Research and development expenditure (% of GDP) 137 62.78 58.07 0 190.0 Producer Price Index (2014-2016 = 100) in agriculture 865 91.01 18.87 15.91 135.9 Share of Cereals Production in total gross production 833 0.226 0.162 0.00313 1.055 Share of Fruits Production in total gross production 832 0.158 0.201 0.00109 1.264 Share of Vegetables Production in total gross production 865 0.127 0.141 0.00296 1.498 58 (6) APE -0.035*** (0.011) -0.053 (0.042) 0.135 (0.139) -0.009 (0.019) -0.028* (0.016) -0.049 (0.042) -0.009 (0.012) -0.017 (0.016) 0.004 (0.014) 0.227* (0.138) -0.002 (0.014) Table 2.4 Baseline estimation results for the farm labor's share of agricultural income Estimation Model GVC indicator OLS PQMLE (2) (1) (3) Coeff. -0.029*** -0.033*** -0.125*** -0.034*** -0.126*** (0.041) (5) Coeff. (4) APE (0.041) (0.011) (0.011) (0.011) GDP per capita (PPP) -0.069*** -0.071*** (0.011) (0.011) Urbanization (% total population) 0.019 (0.017) 0.032* (0.018) Labor Productivity 0.035*** 0.035*** (0.006) (0.006) -0.138 (0.155) 0.233 (0.450) -0.065 (0.062) -0.038 (0.043) 0.064 (0.124) -0.018 (0.017) -0.194 (0.153) 0.489 (0.506) -0.034 (0.072) Producer Price Index -0.040*** (0.010) -0.018 (0.011) -0.147*** -0.041*** (0.049) (0.014) -0.103* (0.059) -0.213 (0.152) -0.038 (0.044) -0.065 (0.059) 0.013 (0.046) 0.920* (0.516) -0.025 (0.048) -0.059 (0.042) -0.011 (0.012) -0.018 (0.016) 0.003 (0.013) 0.253* (0.142) -0.007 (0.013) -0.179 (0.151) -0.035 (0.044) -0.061 (0.059) 0.014 (0.049) 0.826* (0.501) -0.007 (0.051) 0.227** (0.103) 0.062** (0.028) 0.265** (0.110) 0.073** (0.030) 0.080** (0.038) 0.014 (0.036) 0.022** (0.010) 0.004 (0.009) 0.082** (0.038) -0.006 (0.038) 0.023** (0.011) -0.002 (0.010) Net Capital Stocks Share of Cereals Production 0.009*** 0.009*** (0.002) (0.002) 0.003 (0.003) 0.004 (0.003) Share of Vegetables Production -0.017*** -0.018*** (0.004) (0.004) Share of Fruits Production Access to electricity (% of population) R&D expenditure (% of GDP) Government exp. on education (% of GDP) FTAH FAAFH Constant 0.002 (0.002) 0.003 (0.091) -0.003 (0.004) 0.008 (0.014) 0.017 (0.011) -0.004 (0.010) 0.002 (0.002) -0.021 (0.091) -0.003 (0.005) 0.011 (0.014) 0.019* (0.011) -0.007 (0.011) 0.476 (0.416) No No 674 0.287 0.458 (0.420) Yes No 674 0.312 Year FE Time averages Observations R-squared Notes: Robust standard errors in parentheses. APE is the Average Partial Effects.*** p<0.01, ** p<0.05, * p<0.1. Independent variables are in logarithmic terms. R-squared in the PQMLE estimation is the Pseudo R-squared. No Yes 674 Yes Yes 674 -0.288 (1.193) Yes Yes 674 0.201 -0.468 (1.236) No Yes 674 0.134 59 Table 2.5 Baseline estimation results for the farm labor's share of total labor income Estimation Model GVC indicator OLS (1) (2) -0.003 (0.005) -0.003 (0.005) GDP per capita (PPP) -0.059*** (0.005) -0.059*** (0.005) Urbanization (% total population) -0.059*** (0.010) -0.057*** (0.011) Labor Productivity Producer Price Index Net Capital Stocks 0.014*** (0.003) 0.014*** (0.003) 0.0005 (0.005) 0.003 (0.005) 0.006*** (0.001) 0.006*** (0.001) Share of Cereals Production Share of Vegetables Production Share of Fruits Production -0.001 (0.002) -0.004* (0.002) -0.001 (0.001) -0.001 (0.002) -0.004* (0.002) -0.001 (0.001) Access to electricity (% of population) -0.122** (0.051) -0.123** (0.051) R&D expenditure (% of GDP) -0.001 (0.002) -0.001 (0.002) -0.019*** -0.0195*** (0.006) (0.006) PQMLE (3) Coeff. -0.063* (0.038) -0.097 (0.175) -0.184 (1.042) -0.041 (0.079) -0.094 (0.075) -0.119 (0.141) 0.022 (0.048) 0.047 (0.091) -0.025 (0.067) 0.278 (0.355) 0.049 (0.078) 0.066 (0.132) (4) APE -0.006* (0.004) -0.009 (0.017) -0.018 (0.102) -0.004 (0.008) -0.009 (0.007) -0.0117 (0.014) 0.002 (0.005) 0.005 (0.009) -0.002 (0.007) 0.027 (0.035) 0.005 (0.008) 0.006 (0.013) (5) Coeff. -0.056 (0.038) -0.137 (0.180) -0.035 (1.103) -0.016 (0.090) -0.020 (0.077) -0.082 (0.141) 0.028 (0.047) 0.052 (0.089) -0.027 (0.069) 0.189 (0.374) 0.070 (0.090) 0.084 (0.133) (6) APE -0.005 (0.004) -0.013 (0.018) -0.003 (0.108) -0.002 (0.009) -0.002 (0.008) -0.008 (0.014) 0.003 (0.005) 0.005 (0.009) -0.003 (0.007) 0.018 (0.037) 0.007 (0.009) 0.008 (0.013) Government exp. on education (% of GDP) FTAH FAAFH Constant -0.020* (0.011) -0.019* (0.011) -0.043 (0.035) -0.010 (0.008) -0.043 (0.035) -0.010 (0.008) -0.123*** (0.011) -0.123*** (0.011) -0.891*** -0.133*** -0.905*** -0.136*** (0.036) (0.008) (0.037) (0.008) 1.390*** (0.220) No No 674 0.735 1.386*** (0.222) Yes No 674 0.736 2.375** (1.052) No Yes 674 Year FE Time averages Observations R-squared Notes: Robust standard errors in parentheses. APE is the Average Partial Effects.*** p<0.01, ** p<0.05, * p<0.1. Independent variables are in logarithmic terms. R-squared in the PQMLE estimation is the Pseudo R-squared. Yes Yes 674 No Yes 674 2.511** (1.060) Yes Yes 674 60 Table 2.6 Heterogeneity by type of linkage Dependent variable Model Forward Participation I (1) Coeff. 0.035 (0.040) (2) APE 0.009 (0.011) Backward participation -0.094*** -0.026*** II (3) Coeff. (4) APE -0.127*** -0.012*** (0.035) 0.047 (0.039) -0.143 (0.178) -0.057 (1.026) -0.015 (0.091) -0.033 (0.074) -0.110 (0.139) 0.022 (0.048) 0.039 (0.085) -0.030 (0.070) 0.446 (0.369) 0.070 (0.086) 0.090 (0.130) -0.024 (0.033) (0.003) 0.005 (0.004) -0.014 (0.017) -0.006 (0.101) -0.001 (0.009) -0.003 (0.007) -0.011 (0.014) 0.002 (0.005) 0.004 (0.008) -0.003 (0.007) 0.044 (0.036) 0.007 (0.008) 0.009 (0.013) -0.006 (0.008) -0.887*** -0.132*** (0.036) (0.008) (0.033) -0.198 (0.158) 0.435 (0.535) -0.034 (0.072) -0.093 (0.059) -0.169 (0.150) -0.031 (0.044) -0.063 (0.060) 0.015 (0.051) 0.715 (0.501) -0.010 (0.050) (0.009) -0.055 (0.043) 0.120 (0.147) -0.009 (0.019) -0.025 (0.016) -0.047 (0.041) -0.009 (0.012) -0.017 (0.017) 0.004 (0.014) 0.197 (0.138) -0.003 (0.014) 0.252** (0.107) 0.069** (0.029) 0.059 (0.038) -0.027 (0.038) 0.017 (0.011) -0.007 (0.011) GDP per capita (PPP) Urbanization (% total population) Labor Productivity Producer Price Index Net Capital Stocks Share of Cereals Production Share of Vegetables Production Share of Fruits Production Access to electricity (% of population) R&D expenditure (% of GDP) Government exp. on education (% of GDP) FTAH FAAFH Constant Observations 674 Notes: Robust standard errors in parentheses. APE is the Average Partial Effects.*** p<0.01, ** p<0.05, * p<0.1. Independent variables are in logarithmic terms, and I include year FE and time averages in all models. Legends: (I) Farm Labor's Share of Agricultural Income; and (II) Farm Labor's Share of Total Labor Income. 674 -0.409 (1.167) 674 2.318** (1.001) 674 61 Table 2.7 Heterogeneity by income level Dependent variable Percentiles 5% 7,101 $US GVC indicator (1) Farm labor's share of agricultural income Forward Backward Farm labor's share of total labor income Forward Backward (2) (3) (5) (6) GVC indicator (4) -0.041*** (0.014) 0.007 (0.013) -0.029*** (0.010) -0.007 (0.005) -0.015** (0.006) 0.004 (0.005) 25% 19,652 $US -0.036*** (0.012) 0.006 (0.011) -0.025*** (0.009) -0.006 (0.004) -0.012*** (0.003) 0.003 (0.004) 50% 29,047 $US -0.034*** (0.011) 0.006 (0.010) -0.024*** (0.008) -0.005 (0.004) -0.011*** (0.003) 0.003 (0.004) 75% 42,657 $US -0.032*** (0.010) 0.005 (0.009) -0.022*** (0.008) -0.005 (0.003) -0.010*** (0.004) 0.003 (0.003) 95% 63,419 $US -0.029*** (0.010) 0.005 (0.009) -0.020*** (0.008) -0.004 (0.003) -0.009** (0.004) 0.003 (0.003) 674 Observations Notes: Coefficients are average partial effects. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 674 674 674 674 674 62 Table 2.8 Heterogeneity by region Dependent Variable Panel A: Asia GVC indicator Backward Participation Forward Participation Panel B: Europe GVC indicator Backward Participation Forward Participation Panel C: Americas GVC indicator Backward Participation Forward Participation Farm Labor's Share of Agricultural Income (2) (1) Farm Labor's Share of Total Labor Income (3) (4) -0.107** (0.040) 0.047*** (0.017) 0.001 (0.031) 0.228** (0.093) -0.016 (0.011) -0.013 (0.019) 0.150*** (0.038) 0.016 (0.016) -0.021 (0.029) -0.008 (0.006) 0.026*** (0.009) 0.021 (0.022) 0.00824 (0.0104) 0.001 (0.024) -0.004 (0.019) 0.210*** (0.034) 0.131*** (0.019) ✓ ✓ ✓ 0.012 (0.013) ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ Control Variables Year FE Type of value chain Notes: Robust standard errors are displayed in parentheses. Significance levels are denoted as follows: *** p<0.01, ** p<0.05, and * p<0.1. For brevity, control variables have been omitted. Africa was excluded from the analysis due to a lack of sufficient data. The estimates were derived using a linear regression model. The regions include a list of countries as follows: (1) Asia: Australia, Brunei Darussalam, Cambodia, China (Hong Kong SAR), India, Indonesia, Israel, Japan, Kazakhstan, Malaysia, Philippines, Republic of Korea, Russian Federation, Saudi Arabia, Singapore, Thailand; (2) Europe: Austria, Belgium, Bulgaria, Croatia, Cyprus, Czechia, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Italy, Latvia, Lithuania, Luxembourg, Malta, Netherlands (Kingdom of the), Norway, Poland, Portugal, Romania, Slovakia, Slovenia, Spain, Sweden, Switzerland, Türkiye, United Kingdom of Great Britain and Northern Ireland; and (3) Americas: Argentina, Brazil, Canada, Chile, Colombia, Costa Rica, Mexico, Peru, United States of America. 63 Table 2.9 Heterogeneity by global agri-food sector for the farm labor's share of agricultural income Model Sector GVC indicator GDP per capita (PPP) Urbanization (% total population) Labor Productivity Producer Price Index Net Capital Stocks Share of Cereals Production Share of Vegetables Production Share of Fruits Production Access to electricity (% of population) R&D expenditure (% of GDP) Government exp. on education (% of GDP) FTAH FAAFH Constant (1) Coeff. (2) APE Agriculture (3) Coeff. (4) APE Fishing -0.012 (0.031) -0.003 (0.009) -0.026 (0.019) -0.007 (0.005) (6) (5) Coeff. APE Food & Beverage -0.174*** -0.048*** (0.040) (0.011) -0.189 (0.157) -0.052 (0.043) -0.194 (0.157) -0.053 (0.043) 0.464 (0.509) 0.128 (0.140) 0.482 (0.500) 0.133 (0.138) -0.034 (0.074) -0.009 (0.020) -0.033 (0.073) -0.009 (0.020) -0.185 (0.153) 0.515 (0.518) -0.035 (0.072) -0.099 (0.061) -0.027 (0.017) -0.102* -0.028* (0.017) (0.060) -0.102* (0.058) -0.178 (0.152) -0.049 (0.042) -0.181 (0.153) -0.050 (0.042) -0.035 (0.044) -0.009 (0.012) -0.035 (0.044) -0.009 (0.012) -0.075 (0.062) -0.021 (0.017) -0.074 (0.061) -0.020 (0.017) 0.015 (0.052) 0.004 (0.014) 0.017 (0.051) 0.005 (0.014) 0.868* 0.239* 0.890* 0.245* (0.137) (0.492) (0.135) (0.497) -0.012 (0.051) -0.003 (0.014) -0.017 (0.052) -0.005 (0.014) -0.179 (0.149) -0.032 (0.045) -0.052 (0.058) 0.013 (0.051) 0.848* (0.502) -0.002 (0.051) 0.257** 0.071** 0.259** 0.071** 0.253** (0.111) (0.112) (0.112) (0.031) (0.031) 0.052 (0.037) 0.015 (0.010) 0.053 (0.036) 0.015 (0.010) 0.090** (0.039) -0.031 (0.037) -0.009 (0.010) -0.030 (0.036) -0.008 (0.010) 0.002 (0.039) -0.051 (0.042) 0.142 (0.143) -0.009 (0.019) -0.028* (0.016) -0.049 (0.041) -0.009 (0.012) -0.014 (0.016) 0.003 (0.014) 0.234* (0.138) -0.0005 (0.014) 0.069** (0.030) 0.025** (0.011) 0.0005 (0.011) Yes Year FE Yes Time averages 674 Observations Notes: Robust standard errors in parentheses. APE is the Average Partial Effects.*** p<0.01, ** p<0.05, * p<0.1. Independent variables are in logarithmic terms. Yes Yes 674 Yes Yes 674 -0.231 (1.154) Yes Yes 674 -0.220 (1.170) Yes Yes 674 -0.362 (1.184) Yes Yes 674 64 Table 2.10 Heterogeneity by global agri-food sector for the farm labor's share of total labor income Model Sector GVC indicator GDP per capita (PPP) Urbanization (% total population) Labor Productivity Producer Price Index Net Capital Stocks Share of Cereals Production Share of Vegetables Production Share of Fruits Production Access to electricity (% of population) R&D expenditure (% of GDP) Government exp. on education (% of GDP) FTAH FAAFH Constant (1) Coeff. (2) APE (3) Coeff. (4) APE Agriculture Fishing -0.036 (0.032) -0.142 (0.179) -0.061 (1.104) -0.016 (0.091) -0.021 (0.077) -0.079 (0.142) 0.027 (0.047) 0.048 (0.089) -0.027 (0.069) 0.208 (0.378) 0.068 (0.090) 0.082 (0.133) -0.046 (0.033) -0.004 (0.003) -0.014 (0.018) -0.006 (0.108) -0.002 (0.009) -0.002 (0.008) -0.008 (0.014) 0.003 (0.005) 0.005 (0.009) -0.003 (0.007) 0.020 (0.037) 0.007 (0.009) 0.008 (0.013) -0.011 (0.008) -0.026 (0.021) -0.139 (0.181) -0.046 (1.097) -0.017 (0.090) -0.024 (0.077) -0.084 (0.143) 0.029 (0.048) 0.049 (0.089) -0.026 (0.069) 0.222 (0.377) 0.065 (0.091) 0.082 (0.133) -0.053 (0.033) -0.002 (0.002) -0.014 (0.018) -0.005 (0.108) -0.002 (0.009) -0.002 (0.008) -0.008 (0.014) 0.003 (0.005) 0.005 (0.009) -0.002 (0.007) 0.022 (0.037) 0.006 (0.009) 0.009 (0.013) -0.013 (0.008) (5) Coeff. (6) APE Food & Beverage -0.008* (0.005) -0.085* (0.047) -0.132 (0.181) -0.021 (1.108) -0.017 (0.089) -0.019 (0.077) -0.082 (0.141) 0.029 (0.047) 0.057 (0.091) -0.028 (0.069) 0.192 (0.373) 0.075 (0.091) 0.078 (0.134) -0.038 (0.037) -0.013 (0.018) -0.002 (0.109) -0.002 (0.009) -0.002 (0.008) -0.008 (0.014) 0.003 (0.005) 0.006 (0.009) -0.003 (0.007) 0.019 (0.037) 0.007 (0.009) 0.008 (0.013) -0.009 (0.009) -0.91*** -0.137*** -0.91*** (0.036) (0.008) (0.037) -0.14*** -0.900*** (0.008) (0.038) -0.135*** (0.009) Year FE Time averages Observations Notes: Robust standard errors in parentheses. APE is the Average Partial Effects.*** p<0.01, ** p<0.05, * p<0.1. Independent variables are in logarithmic terms. Yes Yes 674 Yes Yes 674 Yes Yes 674 2.566** (1.060) Yes Yes 674 2.505** (1.045) Yes Yes 674 2.472** (1.092) Yes Yes 674 65 APPENDIX B: ENDOGENEITY ISSUES In this appendix, I address endogeneity arising from omitted variable bias, measurement error, and reverse causality. Given the similarity in results between the PQMLE and the linear model estimates in section 2.5, I employ the Blundell-Bond System GMM estimator as a robustness check to address potential endogeneity concerns. The Blundell-Bond dynamic panel data estimation method is used to control for unobserved panel-level heterogeneity and the potential endogeneity of explanatory variables, leveraging the temporal dimension of the data to use lagged values as instruments12. From the table below, the findings suggest that the results are robust to endogeneity, as demonstrated in the baseline estimations. This reinforces the validity of our initial conclusions and enhances the credibility of the estimated effects, confirming that our model's specifications are well-suited to capturing the underlying dynamics of the data. 12 It's usually attributed to Arellano and Bover (1995) and Blundell and Bond (1998) who extended the original difference GMM method proposed by Arellano and Bond (1991). This extended version is commonly called the "Blundell-Bond estimator" or "System GMM" estimator. 66 Table 2.11 Blundell-Bond estimation results Dependent Variable (DV) Lagged DV GVC indicator Forward Participation Backward Participation GDP per capita (PPP) Urbanization (% total population) Labor Productivity Producer Price Index Net Capital Stocks Share of Cereals Production Share of Vegetables Production Share of Fruits Production Access to electricity (% of population) R&D expenditure (% of GDP) Government exp. on education (% of GDP) FTAH FAAFH Constant Farm Labor's Share of Agricultural Income (2) (1) 0.244*** 0.258** (0.120) (0.0944) -0.142*** (0.0203) -0.00348 (0.0199) -0.0996*** (0.0257) 0.000390 (0.0423) -0.113 (0.0887) -0.00259 (0.0104) -0.0492*** (0.0132) -0.00889 (0.0297) -0.00264 -0.00244 (0.0398) -0.150 (0.112) 0.00123 (0.0113) -0.0478*** (0.0136) -0.0190 (0.0342) -0.00379 Farm Labor's Share of Total Labor Income (3) 0.464*** (0.159) 0.0312 (0.0263) -0.0459*** (0.0139) -0.0760** (0.0343) 0.00556 (0.00523) -0.000391 (0.00438) -0.00202 (0.00798) 0.000315 (4) 0.458*** (0.177) -0.0107*** (0.00113) 0.0101 (0.0193) -0.0437*** (0.0142) -0.0964*** (0.0344) 0.00542 (0.00460) -0.00173 (0.00438) 0.00210 (0.0102) -0.000155 (0.00174) 0.00114 (0.00528) -0.000416 (0.00494) 0.282* (0.149) -0.0138** (0.00697) -0.00225 (0.00948) -0.210 (0.221) -0.204 (0.221) -0.273 (0.627) Yes 580 86 (0.00181) -0.00125 (0.00515) -0.00250 (0.00455) 0.287** (0.129) -0.0117* (0.00677) -0.00665 (0.00960) -0.110 (0.195) -0.101 (0.196) -0.406 (0.597) Yes 580 86 (0.00523) 0.0112 (0.0103) -0.00234 (0.0107) -0.0342 (0.110) 0.0237* (0.0135) 0.0325 (0.0217) -0.308 (0.658) -0.269 (0.613) 1.440 (0.955) Yes 580 86 (0.00464) 0.0134 (0.00840) 0.00230 (0.00920) -0.0614 (0.0959) 0.0215* (0.0124) 0.0214 (0.0180) -0.218 (0.627) -0.166 (0.573) 1.245 (0.825) Yes 580 86 67 Year FE Observations Number of panels Notes: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Independent variables are in logarithmic terms. CHAPTER 3: BUFFER OR CONDUIT? GLOBAL AGRI-FOOD VALUE CHAINS AND FOOD PRICE TRANSMISSION 68 3.1 Introduction Chapter 3 shifts attention towards the downstream end of the chain, while Chapters 1 and 2 focus on the distribution of value-added and labor income across different segments of domestic agri-food value chains, respectively. It investigates how participation in global value chains affects the transmission of international food price spikes to domestic consumer markets. Periods of global food price fluctuations prompt policy debate on food price transmission—a process via which international food prices are transmitted to domestic markets. The topic gained relevance thrice in recent decades: amid the 2007-2008 food crisis, the inflationary phase following the COVID-19 pandemic, and the Russia-Ukraine war. The transmission of international food price spikes to domestic markets is a critical policy issue for three reasons. First, poor households allocate a large share of their budgets to food (Muhammad et al. 2011; Zezza et al. 2017). Second, the dependence of farmers on imports or exports for income makes them vulnerable to unexpected global price shocks, which can lead to reduced productivity or commercialization difficulties. Third, spikes in food prices may cause political unrest, such as those seen prior to and during the Arab Spring (Bellemare 2015; Bush and Martiniello 2017). The literature on price transmission between various segments of the agri-food value chain (such as farming, manufacturing, and retailing) is substantial and beyond the scope of this paper (see Lloyd (2017) for a review). This study focuses solely on the global dimension of food price transmission, particularly the role of international market integration. Market integration is defined as the extent to which domestic markets are integrated with each other and with international markets (Mundlak and Larson 1992; Baquedano and Liefert 2014; Minot 2014; Ceballos et al. 2017). Trade policy is the main instrument for integrating into global markets. Over the past half- century, the substantial liberalization of food trade through reduced tariffs, quotas, and other trade barriers has led to greater food price transmission. In particular, Myers and Jayne (2012) find that the transmission of South African maize prices to Zambian prices varies depending on import levels. Similarly, Auer and Mehrotra (2014) show that participation in the pan-Asian cross-border production networks impacts the domestic price dynamics in the Asia-Pacific region. Better infrastructure, such as roads and storage facilities, decreases transaction costs and facilitates the diffusion of information and prices (Fackler and Goodwin 2001). Examples of transaction costs include adjustment, search, and menu costs (Meyer and von Cramon‐Taubadel 2005; Frey and Manera 2007). Also, expectations about future price trends can lead to speculative behavior, 69 affecting domestic food prices as well. Brander, Bernauer and Huss (2023) show that trade policy announcements can increase price volatility in the global food markets. The aforementioned research on market integration and food price transmission uses single equation econometric models and leverages natural experiments for empirical analysis. A recent body of research, however, proposes a two-step regression analysis to examine the determinants of the long-run pass-through (LRPT) of food prices. The LRPT is defined as the extent to which changes in international food prices lead to changes in domestic consumer food prices. Bekkers et al. (2017) find that income per capita predominantly explains the variation in the LRPT, especially in low-income countries. In contrast, Elleby and Jensen (2018) challenge the commonly held idea that the world’s poorest countries are the most vulnerable to spikes in international food prices. They find that food prices in middle-income countries respond more to changes in global food prices. However, the existing literature overlooks the impact of real integration into global agri- food value chains (hereafter GVCs) on the food price transmission. This raises the question: to what extent does participation in global agri-food production networks affect the LRPT of international food prices to domestic markets? The central thesis of investigation in this study posits that global value chains would act as conduits by transmitting global price shocks to the domestic markets. Price changes in the global agri-food input markets may transmit down the chain to countries importing these inputs. As a result, production costs rise, and consequently, the prices at which domestic agri-businesses are willing to sell their final consumer products also increase. The contrasting hypothesis posits that the multiple stages of value addition could act as a buffer against global price spikes. That is, if the agri-food products undergo significant processing before reaching the final consumer, price shocks in the global raw commodity market may not fully transmit to domestic markets. This paper contributes to the literature in three ways. First, I provide novel empirical evidence to an emerging body of research on the impacts of participation in GVCs (Balié et al. 2019; Lim 2021; Lim and Kim 2022; Ndubuisi and Owusu 2023). A recent study by Dalheimer, Bellemare and Lim (2023) finds that participation in GVCs has a negative impact on domestic consumer food prices and a positive impact on price volatility. This paper contributes, however, to understanding this relationship by examining a specific mechanism through which global production networks affect domestic consumer prices (i.e., food price transmission). Second, the 70 literature on food price transmission concerned with specific food commodities and individual countries or regions in isolation, such as the United States (Auer and Fischer 2010), Guatemala (de Janvry and Sadoulet 2010), Ghana (Cudjoe, Breisinger and Diao 2010), Sub-Saharan Africa (Minot 2014), and Nigeria (Hatzenbuehler, Abbott and Abdoulaye 2017), among others. Thus, by using a panel dataset of 173 countries and four episodes of international price spikes: 2000-04, 2005-09, 2010-14, and 2015-22, I also contribute to the existing literature. Third, the literature examining the determinants of food price transmission mainly focuses on trade policy and income (Bekkers et al. 2017; Elleby and Jensen 2018). This study contributes to the literature by examining how participation in global agri-food production networks determines the LRPTs of international prices to domestic consumer prices. I also take the analysis one step further by decomposing the GVC participation indicator into three additive components (pure backward, pure forward, and two-sided participation) to examine the relationship between the different types of global sourcing and food price transmission. Pure forward participation refers to countries participating at the origin of the global value chain, with the exported inputs being entirely produced domestically. Conversely, pure backward participation refers to participation at the end of the chain by exporting agri-food goods to final destinations. The two-sided (or mixed) participation occurs when a country is in a more intermediate position of the value chain, using imported inputs to produce exports that are further re-exported by other trading partners. The approach in this study offers a more granular understanding of how countries contribute to the final value of agri-food products, uncovering nuances that aggregate trade data might overlook (Koopman, Wang and Wei 2014). The empirical findings in this chapter suggest that higher two-sided participation in GVCs is associated with a decrease in food price transmission, while pure backward and forward participations increase the transmission of food price spikes. However, a 1% increase in the overall GVC related-trade indicator decreases food price transmission by 1.5%. This study challenges the dominant view that real integration into global production networks is the primary driver of food price transmission. In sum, this paper demonstrates that GVCs play the role of a buffer against international price fluctuations. One important policy implication of this paper suggests that while the conventional response to international food price volatility has been to increase trade barriers, a more effective strategy, however, should promote mixed integration in global agri-food networks. 71 The rest of the chapter is organized as follows. Section 3.2. presents a simple theoretical model. Section 3.3. presents the empirical strategy used in this paper. Results and concluding remarks are presented in subsequent sections. 3.2 Theoretical Framework: A Simple Model This section establishes the theoretical foundation needed to derive testable predictions for the empirical analysis. I build a simple model based on both the outsourcing model from Feenstra and Hanson (1996, 1997) and the conceptual framework of Bekkers et al. (2017). The model focuses on the relationship between participation in the global production networks and food price transmission. This specific feature is absent from the existing theoretical literature on the determinants of food price transmission. 3.2.1 Producers Consider two agri-food inputs 𝑦𝑖, where 𝑖 = 1,2. Each of these inputs is produced using labor (𝑙𝑖) and capital (𝑘𝑖), with concave and linearly homogeneous production functions, 𝑦𝑖 = 𝑓𝑖(𝑙𝑖, 𝑘𝑖). Let 𝑥1 denote the imports of input 1, and 𝑥2 denote the exports of input 2. The price vector of the traded intermediate inputs is denoted by 𝑝 = (𝑝1, 𝑝2). The production function linking the final food product to inputs is given by 𝑦 = 𝑓(𝑦1 + 𝑥1, 𝑦2 − 𝑥2), where 𝑓(. ) is a concave and is linearly homogeneous. With perfect competition, the revenue function for the industry is maximized subject to the resource constraints: 𝐹(𝑙𝑖, 𝑘𝑖, 𝑝𝑦, 𝑝) ≡ max 𝑥𝑖,𝑙𝑖,𝑘𝑖  𝑝𝑦𝑓(𝑦1 + 𝑥1, 𝑦2 − 𝑥2) − 𝑝1𝑥1 + 𝑝2𝑥2, subject to 𝑦𝑖 = 𝑓𝑖(𝑙𝑖, 𝑘𝑖) 𝑙1 + 𝑙2 = 𝑙, 𝑎𝑛𝑑 𝑘1 + 𝑘2 = 𝑘 (1) where 𝑝𝑦 is the price of the final good, and 𝑙1 + 𝑙2 = 𝑙 and 𝑘1 + 𝑘2 = 𝑘 represent the total factor usage in the food industry. To examine how the price of the final good reacts to a change in input prices, let 𝑐(𝑝1, 𝑝2) denote the unit-cost function that is dual to (1). Log differentiating the price of the final good that satisfies 𝑝𝑦 = 𝑐(𝑝1, 𝑝2) and applying Shephard’s lemma, we get 𝑝ˆ𝑦 = 𝑠1𝑝ˆ1 + 𝑠2𝑝ˆ2 (2) 𝑥𝑖𝑝𝑖 𝑐(𝑝1,𝑝2) is the cost-share of input 𝑖. Eq. (2) states that the relative change in the price of where 𝑠𝑖 ≡ the final agri-food product is a weighted average of the change in the agri-food inputs prices. 3.2.2 Consumers How does a change in the price of traded agri-food inputs affect changes in domestic 72 consumer prices? Let’s define the consumption of food products in a country by 𝑞𝑓 ≡ 𝑦 + 𝑞𝑠, where 𝑦 is the final good and 𝑞𝑠 represents food services such as shipping, processing, storage, and distribution. Suppose 𝑞𝑓 is a homothetic function of y and 𝑞𝑠. We can write the domestic price index of food consumption as follows: 𝑝𝑓 = 𝜃(𝑝𝑦, 𝑝𝑠) (3) where 𝜃(∙) is an aggregation function. Log differentiating Eq. (3) and applying Shephard's lemma, the relative change of the domestic price index can be expressed as a function of the relative change of the final agri-food good and the price of food services: 𝑝ˆ𝑓 = 𝑠𝑓𝑝ˆ𝑦 + (1 − 𝑠𝑓)𝑝ˆ𝑠 (4) where 𝑠𝑓 = 𝑝𝑦y 𝑝𝑓𝑞𝑓 is the share of agri-food product y in the total food consumption. 3.2.3 Theoretical Predictions I deduce two important predictions from the model. First, a rise in the price of imported inputs increases the cost-share of input 1, which leads to a higher price for the final product 𝑦. From Eq. (4), this increase in 𝑝𝑦 enlarges the share of product 𝑦 in the total food consumption, which subsequently increases consumer food prices, 𝑝ˆ𝑓 > 0. This suggests that consumers in countries involved in the latter stages of the global value chain are more susceptible to international price fluctuations. Similarly, an increase in the price of exports also leads to higher consumer food prices. Second, in relative terms, a rise in the relative price of imported inputs (𝑝ˆ1 − 𝑝ˆ2 > 0) decreases the price of the final food product relative to the imported input but increase it relative to the price of the exported input, 𝑝ˆ2 < 𝑝ˆ𝑦 < 𝑝ˆ1. Furthermore, an increase in the relative price of the final product yields 𝑝ˆ𝑠 < 𝑝ˆ𝑓 < 𝑝ˆ𝑦. Equivalently, the consumer prices increase but no more than changes in the price of the imported input, 𝑝ˆ𝑓 < 𝑝ˆ1. In sum, the mix of imports and exports of inputs determines how susceptible a country is to international food price fluctuations. However, countries exclusively importing or exporting inputs experience more direct and immediate impacts on their domestic food prices. Subsequent sections use real-world data to test these predictions on how GVCs affect food price transmission. 3.3 Empirical Framework Food price transmission is unobservable. Therefore, I first estimate the country-specific price transmission measures using Eq. (4). Then, I examine how participation in GVCs impacts 73 the LRPTs to test the theoretical predictions derived in the previous section. 3.3.1 Food price transmission estimation I estimate a parsimonious model including lags of the dependent variable. Variables are expressed as the first difference of the log of food prices following a large body of research on food price transmission (e.g., Cudjoe et al. (2010); Bekkers et al. (2017); and Elleby and Jensen (2018), among others). The model for a given country and period is specified as follows: L 𝑀 𝑁 Δ log 𝑝𝑡 𝑑 = 𝛼 + ∑   𝑙=1  𝜆𝑙Δ log 𝑝𝑡−𝑙 𝑑 + ∑   𝑚=0  𝛾𝑚Δ log 𝑝𝑡−𝑚 𝑤 + ∑   𝑛=0  𝛿𝑛Δ log 𝑒𝑡−𝑛 + 𝜇𝑡 + 𝜀𝑡 (5) 𝑑 represents monthly domestic consumer food prices. I use data on the Consumer Price Index 𝑝𝑡 (CPI) as computed by FAO and available from the FAOSTAT database. The FAOSTAT monthly food CPI data captures the change in the cost of food over time. 𝑝𝑡 𝑤 is the international FAO Food Price Index (FFPI), which records monthly changes in international prices of five commodity groups (e.g., cereals, oils and fats, dairy, meat, and sugar) weighted by the average export shares of each group. The exchange rate 𝑒𝑡 is defined as the price of dollars in local currencies. 𝜇𝑡 are monthly time dummies, used to capture the seasonal effects that are fixed across time. Including time dummies allows more flexibility in time series data de-seasoning without imposing rigid structural assumptions (Lovell 1963). The dataset consists of monthly data series and contains a complete set of time series from January 2000 to December 2022. For each country, I estimate the model for four periods: 2000- 04, 2005-09, 2010-14, and 2015-22. I employ an Autoregressive Distributed Lag (ADL) model to estimate the LRPTs. ADL models are only appropriate to use in the case of unidirectional causality and when food price indices are less likely to be cointegrated (see Ianchovichina, Loening and Wood (2014) for further discussion). For this, I have tested all series for unit roots using Augmented Dickey-Fuller tests prior to estimation. The test fails to reject the null hypothesis of a unit root at the 1% level for only 4.6% of the time series. This paper uses the same lag structure for all variables in (5) for consistency and simplicity in the model structure, 𝐿 = 𝑀 = 𝑁. Previous research uses a fixed number of lags across countries and periods. However, the optimal number of lags is country and period-specific and using a common lag for all series may provide unreliable estimates of the LRPTs. For this reason, the lag structure in this paper is based on the Akaike Information Criterion (AIC) with an average number of lags across all series of 2.67. Next, the formula used for computing the LRPT follows Elleby and Jensen (2018): 74 𝑑 ∂𝑝𝑡 𝑤 + ∂𝑝𝑡  𝑀 ∑ 𝑚=0  𝛾𝑚 L 1 − ∑   𝑙=1  𝜆𝑙 𝜓 can be easily estimated as a nonlinear combination of parameter estimates. The estimated value 𝑑 ∂𝑝𝑡+2 𝑤 + ⋯ ≡ 𝜓 = ∂𝑝𝑡 (6) 𝑑 ∂𝑝𝑡+1 𝑤 + ∂𝑝𝑡 𝜓ˆ represents the extent to which changes in international food prices lead to changes in domestic consumer food prices. Table 3.1 provides descriptive statistics of the estimates. 3.3.2 Measurement of GVCs indicators Interest in trade statistics and supply and demand linkages between countries has increased in response to the growing global fragmentation of production (see Koopman, Wang and Wei (2014) for a review). In this regard, Giunta, Montalbano and Nenci (2022) show that the available inter-country input–output data can be used to compensate for the scarcity of firm-level data for evidence-based GVC analyses. However, standard approaches to measuring GVC participation have two main biases (Borin, Mancini and Taglioni 2021). First, backward linkages are systematically higher than forward linkages, and forward and backward participation do not balance out at the global level. This leads to an overestimation of the degree of backward integration. Second, previous methodologies neglect the producer perspective in favor of the exporter perspective, leading to an underestimation of some sectors' participation, notably services. To address these issues, this paper follows the methodology proposed by Borin, Mancini and Taglioni (2021). Specifically, I decompose GVC related-trade into three additive components (pure backward, pure forward, and two-sided participation). Countries may participate in global agri- food production activities at the beginning of the value chain, where they export domestic value- added that is further re-exported by the partner (pure forward participation). Conversely, pure backward participation refers to participation at the end of the chain, relying on imported inputs to export goods and services that are not re-exported by the partner. The two-sided (or mixed) participation occurs when a country is in a more intermediate position in the value chain, using imported inputs to produce its own exports, which are further re-exported by the partner (see Figure 3.1 for an example). Country 𝑖’s overall participation in GVCs in year 𝑡 with partner 𝑗 ≠ 𝑖 and across all sectors 𝑠 can be computed using the following formula: GVCij = GVCPureForwardij + GVCPureBackwardij + GVCTwoSidedij I compute participation in GVCs using the UNCTAD-Eora Global Value Chain Database for 173 countries from 2000 to 2015. I retrieve data on agriculture, fishing, and food & beverage 75 sectors, as defined by the International Standard Industrial Classification. To span the estimates of LRPT, I use the Asian Development Bank (ADB) estimates of GVC participation to supplement the GVC participation data from EORA. ADB uses nominal multiregional input-output tables (MRIOTs) for 62 economies, from 2007 to 2022 are expressed in constant (base year 2010) USD prices. Table 3.1 presents descriptive statistics of the GVC indicators. 3.3.3 Determinants of the LRPTs I regress the estimated LRPTs from Eq. (6) on a set of explanatory variables as follows: 𝜓ˆ 𝑖𝜏 = 𝛽0 + 𝛽1log (𝐺𝑉𝐶)𝑖𝜏 + 𝛽2 ′ 𝑋𝑖𝜏 + 𝜇𝜏 + 𝜖𝑖𝜏 (7) where 𝑖 indexes the countries and 𝜏 the periods, 𝜓ˆ 𝑖𝜏 is the estimated LRPT for country 𝑖, 𝛽0 is the intercept, 𝛽1 is the coefficient on the log of average GVC participation in period 𝜏, 𝑋𝑖𝜏 is a vector of covariates in logarithmic terms, 𝜇𝜏 are time fixed effects, and 𝜖𝑖𝜏 is the error term with mean zero. I exclude country fixed effects similar to Chapter 1, which is appropriate for several reasons. First, country fixed effects consume degrees of freedom, which might be a concern given the limited number of observations per country in the dataset. In the case where certain countries are underrepresented, including country fixed effects can lead to unreliable estimates. Second, this study’s interest is in examining global trends rather than country-specific effects. That is, the interactions between countries are an important aspect of the study. Including country fixed effects might not capture these cross-country dynamics. The set of covariates includes variables such as the ratio of food imports to food absorption (calculated as food production plus food imports minus food exports) to serve as a proxy for the share of tradable food products in a country. I also control for national-level food consumption as well as income, measured as Gross Domestic Product (GDP) per capita in US dollars. Additionally, I use a dummy variable for landlocked countries (equals 1 if a country is landlocked and 0, otherwise) to account for geographical constraints on trade. Data for the control variables is available from the FAOSTAT and World Bank Indicators platforms and Table 3.1 provides descriptive statistics. This study is interested in testing the null hypothesis 𝐻0 ∶ 𝛽1 = 0 versus the alternative hypothesis 𝐻𝑎: 𝛽1 ≠ 0. To estimate the model, I employ Weighted Least Squares (WLS), where the weights are the inverse of the standard errors of the estimated price transmission. OLS with heteroscedasticity-robust standard errors serves as an alternative to WLS. For a more detailed discussion on this matter see Bekkers et al. (2017). 76 3.4 Results and Discussion 3.4.1 Baseline Results Table 3.2 presents WLS estimation results of the relationship between participation in GVCs and food price transmission. Across models (1) to (5), the coefficient for participation in GVCs is significant but varies in magnitude. Other control variables, such as the GDP per capita, total population, rural population, etc., show expected signs and are significant in various models. Focusing on model (5), the most parsimonious specification, a 1% increase in participation in GVCs is associated with a 1.5% decrease in food price transmission. There are several possible interpretations of these findings. First, open food trade increases efficiency in both the production and distribution of food commodities across countries, leveraging the principle of comparative advantage as articulated by Ricardo. This principle suggests that food is optimally produced in regions where there are relative efficiency gains due to lower opportunity costs. Another source of efficiency gains comes from the fact that countries that are deeply integrated into these chains benefit from global technology spillover (Coe and Helpman 1995; Coe, Helpman, and Hoffmaister 1997; Coe, Helpman, and Hoffmaister 2009) and better production practices (Griffith, Harrison, and van Reenen 2006). These efficiency gains can stabilize prices by reducing costs and minimizing the impact of supply chain disruptions on final product prices. Second, higher participation in GVCs allows countries to diversify their sources of food. This diversification can buffer against local or regional supply shocks (e.g., poor harvests, natural disasters, or political instability) that might otherwise cause significant price fluctuations. For instance, if a country's economy faces economic downturns or inflationary pressures, its integration into GVCs might provide stability in food prices through continued access to global supply chains. 3.4.2 Positioning in GVCs In this section, I analyze the distinct effects of pure backward, mixed and pure forward participation on the food price transmission estimates. I perform a regression analysis similar to the baseline model but break down the total GVC participation indicator into three additive components. The model is re-written as follows: 𝜓ˆ 𝑖𝜏 = 𝛽0 + 𝛽B1log (𝐵𝑎𝑐𝑘𝑤𝑎𝑟𝑑)𝑖𝜏 + 𝛽M1log (𝑀𝑖𝑥𝑒𝑑)𝑖𝜏 + 𝛽F1log (𝐹𝑜𝑟𝑤𝑎𝑟𝑑)𝑖𝜏 + 𝛽2 ′ 𝑋𝑖𝜏 + 𝜇𝜏 + 𝜖𝑖𝜏 (8) Table 3.3 shows that both pure backward and forward participations increase food price 77 transmission by 1.8% and 1.9%, respectively. The positive effects suggest that countries specializing in the initial (or final) stages of the value chain might experience higher food price transmission, possibly due to the limited trade linkages at the early (or late) stages of production. Conversely, across all models, two-sided participation has a negative effect on food price transmission. Specifically, a 1% increase in two-sided participation decreases the food price transmission by 2%. The overall negative effect seen in the baseline estimation results was likely driven by the strong negative association of two-sided participation, which may have masked the positive impacts of pure backward and forward participation. By decomposing GVC participation, we get a more detailed picture of how different types of real integration in the global value chain influence food price transmission. The main finding in this section suggests that countries positioned more intermediately in the global production network—using imported inputs for exports that are subsequently re- exported—experience a decrease in food price transmission. The two explanations I provided in the previous section also apply to the estimate on two-sided participation. Specifically, open food trade through GVC participation enhances production and distribution efficiency and enables diversification of trade relationships. This could be because such countries can better manage global price spikes through diversified trade relations and more balanced import-export activities. Two-sided GVC participants that are both suppliers and users of intermediate agricultural and food inputs (e.g., mixed ingredients, basic packaging of agricultural inputs, etc) can switch suppliers or find alternative markets for their exports of intermediate inputs more easily, potentially dampening the impact of food price spikes in the international market. The argument provided holds true in cases where there are multiple suppliers or clients for commodities, and price spikes affect only some commodities but not all. However, in other cases, such as with highly specialized products such as Nutella (illustrated in Figure 3.1), the dynamics can be different. Nutella relies heavily on specific ingredients such as hazelnuts and cocoa, where the market might be more constrained due to the limited number of producers. In such cases, the ability to switch suppliers or find alternative markets might be limited, making it harder to mitigate the impact of global price spikes on such specialized products. That is, the aggregated nature of the data used in this study might not fully capture the specificity of some unique supply chains such as Nutella. As such, there's a recognized need for further research focusing specifically on these products. However, this study's findings are important for developing strategies at the macro 78 level that leverage global trade to stabilize domestic food prices. Policymakers can use insights in this chapter to encourage two-sided GVC participation, such as supporting trade policies that facilitate easier access to a diversified range of inputs and markets. 3.4.3 Robustness Checks and Discussion (a) Optimal Lag Structure The optimal lag structure in this paper is based on the Akaike Information Criterion (AIC) because the goal is predictive modeling of food price transmission where the out-of-sample prediction is the main concern of this study’s empirical analysis. In contrast, the Bayesian Information Criterion (BIC) is generally preferred in explanatory modeling contexts and is recommended for larger sample sizes as it more heavily penalizes model complexity. To test this claim, I graphed the number of time series with their optimal lag structure for different selection criteria. Figure B.3.1 shows that AIC values are relatively close to the Final Prediction Error (FPE) values across the different numbers of lags. Since FPE directly measures out-of-sample forecast errors, its closeness to AIC suggests that AIC is also capturing the predictive power of the models well. AIC is known for balancing the model fit with the number of parameters. In predictive modeling, it's important not to overfit the data, as this can lead to poor out-of-sample performance. The AIC's penalization for the number of parameters helps to correct overfitting while still aiming for a good fit, which is crucial for the first regression of the analysis in this paper. In sum, the stated goal of predictive modeling supports the use of AIC which appears to be an appropriate criterion for model selection in this study. It finds a middle ground between the complexity and predictive power, which is essential in generating accurate long-run pass-throughs estimates to study the relationship between participation in GVCs and food price transmission. (b) Reverse Causality Previous estimations assume that the direction of causation predominantly goes from GVC participation to food price transmission. To test this, I include lagged dependent variables in the baseline regression model to isolate the effect of GVC participation on current food price transmission. The two lags model shows the highest F-statistic (252.39) and R-squared (0.95). In sum, the estimate of the impact of participation in GVCs on food price transmission in the previous section is robust to reverse causality (see Appendix B for full results). This is natural because we would expect that the process of integrating into or increasing participation in GVCs usually 79 precedes changes in how international prices are transmitted to domestic markets. There can be a time lag between when a country becomes part of a global production network and when this participation starts to impact domestic price transmission. This lag suggests a one-way impact from GVC participation to price transmission. Furthermore, decisions to participate in GVCs are often strategic or policy-driven, influenced by geopolitics, economic policies, or corporate strategies aiming to leverage global markets. These decisions are independent of the food price transmission, which is more of a market-driven outcome than a factor influencing policy or strategic choices. (c) Difference in GVC Measurement Methods To test the effect of using two different databases in estimating GVC participation, I dropped the last period that used data from ADB. Overall, the coefficients for GVC participation and other control variables show similar magnitudes and significance levels to baseline estimates. The coefficients for each type of GVC participation (backward, two-sided, and forward) are also similar in magnitude and significance across both sets of results. The empirical results are robust to the inclusion of ADB data (see Appendix B for estimation tables). 3.5 Concluding Remarks and Policy Recommendations Global food price transmission poses significant policy challenges to governments, particularly during global crises such as the 2007-2008 food crisis, post-COVID-19 inflation, and the Russia-Ukraine war. This issue is crucial due to its impact on poor households that are vulnerable to price spikes. The existing research on the determinants of food price transmission overlooks the role of participation in global agri-food production networks. This study examined whether GVCs act as conduit or buffer against spikes in international food prices. To do this, I first estimated country-specific long-run pass-throughs, defined as the extent to which changes in international food prices lead to changes in domestic consumer food prices. I estimated the long-run pass-throughs for 173 countries and four episodes of international price spikes. The second step regression examined how participation in global agri-food production networks impacts the long-run pass-through estimates. Specifically, I decomposed the GVC related-trade into three additive components. Countries may participate in global agri-food production activities at the beginning of the value chain, where they export domestic value-added that is further re-exported by the partner (pure forward participation). Conversely, pure backward participation refers to participation at the end of the chain, relying on imported inputs to export goods and services that are not further re-exported by the partner. The two-sided or mixed 80 participation occurs when a country is in a more intermediate position of the value chain, using imported inputs to produce its own exports, which are further re-exported by the partner. Using a two-step regression analysis, several key findings emerged from this chapter. First, the overall participation in GVCs is associated with a decrease in food price transmission. Second, both pure backward and forward participations increased food price transmission by 1.8% and 1.9%, respectively. Conversely, I found that two-sided participation has a negative effect on food price transmission. Specifically, a 1% increase in two-sided participation decreases the food price transmission by 2%. The findings suggest that a more central positioning in GVCs contributes to domestic price stabilization, probably due to increased diversification in food sources that act as a buffer against global supply shocks. That is, countries with higher two-sided participation can better manage international price volatility through diversified trade relations and more balanced import-export activities. Recent discussions on agri-food supply chain disruptions suggest that increased real integration in global agri-food production networks has led to greater transmission of international food prices to domestic markets. This study challenges the dominant view that GVCs are the primary driver of food price transmission. Based on the empirical findings, protectionist trade measures aimed at safeguarding domestic markets from the adverse effects of international food price spikes may inhibit the beneficial effects of two-sided participation in global agri-food production networks. Therefore, policy efforts should shift towards promoting a more balanced integration in global agri-food production networks, leveraging the stabilizing effect of two-sided participation in GVCs on domestic food prices. 81 REFERENCES Auer, R., and A.M. Fischer. 2010. “The effect of low-wage import competition on U.S. inflationary pressure.” Journal of Monetary Economics 57(4):491–503. Auer, R.A., and A. Mehrotra. 2014. “Trade linkages and the globalisation of inflation in Asia and the Pacific.” Journal of International Money and Finance 49:129–151. Balié, J., D. Del Prete, E. Magrini, P. Montalbano, and S. Nenci. 2019. “Does Trade Policy Impact Food and Agriculture Global Value Chain Participation of Sub‐Saharan African Countries?” American Journal of Agricultural Economics 101(3):773–789. Baquedano, F.G., and W.M. Liefert. 2014. “Market integration and price transmission in consumer markets of developing countries.” Food Policy 44:103–114. Bekkers, E., M. Brockmeier, J. Francois, and F. Yang. 2017. “Local Food Prices and International Price Transmission.” World Development 96:216–230. Bellemare, M.F. 2015. “Rising Food Prices, Food Price Volatility, and Social Unrest.” American Journal of Agricultural Economics 97(1):1–21. Borin, A., M. Mancini, and D. Taglioni. 2021. “Measuring exposure to risk in global value chains.” Policy Research Working Paper N°9785. Brander, M., T. Bernauer, and M. Huss. 2023. “Trade policy announcements can increase price volatility in global food commodity markets.” Nature Food 4(4):331–340. Bush, R., and G. Martiniello. 2017. “Food Riots and Protest: Agrarian Modernizations and Structural Crises.” World Development 91:193–207. Ceballos, F., M.A. Hernandez, N. Minot, and M. Robles. 2017. “Grain Price and Volatility Transmission from International to Domestic Markets in Developing Countries.” World Development 94:305–320. Coe, D. T., and E. Helpman. 1995. “International R&D spillovers.” European Economic Review 39(5), 859-887. Coe, D. T., E. Helpman, and A. W. Hoffmaister. 1997. “North-south R & D spillovers.” The Economic Journal 107(440), 134-149. Coe, D. T., E. Helpman, and A. W. Hoffmaister. 2009. “International R&D spillovers and institutions.” European Economic Review 53(7), 723-741. Cudjoe, G., C. Breisinger, and X. Diao. 2010. “Local impacts of a global crisis: Food price transmission, consumer welfare and poverty in Ghana.” Food Policy 35(4):294–302. Dalheimer, B., M.F. Bellemare, and S. Lim. 2023. “Global Agricultural Value Chains and Food Prices.” Working Paper. 82 Elleby, C., and F. Jensen. 2018. “Food Price Transmission and Economic Development.” The Journal of Development Studies 55(8):1708–1725. Fackler, P.L., and B.K. Goodwin. 2001. “Chapter 17 Spatial price analysis.” In pp. 971–1024. Feenstra, R., and G. Hanson. 1996. “Foreign Investment, Outsourcing and Relative Wages.” In R. Feenstra, G. M. Grossman, and D. A. Irwin, eds. The Political Economy of Trade Policy: Papers in Honor of Jagdish Bhagwati. MIT Press, pp. 89–127. Feenstra, R.C., and G.H. Hanson. 1997. “Foreign direct investment and relative wages: Evidence from Mexico’s maquiladoras.” Journal of International Economics 42(3–4):371–393. Frey, G., and M. Manera. 2007. “Econometric models of asymmetric price transmission.” Journal of Economic Surveys 21(2):349–415. Giunta, A., P. Montalbano, and S. Nenci. 2022. “Consistency of micro- and macro-level data on global value chains: Evidence from selected European countries.” International Economics 171:130–142. Griffith, R., R. Harrison, and J. Van Reenen. 2006. “How special is the special relationship? Using the impact of US R&D spillovers on UK firms as a test of technology sourcing.” American Economic Review 96(5), 1859-1875. Hatzenbuehler, P.L., P.C. Abbott, and T. Abdoulaye. 2017. “Price Transmission in Nigerian Food Security Crop Markets.” Journal of Agricultural Economics 68(1):143–163. Ianchovichina, E.I., J.L. Loening, and C.A. Wood. 2014. “How Vulnerable are Arab Countries to Global Food Price Shocks?” The Journal of Development Studies 50(9):1302–1319. de Janvry, A., and E. Sadoulet. 2010. “The Global Food Crisis and Guatemala: What Crisis and for Whom?” World Development 38(9):1328–1339. Koopman, R., Z. Wang, and S.-J. Wei. 2014. “Tracing Value-Added and Double Counting in Gross Exports.” American Economic Review 104(2):459–494. Lim, S. 2021. “Global Agricultural Value Chains and Structural Transformation.” Working Paper. Lim, S., and S.W. Kim. 2022. “Global agricultural value chains and employment growth.” Journal of the Agricultural and Applied Economics Association 1(4):402–418. Lloyd, T. 2017. “Forty Years of Price Transmission Research in the Food Industry: Insights, Challenges and Prospects.” Journal of Agricultural Economics 68(1):3–21. Lovell, M.C. 1963. “Seasonal Adjustment of Economic Time Series and Multiple Regression Analysis.” Journal of the American Statistical Association 58(304):993. Meyer, J., and S. von Cramon‐Taubadel. 2005. “Asymmetric Price Transmission: A Survey.” Journal of Agricultural Economics, 55(3):581–611. 83 Minot, N. 2014. “Food price volatility in sub-Saharan Africa: Has it really increased?” Food Policy 45:45–56. Muhammad, A., J.L. Seale, B. Meade, and A. Regmi. 2011. “International Evidence on Food Consumption Patterns: An Update Using 2005 International Comparison Program Data.” International Food Consumption Patterns and Global Drivers of Agricultural Production:1–104. Mundlak, Y., and D.F. Larson. 1992. “On the Transmission of World Agricultural Prices.” The World Bank Economic Review 6(3):399–422. Myers, R.J., and T.S. Jayne. 2012. “Multiple‐Regime Spatial Price Transmission with an Application to Maize Markets in Southern Africa.” American Journal of Agricultural Economics 94(1):174–188. Ndubuisi, G., and S. Owusu. 2023. “Global Value Chains, Job Creation, and Job Destruction among Firms in South Africa.” STEG Working Paper. Zezza, A., C. Carletto, J.L. Fiedler, P. Gennari, and D. Jolliffe. 2017. “Food counts. Measuring food consumption and expenditures in household consumption and expenditure surveys (HCES). Introduction to the special issue.” Food Policy 72:1–6. 84 APPENDIX A: TABLES AND FIGURES Turkey (hazelnuts) Brazil (sugar) Malaysia and Indonesia (Palm Oil) d r a w r o F n o i t a p i c i t r a P Ivory Coast (cocoa) New Zealand (Skim Milk Powder) Madagascar (Vanilla Flavoring) d e d i S - o w T n o i t a p i c i t r a P United States, China, and India: final production, marketing, sales, distribution, and consumer engagement. Australia and Canada: localized marketing strategies, distribution, and sales. d r a w k c a B n o i t a p i c i t r a P Germany and France: basic manufacturing (roasting and grinding of hazelnuts). Netherlands and Poland: additional processing of raw materials, production of packaging materials, … The Netherlands: a logistical hub for ingredients. Italy: the home of Ferrero… makes Nutella and R&D activities. Figure 3.1 Illustration of the Global Value Chain of Nutella Notes: This figure explores the GVC of Nutella, highlighting the sourcing of key ingredients such as Turkish hazelnuts and Indonesian palm oil, and the manufacturing and distribution processes that bring the final product to consumers worldwide. 85 Table 3.1 Descriptive Statistics Variables LRPT N mean s.d. min max 579 -2.844 6.065 -9.637 4.324 Standard Errors of LRPT estimates 579 155.0 84.39 6.35e-05 6.35e-05 GVC participation 570 35.47 7.994 10.58 90.63 Two-sided participation 570 2.665 1.336 0 25.80 Pure backward participation 570 8.528 4.877 0.0727 74.76 Pure forward participation 570 24.01 8.661 0 49.36 GDP per capita in US dollars 579 3,083 6,497 256.4 114,684 Total population in millions 579 18.700 44.001 27,964 1400.00 Rural population in millions 579 10.110 25.223 0 901.50 Agricultural production in million $US 470 4.949 26.993 385 1384.00 Food consumption 442 0.259 0.134 0.0259 1 Share of imports in food consumption 559 0.0888 0.0770 0.00192 0.550 Share of exports in food consumption 546 0.0309 0.0454 Landlocked countries dummy 579 0.202 0.401 0 0 0.641 1 86 Table 3.2 WLS estimation results Variables Participation in GVCs GDP per capita in US dollars Total population Rural population Food consumption Share of imports in food consumption Share of exports in food consumption Agricultural production Landlocked countries dummy Constant (1) -15.95*** (0.811) (2) -6.682*** (0.597) (3) -1.761** (0.727) (4) -1.200*** (0.439) 2.301*** (0.261) 1.240* (0.662) -0.369 (0.532) -3.011*** (0.368) 2.487*** (0.193) -0.669*** (0.062) -0.078 (0.355) 0.910*** (0.226) -30.90*** (4.278) NO 428 0.945 (5) -1.471*** (0.410) 2.122*** (0.234) 2.775*** (0.619) -1.142** (0.480) -3.122*** (0.328) 2.467*** (0.171) -0.247*** (0.068) -1.408*** (0.352) 1.239*** (0.203) -21.65*** (3.931) YES 428 0.957 Time FE Observations R-squared Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All covariate variables are in logarithmic terms. 53.63*** (2.880) NO 570 0.405 17.31*** (2.188) YES 570 0.786 7.230 (5.149) YES 570 0.995 87 Table 3.3 WLS estimation results for GVC components (1) 8.753*** (0.907) -10.12*** (0.731) 5.043*** (1.157) (2) 0.497 (0.571) -3.032*** (0.467) -1.867*** (0.701) (3) -1.291*** (0.374) 0.122 (0.216) -0.490 (0.496) Variables Pure backward participation Two-sided participation Pure forward participation GDP per capita in US dollars Total population Rural population Food consumption Share of imports in food consumption Share of exports in food consumption Agricultural production Landlocked countries dummy Constant Time FE Observations R-squared Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All covariate variables are in logarithmic terms. -27.43*** (4.769) NO 565 0.470 1.115 (2.883) YES 565 0.829 5.055 (4.683) YES 565 0.995 88 (4) 3.130*** (0.581) -2.877*** (0.434) 3.329*** (0.657) 2.470*** (0.259) 0.767 (0.653) -0.117 (0.523) -2.802*** (0.372) 2.332*** (0.193) -0.588*** (0.0622) 0.0780 (0.347) 1.081*** (0.303) -48.73*** (4.534) NO 424 0.949 (5) 1.825*** (0.537) -2.158*** (0.396) 1.873*** (0.611) 2.186*** (0.236) 2.373*** (0.631) -0.951** (0.484) -2.910*** (0.334) 2.316*** (0.173) -0.218*** (0.0667) -1.185*** (0.349) 1.018*** (0.274) -34.42*** (4.318) YES 424 0.960 APPENDIX B: ROBUSTNESS CHECK e r u t c u r t S g a L l a m i t p O >4 4 3 2 1 0 0 100 200 300 Number of Series 400 500 600 Schwarz Bayesian Information Criterion Hannan-Quinn Information Criterion Akaike Information Criterion Final Prediction Error Figure 3.2 Optimal lag structure by selection criterion Notes: This figure graphs the number of time series and their optimal lag structure in terms of the number of lags from 0 to more than 4 for four selection criteria. 89 Table 3.4 Robustness check for reverse causality Variables Lagged dependent variable Second-order Lag Participation in GVCs Pure backward participation Two-sided participation Pure forward participation GDP per capita in US dollars Total population Rural population Food consumption Share of imports in food consumption Share of exports in food consumption Agricultural production Landlocked countries dummy Constant (1) 0.046 (0.085) 0.033 (0.070) -0.907*** (0.277) (2) -0.017 (0.075) -0.008 (0.062) 1.468*** (0.360) -1.303*** (0.202) 1.640*** (0.317) -0.284** (0.130) 0.968*** (0.295) -0.098 (0.225) -1.114*** (0.169) -0.961*** (0.096) -0.333*** (0.032) -0.972*** (0.171) -1.444*** (0.187) -8.451*** (2.125) YES 181 286.21 0.960 -0.432*** (0.146) 0.829** (0.335) 0.042 (0.253) -1.266*** (0.188) -1.018*** (0.112) -0.306*** (0.035) -1.039*** (0.193) -1.546*** (0.183) 3.932* (2.061) YES 183 252.39 0.947 90 Time FE Observations F-statistic R-squared Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All covariate variables are in logarithmic terms. Table 3.5 Robustness check for excluding ABD data Variables Participation in GVCs GDP per capita in US dollars (1) -16.12*** (0.860) (2) -6.793*** (0.635) Total population Rural population Food consumption Share of imports in food consumption Share of exports in food consumption Agricultural production Landlocked countries dummy Constant (3) -1.132** (0.469) 2.337*** (0.283) 1.274* (0.721) -0.402 (0.580) -2.956*** (0.394) 2.530*** (0.206) -0.670*** (0.0676) -0.0639 (0.381) 0.919*** (0.241) -31.44*** (4.612) NO 372 0.946 (4) -1.401*** (0.438) 2.122*** (0.252) 2.860*** (0.670) -1.239** (0.522) -3.082*** (0.351) 2.493*** (0.183) -0.258*** (0.0725) -1.404*** (0.376) 1.238*** (0.216) -21.74*** (4.241) YES 372 0.958 Year FE Observations R-squared Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All covariate variables are in logarithmic terms. 54.23*** (3.053) NO 510 0.409 17.71*** (2.327) YES 510 0.786 91 Table 3.6 Robustness check for excluding ABD data by type of linkage (1) 8.687*** (0.961) -10.10*** (0.775) 4.928*** (1.235) (2) 0.455 (0.603) -3.031*** (0.493) -1.944*** (0.742) Variables Pure backward participation Two-sided participation Pure forward participation GDP per capita in US dollars Total population Rural population Food consumption Share of imports in food consumption Share of exports in food consumption Agricultural production Landlocked countries dummy Constant (3) 3.034*** (0.622) -2.803*** (0.461) 3.242*** (0.701) 2.508*** (0.281) 0.787 (0.713) -0.146 (0.571) -2.745*** (0.398) 2.370*** (0.206) -0.589*** (0.0673) 0.0984 (0.372) 1.068*** (0.323) -48.60*** (4.875) NO NO 369 0.951 (4) 1.760*** (0.574) -2.096*** (0.420) 1.816*** (0.652) 2.187*** (0.254) 2.465*** (0.684) -1.048** (0.527) -2.871*** (0.357) 2.339*** (0.184) -0.228*** (0.0714) -1.186*** (0.373) 1.013*** (0.292) -34.06*** (4.643) YES NO 369 0.961 Year FE Country FE Observations R-squared Notes: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. All variables are in logarithmic terms. -26.95*** (5.080) NO NO 506 0.471 1.445 (3.049) YES NO 506 0.829 92