Master thesis
The impact of deep trade integration on
production fragmentation
Laura Baiker
matriculation no. 3108023
supervised by Prof. Dr. Bart van Ark
Abstract
Contents
1 Introduction 3
2 Literature review and theoretical considerations 5
2.1 Evidence on the relationship between gross trade and RTAs . . . 6
2.2 RTAs and (new) measures of production fragmentation . . . 7
2.3 Motivation and hypothesis . . . 9
3 Data 11 3.1 Valued added to export (VAX) ratio . . . 11
3.2 Trade agreements . . . 12 3.3 Sample statistics . . . 15 4 Methodology 16 4.1 Empirical specification . . . 18 5 Estimation results 18 5.1 Baseline specification . . . 18 5.2 Robustness checks . . . 22
6 A tale of two countries: India and Turkey 22 6.1 Data . . . 26
6.2 Measuring value added exports . . . 26
6.3 Descriptive analysis . . . 28
6.3.1 India . . . 29
6.3.2 Turkey . . . 32
7 Conclusion 35
List of Tables
1 Sample statistics . . . 15
2 Panel regressions with DESTA depth indices . . . 19
3 Panel regressions with number of provisions . . . 21
4 Panel regressions by period . . . 21
5 Panel regressions with first-differenced data . . . 23
6 Average annual growth rates for India (in percent) . . . 31
7 Average annual growth rates for Turkey (in percent) . . . 32
8 List of countries covered in regression analysis . . . 42
9 List of policy areas covered in depth indices . . . 43
10 List of variables used in regression analysis . . . 44
11 List of trade agreements signed by India . . . 45
12 List of trade agreements signed by Turkey . . . 46
List of Figures
1 World VAX ratio, 1970-2009 . . . 122 Average depth of trade agreements, 1970-2009 . . . 13
3 Number of agreements that cover GVC-related provisions . . . 14
4 VAX ratio of middle-income countries . . . 29
5 VAX ratio of India, 1995-2011 . . . 30
6 VAX ratio of Turkey, 1995-2011 . . . 33
1
Introduction
In the last 25 years, we have seen a major boom in the number of trade agreements that were signed between countries. Since the foundation of the World Trade Organization (WTO) in 1995, more than 400 agreements have been notified whereas from 1948 to 1994, only 124 arrangements were in force (WTO 2017). This trend of ever-tighter integration and cooperation across the globe was accompanied by enormously rapid growth in trade volumes. Between 1970 and 2007, exports of goods and services increased by 6.5 times, which is equivalent to an average annual growth rate of 5.6 percent.1
As international trade seems to have become more intense in recent history, the pro-duction of goods and services has also become more spatially dispersed leading to the emergence of the global value chain (GVC). Countries are no longer trading wine against cloth but instead are increasingly engaged in trade with intermediate inputs. Several countries contribute value added to the production process and goods have to cross mul-tiple borders until they reach their ultimate destination. In particular, countries import intermediate goods, they add domestic value added and then re-export the product either as final good or input to the production process.
Due to this fragmentation of production, economists are in need for trade measures that capture the actual value added of countries as gross trade statistics became mis-leading. This implies to follow goods through the GVC from input suppliers to final consumption and to track value added at each stage in the production process. Hummels et al. (2001) were the first that consider vertical specialization, which describes the phe-nomenon that countries specialize in particular stages of a good’s production. It is equal to the amount of imported intermediates embodied in goods that are exported further.
The increasing availability of multiregional input-output tables that trace industrial interdependences worldwide facilitated empirical research and new measures for GVC trade were developed. This paper makes use of a newly available panel dataset on bi-lateral value-added exports which was constructed by Johnson and Noguera (2012). The authors consider the ratio of value-added exports relative to gross exports (VAX) as an indicator for the intensity of production sharing. In particular, it captures the amount of value added from a given source country that is embodied in the final demand in each destination.
Besides this, there is an on-going debate in the literature with respect to the impli-cations of regional trade agreements (RTAs) on production fragmentation since earlier research has mainly focused on gross trade statistics. However, Yi (2003) points to a puzzle in literature: the large increases in aggregate trade in the last decades were ac-companied by only modest tariff reductions. He argues that vertical specialization is the key force as it is more sensitive to trade costs than regular trade. Because production structures are shared, goods have to cross several borders until they reach the destination of final demand, which implies that tariffs are paid multiple times.
With the help of the gravity model and value added data on country pairs, recent studies provide quantitative evidence that common membership in RTAs fosters bilateral production sharing (like it fosters gross trade). However, the present study goes one step further and takes into account that agreements are heterogeneous with respect to the pro-visions that are included in such arrangements. For instance, the European Union (EU) and the North American Free Trade Agreement (NAFTA) contain relatively deep regula-tions such as standards, investments, and competition policy, whereas the Association of Southeast Asian Nations (ASEAN) free trade area (AFTA) fails to promote liberalization in these areas. Since RTAs differ in their depth, estimates of an average treatment effect on production fragmentation might not be informative. Instead, it is hypothesized that shallow agreements, which solely focus on tariff reductions, are less likely to promote ver-tical specialization than deep RTAs. Since international production sharing facilities and business activities require coordination and management across countries, the impact of deep RTAs is more pronounced in the presence of fragmentation as this goes beyond a simple exchange of goods and services.
The contribution of this paper is twofold. First, the dataset from Johnson and Noguera (2012) on bilateral value-added exports is exploited to conduct a comprehensive panel analysis on the implications of deep trade agreements for 42 countries between 1970 and 2009. Second, using data from the World Input-Output Database (WIOD, Timmer et al. (2015)), the regression results are tested and interpreted in the context of a descriptive analysis for two major emerging economies: India and Turkey.
In order to measure the ‘depth’ of RTAs, I draw on two different data sources that have become recently available: The Design of Trade Agreements (DESTA, D¨ur et al. 2014), and a dataset on the content of deep agreements provided by the World Bank (Hofmann et al. 2017). Both comprise information on the provisions that are included in RTAs and thus offer an accurate measure of the degree of trade liberalization. Such provisions are, for instance, competition policy, intellectual property rights or regulations in service trade. The data allows me to investigate whether the adoption of bilateral and multilateral agreements has an influence on the pattern of production sharing between two trading partners. The panel regression is conducted using a gravity equation for bilateral trade that is estimated with a fixed-effects model. An important issue which is addressed in this framework is the potential endogeneity of RTAs. Countries might select into agreements for reasons that are also correlated with current bilateral trade flows. The common solution to that is to use country-pair fixed effects that control for unobserved heterogeneity across countries (Baier and Bergstrand 2007).
economies due to automation (Timmer et al. 2016) are most likely to be responsible for the declines in global fragmentation.
The underlying study provides clear evidence that comprehensive trade agreements have a significant impact on production fragmentation. The regression analysis shows that deep RTAs lower the VAX ratio on average by 2 to 3.5 percent. Moreover, an additional provision included in trade agreements reduces the VAX ratio on average by 0.5 to 0.9 percent. The RTA effect appears to be greatest in the medium run (i.e. between 10 and 15 years) and in the long run (i.e. more than 15 years) after the adoption, which suggests that GVC activity requires large investments for setting up production facilities, and thus RTAs only bear fruit after some time. These findings are also reflected in the case of Turkey and India. The descriptive analysis reveals that production sharing has become more prevalent since the mid-1990s. This trend was supported by the adoption of deep trade agreements that do not only focus on the reduction of industrial tariffs, but also on provisions that facilitate integration into GVCs. However, I also find evidence on a recent slowdown in production fragmentation. Apart from technological advances, the deceleration of deep agreements in Turkey and India is likely to contribute to this outcome as well.
The remainder of this paper is organized as follows. The next section reviews the existing literature on the impact of RTAs on gross trade and value-added trade. The focus lies on studies that use post estimation techniques in a gravity-style setting. Moreover, the disadvantages of using gross trade statistics in our fragmented world are discussed in more detail. Inspired by these findings, I further outline the theoretical considerations underlying my hypothesis. Section 3 describes the data that are utilized and provides descriptive statistics of the sample. In Section 4, I explain the methodology and empirical specification. The fixed-effects model is applied to estimate the gravity equation due to the endogenous formation of the RTA dummy. Section 5 presents estimation results and robustness checks. In Section 6, computations and results of the input-output analysis for India and Turkey are presented in order to check the panel estimates obtained in Section 5 for two emerging economies. I will further examine vertical specialization trends in post-crisis years and discuss the reasons for the current GVC slowdown that were found in the literature. Finally, Section 7 concludes.
2
Literature review and theoretical considerations
2.1
Evidence on the relationship between gross trade and RTAs
There are numerous research papers that investigate the impact of RTAs on bilateral gross trade flows, however, the focus of this paper is on studies that use historical data and ex-post estimations. Since the 1970s, the gravity equation is the ‘workhorse’ to explain variations in cross-sectional and panel data on bilateral trade flows (Cipollina and Salvatici 2010). The results mostly point to a positive relationship between RTAs and trade volumes, but they are sensitive to datasets, sample coverage, and explanatory variables.
Earlier estimations that use cross-sectional data often yield insignificant or infinite small results on the trade impact of RTAs. As summarized by Baier and Bergstrand (2007), the first who applied the gravity model to bilateral trade was Tinbergen (1962). Besides a number of policy variables, he included dummy variables for individual RTAs. However, the coefficients appeared to be insignificant and small. For instance, the British Commonwealth increased countries’ trade flows by only 5 percent. Similarly, Bayoumi and Eichengreen (1997) estimate that the European Economic Community (EEC) and the European Free Trade Association (EFTA) increased trade by only 3.2 and 2.3 percent yearly within their observation period from 1956 to 1973.
Baier and Bergstrand (2007) emphasize that the presence or absence of RTAs is not exogenous to trade flows and thus past research underestimated the RTA variable. When controlling for this endogeneity, it appears that bilateral trade of RTA members doubles ten years after the adoption. The approach by Baier and Bergstrand (2007) is nowa-days commonly used in trade literature and multiple studies confirm their results (see e.g. Grant and Lambert 2008, Baier et al. 2008). Taking into account a vast amount of estimation results in the literature, Cipollina and Salvatici (2010) find a positive RTA mean effect of 40 percent in their meta-analysis after they filtered out potential estimation biases. These biases arise either due to omitted variables, data measurement, misspeci-fication or publication selection. The first two issues are also addressed in the empirical analysis of this paper and are discussed in Section 4.
2.2
RTAs and (new) measures of production fragmentation
The aforementioned studies all investigate the impact of RTAs on gross trade. However, Yi (2003) points out two puzzles that cannot be explained by standard trade models. First, since the 1960s, worldwide tariffs have decreased by only 11 percentage points whereas, at the same time, the manufacturing export share of GDP has grown by 340 percent. Second, the response of trade volumes to tariff reductions has increased since the mid-1980s. Whereas between 1962 and 1985, the elasticity of trade with respect to tariffs was equal to 7, it increased to 50 between 1986 and 1999. According to Yi (2003), the key to these puzzles is the occurrence of vertical specialization in recent decades. Vertical specialization describes the phenomenon that “countries specialize in particular stages of a good’s production. [. . . ] a country imports a good from another country, uses that good as an input in the production of its own good, and then exports its good to the next country.” (Hummels et al. 1998, p. 80). Thus, with production fragmentation, goods cross multiple borders until they reach their ultimate destination and in turn, tariffs are paid several times. Yi (2003) argues that tariff reductions exert a magnification effect on the decrease of production costs and thus on the increase of trade. Additionally, lower tariffs make it more efficient for countries to further specialize in particular stages of the production process which again leads to an increase in trade. Therefore, vertical specialization is more sensitive to trade liberalization than exports of final products, and overall trade growth is larger than predicted by standard trade theory.
Closely related to Yi (2003), several studies argue that gross trade statistics do not reflect the actual value added that countries contribute to the final product. More likely, gross exports overestimate countries’ export performance through double counting of value added as products cross borders multiple times. Thus, new measures are needed to assess the causes and consequences of production sharing. Two methodological approaches have emerged in literature: trade data on parts and components and measures based on national and global input-output tables. The former was initiated by Yeats (1999) and Ng and Yeats (2001) and is a simpler, but also less accurate measure of vertical trade (Amador and Cabral 2016). The big advantage of input-output tables is a finer classification of industries and the possibility to trace their interdependence in GVCs. In a pioneering study, Hummels et al. (2001) compute vertical specialization for ten OECD countries and four emerging economies between 1970 and 1990 with the help of national input-output tables. In their case, this is equal to the import content of goods that are exported further. They estimate that vertical specialization grew by almost 30 percent between 1970 and 1990.
each country participating in the GVC. They show that the ratio of value-added exports to standard exports is lower for intra-regional trade than for extra-regional trade. This might imply that production sharing is regionalized rather than globalised. Moreover, their results confirm the strong role of Asia being the major assembling location in the GVC since there is a large share contained in imports that is re-exported at a later stage. Similar to this approach is the VAX ratio proposed by Johnson and Noguera (2012). It is equal to the amount of domestic value added embodied in foreign final demand relative to gross exports and serves a proxy for the intensity of production sharing. Between 1970 and 2009, the ratio fell by 10 percentage points worldwide whereby the declines were quite heterogeneous across countries. Among nearby trading partners and RTA members, the decreases were largest. The authors show that nearly the entire reduction can be explained by changes in trade frictions whereas changes in endowments and productivity only play a minor role. Approximately 20 percent of the global decline in the VAX ratio can be explained by changes in trade frictions that follow from RTAs.
Koopman et al. (2014) combine these previous attempts in a unified framework and fully decompose gross exports into its various components. The authors show that vertical specialization has risen worldwide and to a particular extent in Asia. Developing countries tend to have the largest foreign value-added content in their gross exports which confirms their role as the assemblers in the GVC. For instance, Mexico and China have foreign value-added shares equal to 47 percent and 56 percent respectively.
In contrast to the vast amount of studies that investigate the impact of RTAs on gross trade flows, evidence for value-added trade is relatively scarce. One reason might be the high data requirements as measures of bilateral production fragmentation are needed, however, there are some studies that overcome this issue.
Following closely the framework of Hummels et al. (2001), Leung (2016) analyses the effect of FTAs on vertical specialization between the US and its trading partners from 1997 to 2012. She uses a simple FTA dummy that indicates whether two countries are members of the same agreement or not. Her results reveal that vertical specialization increases by 155 percent when the US forms an FTA with its trading partners. Moreover, Lopez-Gonzalez (2012) estimates the impact of FTAs on total and intermediate imports of countries involved in GVC and non-GVC activity. It appears that intermediate inputs of bilateral and international value chains increase by 65 percent when FTAs are in place whereas this is only half the size for non-GVC activities. The determinants of interme-diate and final good flows are analysed by Yane (2013) who uses data from the WIOD between 1995 and 2009. When countries are part of the East Asian trading bloc, trade in intermediates and trade in final goods is higher compared to countries belonging to the EU or NAFTA. Furthermore, intermediate goods respond more sensitive to trade costs than final goods.
explain these surprising findings with the fact that tariff concessions exist since the 1960s and that RTA formation substitutes for this rather than generating new incentives for fragmentation. However, the authors find a high significance for AFTA members trade which suggests that it is important to consider not only tariff reductions induced by RTAs but to look at non-tariff barriers that foster deeper integration among members of RTAs.
2.3
Motivation and hypothesis
The discussion above reveals that empirical research regarding the impact of trade liber-alization on production fragmentation is still in its infancy. However, at this point, it is important to acknowledge that there is broad consensus on two main issues in empirical research. First, trade agreements promote production fragmentation between member countries. The implication behind this is straightforward since RTAs lower tariffs (and sometimes non-tariff barriers) and thus reduce the costs of imports relative to domestic products. Therefore, countries specialize more and more in particular stages of the pro-duction chain and exchange intermediate inputs in GVCs with countries belonging to the same RTA.
Second, gross exports are more sensitive to RTAs than value-added exports. As al-ready mentioned, this magnification effect was firstly emphasized by Yi (2003). Providing empirical evidence, Noguera (2012) argues that trade in intermediates becomes more at-tractive when tariff barriers are reduced, but this multiple border crossing is counted only once in value-added exports. In contrast, gross exports are plagued by double counting and thus respond more strongly to trade liberalization than value added. More precisely, Noguera (2012) finds that the adoption of RTAs has raised countries’ countries’ value-added trade by 15 percent whereas gross trade is increased by 23 percent.
What has been neglected so far is that trade agreements are heterogeneous with re-spect to the extent of tariff reductions and provisions aiming at the reduction of non-tariff barriers. This implies that we have treated the EU, an economic union, to be equally im-portant for facilitating production sharing than the bilateral Chile-India PTA that solely lowered tariffs for some goods. Despite the improvements in measuring fragmentation, however, there is not much literature so far that looks on the implications of shallow versus deep RTAs for countries’ trade and production patterns.
decline of 10 to 15 percent is caused by deep RTAs (i.e. CUCMEUs).
In recent years, measurements of the depth of trade agreements became more sophis-ticated as researchers started to focus on the content of the signed contracts. They try to capture the regulations and trade facilitations that are introduced with the arrangement and to illustrate a more realistic view on the heterogeneity of RTAs. For instance, D¨ur et al. (2014) move beyond a simple categorization of RTAs and construct the DESTA data set on the design of trade agreements covering the years 1945 to 2009. For their depth measurement, the authors open the black box of RTAs and check whether provi-sions on intellectual property rights, competition policy, and foreign direct investments (FDI) among others are included in the agreements. They show that ‘shallow’ agree-ments have a smaller impact on gross trade than ‘deeper’ ones. The authors provide two distinct indices that are applied in the empirical analysis of this paper: (1) an additive index that counts the number of core provisions that go beyond tariff reductions; and (2) a continuous depth variable from latent trait analysis on various provisions.
Another main contribution is from Horn et al. (2010) and Hofmann et al. (2017) that construct datasets on the content of deep agreements. The authors collected information on legally enforceable provisions covered by RTAs that have been notified to the WTO. Orefice and Rocha (2014) construct three distinct depth indices for a sample of 200 coun-tries between 1980 and 2007 from these data sources. The first index counts the number of provisions included in trade arrangements, the second one counts the number of WTO+ provisions (commitments that fall under the current mandate of the WTO), and the third one counts the number of WTO-X provisions (obligations outside the current mandate of the WTO). Using import data on parts and components, the authors investigate the re-lationship between production networks trade and deep integration. They conclude that countries signing deeper commitments experience a greater increase in network trade. On the other hand, the authors also show that countries already engaged in production networks are more likely to sign deep RTAs.
The empirical analysis in this paper makes use of these two novel databases and em-ploys the depth indices as measures of main interest; further details are discussed in Section 3. To my knowledge, no study so far has investigated the impact of deep inte-gration on bilateral VAX ratios while exploiting these data sources. Formally, my first hypothesis can be written as follows:
Hypothesis. Deep RTAs exert a greater impact on bilateral production fragmenta-tion than shallow RTAs.
through which deep integration impacts on vertical specialization. First, assuming in-complete information and significant transaction cost, the provision of standards provides information on the quality of inputs which reduce buyers’ costs. Second, a common rule of law is essential to enforce contracts between buyers and sellers, mitigate uncertainties and ‘hold-up’ problems. Thus, deep integration is likely to facilitate input flows and the coordination of business activities, and thus makes production fragmentation more effi-cient. Furthermore, D¨ur et al. (2014) argue that RTAs liberalizing investment policies are likely to attract FDI which in turn promote vertical intra-industry trade. Therefore, besides tariff reductions, it is important to take the design of agreements into account to be able to isolate the actual impact of RTAs on production fragmentation.
Following the above-outlined considerations, the next section presents the data used to estimate RTA effects on production fragmentation between pairs of countries.
3
Data
The purpose of this paper is to estimate the impact of bilateral as well as multilateral trade agreements on the degree of production fragmentation between pairs of countries. This section presents summary statistics and describes the underlying data that is used in the panel analysis.
3.1
Valued added to export (VAX) ratio
The analysis is based on a data set constructed by Johnson and Noguera (2012) that contains bilateral gross exports, value-added exports, intermediate and final exports for 42 countries between 1970 and 2009. The countries are listed in Table 8 in the appendix. The authors combine data from trade statistics, national accounts and benchmark OECD and IDE-Jetro Asian input-output tables to construct a consistent time series of global input-output tables. Additionally, to obtain bilateral trade flows, the authors draw on bilateral commodity trade statistics and multilateral sums. Computations of value-added exports are explained in more detail in Section 6.
Of main interest is the VAX ratio: it captures the intensity of production sharing on the macro level between two countries and is employed as the dependent variable. The ratio is equal to value added exports – the amount of value added embodied in foreign final demand – relative to gross exports.
Figure 1: World VAX ratio, 1970-2009
3.2
Trade agreements
The analysis makes use of the DESTA database on the design of trade agreements that is provided by D¨ur et al. (2014). The authors included 587 RTAs signed between 1945 and 2009 with more than 100 items coded for each agreement. D¨ur et al. (2014) develop two indices for the depth of trade integration with RTAs. The first one is a simple additive index that ranges from 0 to 7. It measures whether tariff reductions and other provisions that aim at decreasing behind the border barriers are included as a clause in the agreement. The provisions of main interest are items on services, investments, standards, public procurement, competition, and intellectual property rights. For the purpose of my analysis, the depth index is rescaled so that bilateral country pairs that have an arrangement, but no important provision included, are not treated the same as countries that have no RTA in force at all. Thus, the index ranges now from 0 to 8, while 1 stands for ‘basic’ trade facilitations such as tariff reductions that induce at least some specific industries to trade more.
However, one drawback is that this index gives each item the same weight and ignores that some provisions are more important for trade than others. Therefore, the second index from D¨ur et al. (2014) is a continuous depth variable that is created from latent trait analysis on 48 variables. The relationship between various items and the depth of agreements is tested in a factor analysis for binary data. This method considers the frequency with which each provision is covered in trade agreements and grades them in a hierarchical way. This gives the degree to which each item is associated with the latent trait, i.e. with the depth of RTAs.
Figure 2: Average depth of trade agreements, 1970-2009
relatively stable but sharply increased thereafter. Thus, the global rise in production fragmentation that started in the early 1990s was accompanied by an expansion of deep RTAs. The descriptive statistics presented here can provide a first indication for the main hypothesis outlined in Section 2.
To meet the objectives of a comprehensive analysis of the relationship between deep trade agreements and production fragmentation, a second data source is used. Bilateral data on the content of deep trade agreements is derived from the World Bank (Hofmann et al. 2017). The complete dataset comprises 52 provisions in 279 RTAs that have been no-tified to the WTO between 1958 and 2015. The coverage of agreements is lower compared to the DESTA database since it excludes RTAs that are not notified to the WTO. In my sample. There are 42 agreements from the World Bank dataset included. The regression analysis based on this panel also serves to test the validity of the DESTA indicators since there was found some noise in the dataset.2
Hofmann et al. (2017) distinguish between WTO+ and WTO-X provisions and define the legal enforceability of each item. WTO+ refers to provisions that fall under the cur-rent mandate of the WTO and that are subject to some form of commitment in WTO agreements. Examples that fall into this category are industrial tariffs, technical barriers to trade or public procurement. Obligations that are outside the current mandate of the WTO are captured by the WTO-X provisions such as competition policy, movement of capital and innovation policies. The legal content of provisions is evaluated by assum-ing that clauses expressed with a clear and precise legal language are more likely to be
enforced. In contrast, loosely formulated legal language might be related to policy areas that are not legally enforceable.
Following the approach in Orefice and Rocha (2014), three additive indices are con-structed from this data. The first one simply counts the number of provisions that are included in trade arrangements and ranges from 0 to 52. The second one solely considers the WTO+ provisions (from 0 to 14) and the third one the WTO-X provisions (from 0 to 38). Only provisions that are legally enforceable are taken into account. A list of policy areas that are covered by the depth indicators from DESTA and World Bank database is provided in Table 9 in the appendix.
Figure 3: Number of agreements that cover GVC-related provisions
Figure 3 shows the number of trade agreements covered in my sample for ten different WTO+ and WTO-X provisions that are most likely to be related to GVCs.3 The graph includes agreements that are covered by the World Bank data and for which information on provisions is available. All RTAs have a protocol on industrial tariffs, and most include WTO+ provisions on customs, the Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPs), and the General Agreement on Trade in Services (GATS). However, fewer agreements include WTO-X provisions, for instance, competition policy is only regulated in 13 agreements and intellectual property rights (IPR) in 18 agreements. In sum, a large share of the RTAs in my sample contain provisions that are related to GVCs. However, provisions that are outside the current mandate of the WTO such as deeper regulations on competition, investment, legislation and property rights are less frequent.
3.3
Sample statistics
The underlying sample contains information on 1,722 country-pairs between 1970 and 2009. It includes 42 countries, from which 29 countries are developed and 13 emerging economies. In total, there are 56,732 pair-year observations. As suggested by Johnson and Noguera (2012), observations with VAX ratios above ten and gross exports below one million are dropped from the sample because of low data quality for the 1970s and early 1980s. Since serial correlation seemed to be an issue in the analysis, I use five-year periods of the data. This decreases the dataset to 11,080 observations.
Table 1: Sample statistics
Variable Obs. Mean Std. Dev. Min. Max.
Without RTA:
VAX ratio 7,400 0.982 0.525 0.362 9.36
Value added exports 7,400 1,235 5,663 1 165,801
Gross exports 7,400 1,491 6,694 1 211,547
With RTA:
VAX ratio 3,680 0.811 0.385 0.245 9.21
Value added exports 3,680 2,485 9,121 1 239,177
Gross exports 3,680 3,707 13,752 1 351,918 Depth index 3,680 3.52 1.87 1 8 Depth latent 3,680 1.21 0.936 0.008 3.21 Number of provisions 1,964 26.4 11.985 0 37 WTO+ provisions 1,964 12 3.152 0 14 WTO-X provisions 1,964 14.5 8.91 0 23
Notes: Value-added exports and gross exports are expressed in million USD.
Descriptive statistics of variables utilized in the analysis are presented in Table 1. For the sake of completeness, value-added exports and gross exports are added. In the underlying panel data, bilateral value-added flows belonging to RTA members are under-represented and make only 33 percent of the sample. Thus, to obtain a better picture of value-added trade and the heterogeneity of RTAs with respect to depth, summary statistics are presented separately for country-pairs with and without agreements.
4
Methodology
The gravity specification is utilized as the ‘workhorse’ to explain bilateral trade flows for more than 40 years. In its simplest form, the gravity equation estimates the determinants of trade flows with the distance between the two trading partners and their real Gross Domestic Products (GDP). Usually, binary control variables such as contiguity, common language or colonial origin are added to proxy for non-policy trade costs.
However, a well-known empirical obstacle in estimating the impact of trade agreements on export and import flows is the endogeneity of the RTA dummy. Baier et al. (2008) argue that the same variables that explain bilateral trade flows – i.e. the right-hand side of the gravity equation are also correlated with the RTA dummy since they might affect the likelihood of signing a trade agreement. For instance, governments select into agreements with neighbour countries or economies that have similar characteristics. Indeed, Baier and Bergstrand (2004) show that large economies – in terms of their GDP – are more likely to adopt trade agreements because of economies of scale and the love-of-variety effect. However, Baier and Bergstrand (2007) argue that the largest bias is caused by countries’ selection into agreements based on time-invariant unobserved heterogeneity. In particular, the error term might capture unobserved characteristics that simultaneously influence the volume of trade between countries and the likelihood that these countries adopt a trade agreement. For instance, when two countries have strong consumer preferences for each other’s products, their bilateral trade might be larger, and at the same time, these countries have a higher incentive to sign an RTA in order to facilitate future trade (Lopez-Gonzalez 2012). Therefore, these unmeasurable preferences are not only correlated with the independent variable (i.e. bilateral trade), but also with the regressor (i.e. engagement in an RTA).
Ordinary least square (OLS) estimates are likely to be biased and inconsistent when the RTA dummy correlates with the error term. In fact, Baier and Bergstrand (2007) demonstrate that traditional estimates are biased downwards by 75 to 85 percent. The most common approach to solve for endogeneity issues with cross-sectional data is the use of instrumental variables. However, it is difficult to identify variables that correlate with the RTA dummy, but not with bilateral trade flows since both statistical terms are affected by the same variables. Consequently, results from studies using an instrumental variable approach are rather mixed. Baier and Bergstrand (2007) argue that panel data offers the opportunity to control for endogeneity with a fixed-effects approach and thus to obtain unbiased estimates of the average treatment effect of the RTA dummy on trade. These estimates appear much larger in size and are more robust to sensitivity analyses than earlier studies that took the exogeneity of the RTA variable as granted.
argue that an omitted variables bias occurs when one ignores “multilateral resistance” in the gravity equation. This term accounts for the fact that bilateral trade is not only de-pendent on trade costs between the two trading countries, but also on barriers with respect to the rest of the world. Thus, the authors suggest using exporter-year and importer-year fixed effects, which do not only capture variation in multilateral price terms but also in countries’ real GDP.
The fixed-effects estimator is preferred over the random-effects estimator for two rea-sons. First, in contrast to the random-effects model that yields consistent estimates only when the unobservable characteristics are uncorrelated with the explanatory variable, the fixed-effects model relaxes this assumption (Cameron and Trivedi 2010). Since the source of estimation bias is likely to be found in the correlation between time-invariant bilat-eral variables with the RTA dummy, pair fixed effects are more appropriate than random effects (Baier and Bergstrand 2007). Second, to double check the applicability of the fixed-effects model, I run a robust version of the Hausman test. The hypothesis, that the random-effects model yields consistent estimates has to be rejected for each specification and thus the fixed-effects approach is chosen.
One disadvantage of using fixed effects is the impossibility to estimate variables that are constant over time as the individual effects are removed from the initial model by the subtraction of time averages. However, since the data spans an observation period that is long enough to show a certain variation in the RTA dummy, the use of fixed effects is appropriate. In contrast, time-constant dummy variables that proxy for trade costs such as distance, language, and colonial heritage cannot be included in the model.
Various additional tests are conducted in order to avoid econometric issues arising from misspecification. Hypothesis testing reveals that the inclusion of exporter-year and importer-year fixed effects is essential. The hypothesis that the fixed effects are jointly equal to zero can be rejected for each specification, and thus dummy variables for each combination have to be included. This strongly supports the suggestion of Anderson and Wincoop (2003) to include fixed-effects that proxy for multilateral resistance.
4.1
Empirical specification
The aim of this paper is to conduct a comprehensive analysis of deep trade integration on bilateral production sharing and to test the main hypothesis outlined in Section 2. The baseline specification can be formally written as:
lnV AXi,j,t= α + β1RT Ai,j,t+ β2DEP T Hi,j,t+ σi,j,t+ γi,t+ γj,t+ i,j,t (1)
where lnV AXi,j,t is the VAX ratio between exporter country i and importer country j
in year t. Since the distribution of the variable is highly skewed, the natural logarithmic transformation is applied to obtain an approximation to the normal distribution.4 RT A
i,j,t
is a dummy variable which is defined to be equal to 1 when country i and j are members of the same RTA in year t, and 0 otherwise.
DEP T Hi,j,t denotes a vector of variables that measure the depth of trade agreements.
To meet the objective of a comprehensive analysis on the impact of deep agreements, several indices are utilized in the panel regression as explained in more detail in Section 3. In sum, the two indices from D¨ur et al. (2014) are: (1) the additive index which, – in its rescaled version – ranges from 0 to 8; and (2) the continuous depth variable that results from latent trait analysis. From the World Bank data (Hofmann et al. 2017), three more depth measurements are constructed. These are: (1) the number of provisions implemented with RTAs ranking from 0 to 52, where 0 means no agreement in force at all; (2) the number of WTO+ provisions ranking from 0 to 14; and (3) the number of WTO-X provisions ranking from 0 to 38.
As previously mentioned, bilateral fixed effects σi,j,t are included to control for
unob-served time-invariant heterogeneity between countries. γi,t and γj,t denote exporter-year
and importer-year fixed effects that account for multilateral price terms. The error term is depicted by i,j,t. Table 10 in the appendix summarises the variables used in the panel
analysis.
5
Estimation results
The estimation results of baseline specification 1 are presented and discussed in this section. I will further test the robustness of these results with first-differenced data as this might have some advantages over the fixed-effects model.
5.1
Baseline specification
Each specification presented here includes a full set of importer-year, exporter-year, and country-pair fixed effects to account for unobserved heterogeneity across countries as
explained above. Robust standard errors that cluster at the country-pair level are reported in brackets below the coefficients.
Table 2: Panel regressions with DESTA depth indices
Variables (1) (2) (3) (4) RTA -0.042*** 0.028 -0.001 -0.017 (0.014) (0.019) (0.017) (0.023) depth index -0.021*** (0.004) depth latent -0.035*** (0.009) deep short -0.005 (0.017) deep medium -0.062*** (0.015) deep long -0.034* (0.020) deep anticipatory -0.009 (0.012) shallow short 0.002 (0.021) shallow medium -0.006 (0.020) shallow long -0.033 (0.025) shallow anticipatory 0.031 (0.021)
Country-pair FE Yes Yes Yes Yes
Country-time FE Yes Yes Yes Yes
Observations 11,080 11,080 11,080 11,080
Country-pairs 1,721 1,721 1,721 1,721
Adjusted R-squared 0.673 0.675 0.674 0.674
Notes: The dependent variable is the log of the VAX ratio. Robust standard errors clustered by country-pairs in parentheses. ***, **, * indicates signifi-cance at the 1%, 5%, 10% level.
Column (1) of Table 2 replicates the analysis in Johnson and Noguera (2016) and shows estimates of the simple binary RTA variable on VAX ratios of country-pairs while ignoring that RTAs differ in their depth. The estimated coefficient shows a negative and highly significant effect: when two countries are members of the same RTA, the bilateral VAX ratio declines on average by approximately 4 percent (e(−0.041) − 1 = 0.04). The
coefficient is slightly lower than in the analysis of Johnson and Noguera (2016) who find an average impact of around 5 percent.
dummy to be equal to 1 only for country-pairs with FTAs, customs unions, common markets and economic unions, the RTA dummy in my analysis is equal to 1 also for bilateral observations with one-way and two-way preferential trade agreements. This implies that I treat countries which are engaged in some form of trade liberalization – even if the extent is minimal – not the same as countries that have no agreement in force at all.
More importantly, columns (2) to (5) take into account that RTAs are heterogeneous with respect to the degree of trade liberalisation and introduce the depth indices discussed above. Column (2) shows estimates for the rescaled additive depth index by D¨ur et al. (2014). Interestingly, the RTA dummy becomes insignificant when the index is added which might indicate that depth matters more for countries’ vertical specialization than simply being part of an agreement. The same holds for the second DESTA indicator – the continuous depth variable from latent trait analysis – that is utilized in column (3). For both indicators, the coefficients appear to be negative and highly significant. In particular, deeper RTAs tend to decrease the VAX ratio by 2 to 3.5 percent.
In column (4), I follow the estimation approach in D¨ur et al. (2014) and replace the depth indices by eight dummy variables that are defined as follows. The deep dummies are equal to 1 for observations that have a depth index value higher than the median across all country-pairs evaluated at different points in time: 1) five years prior to the adoption of the RTA (deep anticipatory); 2) five years after the adoption (deepshort); 3) between five and 15 years after the adoption (deep medium); and 4) between 15 and more years after the adoption (deep long). Similarly, the shallow dummies are equal to 1 when the value of the depth index is lower or equal to the median across all country-pairs.
It appears that only two out of eight dummy variables are statistically significant and have the expected negative sign (deep medium at the 1 percent level and deep long at the 10 percent level). This suggests that the depth of trade agreements matters for bilateral vertical specialization only in the medium and long run. This is contrary to the results for gross trade obtained by D¨ur et al. (2014) since the authors found only shallow short to be insignificant. However, in contrast to trade in finished products, GVC activity requires firms to undertake large investments to set up their production facilities. These investments are not realized overnight and thus deep trade agreements are more likely to bear fruit in the medium and long run after the adoption.
Table 3 shows estimates for the depth indices constructed from the World Bank data on the content of deep agreements (Hofmann et al. 2017). Following closely the approach in Orefice and Rocha (2014), the indicator variables on the number of provisions, WTO+ provisions, and WTO-X provisions are introduced separately in columns (1) to (3).
Table 3: Panel regressions with number of provisions Variables (1) (2) (3) number of provisions -0.005*** (0.001) WTO+ provisions -0.008*** (0.001) WTO-X provisions -0.009*** (0.001)
Country-pair FE Yes Yes Yes
Country-time FE Yes Yes Yes
Observations 7,517 7,517 7,517
Adjusted R-squared 0.693 0.691 0.694
Notes: The dependent variable is the log of the VAX ratio. Ro-bust standard errors clustered by country-pairs in parentheses. ***, **, * indicates significance at the 1%, 5%, 10% level.
appears to be slightly greater than the impact of an additional WTO+ provision. This finding is contrary to the results in Orefice and Rocha (2014) and might be surprising since the WTO-X area covers a broad range of provisions that are not related to GVC trade at all (e.g. health and other social matters).
Table 4: Panel regressions by period
(1) (2) (3) (4) Variables 1970-2005 1980-2005 1990-2005 2000-2005 RTA -0.001 0.017 0.005 0.063 (0.017) (0.025) (0.028) (0.065) depth latent -0.035*** -0.036*** -0.024** -0.021 (0.009) (0.011) (0.011) (0.032)
Country-pair FE Yes Yes Yes Yes
Country-time FE Yes Yes Yes Yes
Observations 11,080 8,823 6,377 3,354
Country-pairs 1,703 1,703 1,703 1,703
Adjusted R-squared 0.674 0.691 0.741 0.867
Notes: The dependent variable is the log of the VAX ratio. Robust standard errors clustered by country-pairs in parentheses. ***, **, * indicates significance at the 1%, 5%, 10% level.
The long-time horizon of the panel data allows me to study the evolution of a relation-ship between deep trade integration and production fragmentation over time. In Table 4, equation 1 is estimated for different time periods of the sample. The four columns cover the years: (1) 1970-2005, (2) 1980-2005, (3) 1990-2005, and (4) 2000-2005.
two columns indicating that trade agreements in the 1970s and 1980s play an important contribution to the degree of bilateral production sharing. In the period 1990 through 2005, the coefficient drops to 2.5 percent and significance is lowered to the 5 percent level. Finally, for the last five years of the sample, the variable is insignificant. In sum, these findings suggest that the elasticity of the bilateral VAX ratio with respect to trade liberalization has decreased over time and even becomes insignificant for the 2000s. This might be a first indication for the analysis on Turkey and India in the second part of this paper where the period under consideration ranges from 1995 to 2014, while the recent slowdown in GVC activity is investigated in more detail.
5.2
Robustness checks
To test the validity of the coefficients obtained in the previous subsection, I re-run equa-tion 1 using first-differenced data. This can be formally written as follows:
∆lnV AXi,j,t = α + β1∆RT Ai,j,t+ β2∆DEP T Hi,j,t+ σi,j,t+ γi,t+ γj,t+ i,j,t (2)
According to Wooldridge (2002), the fixed-effects model is more efficient for panel data with more than two periods when there is no serial correlation in the error term. However, serial correlation in the data was detected by an empirical test specifically for fixed- and random-effects models, which was proposed by Wooldridge (2002). Adding to this, Baier and Bergstrand (2007) deliver an economic reason: the unobserved heterogeneity that drives RTA formation is slow-moving and thus, correlation over time is likely to be the case. As Wooldridge (2002) notes, the inefficiencies of the fixed-effects model increase with the observation period. In the presence of serial correlation, he suggests using the first-differenced panel instead.
Table 5 shows estimation results for the various depth indicators using first-differenced data. Except from WTO+, all variables are significant and have the expected negative sign. However, in general, the coefficients slightly decrease in size and significance. For instance, the impact of the additive depth index decreases to 0.6 percent and is statistically significant at the 10 percent level. Nevertheless, the previous estimates of the fixed-effect model are proven to be robust when using first-differenced data.
6
A tale of two countries: India and Turkey
Table 5: Panel regressions with first-differenced data Variables (1) (2) (3) (4) (5) dRTA 0.008 0.007 (0.015) (0.013) ddepth index -0.006* (0.003) ddepth latent -0.017** (0.007) dnumber of provisions -0.002*** (0.001) dWTO+ -0.003** (0.002) dWTO-X -0.003*** (0.001) Country-pair FE No No No No No
Country-time FE Yes Yes Yes Yes Yes
Observations 9,245 9,245 5,459 5,459 5,459
Country-pairs 1,240 1,240 1,240 1,240 1,240
Adjusted R-squared 0.169 0.170 0.219 0.219 0.219
Notes: The dependent variable is the log of the VAX ratio. Robust standard errors clustered by country-pairs in parentheses. ***, **, * indicates significance at the 1%, 5%, 10% level.
of the financial crisis, but recent numbers suggest a slowdown in trade volumes since then. In fact, yearly export growth remained below 4 percent in the post-crisis period.5
Furthermore, the elasticity of trade with respect to income has undergone a considerable change. Whereas between 1986 and 2000, the trade-income elasticity was on average 2.2, it decreased to 1.3 in the 2000s (Constantinescu et al. 2015).
The explanations for this trend in literature are numerous. A change in the composi-tion of final demand across industrialized and emerging countries appears to be among the most important determinants. The International Monetary Fund (IMF 2016) estimates that three-fourths of the current slowdown can be explained by the overall weakness in economic activity, whereby subdued investment growth accounts for the lion’s share. In particular, relative demand shifted from durable goods, that are primarily manufactured in GVCs, to services that are less trade intensive since they are produced domestically (Bussi`ere et al. 2013).
Beyond the compositional factors, other contributors to the trade decline are structural in nature, most notably the slower expansion of GVCs. The process of de-fragmentation is likely to be caused by technological improvements in the production of goods and services. On the one hand, firms in emerging economies are nowadays able to substitute domestic inputs for foreign inputs because of increasing capabilities to produce upstream products.
This is especially pronounced by Kee and Tang (2016) for the case of China. On the other hand, Timmer et al. (2016) argue that production processes might be re-shored to advanced countries because of automation of labour-intensive tasks.
Evaluating cyclical versus structural effects, Constantinescu et al. (2015) find that the decrease in production fragmentation is most responsible for the stagnation in gross trade. The ratio of foreign value added to domestic value added in gross exports rose by 8.4 percentage points between 1995 and 2005. In contrast, from 2005 to 2012, the ratio only increased by 2.5 percentage points. The authors suggest that this decline can be explained by the fact that the technological shock occurring in the 1990s might already be absorbed by the beginning of the new century.
In a similar analysis, Al-Haschimi et al. (2016) show that compositional and struc-tural effects are both equally responsible for the current trade slowdown. Among the compositional factors, a key determinant of the lower trade-income elasticity is the rel-ative demand shift from industrialized to emerging economies. This is because trade elasticities in advanced nations are typically higher than in emerging economies (Slopek 2015). On the structural side, the authors show that GVC participation has stalled since 2011, accounting for a significant share of the decline in gross trade. This might be at-tributable to an increase in protectionist measures. Because of local content requirements, firms have an incentive to source and produce in the export market instead of importing intermediate inputs for their production process. Moreover, other trade frictions such as transportation costs and trade barriers have already reached low levels, and thus, further improvements have only small impacts on trade volumes. In general, Al-Haschimi et al. (2016) argue that the recent trade trend reflects a normalisation of the trade-income elas-ticity and that previous decades rather present an abnormal situation of excessively high growth rates. This new normal will persist in future if no further positive shock, such as a new wave of trade liberalisation, will move the economy to a new equilibrium.
Timmer et al. (2016) use the WIOD 2016 Release to construct a new measurement of GVC trade for the period from 2000 through 2014. The global import intensity (GII) captures imports required in all stages of international production. The authors show that due to high demand for consumer durables and increasing fragmentation, the GII rose substantially between 2000 and 2008. In contrast, they observe a decline in the GII from 2011 onwards reflecting a slowdown in GVCs and a relative demand shift to less tradable products. However, previous research has shown that most value added is still produced in the domestic market (Los et al. 2015) and that there is potential for further fragmentation, especially in the service sector (Baldwin 2016).
as the ‘bullwhip effect’ (Altomonte et al. 2012) and is additionally magnified through uncertainties. Input suppliers might start to avoid risk and decrease production and trade in intermediate goods because they have less information about the demand decline than exporters of final goods.
The above-outlined research results suggest that not only the pattern of production fragmentation has altered during the last decade, but also the nature of trade agreements has undergone a major change. In the 1990s, the average number of agreements that were signed each year amounted to 30, whereas since 2011, this rate has fallen to ten agreements each year (IMF 2016). However, recently signed RTAs are deeper as they comprise more policy areas and countries. Besides reductions in tariffs, new RTAs aim to promote the development of GVCs by also reducing non-production costs such as technical barriers and product standards (Blyde et al. 2015).
According to Baldwin (2014), regionalism of the 21st century is entirely different from regionalism of the previous century since these agreements are not only about trade but also about production sharing. This means that in the 20th century, regionalism focused on selling goods, whereas nowadays, it also focuses on producing them. Recent RTAs include provisions that underpin GVCs such as the coordination of production facilities and the elimination of risks to tangible and intangible property. In particular, Baldwin (2014) identifies WTO+ and WTO-X provisions that are essential for the promotion of GVCs: Agreement on Trade-Related Investment Measures (TRIMs), TRIPs, competition policy, IPR, investment, movement of capital, an approximation of legislation, industrial tariffs, customs, and GATS.
To summarize, the new century seems to be characterized by a fundamental change in vertical specialization and the nature of trade agreements. On the one hand, RTAs become deeper and are designed in such a way that they support vertical specialization. On the other hand, the expansion of GVCs has slowed down during the post-crisis period due to major technological improvements. This suggests that the relationship between GVCs and trade agreements has undergone an enormous change in recent years. The motivation of this section is to investigate these developments for two emerging economies in more detail: Turkey and India. It allows to test further whether the empirical results obtained in the regression analysis above are reflected in Turkey’s and India’s VAX ratio. In that case, we expect to observe a falling ratio indicating that more production fragmentation occurs after the adoption of deep trade agreements.
certain variation in the data to be able to observe an RTA impact within the observation period.
In the following subsections, I will describe the WIOD, outline computations for value-added exports using input-output analysis and conduct a descriptive analysis illustrating the relationship between the adoption of trade agreements and VAX ratios of India and Turkey.
6.1
Data
The WIOD has published two distinct releases that are both used to calculate VAX ratios of India and Turkey from 1995 to 2014. The focus of the analysis lies on the WIOD release 2013 that includes 40 countries with 35 industries and an estimated model for the rest of the world. It covers the period from 1995 through 2011 and thus allows me to conduct a consistent time-series analysis of earlier years of trade liberalization. However, since the motivation of this section is also to examine more recent trends in production fragmentation, a second time-series is calculated based on the WIOD release 2016 that covers the period from 2000 through 2014.6 Compared to the old version, this
release covers more countries and industries with a finer classification, especially in the manufacturing and service sectors.
The WIOTs are constructed using officially published input-output tables, interna-tional trade statistics, and nainterna-tional accounts data. They can be seen as nainterna-tional input-output tables that are connected by bilateral international trade flows and thus comprise all transactions of the global economy on an industry-level Timmer et al. (2016). Flows of goods and services in the WIOT are intended either as intermediate inputs for the produc-tion process by other industries or final demand by governments, firms, and households. Among other things, the big advantage of the WIOD is the consistency of input-output tables over time, whereby the focus of the construction lies on a high level of data quality rather than quantity.7
Furthermore, information on the coverage and depth of trade agreements is derived either from the DESTA database (D¨ur et al. 2014) or the World Bank (Hofmann et al. 2017) that were explained at length in Section 3. Additional information on the geographical scope of multilateral RTAs and implementation periods is obtained from the RTA database of the WTO.
6.2
Measuring value added exports
The approach outlined in this subsection follows the definition of value-added exports by Johnson and Noguera (2012). In particular, the authors measure the amount of value
6Unfortunately, it was not possible to combine the two releases to one consistent time series since differences in the numbers appeared to be too large. Moreover, the WIOD team strongly dissuades from combining the two releases since data sources to construct the WIOTs have undergone major revisions.
added generated in country i to satisfy final demand in country j. Whereas Johnson and Noguera (2012) compute bilateral value-added exports for each country-pair, the interest of my analysis lies in value-added exports from Turkey and India vis-`a-vis the outside world. Thus, the calculations outlined in this subsection refer to value-added generated in Turkey (or India) to satisfy final demand in all other countries. To obtain a measure of production sharing, Turkey’s and India’s value-added exports are divided by their respective gross exports. This gives us the well-known VAX ratio.
The calculations below are based on the WIOT as explained in the previous subsection and are carried out separately for India and Turkey. Following standard matrix notation, I will use bold capital letters to indicate matrices, bold small letters for vectors, and italicized letters for scalars.
To compute value-added exports, the first step to start with is the typical representa-tion of the input-output model:
x = A + f (3)
Equation 3 implies that total output is split between intermediate and final usage, where x is a 1435×1 vector of gross output; A is a 1435×1435 matrix of input coefficients that capture the direct effects and that are obtained by dividing the technology matrix Z by the vector of gross output, i.e. A = Z · ˆx−1 ; and f is a 1435 × 1 vector of final demand.8
The solution to Equation 3 is given in the following, where gross output is expressed as a function of the Leontief inverse and final demand:
x = (I − A)−1f = Lf (4)
Equation 4 makes use of the well-known Leontief inverse L = (I − A)−1, where I denotes an identity matrix of the dimension 1435 × 1435. The Leontief inverse gives us the amount of output that is needed to produce a certain vector of final demand. It captures the direct as well as indirect input requirements for the production process of f . This means that the necessary intermediate inputs for f are captured by Af (i.e. the direct effect). To produce these direct inputs, A2f intermediate inputs are required. The production of A2f , in turn, requires inputs of the amount A3f , and so on and so forth (i.e. the indirect effects).
To obtain value-added exports from country i that are embodied in foreign final de-mand, Equation 4 is pre-multiplied with the value-added coefficients:
vaexpi = va · Lf (5)
where vaexpi is the total amount of country i’s value-added exports, and va denotes
a 1 × 1435 vector of added coefficients that are obtained by dividing the value-added vector v by the gross output vector, i.e. va = v · ˆx−1. These coefficients capture
the amount of value added required to produce one unit of gross output. Since the interest lies only in the amount of value added that is produced inside Turkey (or India), the coefficients for all other countries are set to zero. Moreover, to obtain value added required for foreign final demand, final demand of Turkey (or India) is set to zero.
Lastly, gross exports, intermediate exports, and final exports are directly observable in the WIOT as bilateral deliveries from country i to country j for each industry. The aggregate VAX ratio of country i can simply be calculated by dividing total value-added exports by total gross exports:
V AXi =
vaexpi
groexpi
(6) where groexpi denotes the total amount of gross exports of country i and V AXi
the ratio of value-added exports to gross exports. The VAX ratio is employed in the next subsection for a descriptive analysis of the impact of deep trade agreements on production fragmentation, whereby a lower VAX ratio implies a higher degree of vertical specialization.
6.3
Descriptive analysis
Turkey and India are appropriate examples to study the effect of trade agreements on production fragmentation in more detail. Both countries did not start to liberalize the bulk of their domestic production to trade until the mid-1990s. Earlier agreements were more shallow in nature and thus unlikely to be correlated with changes in value-added trade. In contrast, more recent RTAs cover a broader scope of policy areas and promising steps for further integration. They include provisions that particularly aim at promoting production sharing across trading partners. The conclusion of the Uruguay Rounds and the foundation of the WTO in 1995 additionally facilitated access to goods and service markets worldwide.
Among other middle-income countries, India and Turkey appear to play in the medium range of GVC activity. Figure 4 shows the VAX ratio for several middle-income countries in 2011.9. Whereas Turkey’s and India’s participation in GVCs seems to be greater
than for most Latin American countries such as Argentina, Brazil, and Peru, they are less integrated than Asian countries such as Vietnam, Cambodia, China, Malaysia, and Thailand.
The WIOD allows us to study the evolution of value-added trade from 1995 through 2014. The descriptive analysis focuses on bilateral and multilateral RTAs covering pro-visions that are necessary for facilitating Turkey’s and India’s integration into GVCs. A full list of agreements signed by India and Turkey are provided in Table 11 and Table 12 in the appendix. I discuss trends in production fragmentation with respect to the VAX ratio, domestic value-added exports embodied in foreign final demand, gross exports as well as exports in intermediate and final goods.
Figure 4: VAX ratio of middle-income countries
Notes: VAX ratios are calculated on basis of the Trade in Value Added (TiVA) database from OECD (2016). GNI per capita is measured in current USD and is derived from World Development Indicators.
6.3.1 India
As shown in Figure 4, India’s integration into GVCs remains weak when compared to other Asian economies. One reason for the relatively low level of production fragmenta-tion is India’s specializafragmenta-tion in services that are less tradable than manufacturing goods. Nevertheless, the service sector has gained more prominence in GVCs since infrastructure and business services are increasingly required for the smooth operation of production sharing across countries. Because India is endowed with a large pool of low-cost, fluent-English speaking and skilled workers, it has emerged as a major offshore service provider, particularly for information technology and business processes (Chen 2012). However, the integration of India into GVCs via the provision of services is not reflected in the analysis below which might point to limitations in the data.10
In the last ten years, India has fostered the adoption of comprehensive bilateral and multilateral agreements, most notably with East Asian countries. So far, India has notified 15 RTAs to the WTO of which nine are bilateral agreements. One notable development is the rising number of arrangements that cover trade in services. In the last ten years, India has signed five RTAs that include provisions on services.
Figure 5 captures the evolution of India’s VAX ratio from 1995 through 2011, while vertical lines denote the adoption of comprehensive RTAs as explained at the bottom of the graph. Table 6 presents corresponding average annual growth rates of the VAX ratio, value-added, intermediate, final and gross exports for three subperiods: 1995-2000,
Figure 5: VAX ratio of India, 1995-2011
Notes: The vertical lines show the adoption of RTAs with Singa-pore (2005), SAFTA (2006), MERCOSUR (2009), South Korea (2010), ASEAN (2010), Japan (2011), and Malaysia (2011).
2000-2005, 2005-2011, and 2011-2014, while the last subperiod is calculated based on the WIOD release 2016.11
Since the early-1990s, India’s trade policies have been characterized by mainly unilat-eral libunilat-eralizations with two reform periods between 1991-1993 and 1998-2004 (Sally 2011). In 1991, India adopted the ‘Look East Policy’ that aimed at trade liberalization towards East and South-East Asian markets. This terminated a long period of import-substitution growth and orientated the economy towards an export-oriented strategy (Mukherjee and Goyal 2015).
These trade reforms are most likely to be responsible for an increase in production fragmentation starting around 1997 as observed in Figure 5. Between 1995 and 2000, the VAX ratio decreased on average by 1.2 percent annually indicating greater fragmentation, while the decline was even more pronounced toward the turn of the century. Further measures of trade liberalization at the beginning of the century and China’s accession to the WTO in 2001 might additionally contribute to the rise in GVC activity. From 2000 to 2005, the VAX ratio declined on average by to 1.31 percent yearly. The largest decrease was observed in 2004 as India’s VAX ratio fell by 4.4 percent, particularly because growth in gross exports exceeded that of value-added exports by more than 6 percentage points. The first bilateral RTA that includes regulatory measures for GVCs and services was
Table 6: Average annual growth rates for India (in percent) 1995-2000 2000-2005 2005-2011 2011-2014 Value-added exports 8.65 16.88 13.09 0.05 Intermediate exports 9.46 19.08 11.38 -0.10 Final exports 10.57 17.52 16.35 -0.45 Gross exports 9.92 18.43 13.55 -0.16 VAX ratio -1.16 -1.31 -0.41 0.21
Notes: Growth rates for 1995-2000, 2000-2005, and 2005-2011 are based on the WIOD release 2013, and 2011-2014 on the release 2016. The complete time series is shown in Figure 7 in the appendix. The discrepancies in the two releases are most likely to arise from differences in the data sources that underlie the construction of the WIOD. For instance, the systems of national accounts, the industrial classification and the bilateral trade in service data have undergone major revisions.
signed with Singapore (2005). It contains provisions on industrial tariffs, customs, tech-nical barriers to trade, TRIMs, GATS, competition policy and capital movement. In the subsequent year, India adopted the South Asian Free Trade Area (SAFTA) that comprises eight Asian countries, but with only three WTO provisions it is rather weakly designed for further GVC integration. When examining the short-run period after the adoption of these two agreements, no clear trend in fragmentation can be observed. Instead, from 2005 to 2011, the VAX ratio stagnated with an average annual decline of just 0.4 per-cent, which was also caused by overall weakness in the Indian economy due to the global financial crisis.
Another motivation of this section was to show that GVC activity has stalled in recent years as found in the research studies outlined above. Table 6 indicates that from 2011 to 2014, the VAX ratio increased on average by 0.21 percent yearly indicating less production sharing. Similarly, growth rates of value-added, intermediate, final and gross exports were close to zero or even negative. This might point to a general contraction of the Indian economy and confirms recent studies that measure trade stagnation in either gross or value-added terms. However, the process of stagnation in India has already started around 2005 as can be observed in Figure 5. The period from 2005 to 2011 shows a substantially slower growth in value-added, intermediate and gross exports compared to the preceding five-year period, only growth in final exports shrunk less noticeable.
In sum, the descriptive analysis of India provides evidence that deep trade integra-tion spurs producintegra-tion fragmentaintegra-tion and thus confirms the regression results outlined in Section 5. Unilateral trade liberalizations in the early-1990s caused a decline in the VAX ratio from 1997 onwards. However, as India failed to engage in comprehensive bilateral and multilateral RTAs until 2010 (except for the commitment with Singapore), the VAX ratio has stagnated in the last decade.
