Detecting the Position of Countries in Global Value Chains Using a Bilateral Approach
Detecting the Position of Countries in Global Value Chains Using a Bilateral Approach
Organisation Tags (3)
University of Malaya (UM)
University of Indonesia (UI)
Central Bank of the Republic of Turkey
Transcription
- Detecting the Position of Countries in Global Value Chains Using a Bilateral Approach O ğuzhan Erdoğan May 2020 Working Paper No: 20/08
- © Central Bank of the Republic of Turkey 2020 Address: Central Bank of the Republic of Turkey Head Office Structural Economic Research Department Hacı Bayram Mh. İstiklal Caddesi No: 10 Ulus, 06050 Ankara, Turkey Phone: +90 312 507 80 04 Facsimile: +90 312 507 78 96 The views expressed in this working paper are those of the author(s) and do not necessarily represent the official views of the Central Bank of the Republic of Turkey.
- Detecting the Position of Countries in Global Value Chains Using a Bilateral Approach O ğuzhan Erdoğan∗ Central Bank of the Republic of Turkey Abstract To detect the position of countries in global value chains in a consistent manner, we propose that export and import upstreamness measures of a country should be varying across its trade partners over time. To formalize our argument, we define t he notion of bilateral upstreamness between any pair of countries and show how its measure is affected from country-specific f actors. M oreover, w e i ncorporate t he v ariables in gravity literature into our estimation equations to account for how the geographical factors can have an impact on their bilateral production line position. Following Antrás et al. (2012), we also consider the hypotheses tested in their paper with our more aggregated and recent data set. Similar to their results, we find t hat b etter rule of law, higher level of financial d evelopment a nd i nvestment i n h uman c apital lead the export composition of countries to be more final g ood-oriented i n international markets. Finally, we portray Turkey’s production line position in comparison with different country blocks and income groups to illustrate our bilateral analysis. Keywords: Bilateral Upstreamness, International Trade, Global Value Chains, Production Line Position. ∗ Email: oguzhan.erdogan@tcmb.gov.tr. 1
- Non-Technical Executive Summary As the procurement process of firms operating in a variety of industries became extensively exposed to cross border linkages coupled with a tremendous rise in the volume of intermediate goods trade , the integration of countries into Global Value Chains (GVCs) has drawn the interests of considerable number of scholars. In search of a proper method to detect at which portion of value chains a given country performs, the measure of upstreamness serves as an important tool to characterize the number of production stages for the outputs of any given industry until reaching to final demand. However; combining the upstreamness measure with product level trade data has been mainly done by unilateral computation methods to detect the position of a country in GVCs against the rest of the world (Antrás, Chor, Fally, & Hillberry, 2012; Chor, Manova, & Yu, 2017). In this paper, on the basis of the empirical regularities that export and import upstreamness measures of a country could be varying across its different trade partners over time, we propose a bilateral computation method and call it as bilateral upstreamness. We highlight the empirical findings to analyze how the bilateral GVC positioning is affected from the country specific variables. Between any given pair of trade partners, as the exporter country governs better quality institutions, has a higher level of financial development and invests in human capital, its export composition towards to destination markets is prone to comprising more downstream products. Conversely, approaching to the same question from the perspective of an importer asserts that if the ratio of private credits to GDP rises and it utilizes more capital relative to labor in production, it has an incentive to participate in GVCs purchasing more upstream materials. Incorporating geographical and institutional variables into our estimation equations, we find that as the trade partners are located further away from each other and speak a common language, this might give an incentive to exporter country to supply more final good oriented commodities. On the other hand, sharing a common border might induce the importer country to demand relatively more upstream products. In line with the notion of bilateral upstreamness, we document salient patterns on the production line position of Turkey. Starting from mid 2000s, Turkey is mostly located in intermediate stages in production line, yet this tendency appears to be losing momentum towards 2011 and Turkish exports are composed of more downstream products. Considering the evolution of Turkish import upstreamness, Turkey on average occupies a closer position to BRICS countries in GVCs compared to European Union and OECD. Decomposing Turkey’s trade partners across different income quintiles, we also find that Turkey has a better capability to enhance the span of its domestic production stages against richer countries. Analyzing our bilateral measure based on a sectoral framework, we contend that chemicals, textile and plastics industries constitute three leading sectors to contribute domestic value added to the content of Turkish exports between 2003 and 2014. 2
- 1 INTRODUCTION 1 Introduction The rising trend in international trade has been the fragmentation of production over the last two decades . The production process now requires sourcing of inputs and components from multiple suppliers. The fact that these multiple suppliers are located in different countries allows different stages of production to be conducted in different countries (WTO, 2019). Such a phenomena is called Global Value Chains (GVC) in the literature of international trade. Countries in need of intermediate goods to complete a series of production stages demand inputs in factor markets. This demand driven outbreak has enhanced worldwide trade volume of intermediate goods and raised new policy questions about the position that countries take along the global value chains: Do countries engage in relatively more upstream or downstream trade activities when different manufacturing stages take place in different countries? Can we find a formal way to determine the bilateral trade pattern between any pair of countries? What are the implications of this disintegration of production in terms of country specific factors and variables? These kinds of questions are particularly important in terms of the strategic trade interactions between countries in international markets. Countries try to bring more value added to the products that they import from the world markets, undertaking a great deal of processing in industrial plants to convert them into high-valued final goods. Therefore, their respective comparative advantage in specializing in different production stages places them at different portions of global value chains, which in turn determines the characteristic of bilateral trade pattern between any pair of them (Costinot, Vogel, & Wang, 2012; Antràs, de Gortari, & Itskhoki, 2017). Countries which import relatively upstream products, implement a sequence of stages on the imported products and eventually supply the goods targeting final customers in international markets are positioned in the final portion whereas those who supply raw materials and cannot conduct a large span of production stages within their domestic borders are positioned in the lower portion of global value chains. Hence, the terms of trade between countries are largely influenced by the actions of the countries located at relatively final portions. In this paper we propose a bilateral computation method to detect the position of countries in global value chains using the notion of upstreamness. Developed as a metric to measure the distance of a product to its final use, the idea of upstreamness has recently occupied the international trade literature and served researchers as a tool to determine the position of countries in GVCs. Based on the input-output tables to account for the input usage in each industry, the measure of upstreamness is used to estimate the number of stages required for a certain industry until it meets the final demand. Therefore, it implicitly ex- 3
- 1 INTRODUCTION plains the characteristics of industries or outputs procured in those industries in terms of their usage as intermediary or final good . The more upstream a product is, the more industries it visits until meeting the final demand. Similarly, the less upstream a product is, the fewer stages are conducted until its usage as a final good. Building on this notion, we compute the (weighted) export and import upstreamness measures of countries with respect to their available number of trade partners in contrast to the common procedure in the literature which computes these measures by weighting the industry specific upstreamness measures with their respective shares in aggregate trade volume. The merit of using such a bilateral framework is to evaluate the position of countries in GVCs via a diversified number of upstreamness values changing on the basis of trade characteristics between any given pair of partner countries. Adopting such a bilateral technique might prevent researchers to be dependent only on unilateral export and import upstreamness measures, which in our defence are not sufficient to detect the position of countries in sliced value chains. Having proposed that, we estimate the impact of country level factors on upstreamness measures computed by employing both procedures. Based on a series of regression results, we observe that upstreamness measures computed within a bilateral set-up might in fact uncover the hidden production line patterns between a source and a destination country. This paper contributes to the literature in two ways. Our first aim is to test the hypotheses outlined in Antrás et al. (2012) using a more aggregate and recent UN Commodity Trade data set covering the period from 2003 to 2014. We examine whether the country specific factors can have an impact on the export and import upstreamness measures using country level and panel regressions. Thus, this paper is in a sense a modification of the results of previous works with an updated data set. Second, more importantly, we deviate from the existing literature via a bilateral analysis to explain how at a point in time trade dependent upstreamness between any pair of countries is affected from the country specific factors such as per capita GDP, financial development level, years of schooling etc. Based on the empirical findings that export and import upstreamness of a certain country can vary across its different trade partners over time, we conceptualize these empirical regularities as bilateral upstreamness between any pair of countries and demonstrate that it serves as a better tool in capturing the theoretical predictions on the relation between country level factors and GVC positioning compared to the unilateral upstreamness measures of a given country against the rest of the world. Moreover, we incorporate gravity variables provided in Head, Mayer, and Ries (2010) into the estimation equations to account for the impact of geographical and institutional variables on the production line position of a country against its trade partners. To best of our knowledge, this will be the first attempt to introduce the gravity variables to upstreamness 4
- 2 A REVIEW OF LITERATURE ON UPSTREAMNESS literature . Hence, harnessing such a bilateral framework allows us to make better predictions on the movement of countries in production chains and answer how the GVC positioning of a certain country alters with respect to its trade partners. 2 A Review of Literature on Upstreamness The upstreamness literature can be traced back to the very first concepts of forward and backward linkages. Blair and Miller (2009) define the forward and backward linkages in terms of the backward or forward production opportunities resulting from an expansion of output in a given sector. An expansion of output in an industry leads that sector to demand more inputs in factor markets. That means it requires a higher level of output produced in the precedent sectors. This is called backward linkages. On the other hand, in case of an expansion, the more output of this sector becomes available for the industries in subsequent stages. Thus, later stages can enjoy an increase in the availability of the factor that they use to complete the production of their own products. Working in the opposite direction to the previous terminology, it is called forward linkages. The more upstream an industry is, the more it could be said to cause forward linkages. Likewise, if an industry shows a downstream pattern, a positive output shift in that industry induces a demand increase in the already few number of goods it uses as an input. In his search for an optimal policy a country should adopt to develop, Hirschman (1958) argues that countries should invest in the sectors which have the largest amount of forward and backward linkages, supporting a renowned theory of "Unbalanced Growth". Instead of subsidizing all industries in the economy in a balanced way, he contends that industries generating maximum linkages ought to be developed first. In this respect, those who create the highest number of linkages work as a "self-propelling" mechanism, elevating the production in other sectors and bringing about the maximum benefit to overall economy. One of the first efforts to properly compute the upstreamness using Input Output Tables is done by Antrás et al. (2012). They construct the upstreamness measure for 426 industries in United States using 2002 Input-Output (I-O) Tables. They list the most upstream and downstream industries in US in terms of their 6-digit I-O codes. Beside that, using OECD STAN database they compute the industry level upstreamness for a sample of countries and provide the rank correlations of industry upstreamness between any pair of countries, asserting that upstreamness measure is a stable attribute of industries across different countries. Moreover, they apply this concept to trade, calculating the weighted export upstreamness measure for 181 countries between 1996-2005 and try to estimate how the export upstreamness is affected from various country specific factors. One important evidence they come up 5
- 2 A REVIEW OF LITERATURE ON UPSTREAMNESS with is that the presence of good institutions and prevailing rule of law lead countries to export in more downstream industries , meaning that the products they export are considered as final goods within the whole production process. Moving to another paper focusing on the position of China in GVCs using a firm-level analysis, Chor et al. (2017) deal with the position of Chinese firms in the global value chains, using firms level customs and balance sheet data. They use the notion of upstreamness and compute the upstreamness measure for 135 IO sectors from Chinese Input-Output Tables, similar to reasoning of previous paper for the sectors in US. However; they calculate the weighted export and import upstreamness measure using the volume of exports and imports of firms in each specific industry category. More importantly, they introduce a concept of the span of production stages, formally U M − U X , to capture the span of production stages conducted by Chinese firms within China. In contrast to the estimation approach used by Antrás et al. (2012), they use firm specific factors in their estimation to account for how the export-import upstreamness and the span of production stages depend on firm specific characteristics. They argue that firm experience, productivity and size are positively correlated with the downstream characteristics of firms’ exports. For the other direction, looking for any impact of the enlargement of the span of production stages on firm specific characteristics, they find that firms adopting more production stages in domestic economy, i.e. bringing more value added to the imported goods, are exposed to higher level of fixed production costs. In a macro sense, their empirical findings show that Chinese import upstreamness steadily increases between 1992 and 2011 though export upstreamness follows a moderate pattern, which implies that there is a rapid expansion in production stages conducted in China. Antrás and Chor (2017) take a formal step to combine the empirical exercises on the GVCs with the theoretical models of input flows across countries. They first introduce four different measures to elaborate on the methodologies of upstreamness and downstreamness and then provide an empirical investigation on the position of countries between 1995-2011 using 2013 edition of World Input Output Database (WIOD). For 41 countries and 35 industries provided in WIOD for whole period, they present how the GVC positioning of countries changes over time. In their empirical assessments, they find that countries whose export composition mainly targets the final consumers, i.e. having a low level of export upstreamness, are indeed those who can contribute much value added to the products that they process in their domestic industries, ie., having a higher level of import upstreamness. Finally, they conduct a series of counterfactual analyses to look for how the correlation between export and import upstreamness changes within the context of alternative scenarios such as trade cost reduction and increasing share of spending on service sectors. 6
- 3 METHODOLOGY AND DATA DESCRIPTION The rest of the paper is organized as follows : In the next section, we explain the concept of upstreamness and the reasons why the usptreamness values we calculate across countries are reliable predictors to measure the production line positions. Moreover, we describe the data sets used to obtain the weighted export and import upstreamness values and explain the procedure of combining data sets subject to our analysis. Fourth section discusses the empirical findings and how country specific factors are correlated with export and import upstreamness measures using basic correlation plots of average upstreamness values for each country. Moving to a more formal assessment, we employ both unilateral and bilateral upstreamness measures by estimating a variety of model specifications to answer whether country specific and geographical-institutional factors are effective in changing the position of countries in GVCs. A separate subsection is also included to point out where Turkey is located in GVCs and moves over time compared to a subset of developed countries (OECD and EU15) and developing countries (BRICS and NewEU). We also conduct a similar analysis by categorizing the trade partners of Turkey across five different income groups and then employ a sector-based framework to better understand the dexterity of each industry subject to our analysis in enhancing the span of production stages operated domestically. To enrich our discussion on the relation between value added and upstreamness, we make use of the indicators provided by OECD’s Trade in Value Added (TiVA) Database and figure out certain similarities and discrepencies between both empirical methodologies. Last section presents our concluding remarks. 3 3.1 Methodology and Data Description Upstreamness Measure Following Fally (2011) and Antrás et al. (2012), we begin by considering N industry closed economy and extend this to an open economy framework. For each industry i ∈ 1, 2, ..., N the value of gross output (Yi ) equals the sum of its final use as a final good (Fi ) and its use as an intermediate input to other industries (Zi ) N Yi = F i + Z i = F i + dij Yj (1) j=1 where dij is the dollar amount of sector i’s output needed to produce one dollar’s worth of industry j’s output. As this is a recursive formula, we can also write down the gross output in sector j in terms of its use as final and intermediate good (Yj = Fj + Zj ). Plugging this 7
- 3 METHODOLOGY AND DATA DESCRIPTION back into the previous expression : N Yi = F i + dij (Fj + Zj ) (2) j=1 N = Fi + N dij Fj + j=1 (3) dij Zj j=1 Defining another industry k ∈ {1, ..., N } \{i, j} for which Yk = Fk + Zk and using the required conversion parameter djk , N Yi = F i + N N dij Fj + j=1 N N dij djk Fk + j=1 k=1 (4) dij djk Zk j=1 k=1 Iterating this identity, we have N N N j=1 N N N dij djk dkl Fl + ... dij djk Fk + dij Fj + Yi = Fi + (5) j=1 k=1 l=1 j=1 k=1 Thus, we can express the output of industry i in terms of an infinite sum of the contribution of each stage in value chains. Starting from the usage of the output of industry i as a final good, we can see its usage as a direct input by the next sector in the production of final goods and its usage as a direct input of the direct input of the production of final goods in the subsequent sectors, and so on. Building on this identity, Antrás and Chor (2013) propose the following upstreamness measure by multiplying each term in the previous expression with their respective distance from final use plus one and dividing by the total output to obtain an average position of a certain industry i in value chains: Fi +2× Ui = 1 × Yi N j=1 dij Fj +3× Yi N j=1 N k=1 dij djk Fk Yi +4× N j=1 N k=1 N l=1 Yi dij djk dkl Fl + ... (6) To implement the open economy adjustment, Antrás et al. (2012) propose the following expression (see the paper for further discussion), replacing dij with: dˆij = dij Yi Yi − Xi + Mi (7) where Xi and Mi are exports and imports of sector i, respectively. If industry i has a high upstreamness value, that means it requires a great deal of produc8
- 3 METHODOLOGY AND DATA DESCRIPTION tion stages until meeting the final demand . Similarly, if it has a low value, its output tends to be directly served to final consumers rather than being used as an intermediate goods in the production of final goods by other sectors. For instance, petrochemical product (325110) 1 can be refined in petroleum industry and served to the consumers (2 stages to final use) or it can be used in the Plastic industry (325211) and then used as an input to Alumina refining (33131A) and then finally sold to Automobile industry (336111) to be consumed as a final good (4 stages to final use). On the other hand, Breakfast Cereal (311230) can be directly served to the customers (one stage to final use). 3.2 Data The main data set we use to classify the position of countries along the GVC is United Nation’s Commodity Trade Data (UNComtrade, 2019), which provides the export and import value of HS4-code products available for each country from 2003 to 2014. Furthermore, we use the industry-level upstreamness measures computed from 2002 US I-O Tables, which are provided in 6-digit I-O codes for 426 industries. As Antrás et al. (2012) put forward, these upstreamness measures are not considered only as a US specific measure. They compute a spearman rank correlation test for industry level upstreamness of US with countries in European Union and OECD. They find that the rank correlation is always large and positive in all country pairs, verifying the consistency of industry upstreamness across countries. In aggregate terms, this finding makes sense when we consider the share of US in world trade volume as it exports and imports a large set of products, providing a better coverage of industries. US is also assumed to be governing frontier technologies, which makes US based measures more standardized compared to the measures constructed based on other countries’ data. 2 In terms of the estimation approach, harnessing US based upstreamness measure alleviates any sort of endogeneity concern in the section for econometric analysis. Therefore, these findings allow us to use US industry upstreamness measure to reach a more generalized result for the export and import upstreamness of any country in the world. Formally, we compute the import and export upstreamness measure for each country c in year t ∈ 2003, ..., 2014 following the common procedure in the literature 1 The value in parenthesis denotes 6-digit I-O code It is not uncommon to construct standard industry-level measures using U.S. data. For instance, Rajan and Zingales (1998) construct an industry-level index of external finance dependence using balance sheet data on U.S. firms. 2 9
- 3 METHODOLOGY AND DATA DESCRIPTION (Antrás et al., 2012; Chor et al., 2017): N N UctM Mcit Ui , = i=1 M ct UctX = Xcit Ui i=1 Xct (8) and N UctM − UctX X M cit − cit Ui = Mct Xct i=1 (9) where N is the number of available HS4 products provided in UN Comtrade Data. Mcit (Xcit ) denote the import (export) value for product i in year t from destination (for source) N country c. Mct = N i=1 Xcit are total value of imports and exports for i=1 Mcit and Xct = that product i in question. The lower value of export upstreamness shows that export composition of countries tends to include more downstream products, ready to be directly served to final consumers. On the other hand, the greater value of that is associated with more upstream exports to the international markets. In expression (8), import upstreamness is defined as the weighted average of the upstreamness of imported products, describing whether the import volume comprises relatively intermediate or final goods. The higher value of import upstreamness shows that countries purchase more upstream products which require a series of production stages to be converted to final use. Therefore, countries having a higher import upstreamness can seize the opportunity to bring more value added on their imported products. As the import upstreamness measure is an analytical tool to characterize the distance of imported goods from their final usage, the number of stages operated domestically does not necessarily correlate with the amount of value added created on different types of imported materials. Not all raw material importers could have the essential technology or factors of production to implement value added on the purchased materials from the foreign markets. To rely on a better ground, using the definition in Chor et al. (2017), we suppose that the expression (9) captures the span of production stages operated within the domestic borders, showing the depth of the production process for any type of commodity 3 . Hence, a country having a positive UM − UX is said to process its imported products at a sufficiently great number of industries within its domestic plants before supplying to the world markets. Thus, that country will have a better chance to create more value added on those imported materials at quite a few number of industries, which will subsequently increase the amount of value added in domestic economy. 3 Henceforth, we assume that the difference between import and export upstreamness measures is used as a proxy for value added produced domestically. 10
- 3 METHODOLOGY AND DATA DESCRIPTION Note that Ui is the industry level upstreamness for each i from 426 different industries . It might be the case that the data for HS4 products does not exactly match with the upstreamness data. Thus, we use a concordance table to merge the upstreamness data with UN Comtrade data.4 Then, we obtain a new data set consisting of HS6 product codes and their respective upstreamness values. In order to use this new data set together with UN Comtrade data, we reduce HS6 to HS4 and merge that with UN Comtrade data. 5 Finally, we obtain a generic data set which shows the upstreamness value of exported and imported products recorded in HS4 category for each country in year t. From this data set, we can readily compute the export and import upstreamness using (8). Based on these usptreamness measures, we can compare the position of countries along GVCs. For instance, between 2003 and 2014 average export upstreamness of Saudi Arabia is 3.28 whereas Japan has an average export upstreamness of 2.03. We can interpret these numbers as the number of stages required for the exported products to be converted to final usage. Thus, we can verify that while Japanese exports are processed at most two industries in the importer countries, products purchased from Saudi Arabia visit at least three sectors until they meet the final demand. This result points out that Japanese exports show a more downstream pattern in world markets. Furthermore, the import upstreamness of Saudi Arabia is 1.98, which is relatively below the value of Japanese import upstreamness, 2.48. On the basis of these numbers, we can assert that Japan has a wider span of production stages, meaning that it has a better capability to induce value added on the imported products compared to what Saudi Arabia can do, which places Japan at a higher position along the global value chains. In order to account for the variations in export and import upstreamness of each country, we use the following country characteristics. To look for how the education attainment affects the export and import upstreamness, we use Barro and Lee (2013)’s education attainment data, which is available from 1950 to 2010 in 5-year intervals. Since our main data set covers the period from 2003 to 2014, we fill this period paying attention to the proximity of the years available. For instance, we use the estimate years of schooling in 2005 for the interval [2003,2007] and the estimate years of schooling in 2010 for the interval [2008,2014]. To understand how the ratio of private credit to GDP as an instrument for financial de4 This concordance table gives us a mapping between HS6 type product categories and IO codes. When we merge the upstreamness data with concordance table, there are 122 IO codes in master data which do not match with using data and there are 6 HS6 codes in using data which do not match with master data. Thus, we take the remaining number of industries which are exactly matched. 5 HS6 is a broader product-level classification than HS4. Thus, when reducing HS6 to HS4, there might be more than one upstreamness value corresponding to each HS4 code. To circumvent this multiplicity problem, we take the average of upstreamness measure and obtain a new data set in which each HS4 product code has a unique upstreamness value. 11
- 4 EMPIRICAL RESULTS velopment has an impact on the span of production stages , we use World Global Financial Database from Bartelsman, Becker, and Levine (2013). This is also available between 2003 and 2014. Additionally, we include real GDP per capita from World Development Indicators (WorldBank, 2019) in our analysis to understand how the trade upstreamness evolves over time in response to the changes in per capita real GDP. Using the same data set, we also incorporate total factor productivity to observe how the productivity gains could have an impact on country level upstreamness. Antrás et al. (2012) augment the physical capital per worker, which is constructed using perpetual inventory method, from Penn World Tables (Feenstra, Inklaar, & Timmer, 2015), into their estimation approach to measure the response of export upstreamness to the changes in factor endowments. We also use the same variable from PWT. As the production stages are largely affected from the quality of the contracts and institutions, we use the estimate of rule of law in the specified time period using World Governance Indicators from Kaufmann, Kraay, and Mastruzzi (2017). 4 4.1 Empirical Results An Overview on Initial Findings Table 1: Summary Statistics All goods Number of Observations Number of HS4 Products Number of Countries Log (Trade Volume), Mean Export Upstreamness (UX ), Mean Import Upstreamness (UM ), Mean UM − UX , Mean Manufacturing goods 2003-2014 2003 35,343,621 1244 221 4.24 [2.62] 2.37 [0.56] 2.13 [0.24] -0.24 [0.61] 2,425,154 1244 215 4.06 [2.52] 2.27 [0.59] 2.08 [0.23] -0.19 [0.63] 2014 2003-2014 3,176,235 18,912,807 1242 555 219 221 4.33 4.38 [2.68] [2.67] 2.41 2.12 [0.57] [0.49] 2.12 1.91 [0.22] [0.20] -0.29 -0.21 [0.62] [0.21] 2003 2014 1,293,303 555 215 4.20 [2.57] 2.06 [0.46] 1.87 [0.20] -0.18 [0.46] 1,705,942 555 219 4.47 [2.73] 2.15 [0.50] 1.92 [0.18] -0.23 [0.52] Notes: Summary Statistics are reported for all goods and manufacturing goods available in UN Comtrade Database. The values in square brackets denote standard deviations. The values for manufacturing goods are generated using NBER-CES Manufacturing Industry Database from Becker, Gray, and Marvakov (2016). It comprises a list of 473 NAICS industries similar to 6-digit IO codes. Using a concordance mapping provided in BEA website, we create another concordance table between HS6 and available NAICS codes. Merging this regenerated concordance table with upstreamness values, the upstreamness values of manufacturing goods are expressed in terms of HS6 codes. Finally, after reducing HS6 to HS4 and circumventing the multiplicity problem, upstreamness values for HS4 manufacturing products (555 in total) are obtained. The mean of log (trade volume) is calculated only within the sample of manufacturing goods. 12
- 4 EMPIRICAL RESULTS Table 1 presents the summary statistics on the UN Comrade data set and upstreamness measures we compute using the procedure in the data section with respect to the entire period (2003-2014), initial year (2003) and terminal year (2014). The average log world trade volume has drastically increased during this 12-year period and its dispersion around the mean level has also experienced a significant rise. The first implication for the overall position of countries in GVC comes from the fourth to six rows in Table 1. On average export pattern is relatively more upstream than import pattern for the entire time period. That is, countries mostly sell products, which are also processed by others to enhance the amount of value added. However, the countries are more scattered around the mean level of export upstreamness than they are around that of import upstreamness. This could happen if there is a significant difference between export upstreamness of countries. While a group of countries produces goods which are directly put into final use, others might export products necessarily used in intermediate stages of the production process. This discrepancy can account for the pattern of trade between a raw material supplier who cannot add much value added to its exports and an importer who contributes to the imported products and converts them to final usage after a certain number of stages. For the case of import upstreamness, the position of countries in production line is more moderate. Figure 1: The Movement in World’s Production Line Position To understand how on average the production line position of the countries changes over time, Figure 1 presents the world export and import upstreamness weighted by the countries’ income level for all goods provided in UN Comtrade. Thus, the export and import 13
- 4 EMPIRICAL RESULTS upstreamness of high income level countries have a larger impact in determining the shape of the graphs in Figure 1 . Similar to what we observe in Table 1, exports are on average more upstream than imports. However; they almost move parallel to each other during the specified time period. Both of the lines show an increasing trend up to 2008. However; with the advent of the global financial crisis, we could see a fall both in both of the upstreamness measures, implying that the exported products show a tendency to be supplied more downstream and the share of intermediate goods in the trade volume of countries declines as well. Because the production level contracts all over the world, this reflects the decrease in the volume of trade for input materials. Though import upstreamness recovers after the crisis period, it steadily falls in the period starting with 2011. Exports, on the other hand, become more upstream after the crisis period, but start targeting final customers especially towards the end of our time period. Table 2: Upstreamness of Manufacturing Exports by Country Income Quartiles Income quartile Bottom 2nd 3rd Top Mean 2.11 2.22 2.17 2.08 S.D. Min Max 0.58 1.12 4.15 0.47 1.07 4.16 0.48 1.09 3.71 0.41 1.04 3.06 Table 3: Upstreamness of Manufacturing Imports by Country Income Quartiles Income quartile Bottom 2nd 3rd Top Mean 1.90 1.96 1.93 1.88 S.D. Min Max 0.20 1.14 2.92 0.18 1.22 2.44 0.18 1.38 2.47 0.20 1.19 2.65 Table 4: Span of Production Stages (UM − UX ) of Manufacturing Goods by Country Income Quartiles Income quartile Bottom 2nd 3rd Top Mean -0.21 -0.26 -0.24 -0.20 14 S.D. Min Max 0.55 -2.29 0.66 0.51 -2.25 0.63 0.51 -1.83 0.65 0.45 -1.38 0.71
- 4 EMPIRICAL RESULTS Following the empirical strategy of Antr ás et al. (2012), we consider the summary statistics of export and import upstreamness by country income groups, as described in Table 2 and 3. Additionally, we extend this analysis including the summary statistic of the difference between import and export upstreamness to drive some implications on the number of production stages operated domestically (Table 4). Countries are grouped into income quartiles based on the average log PPP-adjusted Real GDP per capita over 2003-2014, from World Bank World Development Indicators. The average upstreamness of all these three measures together with their respective standard deviations and min-max values within each income quartile are reported in Table 2, 3 and 4, respectively. 6 We would like to highlight two results. First, poorer countries appear to be exporting in more upstream industries compared to richer ones. Therefore, richer countries tend to export products aimed to be directly served in final markets. Second, the standard deviation of export upstreamness in each income group decreases as it moves from bottom to top. Countries in the bottom quartile thus vary in terms of the average position they occupy in global production lines. To illustrate, Haiti and Zambia are both included in bottom income group. However; based on the export composition of goods they supply to world markets in 2014, we can see that their respective export upstreamness demonstrates a large difference (1.14 vs 3.98). This difference can be attributed to the types of products which occupy a larger share in the export content of both countries. While coffee, as a perishable food served to the final customers after few number of stages, constitutes a great portion of the exports of Haiti, Zambia is a major exporter of unrefined copper, which is processed by a greater number of industries until meeting the final demand. When we consider upstreamness of manufacturing imports by each income quartile, we cannot observe such a pattern. If we have a look at Table 4, we also see a slight increase in mean level of UM − UX , arguing that richer countries might have a potential to process the upstream import materials until targeting the foreign markets. The finding that top group has the maximum possible value of this difference is also in line with the observation that considerable portion of the procurement process of manufacturing goods occurs in the rich countries. 6 When we download the PPP-adjusted GDP per capita from World Bank’s Data Bank and merge it with the upstreamness data, there are 385 observations in the upstreamness data that cannot be matched with the GDP per capita data set. Similarly, 644 observations from the second data set are missing in the fist one. Thus, we move on our analysis using the remaining number of 2216 observations, which comprise 189 countries between 2003-2014. Distributing the export and import upstreamness into 4 income group in an ascending order yields almost 47-48 countries for each quartile. 15
- 4 EMPIRICAL RESULTS 4 .2 Correlation with Country Characteristics This subsection mainly depicts how the average export and import upstreamness are correlated with the average country characteristics between 2003-2014. 7 Figure 2 8 indicates how average export and import upstreamness are correlated with the estimates of rule of law. Countries who have a higher quality of governance index tend to export in more downstream industries, consistent with the findings in Antrás et al. (2012). On the other hand, there is not a certain pattern for average import upstreamness. Another important feature in characterizing the position of countries in production line is to look at how upstreamness measures respond to changes in real GDP per capita. Figure 3 gives us a blurring picture between the average upstreamness and log of Real GDP per capita, PPP adjusted. Though we can observe a slightly downward sloping line for export upstreamness, such a trend disappears for import upstreamness. Thus, consistent with what we find in Table 2, rich countries have a propensity to export more downstream products. Furthermore, the years spent in school on average have a positive impact on the downstreamness of the exported products (Figure 4), implying that investment in education might play a role in changing the position of a country’s production line position. The observation that export upstreamness is more responsive to the country level characteristics than import upstreamness might suggest that country characteristics mostly affect the span of production stages from production channel rather than demand channel. Figure 2: Export-Import Upstreamness and Rule of Law 7 We take the average of export and import upstreamness between 2003-2014. The same procedure is also applicable for the estimates of rule of law and log of Real GDP per capita. Here the fact that Barro-Lee’s education attainment data is available for 5-year intervals drives us to calculate average years of schooling between 2003 and 2014 using the values in 2005 and 2010, using the procedure described above. 8 Dots stand for the average values for each country in our data set. 16
- 4 EMPIRICAL RESULTS Figure 3 : Export-Import Upstreamness and Real GDP per capita Figure 4: Export-Import Upstreamness and Years of Schooling 4.3 Econometric Analysis The last subsection provides suggestive evidence that country characteristics on average may affect upstreamness values. This section embraces a more formal approach in analyzing the upstreamness measures and country level characteristics. Our estimation consists mainly of two approaches: 4.3.1 Estimation with Country Level Factors The first approach comprises two main specifications, in the first of which we aim to do a cross country regression following estimation procedure of Antrás et al. (2012), taking the average of all variables over whole period, to examine how the export and import upstreamness are related to the country characteristics and in second of which we intend to do a fixed-effect panel OLS estimation. Formally, Yc = α + βZc + 17 c (10)
- 4 EMPIRICAL RESULTS where Yc is one of the average country upstreamness variables in UcM , UcX , UcM − UcX . Zc is a combination of variables, including the average of log of Real GDP per capita, private credits/GDP, years of schooling, log of physical capital per worker and the estimate of rule of law during the specified time period. 9 c is the error component for any country c. For the second specification, we consider the following panel regression for all t ∈ 2003, ...2014: Yct = α + δc + γt + ΓZct + ct (11) where the dependent variables are the upstreamness measures (UctM , UctX and UctM − UctX ) for each country c at time t. δc and γt are country and year specific fixed effects. Z is the same generic vector depending on time, additionally including total factor productivity from PWT. The last term is the usual cluster error by country and year. Table 5 tests the same hypotheses put forward in Antrás et al. (2012) to understand in which direction the country level characteristics can have an impact on the export upstreamness for all and manufacturing goods. Considering the second column, we can verify that as the strength of contracting institutions rises, countries export relatively more downstream though per capita GDP works in the opposite direction. As countries move to the higher stages of income, their exports become relatively more upstream. The usage of private credits over GDP, as a proxy for financial development has a negative and significant effect only in the third column when we do the estimation for all goods. However; when we pay a close attention to the estimation done only using manufacturing goods, we can observe that it becomes significant and theoretically consistent in all models, suggesting that export upstreamness for manufacturing goods is more sensitive to the financial development level than it is for all goods, the rest of which consists of mostly agricultural and service products. One possible explanation is that as the producers increase the amount of credit they borrow, they might be willing to invest loans in the manufacturing sectors, which have relatively more favorable terms of trade than the sectors of agriculture or service. This upward investment shift can create an incentive for the manufacturers to operate the final stages of production in domestic plants before supplying to foreign markets during the intermediate stages. In contrast, they might have less incentive to export agricultural goods after conducting the final stages in their home countries or since the output of the service sector is mainly served to domestic markets, they might not be willing to supply service goods to foreign markets. Thus, the pure effect of manufacturing goods can be more precise in reducing the export 9 Here notice that the variable for years of schooling is again calculated as the average of the filled values between 2003 and 2014 using the values of 2005 and 2010. The log of physical capital per worker is computed dividing the real capital stock with the number of employed population. 18
- 4 EMPIRICAL RESULTS Table 5 : Export Upstreamness and Country Characteristics All Exports Log (Real GDP per capita) (1) UX (2) UX (3) UX (4) UX (5) UX -0.0476 (0.0359) 0.129*** (0.0458) -0.305*** (0.0481) 0.0334 (0.0529) -0.368** (0.146) 0.141*** (0.0486) -0.263*** (0.0621) -0.144 (0.101) 0.109 (0.165) -0.207*** (0.0749) -0.122 (0.0974) 0.0715 (0.145) -0.0377 (0.0243) Rule of Law Private Credit / GDP Log (Capital per worker) (PWT) Years of Schooling N R2 Manufacturing Exports Log (Real GDP per capita) 189 0.01 187 0.15 178 0.08 178 0.15 130 0.13 -0.00748 (0.0292) 0.125*** (0.0365) -0.230*** (0.0407) 0.0644* (0.0333) -0.360*** (0.0821) 0.125*** (0.0375) -0.145*** (0.0508) -0.233*** (0.0608) 178 0.10 178 0.13 0.114 (0.130) -0.102 (0.0642) -0.246*** (0.0618) -0.00223 (0.120) -0.0181 (0.0181) 130 0.12 Rule of Law Private Credit / GDP Log (Capital per worker) (PWT) Years of Schooling N R2 189 0.0004 187 0.12 Notes: ***,**, and * denote significance at the 1%, 5% and 10% levels respectively. The values in parentheses are robust standard errors. When we conduct the estimation with all country level variables excluding years of schooling, sample size drops from 178 to 161 and then to 130 if we include years of schooling. The significant drop in the number of our observations is attributed to the non-availability of the education attainment data from Barro and Lee (2013) for 48 countries subject to our estimation. 19
- 4 EMPIRICAL RESULTS upstreamness . The estimate of rule of law has a significant and negative effect on export upstreamness of manufacturing goods though the statistical significance of its coefficient considerably falls in the last model. Model 5 also allows us to make some inference on how investing in human capital can play a role in a country’s production line position. Though we cannot observe a statistically significant impact, as the years of schooling increases, the export activities on average seem to be targeting final demand. However; the significance level is not sufficient to reach this conclusion. Results in Table 5 are consistent with what Antrás et al. (2012) find testing the same specifications. Though the explanatory power of the coefficients of private credits over GDP is slightly lower compared to their results, the coefficient of rule of law is more significant in each of the models tested above. Table 6: Upstreamness and Country Characteristics (1) UX All Goods log (Real GDP per capita) Rule of Law Private Credit / GDP Log (Capital per worker) (PWT) Years of Schooling Constant N R2 (2) UM 0.109 -0.0882* (0.165) (0.0461) -0.207*** 0.0424 (0.0749) (0.0363) -0.122 -0.112 (0.0974) (0.0931) 0.0715 0.0620* (0.145) (0.0362) -0.0377 0.00508 (0.0243) (0.00864) 1.975 3.184*** (1.832) (0.493) 130 130 0.13 0.08 (3) UM − UX -0.197 (0.180) 0.250*** (0.0952) 0.00995 (0.180) -0.00953 (0.158) 0.0428 (0.0280) 1.208 (1.993) 130 0.10 Notes: ***,**, and * denote significance at the 1%, 5% and 10% levels respectively. The values in parentheses are robust standard errors. Table 6 introduces the import upstreamness and the span of production stages, UM −UX , in addition to the last specification of Table 4. The only variables that could have an impact on import upstreamness are GDP per capita and capital per worker though their explanatory powers are less precise. The increase in capital per worker is associated with an rise in import upstreamness but model (3) does not support our previous claim that high income countries tend to contribute more value added to the imported products. Still, it yields an important implication for the production stages conducted within domestic industries. The better institutions allow countries to process the intermediate stages within their own boundaries, 20
- 4 EMPIRICAL RESULTS which might trigger an increase in the value of production . Table 7 reveals the estimation results for specification (11), tested with country and year specific effects. It also includes total factor productivity from PWT. The impact of GDP per capita on upstreamness measures is still in line with the results of Table 6 and does not match with our prior claim. The coefficient of rule of law in the model for export upstreamness is theoretically meaningful, but insignificant. We also do not see the effect of total factor productivity in explaining the variations in production line position. The models except from (2) produce significant estimates for years of schooling but they are inconsistent with theoretical predictions. The model for import upstreamness fails to truly capture the impact of rule of law on import upstreamness but still they make sense considering the finding that higher capital intensity per worker could lead a certain country to import relatively intermediate goods from the world markets (Model 2). Yet, the second models of Table 6 and 7 should be interpreted cautiously as "the relevance of these country variables in explaining trade patterns is thus specific to the supply side, and does not appear to be driven by differences in the composition of demand" (Antrás et al., 2012:16 in longer version). Therefore, country-level characteristics are more meaningful and prevailing in production channel instead of demand channel. 4.3.2 The Impact of Country Level Characteristics on Bilateral Upstreamness Theoretically inconsistent and imprecise estimates of Table 6 and 7 do not bring convincing results in analyzing the direction of correlation from country level variables to upstreamness measures. Thus, rather than looking at the overall variation in upstreamness measures in response to country characteristics, we propose that a country’s export and import upstreamness measure could be varying across its different trade partners. For instance, Turkey could concentrate on the exports of final goods to less developed countries while its exports tend to comprise more raw materials when it exports to more developed countries, necessarily increasing its export upstreamness. On the other hand, it could bring more value added to the imported products from the poorer countries, but there might not exist much opportunities to process the goods imported from richer countries. This fact could also be sensitive to the changes in country level factors. The relative production line position of a country might alter with respect to its trade partners in case of a change in not only its own but also its trade partners’ characteristics. This finding leads us to employ a bilateral approach in examining such a direction of causality. In what follows, we define the bilateral upstreamness measure from the perspective of exportation. For any source country i and 21
- 22 (3) UM − UX -0.299*** (0.0882) -0.0361 (0.0477) -0.00981 (0.0400) 0.0986 (0.0840) -0.146 (0.114) -0.0553*** (0.0180) Yes Yes 1240 0.93 (4) UM − UX -0.250*** (0.0805) -0.0533 (0.0447) 0.00918 (0.0367) 0.0826 (0.0728) -0.147 (0.115) -0.0468*** (0.0143) Yes No 1240 0.93 Notes: ***,**, and * denote significance at the 1%, 5% and 10% levels respectively. The values in parentheses show robust standard errors. While model (1) takes the export upstreamness as a dependent variable, model (2) considers the variation in import upstreamness. We report the estimation results for the span of production stages, UM − UX , for both country and year fixed effects (Model 3) and only for country fixed effects (Model 4) because in checking the diagnostics of third model we do not reject the null hypothesis that there is no year fixed effect. Country Fixed Effects? Year Fixed Effects? N R2 Years of Schooling TFP at constant national prices (2011=1) Log (Capital per worker) Private Credit / GDP Rule of Law Log (Real GDP per capita) (1) (2) UX UM 0.131* -0.169** (0.0705) (0.0658) -0.0135 -0.0496** (0.0424) (0.0213) 0.000493 -0.00931 (0.0327) (0.0424) 0.160** 0.258*** (0.0743) (0.0424) 0.0588 -0.0877 (0.100) (0.0833) 0.0666*** 0.0113 (0.0162) (0.00945) Yes Yes Yes Yes 1240 1240 0.93 0.85 Table 7: Panel Regressions 4 EMPIRICAL RESULTS
- 4 EMPIRICAL RESULTS destination country j at time t N Uijt = Xijkt Uk X ijt k=1 (12) where Xijkt is the volume of exports from the source country i to the destination country j for a specific product category k at time t. Xijt is the total volume of exports. Similar to the formula we use to define the export and import upstreamness, we weight the volume of exports from i to j by the industry upstreamness measure of N -available number of product k. Similarly, we can also define the bilateral upstreamness from the perspective of importation: N Uijt = Mjikt Uk k=1 Mjit (13) where Mjikt is the volume of imports to the destination country j from the source country i for a specific product category k at time t. Mjit is the total volume of imports. To illustrate our findings, we return to comparison of Japan and Saudi Arabia in global value chains. Previously, we examined the average position of each of these countries taking into account their export and import pattern with the rest of the world. Now we consider the bilateral trade pattern between Japan and Saudi Arabia at a point in time, particularly in 2014. Import upstreamness of Saudi Arabia to Japan (export upstreamness of Japan to Saudi Arabia) is around 1.73 whereas the import upstreamness of Japan to Saudi Arabia (export upstreamness of Saudi Arabia to Japan) is around 3.31. This implies Japan can contribute to the value of the materials it purchases from Saudi Arabia within more than three sectors while Saudi Arabia mostly purchases goods from Japan ready to be sold to final customers and cannot bring much value added to its imports from Japan. Our approach in this section consists of the regression equations accounting for the bilateral upstreamness at a point in time between a source and a destination country. We aim to test bilateral specifications on the basis of different estimation strategies. The main purpose of conducting the first bilateral analysis is to explain how the upstreamness between a source and a destination country changes in response to the variables in source or destination countries, controlling for destination-year and source-year fixed effects. In addition to that, our estimation equations consist of gravity variables 10 to account for the impact of geographical 10 CEPII Gravity Database is harnessed to include the gravity variables for estimation procedure from Head et al. (2010). The variables of interest are the distance (km) between the source and destination country, a dummy variable for common language to explain whether the same language is spoken between the trade partners, a dummy variable for contiguity to show if there is a common border and a dummy variable for the presence of a free trade agreement. 23
- 4 EMPIRICAL RESULTS and institutional factors on bilateral production line position between two trade partners . Thus, we can explain how GDP per capita, rule of law, the ratio of private credits to GDP, capital stock per worker, total factor productivity and years of schooling in a certain country could affect the bilateral upstreamness between any trade partners, controlling for the country fixed effects and gravity terms. Formally, Uijt = α0 + α1 Xit + α2 Zij + Γjt + ijt (14) Uijt = β0 + β1 Xjt + β2 Zij + Γit + uijt (15) and where the dependent variable is the bilateral upstreamness measure between a source country i and a destination country j at time t. Xit (Xjt ) is a vector of country specific factors in source country i (destination country j). Zij is a vector of gravity variables between a source i and destination j. Γjt and Γit denote the destination and source fixed effects at time t, respectively. Table 8 displays the estimates of coefficients in both specifications above. Models (1-2) use robust standard errors while (3-4) cluster the standard errors by source-destination country pairs to take into account any inflation in standard errors resulting from the unobservable impact of possible clustering of trade partners. One of the stark findings from models (1) and (3) is that an increase in per capita real GDP in a source country is associated with an increase in its export upstreamness, i.e., import upstreamness of the destination country given that destination specific characteristics are fixed. Thus, as the income level of the exporter country increases, its trade partner tends to import more upstream products. The strength of contracting institutions and financial development in exporting countries can induce them to specialize in the production of goods directly targeting the final markets in importer countries. If the productivity rises in the source country, import upstreamness of the destination country also increases, meaning that the share of upstream intermediate goods in importer’s trade volume significantly rises. The investment in human capital in exporter country slightly changes its export pattern, making the export composition more downstream. Here the gravity terms are also effective in changing the bilateral position between two trade partners. As the source and destination countries are located further away from each other, this can make the types of products exported by the source country more downstream. This result is consistent with the implications of gravity models on the impact of distance on the value of products exported to foreign markets. Increasing distance can lead the producers in the source country to change the composition of exports towards high 24
- 4 EMPIRICAL RESULTS Table 8 : The Impact of Source and Destination Specific Variables on Bilateral Upstreamness Log (Real GDP per capita) Rule of Law Private Credit / GDP Log (Capital per worker) TFP at constant national prices (2011=1) Years of Schooling Log (Distance) Common Language (=1) Contiguity (=1) Free Trade Agreement (=1) Destination - Year Fixed Effects? Source - Year Fixed Effects? N R2 (1) Uijt 0.129*** (0.00537) -0.124*** (0.00258) -0.138*** (0.00350) -0.0148*** (0.00448) 0.225*** (0.0238) -0.00397*** (0.000928) -0.0715*** (0.00215) -0.00585 (0.00379) 0.0400*** (0.00740) -0.0265*** (0.00435) Yes No 184948 0.17 (2) (3) (4) Uijt Uijt Uijt -0.0230*** 0.129*** -0.0230** (0.00425) (0.0137) (0.00983) -0.0455*** -0.124*** -0.0455*** (0.00267) (0.00616) (0.00620) 0.0901*** -0.138*** 0.0901*** (0.00483) (0.00809) (0.0119) 0.0598*** -0.0148 0.0598*** (0.00379) (0.0111) (0.00867) -0.285*** 0.225*** -0.285*** (0.0209) (0.0408) (0.0338) -0.0152*** -0.00397* -0.0152*** (0.000810) (0.00229) (0.00182) -0.0303*** -0.0715*** -0.0303*** (0.00228) (0.00549) (0.00539) -0.0236*** -0.00585 -0.0236** (0.00434) (0.00908) (0.0102) 0.106*** 0.0400* 0.106*** (0.00923) (0.0206) (0.0262) 0.0108** -0.0265** 0.0108 (0.00451) (0.0108) (0.0105) No Yes No Yes No Yes 179004 184948 179004 0.24 0.17 0.24 Notes: ***,**, and * denote significance at the 1%, 5% and 10% levels respectively. The values in parentheses show robust standard errors, except in column (3) and (4) where they are pairwise clustered by source and destination country. 25
- 4 EMPIRICAL RESULTS value commodities , for which it is more profitable to incur fixed and variable trade costs of servicing the remote markets. Thus, the incentive of exporting high value commodities to remote destinations can increase the number of subsequent stages of production undertaken in domestic plants and induce a more downstream pattern for the types of products subject to exportation. Furthermore, free trade agreements can also allow the exporter countries to supply more downstream goods in the destination markets. If there is a common border between the trading partners, we could observe a rise in the import upstreamness of destination countries relative to exporters. The other models in Table 8 (2 and 4) consider the impact of destination specific factors on bilateral upstreamness, taking the characteristics of source country at time t fixed. As the destination country gets richer and the quality of institutions increases, it imports more downstream products from its exporter partner. The share of private credits in GDP significantly improves its ability to import more upstream products. While the intensity of capital per worker in importer country might create an impetus to enlarge the number of production stages conducted within its home plants, the productivity increase works in the other way around. The increase in years of schooling in the destination country seems to be leading the source country to concentrate on the sale of final goods. The inclusion of gravity terms also yields interesting results for the second specification. The increase in the distance between source and destination country allows the source country to export more downstream products. Unlike the first specification, the coefficient for common language dummy is significant for this model. If the trading partners share the same language, this could allow the exporter country to sell final products to the importer. In case of a common border, we can observe a rise in the import upstreamness of the destination country. When we examine the effect of signing a free trade agreement between two trade partners, we see alternating results depending on the fixed effects we use in the estimation. If we fix the destination-year effects, we find a relatively more downstream characteristics for the products of exporter country. However; fixing the source-year effects, free trade agreement might bring about a rise in the import upstreamness of the destination country. We also test the following specification to combine the bilateral estimation equations above to examine the joint impact of source and destination specific variables on bilateral upstreamness in one specification: Uijt = δ0 + δ1 Xit + δ2 Xjt + δ3 Zij + Γi + Γj + γt + eijt (16) where Xit and Xjt are the same source and destination country specific variables in source i and destination j at time t. Zij is the gravity variables in the first specification. We also 26
- 4 EMPIRICAL RESULTS include Γi and Γj to control for the source and destination country fixed effects 11 . Time fixed effect γt is also added to our specification. eijt is the usual error term. Table 9 displays the estimation results of this specification. Using a similar strategy to the previous bilateral analysis, we also test the model with clustered standard errors to account for the impact of any unobserved factors regarding with source destination country pairs. Even if the standard errors are inflated due to the usage of clustering, the coefficients do not lose their statistical significance tough there is a mild fall in the explanatory power of TFP for the source country. The increase in income level and years of schooling in source country appear to be leading to more upstream exports while the strength of institutions and productivity gains bring about more downstream exports. By definition of bilateral upstreamness, the same variables affect the import upstreamness of the destination country in the other way around. When we examine the impact of destination characteristics on the bilateral upstreamness, real income level and capital intensity per worker are the only effective variables. The higher income level of the destination causes it to import less upstream products whereas the higher capital intensity allows it to import more upstream products. When we consider the gravity variables, they are totally consistent with the results previous table. The distance between two trade partners, the presence of a common language and a free trade agreement reduce the export upstreamness of the source country. If trade partners are two neighbor countries with a common border, this would help the destination country import relatively more upstream products. In order to observe to what extent the bilateral relationships between the country level variables and upstreamness measures are sensitive to the empirical strategy in action, we estimate the first (main) bilateral model in our analysis using weighted least squares (WLS), controlling for source and destination-year fixed effects. Table 14 of the Appendix section presents the estimation results of this method using the bilateral share of exports (Panel A) and imports (Panel B) as weights. We first take a look at the results in Panel A. Apart from the coefficient of GDP per capita, we see a discernable improvement in the magnitude of the bilateral relationships. In contrast to the empirical findings in Table 8, distance variable turns out be changing sign. Transforming the distance between any given pair of trade partners so as to have that also reflect the relative importance of the destination country in origin country’s gross exports, the exporters of origin country tend to supply more upstream products if both trade partners are located further away from each other controlling for any variation in destination markets. As being another variable opposed to the results of our first bilateral model, the coefficient for capital per worker implies that if the exporters utilize more capital relative to worker in their production, this would lead them to sell more upstream 11 Notice that they do not vary across time unlike the specifications considered in the first analysis. 27
- 4 EMPIRICAL RESULTS Table 9 : The Joint Impact of Source and Destination Specific Variables on Bilateral Upstreamness Log (Real GDP per capita) in Source Rule of Law in Source Private Credit / GDP in Source Log (Capital per worker) in Source TFP (2011=1) in Source Years of Schooling in Source Log (Real GDP per capita) in Destination Rule of Law in Destination Private Credit / GDP in Destination Log (Capital per worker) in Destination TFP (2011=1) in Destination Years of Schooling in Destination Log(Distance) Common Language (=1) Contiguity (=1) Free Trade Agreement (=1) Destination Fixed Effects? Source Fixed Effects? Year Fixed Effects? N R2 (1) Uijt 0.0903*** (0.0314) -0.0299** (0.0150) 0.00104 (0.0120) 0.0131 (0.0262) -0.106** (0.0432) 0.0221*** (0.00568) -0.136*** (0.0305) 0.00606 (0.0132) -0.00498 (0.0131) 0.118*** (0.0239) -0.0149 (0.0356) -0.001000 (0.00644) -1.36e-05*** (5.19e-07) -0.0343*** (0.00555) 0.0710*** (0.00817) -0.0242*** (0.00484) Yes Yes Yes 109558 0.35 (2) Uijt 0.0903*** (0.0344) -0.0299* (0.0155) 0.00104 (0.0122) 0.0131 (0.0292) -0.106** (0.0480) 0.0221*** (0.00621) -0.136*** (0.0352) 0.00606 (0.0140) -0.00498 (0.0137) 0.118*** (0.0273) -0.0149 (0.0420) -0.001000 (0.00682) -1.36e-05*** (1.24e-06) -0.0343*** (0.0132) 0.0710*** (0.0230) -0.0242** (0.0111) Yes Yes Yes 109558 0.35 Notes: ***,**, and * denote significance at the 1%, 5% and 10% levels respectively. The first model is estimated using robust standard errors whereas in the second model they are clustered by source destination country pairs instead. 28
- 4 EMPIRICAL RESULTS products keeping the destination specific effects constant . The reverse case happens if we fix the source specific effects. Considering Panel B, among the significant relationships, we observe a fall in the magnitude of the coefficients for TFP as a country level variable and contiguity as a gravity term. These are also the only two variables dissimilar to the results of the first bilateral model. In both of the WLS estimation techniques, accounting for the impact of clustering around trade partners inflate the standard errors in a way that most of the bilateral relationships lose their significance enormously. 4.4 4.4.1 The Production Line Position of Turkey The Movement relative to OECD, BRICS and EU members The computation of bilateral upstreamness between any pair of countries allows us to keep track of how production line position of a certain country or a block of countries changes over time. In this section, we compare the evolution of Turkey’s production line position with both developed and developing country groups. For developed group, we consider the countries in OECD 12 and 15 members of European Union 13 . For developing country groups, we consider BRICS 14 and new members of European Union 15 . Taking the average of all countries involved in each group at a point in time, we sketch how Turkish position on average evolves over time against these blocks. This will allow us to portray the changes in relative position of Turkey in global value chains with respect to two different income groups. Figure 5 illustrates the movement of Turkish export upstreamness compared to both developed and developing groups 16 . On average Turkish exports to the rest of the world are processed by at least two industries and it shows the most drastic change among all the other country groups during the period between 2006 and 2011. Turkish exports, especially after 2006, become more upstream, concentrating on the sale of intermediate goods in world 12 Australia, Austria, Belgium, Canada, Chile, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hungary, Iceland, Ireland, Israel, Italy, Japan, Korea, Latvia, Lithuania, Luxembourg, Mexico, Netherlands, New Zealand, Norway, Poland, Portugal, Slovak Republic, Slovenia, Spain, Sweden Switzerland, the United Kingdom and the United States. Since Turkey is also a member of OECD, we exclude that to obtain the average of remaining countries. OECD (2019). 13 Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, Spain, Sweden and the United Kingdom. European-Commission (2019) 14 Brazil, Russia, India, China and South Africa. 15 EU countries which became members after 2004 enlargement: Czech Republic, Estonia, Cyprus, Latvia, Lithuania, Hungary, Malta, Poland, Slovakia, Slovenia, Romania, Bulgaria and Croatia (European Commission, 2019). 16 Readers might be alerted about the drastic variation in Turkish export upstreamness compared to the variation in other country group measures over time. When we compute the variation in individual country level export upstreamness over time, Turkish export upstreamness has the 122nd biggest standard deviation out of a total of 221 countries. Averaging country measures to obtain the movement in country blocks or treating them as a single economy might reduce the variation in upstreamness measures. 29
- 4 EMPIRICAL RESULTS (a) Panel A: Developed Group (b) Panel B: Developing Group Figure 5: Evolution of Export Upstreamness markets. While this pattern increasingly continues until 2011, Turkish exports start targeting final demand during the last three years in our data set. Turkey together with BRICS is located in the intermediate stages of production chains compared to the other country blocks on average. As we observe a relatively flat movement of BRICS countries, EU15 and OECD countries have a tendency to follow a more downstream pattern in exporting. The fact that EU15 countries follow the most downstream pattern highlights an important implication: export composition of the most developed EU members is mostly final good oriented in international markets. (a) Panel A: Developed Group (b) Panel B: Developing Group Figure 6: Evolution of Import Upstreamness The picture becomes more interesting when we analyze how Turkish import upstreamness on average moves compared to the same set of countries (Figure 6) and investigate which one of these country groups resembles Turkey in terms of the production line position. While OECD and EU15 members move parallel to each other, Turkey and BRICS countries follow an exactly opposite pattern. Turkey similar to BRICS average imports mostly upstream products from world markets. Towards 2014, as both of them experience a fall in export 30
- 4 EMPIRICAL RESULTS upstreamness , accompanied by a rise in import upstreamness, they could have increased the number of production stages conducted domestically. Thus, when we consider the position of Turkey in GVCs, it is more similar to BRICS countries than EU15 or OECD members. Apart from this, Turkish imports on average are slightly more upstream than those of BRICS. Even though this pattern vanishes during the period between 2011-2012, Turkey earns its previous position back, pursuing a more upstream pattern 2013 onward. The intuition behind using Turkish export and import upstreamness is to rely on an analytical tool in interpreting the amount of domestic value added created within the boundaries of Turkey. As the import upstreamness is a composite measure of the imported materials from the rest of the world, we deduct that from the export upstreamness and obtain a measure of the span of production stages to see to what extent the imported materials are processed within the domestic plants and exported to the foreign countries to reap the benefit of that processing in terms of monetary units. As the upstreamness in a sense shows the number of stages for any type of commodity, the composite difference we compute in this paper between import and export upstreamness measures delineates the number of industrial stages in which a certain product gains value added before reaching the foreign final or intermediate goods demand. Another way of conducting such an analysis is to benefit from OECD’s TiVA (2018) Database which provides a great deal of GVC measures to characterize the position of countries in GVCs. Based on the Inter-Country Input and Output Tables for a list of 64 economies 17 and 36 industrial sectors in terms of ISIC Rev.3 nomenclature between 2005 and 2016, it gives insights on the domestic value added content of gross exports between any given trade partners. To enrich our discussion on the domestic value added creation and its relation to global value chains, we make use of the TiVA indicators to see how the results differ from our empirical findings. As defined in TiVA Database, the domestic value added content of exports by industry k in home country i to a partner country j represents the exported value added that has been generated anywhere in the domestic economy. In following this definition, we use the share of domestic value added in gross exports available between a given country against its trade partners.18 It is a domestic value added measure reflecting 17 In addition to individual OECD and non-OECD members, it also provides data on a selection of country groups such as OECD, European Union, ASEAN etc. 18 EXGR_DV ASH is the variable name attributed to this measure in OECD TiVA Guide. One caveat that OECD emphasises in using TiVA Database is that the main benefit of integration into value chains does not come from the share of domestic value added within the gross exports but the total amount of exported domestic value added in monetary units. To illustrate this viewpoint, it is noted that even though Vietnam’s domestic value added share has declined from 64 to 53 %, it has increased the total amount of domestic value added exported to the rest of the world by four times between 2005-2016. With an increase in the absolute value of gross exports, Vietnam might have benefited from attending the value chains from the lower portions but this fact might be sensitive to the foregone opportunity cost of benefiting from value 31
- 4 EMPIRICAL RESULTS how much value added generated anywhere in the domestic economy is embodied per unit of total gross exports . (a) Panel A: Developed Group (b) Panel B: Developing Group Figure 7: Evolution of Domestic Value Added Share of Gross Exports Invoking our previous analysis to outline Turkey’s movement in global value chains in comparison with developed and developing country groups, we also sketch the share of domestic value added embodied in gross exports across different years preserving the same country blocks ordering. Comparing the results of both empirical strategy, we observe striking similarities and differences. Note that the drastic rise in Turkish export upstreamness between 2006 and 2011, accompanied by a steady course of import upstreamness might bring about a fall in the capability of creating domestic value added. During the same period, the share of Turkish domestic value added content in gross exports also follows a decreasing pattern, suggesting that the reduction in the number of stages operated domestically can be further explained from TiVA indicators. The only outlier movement within the specified time period occurs during 2008 global financial crisis. However; considering the weakening global demand conditions which contracted the worldwide trade volume to some extent, the fall in gross exports might have well resulted in the sparking shape in domestic value added share, which could be also observed by the decelerating pattern of Turkish export upstreamness in the crisis period in which the global demand for intermediate products relative to final goods had also slumped significantly. As the domestic value added share recovers after 2011 following a moderate course, Turkish export upstreamness also decreases with a firming response of import upstreamness, which we ascribe to a possible enhancement in the domestic span of production stages. chains at final portions. Extracting a newly found natural resource and exporting that with little amount of processing to international markets might elevate the amount of export revenues and thus national income but leaving out the advanced stages in processing to other countries could well deprive the exporting country of the foregone profits earned in the subsequent stages. Hence, we take into account this factor in interpreting our empirical regularities. 32
- 4 EMPIRICAL RESULTS Panel B of the same figure depicts the movement of domestic value added share of Turkey compared to developing country groups . As in the case of import upstreamness, Turkey’s domestic value added share moves closer to BRICS countries on average, above the new EU members but below both the OECD and EU15 average. Turkish import upstreamness moves at a higher level than BRICS average with a lower level of export upstreamness whereas the domestic value added share for BRICS countries follows above Turkish level up to 2012. The difference could be attributed to the reason that even if the imported materials to Turkey from the rest of the world might visit a higher number of stages in domestic plants, the composition of gross exports by BRICS countries might generally be made up of products on which greater amount of value added have been induced compared to Turkish gross exports. 4.4.2 Upstreamness of Turkey by Income Quintiles Figure 8: Turkey’s Export Upstreamness by Income Quntiles Using the bilateral upstreamness between Turkey and its trade partners, we can keep track of how Turkey’s export and import upstreamness changes over time across different income groups. Here we divide the income level of Turkey’s trade partners into 5 different group, each of which yields one quintile for a given year. While Q1 includes the poorest trade partners of Turkey, Q5 consists of the richest group of countries. Figure Figure 8 captures the 33
- 4 EMPIRICAL RESULTS Figure 9 : Turkey’s Import Upstreamness by Income Quintiles evolution of Turkey’s export upstreamness against these different income groups over time. It is immediate to observe that Turkey’s export composition shows the most downstream pattern against the richest country groups on average. Hence, Turkish exporters mainly target the final customers when they export to high income countries. Figure 9 displays the picture for the import upstreamness. Even if Turkish production line position cannot be easily disentangled in the way we analyze its export upstreamness, the import upstreamness of all income quintiles converge to the same value when we move to the end of our data period. Based on this convergence, we can infer that the upstreamness of Turkey’s import composition does not differ significantly across the income level of its trading partners towards 2014. When we combine this result with our previous finding that Turkey mostly sells final goods to high income countries, we can observe that Turkey succeeds in deepening the span of production stages for the product types supplied to richer countries. In other words, consistent with the key findings outlined in OECD-WTO (2015) reporting a significant improvement in the content of Turkey’s value added exports embodied in exports of final products between 1995 and 2011, Turkey’s integration to GVCs might be more oriented towards downstream activities. Hence, recalling the results of second model tested in Table 8, we find another empirical support to the hypothesis that if one of the trade partners attains a higher income group, the composition of its imports from the other partner will show a much more downstream characteristic. 34
- 4 EMPIRICAL RESULTS 4 .4.3 Sectoral Decomposition of Upstreamness of Turkey When we examine the export and import upstreamness of Turkey across its trade partners, we are able to interpret whether Turkey could have accomplished to enlarge the span of production stages operated within its domestic borders between 2003 and 2014. However; the export and import upstreamness values to reach such a result are the aggregate measures computed using all goods available in our data set. This process leaves out the individual impact of each product type on the enlargement of production stages. In other words, the relative importance of different sectors in creating the aggregate value added still remains uncovered. Hence, the interpretation of how in aggregate terms Turkey can contribute more value added to the production of export goods requires an understanding of how this contribution takes place within each industry category and how Turkey’s capability to create value added varies across different sectors. In this section, we employ a sector-based analysis of Turkey’s export and import upstreamness measures to outline the capability of each sector in enhancing the span of production stages. To do that, we decompose our upstreamness measures with respect to the sectoral classification of HS4 goods. Table 10: The First Two Digits of Harmonized System Codes (HS4) and Their Corresponding Industry Names HS Code 01-05 06-15 16-24 25-27 28-38 39-40 41-43 44-49 50-63 64-67 68-71 72-83 84-85 86-89 90-97 Industry Name Animal and Animal Products Vegetable Products Foodstuffs Mineral Products Chemicals and Allied Industries Plastics and Rubbers Raw Hides, Skins, Leather and Furs Wood and Wood Products Textiles Footwear and Headgear Stone and Glass Metals Machinery and Electrical Transportation Miscellaneous The fact that each product type we use in our analysis is enumerated according to a harmonized system allows us to keep track of the origin industry to which that product in question belongs and to characterize the production line position of its origin industry. 35
- 4 EMPIRICAL RESULTS Using Table 10 , we can restrict the sample of products to each HS category and compute the following sector-based export and import upstreamness measures: UTMU R,t,S = 1 J j k∈S MT U R,j,k,t Uk MT U R,j,t UTXU R,t,S = 1 J j k∈S XT U R,j,k,t Uk XT U R,j,t (17) where S is the respective set of HS4 commodities (k) covered by each industry classified according to the first two digits of HS4 codes in Table 10. J denotes the available number of Turkey’s trade partners. Note that we weight the upstreamness values for each product with Turkey’s bilateral trade volume with a given country j and then divide the ultimate result with the number of trade partners to obtain an average measure. As we described previously, the difference between export and import upstreamness values is used as a proxy for value added produced domestically. Following this definition, sector based export and import upstreamness measures can allow us to observe if the (weighted) number of production stages operated within Turkey expands or contracts for a given industry at a point in time. Thus, we end up with the measure of span of production stages for each industry at time t: SP ST U R,t,S 1 = J j X M T U R,j,k,t − T U R,j,k,t Uk MT U R,j,t XT U R,j,t k∈S (18) Using this formula, we can see how the production patterns within each specific industry changes over time. To have an accurate interpretation on how the overall capability of Turkish value added creation moves over time, we sum this measure across all possible years. This method helps us understand how much value added in total is created within the production process of each industry between 2003-2014. A positive sign denotes a cumulative enhancement in the span of production stages whereas a negative sign stands for any possible contraction in the depth of production stages. Figure 9 indicates the cumulative value of SP S for each sector between 2003 and 2014. Note that all sectors have a positive value in this period, meaning that the accumulated number of production stages conducted in domestic plants has risen in each industry. Thus, even if we see a contraction in the depth of the production stages in some years, the overall impact turns out to be positive. Nevertheless, the magnitude of the enlargement of production stages varies a lot across different sectors. To clarify, using Figure 10, we observe that Chemical and Allied Industries (28-38) have the greatest capacity to create value added in Turkey. Therefore, it has the highest potential to convert high-upstream imported materials into downstream export products, maximizing the distance between sector based import 36
- 4 EMPIRICAL RESULTS Note : Cumulative span of production stages is a measure of how much domestic value added is accumulated within the production process of each industry during the specified time period. Figure 10: Cumulative Span of Production Stages for HS4-defined Sectors and export upstreamness measures. Another result is that textile products (50-63) constitute the second leading industry in which Turkish manufacturers engage in high amount of processing to transform imported inputs into final goods to resupply to the world markets in the form of downstream textile products, which is in line with the common finding in the literature that textile industry is one of the integral powerhouses of Turkish economy whose capacity of creating value added has substantially been influenced by the export-oriented industrialization attempts starting from the early 1980s (Taymaz & Voyvoda, 2012). Similar to our analysis we embraced to point out certain differences and similarities in empirical results between the method of upstreamness and value added indicators by Trade in Value Added (TiVA) Database, we also consider the relative share of the domestic value added in an industry’s exports within the gross exports, which captures the domestic value added (DVA) contribution of a certain industry to gross exports compared to other industries. Table 11 in Appendix section yields a list of selected industries ordered with highest to lowest level of domestic value added content of gross exports (as a percentage of gross exports). On the basis of average contribution of selected industries in TiVA Database, Wholesale industry generates the greatest source of domestic value added content of exports, accounting 37
- 5 CONCLUSION for almost 9 .87 percent of gross exports between 2005-2015. Though definitions of industry categories are slightly different, we can still notice the resembling features between both analysis. Consistent with Figure 10, Textile and Chemical Industries constitute a numerous amount of the share of domestic value added in gross exports. Thus, the cumulative enhancement of production stages in these industries can be concurrently observed by their relative importance in creating domestic value added to be exported to partner countries. 5 Conclusion Employing the notion of upstreamness together with UN Comtrade data for the period 20032014 and industry specific upstreamness measures from Antrás et al. (2012), we develop the notion of bilateral upstreamness and empirically present how it is correlated with country level and gravity variables. Fundamental rational behind adopting such a bilateral framework is not to be confined only on unilateral upstreamness measures but to create a versatile measure of upstreamness notion which varies in accordance with trade characteristics between any given pair of partner countries. Computation of export and import upstreamness on the basis of this bilateral set-up can allow us more consistently locate the position a certain country occupies in global value chains compared to the unilateral methodologies. We find that countries differ particularly in terms of the export upstreamness, meaning that while some specialize in the production of final goods, the others are more inclined to stay at intermediate stages of production. Moreover, we show how export and import upstreamness are related to strength of institutions, changes in per capita real GDP and years of schooling. Even if we do not see a clear pattern for import upstreamness, the figures imply that as the countries get richer, adopt new institutions making the rule of law a predominant factor in administration and invest in policies increasing the years of schooling, they have a propensity to be exporting in more downstream sectors. In addition to basic correlation plots, we formally assess the impact of country specific characteristics on upstreamness measures, consisting of mainly two approaches. In the first approach, we test theoretical predictions using the unilateral upstreamness measures. Invoking the empirical results in Antrás et al. (2012); better rule of law, higher level of financial development and investment in human capital lead the export composition of countries to be more final good-oriented in international markets. Furthermore, country level characteristics are more prevalent in explaining the variation in upstreamness measures from supply channel rather than demand channel. Second, depending on the idea that a country’s position in global value chains could vary across its different trade partners, we define the notion of bilateral upstreamness and test how upstreamness values computed within a bilateral 38
- 5 CONCLUSION framework are affected from country specific factors . We find that as the income level of an exporter country rises, this might allow its trade partner to contribute more value added to the imported products. Similar to results with unilateral methodology, better institutions, a higher level of financial development and human capital in a country may allow that to export more downstream products to its trade partner. Likewise, a rise in the income level and quality of institutions in a destination country might change the exporting behavior of its partner in the same fashion, making it supply more downstream products. A higher ratio of private credits to GDP and increasing intensity of capital per worker can induce an importer country to have a bigger import upstreamness whereas productivity gains and improvement in years of schooling lead it to import more downstream products. Introducing the gravity variables to our bilateral analysis, we find a strong evidence bearing resemblance to the other empirical GVC studies (Ignatenko, Raei, & Mircheva, 2019) that geographical and institutional factors can have an impact on the bilateral upstreamness between a source and a destination country. According to our empirical results, as the distance between two trading partners increases, the exporter country is more involved in selling final goods to the importer country. If trade partners speak the same language, then this would reduce the export upstreamness of the source country. While having a common border allows the destination country to import more upstream products, signing a free trade agreement with the origin country might either raise or reduce its import upstreamness depending on the fixed effects in action. Estimating the bilateral specifications using a method of weighted least squares taking the share of exports or imports in total trade volume as weights might also play a role in magnifying the bilateral relationships at hand. To illustrate our bilateral analysis using the evolution of upstreamness measures, we describe the production line position of Turkey. We are first concerned with how Turkey’s export and import upstreamness move over time against developing and developed country blocks. Starting from mid 2000s, Turkey is mostly located in intermediate stages in production line, yet this tendency appears to be losing momentum towards 2011 and Turkish exports are composed of more downstream products. Moreover, Turkey’s import upstreamness follows a pattern which is more similar to the average of BRICS members than the average of EU15 and OECD members which follow a parallel trend over time. This type of resemblance between Turkey and BRICS countries might suggest that Turkey occupies a closer position to BRICS countries in global value chains. Generally speaking, they have a higher level of import upstreamness than EU15 and OECD average over the whole period. Towards the end of our period, their export upstreamness measures have a propensity to diminish at a larger extent, suggesting that the span of domestic production stages has inevitably amplified. The second subsection approaches to the same question decomposing 39
- 5 CONCLUSION Turkey ’s trade partners across five different income groups. As its trade partners get richer, Turkey mostly sells downstream products. In case of import upstreamness, we cannot easily determine its average production line position across different income quintiles. Still, towards 2014, Turkey’s import upstreamness values across different groups converge to the same level regardless of the income level of countries with which it shares a trade activity. To enrich our discussion on value added and upstreamness in previous two subsections, we also benefit from the principal indicators of OECD’s Trade in Value Added (TiVA) Database. We observe that certain types of similarities and differences emerge in comparing both of these GVC positioning methods. Decomposing the aggregate upstreamness measures with respect to each of industries as provided in Harmonized System (HS), we construct sector based upstreamness measures. Using the definition on the amount of value added created domestically, we define a sector based span of production stages (SPS) measure to see how the creation of value added varies across different sectors. Our results indicate that chemical, textile and plastics constitute the three leading industries which can feature the greatest amount of cumulative value added in Turkey between 2003 and 2014. 40
- REFERENCES References Antr ás, P. & Chor, D. (2013). Organizing the Global Value Chains. Econometrica, 81 (6), 2127–2204. Antrás, P. & Chor, D. (2017). On the Measurement of Upstreamness and Downstreamness in Global Value Chains. Working Paper. Antrás, P., Chor, D., Fally, T., & Hillberry, R. (2012). Measuring the Upstreamness of Production and Trade Flows. American Economic Review Papers and Proceedings, 102 (3), 412–416. Antràs, P., de Gortari, A., & Itskhoki, O. (2017). Globalization, inequality and welfare. Journal of International Economics, 108 (100), 387–412. doi:10.1016/j.jinteco.2017.07 Barro, R. & Lee, J. W. (2013). A New Data Set of Educational Attainment in the World, 1950-2010. Journal of Development Economics, 104, 184–198. Bartelsman, E. J., Becker, R. A., & Levine, R. (2013). A New Database on Financial Development and Structure. Becker, R., Gray, W., & Marvakov, J. (2016). NBER-CES Manufacturing Industry Database. Blair, P. D. & Miller, R. E. (2009). Input-Output Analysis: Foundations and Extensions. Second Edition. Cambridge University Press. Chor, D., Manova, K., & Yu, Z. (2017). The Global Production Line Position of Chinese Firms. Unpublished. Costinot, A., Vogel, J., & Wang, S. (2012, May). An Elementary Theory of Global Supply Chains. The Review of Economic Studies, 80 (1), 109–144. doi:10.1093/restud/rds023 European-Commission. (2019). European Neighbourhood Policy And Enlargement Negotiations. Retrieved from https://ec.europa.eu/. Fally, T. (2011). On the Fragmentation of Production in the US. Unpublished. Feenstra, R. C., Inklaar, R., & Timmer, M. P. (2015). The Next Generation of Penn World Table. American Economic Review, 105 (10), 3150–3182. Retrieved from https://ggdc. net/pwt/. Head, K., Mayer, T., & Ries, J. (2010). The Erosion of Colonial Trade Linkages After Independence. Journal of International Economics, 81 (1), 1–14. Hirschman, A. O. (1958). The Strategy of Economic Development. Volume 10. Yale University Press. Ignatenko, A., Raei, F., & Mircheva, B. (2019). Global Value Chains: What are the Benefits and Why Do Countries Participate? IMF Working Paper. International Monetary Fund WP/19/18. 41
- REFERENCES Kaufmann , D., Kraay, A., & Mastruzzi, M. (2017). World Governance Indicators: Methodology and Analytical Issues. World Bank dataset. OECD. (2019). List of OECD Member Countries - Ratification of the Convention on the OECD. Retrieved from https://http://www.oecd.org/. OECD-WTO. (2015). Trade in Value Added: Turkey. Rajan, R. G. & Zingales, L. (1998). Financial Dependence and Growth. American Economic Review, 88 (3), 559–586. Taymaz, E. & Voyvoda, E. (2012). Marching to the beat of a late drummer: Turkey’s experience of neoliberal industrialization since 1980. New Perspectives on Turkey, 47, 83–113. doi:10.1017/S0896634600001710 TiVA, O. (2018). Guide to OECD’s Trade in Value Added (TiVA) Indicators. OECD, Directorate for Science, Technology and Innovation. UNComtrade. (2019). United Nations Commodity Trade Data. Retrieved from https : / / comtrade.un.org/. WorldBank. (2019). World Development Indicators. Retrieved from https : / / databank . worldbank.org/. WTO. (2019). Global Value Chain Development Report: Technological Innovation, Supply Chain Trade, and Workers in a Globalized Economy. World Trade Organization. 42
- Appendices Table 14 : Weighted Least Square (WLS) Estimation Results for the Impact of Country Level and Gravity Variables on Bilateral Upstreamness (1) (2) (3) (4) 0.0283*** (0.00381) -0.141*** (0.00259) -0.200*** (0.00477) 0.0714*** (0.00345) 0.688*** (0.0184) -0.00473*** (0.000810) 0.0721*** (0.00824) -0.121*** (0.00327) 0.253*** (0.00441) -0.107*** (0.00643) -0.317*** (0.0355) -0.0289*** (0.00145) 0.0283 (0.133) -0.141** (0.0674) -0.200* (0.105) 0.0714 (0.110) 0.688* (0.395) -0.00473 (0.0220) 0.0721 (0.146) -0.121 (0.0873) 0.253* (0.130) -0.107 (0.118) -0.317 (0.429) -0.0289 (0.0204) 0.0962*** (0.00201) -0.115*** (0.00357) 0.160*** (0.00454) -0.146*** (0.00388) YES NO 0.0109*** (0.00219) -0.0980*** (0.00389) 0.114*** (0.00612) -0.178*** (0.00436) NO YES 0.0962 (0.0643) -0.115 (0.0986) 0.160** (0.0644) -0.146 (0.118) YES NO 0.0109 (0.0415) -0.0980 (0.0692) 0.114** (0.0556) -0.178** (0.0691) NO YES 184,948 0.928 179,004 0.904 184,948 0.928 179,004 0.904 0.334*** (0.00453) -0.166*** (0.00178) -0.245*** (0.00262) -0.165*** (0.00363) -0.0471** (0.0195) -0.0142*** (0.000769) -0.0712*** (0.00242) 0.00696*** (0.00165) 0.0226*** (0.00305) 0.0687*** (0.00220) -0.105*** (0.0117) -0.00665*** (0.000517) 0.334*** (0.116) -0.166*** (0.0528) -0.245*** (0.0742) -0.165* (0.0964) -0.0471 (0.224) -0.0142 (0.0197) -0.0712 (0.0603) 0.00696 (0.0289) 0.0226 (0.0647) 0.0687 (0.0472) -0.105 (0.146) -0.00665 (0.00972) -0.0843*** (0.00124) 0.0192*** (0.00210) -0.0313*** (0.00318) -0.00167 (0.00243) YES NO 0.0113*** (0.00137) -0.0904*** (0.00234) -0.00342 (0.00280) -0.0441*** (0.00246) NO YES -0.0843*** (0.0247) 0.0192 (0.0402) -0.0313 (0.0321) -0.00167 (0.0392) YES NO 0.0113 (0.0350) -0.0904** (0.0388) -0.00342 (0.0539) -0.0441 (0.0491) NO YES 184,948 0.954 179,004 0.963 184,948 0.954 179,004 0.963 Dependent Variable Uijt Panel A Log(Real GDP per capita) Rule of Law Private Credit / GDP Log (Capital per worker) TFP (2011=1) Years of Schooling Log (Distance) Common Language (=1) Contiguity (=1) Free Trade Agreement (=1) Destination-Year Fixed Effects? Source-Year Fixed Effects? N R2 Panel B Log (Real GDP per capita) Rule of Law Private Credit / GDP Log (Capital per worker) TFP (2011=1) Years of Schooling Log (Distance) Common Language (=1) Contiguity (=1) Free Trade Agreement (=1) Destination-Year Fixed Effects? Source-Year Fixed Effects? N R2 Notes: ***,**, and * denote significance at the 1%, 5% and 10% levels respectively. The weight used to estimate the regressions in Panel A is the bilateral share of exports in total trade volume for each individual country whereas in Panel B we use the bilateral share of imports as a weight. Standard errors of model (3) and (4) are clustered around source and destination country pairs to account for any possible clustering between trade partners. 43
- Figure 11 : The World Log Trade Volume 44
- Table 11 : The Percentage of DVA Contribution of Selected Industries in TiVA Database (2005-2015) Industry Percentage Wholesale Textiles and wearing apparel Basic metals and fabricated metal products Transportation and storage Transport equipment Chemicals and non-metallic mineral products Motor vehicles, trailers and semi-trailers Basic metals Accommodation and food services Food products, beverages and tobacco Computers, electronic and electrical equipment Information, finance, real estate and other business services Electrical equipment Agriculture Fabricated metal products Chemicals and pharmaceutical products Machinery and equipment Rubber and plastic products Other non-metallic mineral products Information industries Mining and quarrying Other social and personal services Arts, entertainment, recreation and other service activities Computer, electronic and optical products Coke and refined petroleum products Real estate activities Wood and paper products; printing Other transport equipment Financial and insurance activities Public admin, defence; education and health Information and communication Paper products and printing Telecommunications Education Wood and products of wood and cork Human health and social work Construction Publishing, audiovisual and broadcasting activities Public admin. and defence; compulsory social security 9.87 9.61 8.16 7.41 6.86 6.46 5.99 5.95 5.88 3.94 3.63 2.75 2.64 2.48 2.22 2.02 1.95 1.79 1.71 1.61 1.58 1.03 1.03 0.99 0.95 0.9 0.88 0.87 0.64 0.64 0.62 0.61 0.48 0.33 0.27 0.23 0.22 0.12 0.08 The industries selected from TiVA Database are denoted in ISIC REV.4 classification. Percentages are the averages across 2005-2015. 45
- 46 8 .95 [1.20] 3.94e-09 [1] 0.37 [0.36] -2.72 [1.37] 0.94 [0.13] 7.85 [2.84] 2003 8.99 [1.20] 7.62e-09 [1] 0.38 [0.36] -2.71 [1.37] 0.96 [0.12] 7.85 [2.84] 2004 Years 2006 2007 2008 2009 9.03 9.07 9.11 9.14 9.12 [1.20] [1.20] [1.20] [1.19] [1.16] -1.67e-08 1.62e-08 9.52e-08 1.48e-08 -5.19e-09 [1] [1] [1] [1] [1] 0.40 0.42 0.45 0.50 0.54 [0.39] [0.44] [0.47] [0.55] [0.66] -2.69 2.67 -2.65 -2.62 -2.61 [1.36] 1.36 [1.35] [1.35] [1.35] 0.97 0.99 1.00 1.00 0.98 [0.10] [0.105] [0.09] [0.09] [0.05] 7.85 7.85 7.85 8.34 8.34 [2.84] [2.84] [2.84] [2.81] [2.81] 2005 2011 9.15 9.17 [1.15] [1.15] -8.49e-09 6.54e-09 [1] [1] 0.54 0.48 [0.76] [0.41] -2.58 -2.56 [1.34] [1.33] 0.99 1 [0.02] [0] 8.34 8.34 [2.81] [2.81] 2010 9.20 [1.14] 9.81e-09 [1] 0.49 [0.41] -2.53 [1.33] 1.00 [0.03] 8.34 [2.81] 2012 9.22 [1.13] 1.54e-08 [1] 0.49 [0.41] -2.51 [1.32] 1.00 [0.06] 8.34 [2.84] 2013 9.23 [1.13] -1.44e-09 [1] 0.51 [0.40] -2.48 [1.31] 1.00 [0.07] 8.34 [2.84] 2014 9.12 [1.17] 3.64e-09 [0.99] 0.46 [0.49] -2.61 1.34 0.98 [0.08] 8.14 [2.82] 2003-2014 Notes: The unit of measurements are as follow: Log of Real GDP per capita (2010 constant US Dollars), Rule of Law (the estimate from World Governance Indicators), Private Credit / GDP (the ratio of private credits by deposit money banks and other financial institutions to GDP), Log of Capital per worker (Capital stock per employment calculated using Penn World Tables (PWT)), TFP (constant national prices 2011=1 from PWT), Years of Schooling (Barro-Lee’s education attainment data). Note also that since 2011 is the base year, all TFP values regarding with this year are taken as 1, naturally having a zero standard deviation. Years of Schooling TFP Log (Capital per worker) Private Credit / GDP Rule of Law Log (Real GDP per capita) Country Characteristics Table 12: Summary Statistics for Country Characteristics
- 47 Log (Real GDP per capita) Rule of Law Private Credit / GDP Log (Capital per worker) TFP Years of Schooling Country Characteristics 237 203 179 173 116 146 2003 237 210 180 173 116 146 2004 Years 237 210 181 173 116 146 2005 237 210 182 173 116 146 2006 238 210 182 173 116 146 2007 238 209 183 171 116 146 2008 238 212 179 169 116 146 2009 238 212 181 169 116 146 2010 242 214 179 169 116 146 2011 239 214 179 169 116 146 2012 Table 13: Number of Observations for Country Characteristics 239 214 178 169 116 146 2013 238 209 175 169 116 146 2014 2858 2527 2158 2050 1392 1752 2003-2014
- Figure 12 : Evolution of Turkish Export Upstreamness Figure 13: Evolution of Turkish Import Upstreamness 48
- Figure 14 : Standard Deviation of Turkey’s Export Upstreamness for each Income Quintile Figure 15: Standard Deviation of Turkey’s Export Upstreamness for each Income Quintile 49
- Figure 16 : The Correlation between Export-Import Upstreamness and Real GDP per capita varying across different years Figure 17: The Correlation between Export-Import Upstreamness and Rule of Law varying across different years 50
- Figure 18 : The Correlation between Export-Import Upstreamness and Private Credit / GDP varying across different years Figure 19: The Correlation between Export-Import Upstreamness and Capital per worker varying across different years 51
- Figure 20 : The Correlation between Export-Import Upstreamness and TFP varying across different years Figure 21: The Correlation between Export-Import Upstreamness and Years of Schooling varying across different years 52
- Central Bank of the Republic of Turkey Recent Working Papers The complete list of Working Paper series can be found at Bank ’s website (http://www.tcmb.gov.tr) A Measure of Turkey's Sovereign and Banking Sector Credit Risk: Asset Swap Spreads (Abdullah Kazdal, Halil İbrahim Korkmaz, Doruk Küçüksaraç, Yiğit Onay Working Paper No. 20/07, May 2020) Nowcasting Turkish GDP Growth with Targeted Predictors: Fill in the Blanks (Mahmut Günay Working Paper No. 20/06, May 2020) Bank Lending and Maturity: the Anatomy of the Transmission of Monetary Policy (Selva Bahar Baziki, Tanju Çapacıoğlu Working Paper No. 20/05, March 2020) Do Local and Global Factors Impact the Emerging Markets’s Sovereign Yield Curves? Evidence from a Data-Rich Environment (Oğuzhan Çepni, İbrahim Ethem Güney, Doruk Küçüksaraç, Muhammed Hasan Yılmaz Working Paper No. 20/04, March 2020) The Role of Imported Inputs in Pass-through Dynamics (Dilara Ertuğ, Pınar Özlü, M. Utku Özmen, Çağlar Yüncüler Working Paper No. 20/03, February 2020) Nowcasting Turkish GDP with MIDAS: Role of Functional Form of the Lag Polynomial (Mahmut Günay Working Paper No. 20/02, February 2020) How Do Credits Dollarize? The Role of Firm’s Natural Hedges, Banks’ Core and Non-Core Liabilities (Fatih Yılmaz Working Paper No. 20/01, February 2020) Hidden Reserves as an Alternative Channel of Firm Finance in a Major Developing Economy (İbrahim Yarba Working Paper No. 19/36, December 2019) Interaction of Monetary and Fiscal Policies in Turkey (Tayyar Büyükbaşaran, Cem Çebi, Erdal Yılmaz Working Paper No. 19/35, December 2019) Cyclically Adjusted Current Account Balance of Turkey (Okan Eren, Gülnihal Tüzün Working Paper No. 19/34, December 2019) Term Premium in Turkish Lira Interest Rates (Halil İbrahim Aydın, Özgür Özel Working Paper No. 19/33, December 2019) Decomposing Uncertainty in Turkey into Its Determinants (Emine Meltem Baştan, Ümit Özlale Working Paper No. 19/32, December 2019) Demographic Transition and Inflation in Emerging Economies (M. Koray Kalafatcılar, M. Utku Özmen Working Paper No. 19/31, December 2019) Facts on Business Dynamism in Turkey (Ufuk Akçiğit, Yusuf Emre Akgündüz, Seyit Mümin Cılasun, Elif Özcan Tok, Fatih Yılmaz Working Paper No. 19/30, September 2019) Monitoring and Forecasting Cyclical Dynamics in Bank Credits: Evidence from Turkish Banking Sector (Mehmet Selman Çolak, İbrahim Ethem Güney, Ahmet Şenol, Muhammed Hasan Yılmaz Working Paper No. 19/29, September 2019) Intraday Volume-Volatility Nexus in the FX Markets: Evidence from an Emerging Market (Süleyman Serdengeçti, Ahmet Şensoy Working Paper No. 19/28, September 2019) Is There Asymmetry between GDP and Labor Market Variables in Turkey under Okun’s Law? (Evren Erdoğan Coşar, Ayşe Arzu Yavuz Working Paper No. 19/27, September 2019) Composing High-Frequency Financial Conditions Index and Implications for Economic Activity (Abdullah Kazdal, Halil İbrahim Korkmaz, Muhammed Hasan Yılmaz Working Paper No. 19/26, September 2019) A Bayesian VAR Approach to Short-Term Inflation Forecasting (Fethi Öğünç Working Paper No. 19/25, August 2019) Foreign Currency Debt and the Exchange Rate Pass-Through (Salih Fendoğlu, Mehmet Selman Çolak, Yavuz Selim Hacıhasanoğlu Working Paper No. 19/24, August 2019)
Create FREE account or Login to add your comment