Do trade agreements stimulate participation in global value chains?

(1)

Do trade agreements stimulate participation

in global value chains?

University of Groningen

Faculty of Economics and Business

MSc Thesis International Economics and Business

Name student: Jeroen Polman Student ID number: S2598108

Student e-mail: j.h.polman@student.rug.nl Date: January 5, 2016

(2)

2

Abstract

This study empirically tests the effect of economic integration agreements (EIAs) on value added exports and foreign direct investment (FDI), using the gravity model. It seems that EIAs increase value added exports with at least 4 percentage points and this effect doubles with a five years ‘phase-in’ period. These measured effects are low, relative to other studies. A possible explanation for these relative low effects, is that EIAs also promote FDI and horizontal FDI is a substitute for exports. EIAs enhance FDI with at least 3 percentage points and ‘deep integration’ agreements enhance FDI between 5 and 17 percentage points.

(3)

3

1 Introduction

Fragmentation of production processes has accelerated since 1985 (Baldwin and Gonzalez 2014). Supported by low communication and coordination costs that makes it profitable to split the production process, with each task performed at its lowest-cost location (Timmer et al. 2014). The globally fragmented production processes are called global value chains (GVCs). Participation in these GVCs may increase growth and productivity by giving access to foreign knowledge and technology, scope for specialization and economies of scale, and access to cheaper or higher quality intermediate inputs (OECD 2013).

Policy makers try to stimulate participation in GVCs by economic integration, in particular with economic integration agreements (EIAs). This could explain the accelerated participation in EIAs since 1990. As Krishna et al. (2011) show, the number of EIAs have grown from 70 in 1990 to 300 in 2010. Since 1962 (Tinbergen), the effect of EIAs has been empirically tested, mostly with gross exports- and otherwise (FDI) as dependent variable.

The increasing fragmentation of production processes causes the rise of trade in intermediaries, where inputs pass borders many times. Therefore, gross trade data include substantial double-counting, which leads to overestimation of value-added trade, whereby value added trade has become increasingly different from gross trade. This will affect policy implications with respect to trade, growth, competitiveness, global imbalances, and the impact of macro-economic shocks. Trade in value added is a better information source for these policy subjects. For example, Pascal Lamy, the Director-General of the World Trade

Organization (WTO), noted that “the statistical bias created by attributing commercial value to the last country of origin perverts the true economic dimension of the bilateral trade imbalances”.

This study uses value added exports as a measurement for global value chains instead of gross exports. For a long time there has been a problem with the availability of value added trade data, but thanks to national input-output tables (IOTs), which are combined to the World Input Output Database (WIOD) (Timmer et al. 2015) and the Trade in Value Added (TiVA) (OECD 2012) database, this gap is diminishing. This study uses these databases to measure the effect of EIAs on value added exports. Beside value added exports, FDI is a good

instrument for GVCs. In particular vertical FDI, which leads to international fragmentation of production processes. On the other hand, horizontal FDI is a substitute for exports, so it will affect the effect of EIAs on exports negatively.To provide policy makers with a better information source for the above noted policy areas, this study will use value added exports and FDI as measurements for GVCs.

Kohl (2014) showed that the effect of EIAs is heterogeneous. He found that about a quarter of the EIAs are really trade promoting. Therefore the insightfulness of empirical findings is limited, based on EIAs in general. To provide policy makers with more information than just the general effect of EIAs on global value chains, it is also studied which EIAs affect GVCs most. Thereby, this study provides several contributions to the existing literature. First, it tests the effect of EIAs on value added trade with a strict

(5)

5 study shines more light on the differences between the WIOD and TiVA database.

The remainder of this paper is structured as follows. Section 2 provides insight into the difference between gross trade and value added trade. Then it reviews the effect of EIAs on FDI, trade and GVCs. Section 3 presents the methodology. Section 4 provides the data sources and tests the data. Section 5 presents the results. Section 6 gives the conclusions, limitations of this research and recommendations for future research.

2 Literature review

2.1 Economic fragmentation

North-North supply chains have existed for multiple decades. They have led to the 1965 Auto Pact between the US and Canada. The pact supported the process, where the US exported auto parts to Canada and reimported assembled cars. This improved the auto industry in North America; because nations can specialize in the manufacturing stages they do relatively best and achieve economies of scale. Since 1985 quite a number of developing countries have become more and more involved in international value chains. This involvement started with developing countries performing one or more stages in the supply chain. These international value chains led to trade in goods that became inputs into production processes in other countries. This trade in intermediaries is also called supply-chain trade. An important driver behind international value chains is that multinationals offshore low skilled labor tasks to these developing countries. This has led to an increasing export share for developing countries in several product groups. For example the manufacturing industry where most activities are labour intensive and not too complicated. This has resulted in a shift in world manufacturing GDP from the G7 to China and other developing countries, as showed in the table below which is sourced from Baldwin and Gonzalez (2014).

Figure 1: Seven risers and seven losers: Manufacturing reversal of fortunes.

(6)

6 Developing and developed countries are performing different stages in these supply chains, also called North-South supply chains. North-South supply chains have much more impact on trade patterns than North-North chains, especially due cost reductions. That is due to

differences in factor prices. Wages in particular are much cheaper in developing countries, which makes offshoring of labour intensive activities profitable.

North-south supply chains are sliced up in tasks. The High-tech and knowledge endowed countries mostly perform the relative knowledge intensive tasks and deliver managerial, technical and marketing know-how. Low wage countries fulfill the labour intensive tasks. This slicing up of the value chain, also called fragmentation, is profitable thanks to technological improvements as Internet, telephony and sophisticated software. These innovations have lowered the communication and coordination costs. These low communication and coordination costs make it profitable to split the production process, with each task performed at its lowest-cost location (Timmer et al. 2014).

Los et al. (2015) have found that for all 40 WIOD countries, except for Canada, the share of domestic value added content in exports has declined between 1995 and 2008 and in some countries even by up to 20 percentage points. Los et al. (2015) focus on the difference between foreign value added inside and outside the region. The regions are EU-27, NAFTA and East Asia. They find that both shares were increasing for almost all

countries-of-completion. Global foreign value added shares increased for 35 out of the 40 countries. The foreign value added shares for EU-27, NAFTA and East Asia have increased from 9.5% to 17.3%, 9.9% to 15.8% and 8.2% to 17.5% respectively, between 1995-2008. This shows that International value chains have been evolving from offshoring of production stages to

complex GVCs, as in the iPod case below. However, trade is still more regionalised than globalised, so distance still matters.

(7)

7 fragmentation is relative low, therefore it suffers less from the double counting problem. In addition, services are often used to support other industries where added value of services is charged in products.

A good example of a north-south supply chain is Apple’s iPod supply chain, which is studied by Dedrick et al. (2009). In particular Japan (around $100 per iPod) but also the US, Korea and other countries deliver parts to China where the assembling of the iPod takes place. While China dispatches assembled iPods for $144 to the US, less than 10 percent of the value is added by China. Apple is performing several high value added stages as coordinating the supply chain, delivering the software, product innovations, process developments, marketing, and customer service. To summarize the main stream of products: intermediates go from Japan to China, were the low-value added but labour intensive assembling stage takes place, after which they go to the US. This is displayed below.

Figure 2.1 Main stream of exports in the iPod case

Table 2.1 Differences gross exports and value added exports Measurement differences

exports

Gross exports Value added exports

Japan $100 to China $100 to US

Rest of the World (ROW) $34 to China $34 to US

China $144 to US $10 to US

US Only if it sells the final

product to other markets.

Only if it sells the final product to other markets.

The model above displays a simplification for the real Apple iPod Supply chain with a rough estimate of the real figures. The figure is not about the exact value of the numbers, but to illustrate the difference between value added exports and gross exports. Japan sends her products to China to assemble it. Hereby is Japan’s output dependent on demand from the US as showed in value added exports. China dispatches assembled iPods for $144 per piece, but the value added of China is $10 in this example. To sum up, measuring in gross exports changes bilateral trade positions and there is a double counting problem, because $278 (34+100+144) worth of exports is counted instead of $144 per dispatched iPod. This double counting problem will only increase if we also measure the foreign content of the

(8)

8 trade overestimations and changes in bilateral trade positions are heterogeneous between countries and industries. This all gives a misleading picture about the trade magnitude of nations and industries.

However, a central problem in measuring value added exports is the missing

information about the final-or-intermediate usage when it comes to data. This problem can be (partly) solved in three ways: the first is the Harmonized Commodity Description and Coding System (HS system), which gives every product its own code that divides products into sections and chapters. Every product has his own code which description can also contains ‘parts’, if it is normally not used as a final product. This provides insight in the final-or-intermediate usage. However this is not fully satisfactory. Some products, as batteries and nails, can be used as intermediate as well as final good and this might make the distinction between intermediates and final products not very reliable. A second approach is to use data from ‘processing trade’ in special custom regimes. These custom regimes note, among others, of goods are consumed or re-exported (used as intermediaries and exported again). This is done to determine import tariffs; re-exported imports have lower (mostly zero) import tariffs. This data is mostly available in China and other developing countries that do a lot of

assembling stages in production chains. However, also in these developing countries are significant gaps in the data, so processing trade data is too incomplete. The detailed

explanation of these solutions is shown in Baldwin and Gonzalez (2014) section 2. Finally, input-output tables (IOTs) measure the means of commodities that are used to make

commodities or services. So they measure all the inputs, as well domestically as foreign, that are used to produce an output. However it is still hard to divide all the value added to the right country of origin and industry. To do this, researchers must make some assumptions. The assumptions for WIOD are discussed by Dietzenbacher et al. (2015) and for the TiVA database by OECD (2012). These IOTs are designed as the matrix below, which has the lay-out of the TiVA database. In the empty squares you fill in the used inputs of the horizontal axes to produce the products on the vertical axes. So at X1 you fill in all ‘agriculture, hunting, forestry and fishing’ that is used as input to produce all the mining and quarrying for a year. This is done for all countries and industries which are present in the TiVA or WIOD database. In this example, X1 up to X5 show exactly how much inputs per industry are used in de mining and quarrying industry. Table 2.2 displays a part of an IOT to show their design and working, for a more detailed explanation and complete example see Timmer et al. (2015). The advantages of IOTs are that the information is more exact than HS codes because there is no doubt about the final-intermediate usage. In addition, it is more detailed than processing trade because it shows exactly where inputs come from and where they go. Finally, the most complete databases are available for IOTs. Almost every developed country has a national IOT, and also increasingly developing countries have one. These IOTs are becoming

(9)

9

Table 2.2 Example Input-output model Input output example

(millions of US$) Agriculture, Hunting, Forestry and Fishing Mining and Quarrying Food, Beverages and Tobacco Textiles and Textile Products Leather and Footwear Agriculture, Hunting, Forestry and Fishing

X1

Mining and Quarrying X2

Food, Beverages and Tobacco

X3 Textiles and Textile

Products

X4

Leather and Footwear X5

2.2 Trade policy

In the past, gross trade has been an important tool for policy and in particular trade policy. By then, or at least before 1980, gross exports were more than 90% domestic value added

(Hummels et al. 2001). Since 1985, international value chains have been developing. These international value chains need trade in intermediates. This causes problems as explained above, in particular the double counting problem for gross trade. Also, the role of

competitiveness has changed. Competitiveness is no longer solely determined by domestic clusters of manufacturing firms, but relies increasingly on the successful integration of other tasks in the chain, both domestic and foreign ones (Timmer et al, 2014). This idea is also supported by Robert C. Johnson (2014); Ahmad & Ribarsky (2014) who concluded that the growing difference between gross exports and value added exports has some important policy implications. They give respectively five and seven policy areas which will be better informed with trade in value-added. We will elaborate more on policy implications with respect to trade policy, the example of the iPod case (Dedrick et al. 2009) is used to make some points clear.

In a world with exports in just final goods, countries try to decrease tariffs to important export markets and hold tariffs for importing sectors. Mainly to recover competitiveness or to let competitiveness of international firms rise. Tariffs on imports are also a good protection against a negative trade account. At the time, policy makers were mostly focused on tariffs with bilateral trading partners, to give direction to competitiveness and the national trade account. The rise of global value chains makes the effects of tariffs different and more complex. By now policy makers should look at where value of imports and exports comes from. In the case of the iPod, the US would (in a gross exports view) just look at the pros and cons of the import tariff with China. Looking at the supply chain it becomes clear that a lower tariff for China’s inputs from Japan and also from the US can lower the input prices of

assembled iPods and so the price of the iPod. Low tariffs between all trading partners within a supply chain can lower total transaction costs. This makes the whole iPod supply chain more competitive, which is an important export product of the US. Another example is Japan, if Japan for example tries to improve the trade balance, it can increase the import tariff on iPods from the US, but it must keep in mind that the imported iPods have a high domestic content. If it stops importing iPods, it will decrease profits in the industries supplying the

(10)

10 US will improve the number of iPods sold and will increase profits for Japan. So, countries should not only look at preferential tariffs between bilateral trading partners, but between all partners in the value chain. This favours the choice to multilateral EIAs. Countries should be cautious with stimulating exports and holding back imports. Exports can have a high foreign value added content, as in the iPod case for China, so the effect of exports can easily be overrated. Holding back imports can be a disaster for related export industries. In addition, imports can have a domestic value added content what can lead to overestimation of the negative effect for the national trade account.

2.3 FDI and deep integration

Global value chains can be originated in two ways: companies search for foreign partners to work with, or they invest in new business abroad. EIAs with ‘deep integration’ provisions are expected to improve the overall economic climate for multinationals to restructure their operations internationally through outsourcing and offshoring of activities (Cardamone and Scoppola 2012). This is basically equivalent to the establishment of GVCs. ‘Deep integration’ provisions are the non-trade provisions including trade related intellectual property rights (TRIPS), capital movement, competition policy, investment liberalisation, dispute settlement, and services liberalisation (Cardamone and Scoppola 2012). These ‘deep integration’

provisions make investments in foreign markets possible and protect companies intangible assets. This is crucial because there is an increasing importance of intangible capital in GVCs. Examples are software, brand names and organizational firm-specific capital. These

intangible assets give international firms a competitive edge and are also the explanation why Apple is able to absorb so much profit from the iPod chain (dedrick et al. 2009). EIAs

(11)

11 platform to save transaction costs. The highest savings are ‘tariff jumping’ and transportation costs. So ‘deep integration’ will boost FDI and thereby the rise of global value chains. Since vertical FDI is trade promoting and horizontal FDI is a substitute for exports, it is the question if this will increase or decline total value added exports.

2.4 The effects of EIAs on trade

The international trade literature recognizes the potential of EIAs to increase trade among its members. Studies mostly use the gravity equation to test the effect of EIAs on bilateral trade levels. The gravity model equation predicts that the volume of trade between two economies should increase with their size and decrease with transaction cost. Regional EIAs lower the transaction cost in particular with tariff reductions. Size is measured as national GDP. Most studies also use common borders, language, and colonial heritage as dummy variables

Since Tinbergen (1962), who showed that the Commonwealth and Benelux

preferences had trade-creating properties, the effects of EIAs on the level of trade have been tested empirically. Important and relatively old studies were Aitken (1973), Bergstrand (1985) and Thursby and Thursby (1987) which show that the European trade blocs increased trade during the 1960s and 1970s. Frankel et al. (1995) see a positive relationship between integration and the level of trade in North Amerika. On the other hand they found that the increasing trade in East Asia during the 1980s is totally due to economic growth. Since 2000, many studies examined the effect of EIAs for more regions across the world. Examples are Rose (2000), Feenstra et al. (2001), and Frankel and Rose (2002), who find that RTAs, in general, are trade creating. However, these preferential trading arrangements (PTA) have a two-sided quality: they liberalize commerce among members, while discriminating against third parties. This diversion mechanism is a downside of preferential EIAs, because if you give some countries a preferential treatment, other countries must be discriminated.

That’s why Ghosh & Yamarik (2004) question whether regionalism enhances the volume of trade within the bloc, or simply diverts bilateral trade away from countries outside the bloc. Trade creation especially occurs as low-cost member countries displace high-cost domestic producers. Trade diversion, on the other hand, occurs when members of a trading bloc reorient their trade away from low-cost non-member countries towards higher-cost member countries. They did several extreme bound analyses to test the robustness of earlier results, and concluded that the trade creation effect is overstated. Their extreme bounds analyses show that the trade creation effect of regional EIAs is close to zero. This is in line with Rose (2002, 2004a) who concludes that there is no evidence that the WTO has increased world trade. If preferential EIAs only divert trade to less efficient countries, then these

(12)

12 countries between 1950 and 2000, he gives it a qualified yes and notes that economic

integration agreements (EIAs) promote trade by at most 50%. Surprisingly, more than half of the EIAs investigated have had no discernible impact on trade at all, while only about one quarter of the agreements are trade promoting. Following Kohl (2014) extensive and more enforceable agreements tend to be more trade promoting. Further characteristics of these agreements, such as their institutional quality, design, and their members’ involvement in the World Trade Organization, also influence the effect of EIAs. To summarize, the effect of EIAs is mixed across studies, but the consensus is rather positive. Further, EIAs are heterogeneous and it appears that extensive and enforceable agreements have more effect.

2.5 The effects of EIAs on global value chains

Global value chains are more than importing and exporting products, firms have to work together. That is why international supply chains need deeper integration than just tariff reductions for products that cross borders. The depth of an agreement is dependent of the number of enforceable provisions covered by an agreement (Orifice and Rocha 2014). Lawrence (1996) is the first who indicates the connection of global value chains and deep integration. He argues that this deep integration is mainly driven by multinationals. They represent efforts to fill the functional needs of international trade and investment. Firms that plan to source in one country and sell in others need security about the rules and mechanisms governing trade. Such firms also prefer secure intellectual property rights as well as technical standards and regulations that are compatible. Orefice and Rocha (2011, 2014) also note that production networks ask for deeper integration. Disciplines such as infrastructure, institutions, competition policy and the standardization and harmonization of product regulations would make production-sharing activities more secure and efficient. This is in line with Antras and Staiger (2012) who conclude that trade flows involving the exchange of customized inputs, incomplete contracts and costs associated with the search for suitable foreign input suppliers create new forms of cross-border policy effects. According to them, this can be mostly explained with the increasing importance of the holdup problem: The share of differentiated inputs in world trade has more than doubled between 1962 and 2000, when distinguishing between differentiated and homogeneous goods. These differentiated/customized inputs often need relationship specific investments. These investments have to be made before the

intermediate products can be sold. Therefore, these investments will be sunk costs when bargaining with a possible buyer, whereby they will decline bargaining power. This will cause underinvestment and lead to lower levels of trade and welfare loss. This despite the fact that two parties were able to work together and could maximize profits by cooperating. Such problems can be solved by enforceable contracts, which could be arranged through EIAs. Also, Johnson (2014) concludes that the rise of global value chains should have changed trade patterns and the role of EIAs. He compares value added trade with gross trade, but doesn’t look at the effects of EIAs on global value chains or value added trade.

(13)

13 a positive effect on production networks. This two-way link between production networks and deep integration is empirically tested, but also this is done with gross trade data.

2.6 The effect of EIAs on manufacturing- and service sectors

Manufacturing industries are more fragmented and international oriented than service industries (Baldwin and Gonzalez 2014). In addition, there is also more growth potential in the sense that all manufacturing products can be offshored, while just about 25% of the service jobs are offshorable (Blinder, 2009). Blinder (2009) concludes that most service jobs cannot be offshored because they need personal interaction. But will this mean that the total value added exports share of services decreases? Not for sure, Timmer et al (2014) note that in almost all high-income countries, the number of services jobs related to manufacturing production increased from 1995 till 2008. This increased the value added share of services related to manufacturing exports. Additionally, trade in products is already more liberated and thereby probably closer to its natural boundaries. This could imply that there is more progress to be made for service sectors, which could be achieved with information- and

communication technology and trade policy. It is questionable if EIAs also promote trade in manufactures more than in services. As far as we know is there little empirical research done about the difference in impact of EIAs on manufacturing- and service sectors. This can be tested with the WIOD and TiVA dataset, because all value added is distributed between sectors and Timmer et al (2014) have divided all these sectors to manufacturing sectors, service sectors, and primary sectors.

2.7 Goal of this paper

The goal of this paper is to examine the effects of EIAs on GVCs. This study will look at the general effect of EIAs and at the heterogeneity of EIAs on GVCs. The amount of global value chains is hard to measure, therefore FDI and value added exports are used as proxies for GVCs. Additionally will be explored, whether the impact of EIAs is heterogeneous between manufacturing- and service sectors. This study uses the WIOD and TiVA database for value added trade data and the FDI data is sourced from UNCTAD (2015).

(14)

14 2.8 Hypotheses

Based on the literature review discussed above the following research questions and hypotheses are formulated:

Do EIAs positively affect global value chains?

H1a: EIAs have a positive effect on the level of value added exports. H1b: EIAs have a positive effect on FDI.

Is the effect of EIAs on global value chains heterogeneous? H2a: The effect of EIAs on value added trade is heterogeneous H2b: The effect of EIAs on FDI is heterogeneous.

Is the impact of EIAs heterogeneous between manufacturing- and service sectors? H3: EIAs affect trade in manufacturing sectors more than trade services.

3 Method

This study will test the effect of EIAs on value added trade and FDI by means of the gravity equation. The gravity equation is used because it is a suited and reliable manner to test the hypotheses above. Since Tinbergen (1962) the gravity equation has been the most common method to measure the effect of EIAs on trade. Since his research there have been multiple researches about the effects of tariff reduction/ EIAs (Aitken, 1973; Bergstrand, 1985; Ghosh & Yamarik, 2004). The gravity equation is a model of bilateral interactions that is able to measure the effects of EIAs on trade. It is reliable, because it has been used for more than 50 years and has been developing over the years. Below the main idea and developments of the gravity equations are given. The models are basically the same for (value added) trade and FDI, only the dependent variable changes. Below, trade is used as the dependent variable.

Gravity equations are a model of bilateral interactions in which size and distance effects enter multiplicatively. Head and Mayer (2013) tested the effect of economic size (measured as GDP) on trade, they compare trade levels of all European countries with Japan. This is a good research design, because all other variables that influence trade levels are the same. Namely, none of these countries share their language, religion, currency or colonial history with Japan and the distance to Japan is comparable. Exports seem to rise

proportionately with the economic size of the destination. Next, Head and Mayer (2013) continue with distance, where they look at the effect of distance on exports and imports. They find that trade is inversely proportional to distance. In the simplest form of the gravity

equation, only size (measured as GDP) and distance affect the level of trade. This regression is expressed below in a panel data format. Logarithms are used if variables are not normal distributed.

lnTijt = β0 + β1(lnGDPit) + β2(lnGDPjt) + β3(lnDISTij) + Ɛijt (1)

Tijt indicates trade between exporter i and importer j in year t, β0 is the constant. GDPit and

GDPjt are proxies for economic size. DISTij is the physical distance in kilometers, between the

(15)

15 However, a better measurement for distance is transaction cost. As a high import tariff is probably more trade declining than a slightly greater distance. So the combination of size and transaction costs better explain trade levels. Distance is obviously an important parameter for transaction costs. So the gravity equation predicts that the volume of trade between two economies should increase with their size and decrease with transaction cost. EIAs, the explanatory variable in my model, should lower the transaction cost, in particular with tariff reductions.

lnTijt = β0 + β1(EIAijt) + β2(lnGDPit) + β3(lnGDPjt) + β4(lnDISTij) + β5 (BORDij)

β6(LANGij) + Ɛijt (2)

The added variable EIAijt is a binary variable that is 1 if both countries in a country-pair

belong to the same economic integration agreement and 0 otherwise. Furthermore, there are several other variables that influence the level of trade, mostly because they affect the

transaction costs. Therefore there are some dummy variables added, namely BORDij that turns

to 1 of a country-pair share a common border, and LANGij that is 1 if a country-pair share a

common language. This to enlarge the explanatory power of the model.

Later on, it becomes clear that this model still has some weak spots, in particular omitted variables that can lead to endogeneity. Anderson and Wincoop (2003) note that multilateral resistance is an important omitted variable. That is because if a country has high trade barriers to the rest of the world it is pushed stronger to trade with a given country. Australia for example is geographically remote from the rest of the world. This will raise transport costs and thereby prices in comparison to the rest of the world. So it raises the multilateral resistance to the rest of the world. This will lead to relatively lower import and export prices with New Zeeland and will favor trade levels between Australia and

New-Zeeland. The econometric solution for this endogeneity is fixed effects. For example, Redding and Venables (2004) note that importer and exporter fixed effects could be used to capture the multilateral resistance terms that emerged in different theoretical models. This leads to the regression below, which this study calls the Anderson and van Wincoop (AW) model.

lnTijt = β0+ β1(EIAijt)+ β2(lnGDPit) + β3(lnGDPjt) + β4(lnDISTij) + Fi + Fj + Ft + Ɛijt (3)

Where Fi and Fj are included to account for fixed effects from the importing and exporting

country. Ft is included to account for trends and shocks over time, as the financial crisis in

(16)

16 country and time fixed effects but interaction time- and country fixed effects, regression 4. Bilateral fixed effects (Fij) account for endogeneity bias and the variation in distance, borders,

and common languages. Country-and-time (Fit and Fjt) fixed effects account for variation in

real GDP and the multilateral price terms. The explanatory effect of all dummy variables is captured by the fixed effects, making dummies for size and distance redundant. This leads to regression 5, which this study calls the BB model. A second advantage of this model is that the GDP variables are out of the regression, which is build-up of the same components as value added exports. This gives the models below.

lnTijt = β0 + β1(EIAijt) + β2(lnGDPit) + β3(lnGDPjt) + β4(lnDISTij) + Fit + Fjt + Fij + Ɛijt (4)

lnTijt = β0 + β1(EIAijt) + Fij + Fit + Fjt + Ɛijt (5)

Regression 5 will be the main model for testing the hypotheses, because it performs best against endogeneity. This makes it the most reliable and least biased model. Additionally, redundant variables are omitted.

However, the effect of EIAs holds longer than one year. Following Baier and Bergstrad (2007): every trade agreement is “phased –in”, typically over 10 years. Unfortunately, the TiVA dataset only includes data for the years 1995, 2000, 2005, and annually from 2008 to 2011, and is therefore not suited for time lags. WIOD contains annually data between 1995 and 2011, this is too short for a time lag of ten years, so the model will be estimated using lag variables for the first 5 years. FDI data is even shorter, from 2001 until 2012. Additionally, some years are missing and therefore this dataset is also not suited for time lags. Beside this, the combination of PPML estimation technique and time lags is not working. Estimation with phase-in effects yields:

Tijt = β0 + β1(EIAijt) + β2(EIAij,t-1) + β3(EIAij,t-2) + β4(EIAij,t-3) +β5(EIAij,t-4) + β6(EIAij,t-5) +

Fij + Fit + Fjt + Ɛijt (6)

Where EIAij,t-1 till EIAij,t-5 are lagged levels of the EIAijt dummy.

3.1 Strict exogeneity

Regression 5 is strict exogenous if EIAs cause trade and not the other way around. Following Wooldridge (2002,p. 285) we can test for strict exogeneity by adding a future level of EIAs. If FTA changes are strictly exogenous to trade, flow changes should be uncorrelated with the concurrent trade flow. They are uncorrelated if we see no significant effect for (FTA, t+1)

(17)

17 Tijt = β0 + β1(EIAijt)+ β2(EIAij,t+1) + Fij + Fit + Fjt + Ɛijt (7)

Table 3.1 Estimates to Test for strict exogeneity

(1) (2) (3) (4) (5) VARIABLES ESTIMATION TECHNIQUE TiVA OLS WIOD OLS WIOD OLS FDI OLS FDI PPML EIAs -0.304*** 0.0591 0.0992** 2.187* 0.151*** (0.0779) (0.0362) (0.0474) (1.163) (0.0346)

5 Extra phase-in years 0.0259

(0.0338)

EIAij, t+1 0.358*** -0.0329 -0.0258 -1.986* -0.118***

(0.0688) (0.0249) (0.0232) (1.131) (0.0297)

Observations 25,619 26,513 18,719 23,505 23,148

Adjusted R-squared 0.958 0.972 0.977 0.750

Ft and Fij used NO NO NO NO YES

Fij, Fit and Fjt used YES YES YES YES NO

Estimates marked ***/**/* are significant at the 1/5/10 % level. Robust standard errors (clustered by country-pair) are in parentheses. The dependent variable is the natural log of the real bilateral trade flow- or FDI out stock from i to j as noted above each regression. The variable ‘EIAij, t+1’ is added to the main regressions of the results section.

3.2 Poisson Pseudo-maximum likelihood (PPML) estimation technique

Silva and Tenreyro (2006) claim that the error term is affected by other right-hand side variables. They empirically show that in the Anderson and Wincoop (2003) and most other log-linearized gravity equation suffer due the presence of heteroscedasticity.

Heteroscedasticity leads to inconsistent estimates. To overcome this heteroscedasticity they use the PPML estimation technique. In addition, this model can deal with zero observations in a natural way, while the gravity equations above have a problem with zero observations. This problem arises because the gravitational force can be very small, but never zero, whereas trade and in particular FDI between several pairs of countries can be zero. As displayed in Table 4.1 the FDI dataset does include many zeros. The main model by FDI as dependent variable will be the PPML estimation technique, because the PPML model outperforms other estimation techniques when using a dataset including many zeros (Siva and Tenreyro 2011).

4 Data

4.1 Data sources

The TiVA database is sourced from OECD (2012). It contains a balanced panel with the value added exports of 61 countries covering OECD, EU28, G20, most east and south-east Asian economies, and some South American countries (appendix, Table A.1). The original database covers 34 unique industrial sectors, which are combined to TiVA total (all industries), TiVA manufacturing (16 industries), and TiVA services (17 industries). Data is available for 1995, 2000, 2005, and 2008 up to 2011.

(18)

18 including 27 European countries and 13 other major economies, which covers more than 85% of world GDP (appendix, Table A.1). The dataset contains annual data from 1995 to 2011 with figures of 35 different industries covering the overall economy. This dataset is balanced, so data is annually available for every sample country and industry between 1995 and 2011.

The FDI data includes data from 2001 till 2012 and is sourced from UNCTAD (2015). The database is based on multiple National Statistical Offices, where information is mostly taken from the balance of payments. UNCTAD has combined these national data sources to form this database. FDI is contained in the Balance of Payments if an investment is made to acquire lasting interest in enterprises operating outside of the economy of the investor (Gugler 2015). A lasting interest in general means at least 10% of voting power. This study only uses the FDI figures of outward stock from the TiVA countries (appendix, Table A.1). FDI outward stocks are presented at book value or historical cost, reflecting prices at the time when the investment was made.

EIAs are sourced from Kohl et al (2015). This dataset contains 296 agreements across the world. For all agreements it is indicated which provisions are covered. In this study, Provisions are only counted as covered if they are enforceable. The provisions are listed in the appendix (Table A.2).

The gross domestic product (GDP) figures are from the World Bank (2013). GDP is calculated in purchaser's prices, GDP is the sum of gross value added by all residents-, firms- and government in the economy.

Data on bilateral distance, common borders, and languages are from Mayer and Zignago (2011). Bilateral distance measures the distances between capitals in kilometers. The dummy for common borders turns to 1 if two countries share a common border and 0

otherwise. The dummy for languages is one if two countries share the same mother language and 0 otherwise.

4.2 Testing data

First the means, minimum value and maximum value of all variables are calculated. After noticing that the minimum value of most dependent variables was negative or zero, it was tested how much observations are zero or negative. See appendix (Table A.3) for the summary statistics of all observations.

Table 4.1 Negative and zero observations datasets

Dependent variable Number of

observations

Number of zero’s Number of negative

values TiVA total 25620 71 0 TiVA manufactures 25620 264 1 TiVA services 25620 72 0 WIOD total 23400 0 0 WIOD manufactures 23400 0 0 WIOD Services 23400 0 0

Gross trade imports 24450 619 0

Gross trade exports 24450 676 0

(19)

19 TiVA manufacturing has one negative observation, which is probably because something is exported with a little loss or a little accounting mistake. Since this is just one observation, it will not significantly influence the results. Then there are the zeros of the TiVA database, except for TiVA manufacturing, the number of zeros is less than 0.3 percent points of the measured observations. Therefore, no serious problems are expected, but it is good to keep in mind that there are some zero observations. Then FDI, the FDI dataset is an unbalanced panel with 23511 out of 43920 (61*60*12) measured observations. Additionally, the number of zeros is huge and there are also quite some negative values. Missing observations are not random, because missing observations are more likely to occur for small and distant countries (Silva and Tenreyro 2006). The zero values are a result of no outward FDI, or of a value less than $500.000, since values are rounded in millions of US dollars. Finally, it is possible to get negative values if a domestic country acquires an affiliate of a foreign company operating domestically (Gugler 2015). So if a Dutch company acquires an affiliate of a foreign country that is operating in the Netherlands, this is counted as inward divestment, also called negative sales. The non-random missing observations, zero values, and negative values can make the OLS model severely biased, distorting the interpretation of the model. That is why the FDI model will also be estimated with the pseudo-maximum-likelihood (PML) estimation technique. This model provides a natural way to deal with zero values of the dependent variable (Silva and Tenreyro, 2006).

4.2.1 Correlations

To test for multicollinearity all correlations are tested. The results are displayed in the appendix (tables A.4, A.5, and A.6. There are two high correlations. First, GDP is correlated with all dependent variables, close to 0.3. This is not really surprising because value added exports and GDP are constructed with some similar components. As noted by data sources, GDP is the sum of gross value added by all residents-, firms- and government in the economy, which is of course correlated with value added exports. Second, EIAs are negatively

correlated with distance -0.68 and even -0.8 for WIOD observations. This is also not

surprising; it is more likely to have a trade agreement with your neighbour than with a country on the other side of the world. However the figures are higher than in most other studies that test the effect of EIAs on trade. This correlation is probably boosted by European biased samples, because the European countries are highly integrated and located close to each other. This also declares the even higher correlation for WIOD, which suffers more from a European bias.

Table 4.2 correlations variables

tivatot EIA Distance GDPi

tivatot 1

EIA 0.0698 1

Distance -0.126 -0.6806 1

(20)

20 4.2.2 Normal distributions

After performing Skewness and Kurtosis tests for all continuous variables it seems that none of the continuous variables are normal distributed, as showed in the appendix (Table A.10). Therefore logarithms are taken, which causes a better distribution as showed in Table 4.1 and for all other variables in appendix (tables A.7, A.8, and A.9). However, following the

Skewness and Kurtosis test, the variables are still not normally distributed (appendix, Table A.10). This is a shortcoming in the data.

Figure 4.1 Distribution TiVA exports

4.2.3 Differences between the TiVA and WIOD database

Since the rise of global value chains, the foreign value added content of exports has been growing. It is a challenge to disentangle domestic and foreign value added for fragmented exports and imports. When found how many value is added in other countries, it is still difficult to determine the source country. Researchers try to harmonize IOTs of different countries to come closer to the real value added trade in countries. This creates combined international value added trade data, but still includes measurement errors. This also becomes clear after comparing observations from WIOD and TiVA. Which shows that WIOD values are on average 61% higher and sometimes more than 10 times the TiVA value. Further, the total of value added exports for all service sectors is comparable between both databases, while the value of manufacturing value added exports is twice as much in the WIOD

database. This comparison is made from the 10920 corresponding observations, which are the TiVA years (1995, 2000, 2005 and annual from 2008 and 2011) and all the 40 WIOD

countries.

(21)

21 methods provided by different researchers and institutions. This creates differences between both datasets. Several reasons which can cause differences in data are noted in the table below. It is difficult and outside the scope of this study to find out the precise impact of every cause. The results section will show how much the differences influence our estimates. Table 4.3 differences in composing datasets

Reasons which can cause differences in data.

WIOD TiVA

Trade statistics only contain first tier suppliers. How is more detailed information collected?

Based on Broad economic categories (BEC) codes

'proportionality' assumption

Mains source for database Supply use tables from official national accounts

Input output tables SUT are available for a

limited set of years (often every 5 years). Estimation is needed for non-benchmark years.

Estimation following the SUT-RAS method

Leaving out several non-benchmark years

Some trade statistics are missing

Gaps are filled with data from various sources or have been Gathered from the respective National Statistical Institutes (NSIs) on request.

Missing flows are currently estimated using

econometric model estimates.

Exports from country A to country B should be symmetric to imports of B from A, but this is mostly not the case.

Exports are mirrored to import statistics

Conversion of c.i.f. price based import figures to f.o.b. price based imports to reduce the inconsistency issues of mirror trade. Bilateral trade statistics in

services is scarce

Various sources including UN, OECD, and Eurostat

TEC database

The WIOD database has proved its value for several years in multiple studies as Timmer et al. (2015), and Baldwin and Gonzalez (2014). In Contrast to the TiVA database, where we use the initial release. In addition, only the WIOD database is able to prove causality in a strict exogenous model as showed in table 3.1. Therefore, this study attaches more importance to WIOD results, but will be cautious when there are mixed results.

4.3 Shortcomings of the data

The used data has several shortcomings. First, the variables are not normally distributed. Second EIAs are negatively correlated with distance, and GDP is correlated with all

(22)

22 several countries. This will influence results, in particular when using time lags. Finally, both WIOD and TiVA measure value added exports, but differences are enormous. This indicates measurement- and harmonize inaccuracies.

5 Results

5.1 EIAs positively affect global value chains

The main objective of this paper is to test whether EIAs positively affect global value chains. This study uses two instruments to measure global value chains, value added exports and FDI. The TiVA database as well as the WIOD are used to test the effect on value added exports. As expressed in the data section, there are some significant differences between the datasets. We will first find out how much this is influencing our results. Therefore we do our main

regression, with the 10920 observations that are present in both data sets. The first three columns in table 5.1 are done with TiVA observations and the last three columns with WIOD observations. Results out of column 1, 2, and 3 should be comparable with the results of column 4, 5, and 6 respectively. The effect of a trade agreement is slightly higher in column 1 than column 4, with significant positive coefficients of 1.48 and 1.38 respectively, while the coefficients for GDPi and bilateral distance are higher and more significant with TiVA values. In the estimations, with interaction country and time fixed effects, the coefficients in column 2 and 3 are more than 3 times higher than those in column 5 and 6, while the errors are just slightly higher. Therefore, the TiVA regressions are more prone to get significant results. This has appeared in the regression with standard errors (column 3 versus 6), where only TiVA is able to get a significant positive result, at the 5 percent level. To conclude, the TiVA- and WIOD databases are not interchangeable, because data differences can influence the results. Table 5.1 Estimates with the 10920 observations that are present in both datasets

(1) (2) (3) (4) (5) (6)

VARIABLES TiVA TiVA TiVA WIOD WIOD WIOD

EIAs 1.483*** 0.0834 0.0834** 1.378*** 0.0235 0.0235 (0.0672) (0.0585) (0.0331) (0.0715) (0.0464) (0.0296) lnGDPi 0.0108** 0.00705* (0.00506) (0.00387) lnGDPj -2.92e-05 -0.000939 (0.00379) (0.00349) lnDIST -0.322*** -0.00512** (0.0250) (0.00228) Observations 10,920 10,920 10,920 10,918 10,918 10,918 Adjusted R-squared 0.817 0.967 0.967 0.816 0.970 0.970

Fi, Fj, and Ft used YES NO NO YES NO NO

Fij, Fit, and Fjt used NO YES YES NO YES YES

(23)

23 5.1.1 H1a: EIAs have a positive effect on the level of value added exports

EIAs are expected to have a positive effect on value added exports for two reasons. First, transaction costs decline, which makes exporting more profitable. Second, it will boost vertical FDI, which affects trade positively. The second set of regressions test the effect of EIAs on value added exports (Table 5.2). The first three regressions are done with the TiVA database and the last three with WIOD. The Anderson and van Wincoop (AW) regressions with country and time specific fixed effects (column 1 and 4) show a positive and significant effect for EIAs. Additionally, exporter GDP is positive, and distance is negative, except for GDPi, all at a 1 percent level. The main model of this study is the BB version with interacting time and country fixed effects, because it suffers least from endogeneity. Using standard error terms (column 2 and 5), the effect of EIAs is positive and significant at a 5 percent level, with TiVA as well as WIOD observations. The coefficients are 0.054 and 0.034 which can be interpreted as a 6% (e^0.054) and 4% (e^0.0439) increase in value added exports. However, this model suffers from endogeneity. The estimations in columns 3 and 6 are identical to 2 and 5, but then with robust standard errors. Using robust standard error terms clustered by Fij,

(24)

24

Table 5.2 The effect of EIAs on value added trade

(1) (2) (3) (4) (5) (6)

VARIABLES TiVA TiVA TiVA WIOD WIOD WIOD

EIAs 1.354*** 0.0538** 0.0538 1.116*** 0.0349** 0.0349 (0.0474) (0.0246) (0.0379) (0.0743) (0.0170) (0.0400) lnGDPi 0.0136*** 0.00642* (0.00407) (0.00365) lnGDPj 0.00322 -0.00290 (0.00363) (0.00298) lnDIST -0.454*** -0.0430*** (0.0246) (0.00299) Observations 25,620 25,620 25,620 26,514 26,514 26,514 Adjusted R-squared 0.828 0.958 0.958 0.812 0.972 0.972

Robust standard errors YES NO YES YES NO YES

Fi, Fj, and Ft used YES NO NO YES NO NO

Fij, Fit, and Fjt used NO YES YES NO YES YES

Estimates marked ***/**/* are significant at the 1/5/10 % level. Robust standard errors (clustered by country-pair) are in parentheses. The dependent variable is the natural log of the real bilateral trade flow from i to j. The dependent variable in column 1 and 2 is sourced from the TiVA data set and for column 3 and 4 from WIOD.

5.1.2 ‘Phased-in’ effect EIAs

According to Baier and Bergstrand (2007), the effect of EIAs will hold longer than one year, and will ‘phase-in’ in ten years. Since de WIOD is the only useful dataset for time lags, the results with WIOD data are displayed below (Table 5.3 and 5.4). Table 5.3 shows estimations with standard errors and cumulative phase-in effects and Table 5 shows estimations with robust standard errors clustered by Fij. Just the main model with interaction country and time

fixed effects is used. Considering the fact that EIAs only affect value added exports

significantly with standard error terms, we will discuss this model. Comparable regressions with robust standard errors clustered by Fij are listed in appendix (Table B.1). The

comparable model with time lags separated per year, is also displayed in appendix (Table B.2).

(25)

25

Table 5.3 The effect of EIAs with phase-in effects with cumulative phase-in effect

(1) (2) (3) (4) (5) (6)

VARIABLES WIOD WIOD WIOD WIOD WIOD WIOD

EIAs 0.0349** 0.0470* 0.0526** 0.0430** 0.0550*** 0.0810***

(0.0170) (0.0280) (0.0229) (0.0211) (0.0207) (0.0209)

1 Extra phase-in year 0.00126

(0.0273)

(0.0217)

(0.0197)

4 Extra phase-in years 0.0353*

(0.0191)

Observations 26,514 24,956 23,399 21,839 20,279 18,720

Adjusted R-squared 0.972 0.973 0.974 0.975 0.976 0.977

Fi, Fj, and Ft used NO NO NO NO NO NO

Fij, Fit, and Fjt used YES YES YES YES YES YES

Estimates marked ***/**/* are significant at the 1/5/10 % level. Standard errors are in parentheses. The dependent variable is the natural log of the real bilateral trade flow from i to j.

Table 5.4 The effect of EIAs with phase-in effects

(1) (2) (3) (4) (5) (6)

VARIABLES WIOD WIOD WIOD WIOD WIOD WIOD

EIAs 0.0349 0.0184 0.0185 0.0201 0.0201 0.0160 (0.0400) (0.0370) (0.0367) (0.0365) (0.0363) (0.0361) 1e Phase-in year 0.0236 0.0222 0.0224 0.0224 0.0217 (0.0248) (0.0162) (0.0162) (0.0162) (0.0162) 2e Phase-in year 0.00172 -0.0101 -0.0101 -0.00984 (0.0242) (0.0165) (0.0165) (0.0165) 3e Phase-in year 0.0148 0.0151 0.0148 (0.0241) (0.0152) (0.0152) 4e Phase-in year -0.000395 0.0233 (0.0246) (0.0149) 5e Phase-in year -0.0303 (0.0258) Observations 26,514 26,513 26,512 26,511 26,510 26,509 Adjusted R-squared 0.972 0.972 0.972 0.972 0.972 0.972 Fi Fj and Ft used NO NO NO NO NO NO

fij fit fjt used YES YES YES YES YES YES

Estimates marked ***/**/* are significant at the 1/5/10 % level. Robust standard errors clustered by Fij are in

(26)

26 5.1.3 H1b: EIAs have a positive effect on FDI

The second proxy for global value chains is FDI. The expectation is that EIAs boost FDI. It boosts FDI by improving the overall economic climate for multinationals to restructure their operations internationally (Cardamone and Scoppola 2012). Due the fact that there are so many missing observations and zero values, apart from the OLS method, the PPML model is also used because this method deals in a natural way with zero observations (Silva and Tenreyro 2006). The first two columns are regressions with the OLS method, and the regressions in the last three columns are done with PPML estimation. The coefficient in the first column with specific country and time fixed effects is 2.542. This means that EIAs increase FDI with more than 1000%, so this model really overestimates the effect of EIAs. While Column two has no significant result possibly due to all the missing observations. Therefore, in the last three columns the Poisson model with PML estimation is used. The Poisson model works with a maximum of two fixed effects, therefore the AW model (column 3) is estimated with only specific importer- and exporter fixed effects. The BB model (column 4) is done with bilateral and time fixed effects and column 5 uses importer and time fixed effects as in Gómez and Milgram (2010). Due to the fact that the BB model is estimated without interacting time- individual effects, in theory the explanatory power of GDPi and GDPj is not captured, so we let this stand in the regression. The PPML estimation in column 3 gives all the expected signs. EIAs are positive with a coefficient of 0.14, also the size

(27)

27

Table 5.5 The effect of EIAs on FDI

(1) (2) (3) (4) (5)

VARIABLES OLS OLS PPML PPML PPML

EIAs 2.542*** 0.203 0.142*** 0.0309** 0.104*** (0.304) (0.409) (0.0100) (0.0140) (0.00982) lnGDPi -0.0117 0.0174*** 0.00141 0.0791*** (0.0290) (0.00189) (0.00126) (0.00487) lnGDPj -0.00170 0.0206*** 0.00171 0.00276 (0.0360) (0.00280) (0.00162) (0.00276) lnDIST -0.126*** -0.00785*** -0.00990*** (0.0194) (0.000951) (0.00104) Observations 23,511 23,506 23,511 23,149 23,511 Adjusted R-squared 0.425 0.750

Fi used YES NO YES NO NO

Fj used Ft used Fij used Fit used Fjt used YES YES NO NO NO NO NO YES YES YES YES NO NO NO NO NO YES YES NO NO YES YES NO NO NO

clustered by Fij YES YES NO NO NO

Estimates marked ***/**/* are significant at the 1/5/10 % level. Robust standard errors (clustered by country-pair) are in parentheses. The dependent variable is the natural log of FDI from i to j.

5.2 The effect of EIAs on global value chains is heterogeneous

The second objective of this paper is to test of the effect of EIAs on global value chains is heterogeneous. The literature predicts that more extensive and enforceable EIAs have more effect on value added trade, that is why we have made a dummy variable for deep integration. 5.2.1 H2a: The effect of EIAs on value added trade is heterogeneous

The TiVA and WIOD datasets contain 17 different provisions that could occur in EIAs, which are displayed in appendix Table A.2. EIAs are counted as deep integration agreements if 15 of these 17 categories contain enforceable provisions. With TiVA as dependent variable (column 1 and 2), we see that the impact of extensive EIAs is positive and significant, even at a 1 percent level. Additionally, the coefficient for deep integration is 0.18 (20%), which is more than three times as much than of EIAs in comparable estimation without the coefficients for deep integration (Table 5.2 columns 2 and 3). This suggest that the effect of EIAs is

(28)

28

Table 5.6 Estimating of EIAs affect value added trade heterogeneous

(1) (2) (3) (4)

VARIABLES TiVA TiVA WIOD WIOD

EIAs 0.00253 0.0497 (0.0398) (0.0408) Deepintegration 0.183*** 0.180*** -0.0674 -0.0177 (0.0466) (0.0509) (0.0422) (0.0523) Non-deepintegration -0.00744 0.0497 (0.0390) (0.0408) Observations 25,620 25,620 26,514 26,514 Adjusted R-squared 0.958 0.958 0.972 0.972

Fij, Fit, and Fjt used YES YES YES YES

Estimates marked ***/**/* are significant at the 1/5/10 % level. Robust standard errors (clustered by country-pair) are in parentheses. The dependent variable is the natural log of the real bilateral trade flow from i to j. The dependent variable in column 1 and 2 is sourced from the TiVA data set and column 3 and 4 from WIOD. The variable deep integration turns to on if at least 15 provisions are enforceable and non-deep integration if less than 15 provisions are enforceable. The boundary of 15 provisions gives the most equal distribution as expressed in appendix Table A.11

5.2.2 H2b: The effect of EIAs on FDI is heterogeneous

According to Cardamone and Scoppola (2012), FDI is mostly promoted by non-trade provisions. This dataset contains 5 non-trade provisions: TRIPS, capital movement,

competition policy, investment liberalisation, and service liberalisation. The dummy variable deep integration turns to 1 if 4 of the 5 non-trade provisions are covered and enforceable. The OLS regressions are not in the model below, because we have seen that results are unreliable (Table 5.5 columns 1 and 2). Additionally, there were no significant results when using OLS in combination with the deep integration dummy.

(29)

29

Table 5.7 Estimating of EIAs affect FDI heterogeneous

(1) (2) (3) (4) (5) (6) VARIABLES PPML PPML PPML PPML PPML PPML EIAs 0.0847*** -0.0792** 0.00276 (0.0249) (0.0310) (0.0222) DeepintegrationFDI 0.0764*** 0.154*** 0.123*** 0.0452*** 0.149*** 0.151*** (0.0260) (0.0106) (0.0310) (0.0142) (0.0228) (0.00951) lnGDPi 0.0174*** 0.0171*** 0.00136 0.000755 0.0822*** 0.0820*** (0.00189) (0.00188) (0.00125) (0.00120) (0.00492) (0.00492) lnGDPj 0.0205*** 0.0202*** 0.00175 0.00108 0.00261 0.00225 (0.00280) (0.00280) (0.00162) (0.00162) (0.00275) (0.00275) Non-deepintegration -0.0322 -0.146*** -0.0687*** (0.0223) (0.0235) (0.0206) Observations 23,511 23,511 23,149 23,149 23,511 23,511

Fi used YES YES NO NO NO NO

Fj used YES YES NO NO YES YES

Ft used NO NO YES YES YES YES

Fij used NO NO YES YES NO NO

Fit used NO NO NO NO NO NO

Fjt used NO NO NO NO NO NO

Estimates marked ***/**/* are significant at the 1/5/10 % level. Robust standard errors are in parentheses. The dependent variable is the natural log of FDI from i to j. The variable Deep-integration turns to one if at least 4 of the 5 non trade provisions are enforceable. Non deep-integration turns to one if less than four provisions are enforceable. There were 100, 834, 792, 7481, and 934 EIAs with respectively 1, 2, 3, 4, and 5 non trade provisions. So the boundary of four non-trade provisions gives the most equal distribution.

5.3 H3: EIAs affect trade in services more than trade in products?

To test whether the effect of EIAs on service sectors is different from manufacturing sectors, all the value added exports of the manufacturing and service sectors are combined to value added manufacturing exports and value added service exports.

Table 5.8 displays the regressions with TiVA data and Table 5.9 with WIOD data. First the estimations in Table 5.8 are analysed. The first two columns in Table 5.8 measure the effect of EIAs with country and time specific fixed effects. These estimations suggest that the effect of EIAs on manufacturing- and service sectors are comparable with significant positive coefficients of 1.38 and 1.34. Looking at the main model from BB (Table 5.8 columns 3 and 4) we see that manufacturing is insignificant, while the effect of EIAs on services is positive and significant. This suggests that just service industries are positively affected by EIAs.

However the results are totally different with the WIOD data, as expressed in table 5.9. With the AW model the coefficient for EIAs is almost twice as high for manufacturing

(30)

30

Table 5.8 Estimating the effect of EIAs on TiVA service- and manufacturing sectors

(1) (2) (3) (4) VARIABLES Manufacturin g exports Service exports Manufacturing exports Service exports EIAs 1.385*** 1.340*** -0.00955 0.104** (0.0489) (0.0528) (0.0457) (0.0440) lnGDPi 0.0199*** 0.00834* (0.00441) (0.00447) lnGDPj 0.00404 0.00430 (0.00434) (0.00443) lnDIST -0.433*** -0.513*** (0.0245) (0.0268) Observations 25,620 25,620 25,620 25,620 Adjusted R-squared 0.825 0.791 0.944 0.948

Fi, Fj, and Ft used YES YES NO NO

Fij, Fit, and Fjt used NO NO YES YES

Estimates marked ***/**/* are significant at the 1/5/10 % level. Robust standard errors (clustered by country-pair) are in parentheses. The dependent variable is the natural log of the real bilateral trade flow from i to j. The dependent variable in column 1 and 3 are the combined value added exports of all manufacturing sectors, and t the dependent variable in column 2 and 4 are the combined value added exports of all service sectors

Table 5.9 Estimating the effect of EIAs on WIOD service- and manufacturing sectors

(1) (2) (3) (4) VARIABLES Manufacturing exports Service exports Manufacturing exports Service exports EIAs 1.372*** 0.716*** 0.0442** -0.0193 (0.0229) (0.0299) (0.0217) (0.0280) lnGDPi 0.00790*** -0.00356 (0.00267) (0.00350) lnGDPj -0.00334 -0.00268 (0.00266) (0.00349) lnDIST -0.0470*** -0.0383*** (0.00262) (0.00343) Observations 26,564 26,504 26,564 26,504 Adjusted R-squared 0.815 0.692 0.962 0.938

Fi, Fj, and Ft used YES YES NO NO

Fij, Fit, and Fjt used NO NO YES YES

(31)

31

6 Conclusions

The purpose of this paper was to test if EIAs positively affect GVCs and to test if this effect is heterogeneous. This information could be valuable for policymakers who want to increase economic growth and productivity by stimulating participation in GVCs with economic integration. Therefore it is estimated if EIAs positively affect value added exports and FDI out stock. It seems that the effect of EIAs on value added exports and FDI is at least 4% and 3%, respectively. These are minimum effects since the effect of the lowest coefficients are taken without phase-in effects. This study was not able to test ‘phase-in’ effects properly due to limited data availability. However, according to the WIOD observations, the effect of EIAs will double if we take a 5 years ‘phase-in’ period into account. So EIAs enhance value added trade with 8 percentage points within 5 years ‘phase-in’ period. The effects of EIAs are heterogeneous. More extensive and enforceable EIAs promote value added exports more, according to the TiVA observations. While no significant effect is measured with the WIOD observations. With FDI, we see that the impact of EIAs with at least 4 non trade provisions is higher than the general effect of EIAs. EIAs with at least 4 non trade provisions enhance trade between 5 and 17 percentage points. The gap between 5 and 17 percentage points is quite big, future research probably can make a more accurate estimation. Further, no positive effect is measured for the 1726 observations where a country pair shares an EIAs with less than 4 nontrade provisions. A possible explanation is that exports, a substitute for horizontal FDI, are more stimulated than FDI. Due to data differences between TiVA and WIOD, this study was not able to draw conclusions about the effect of EIAs on manufacturing sectors compared with service sectors. Meanwhile, it has shined some light on the big differences between TiVA and WIOD values. Future research could figure out the main causes of these

differences, improve data quality, and find out how data inaccuracies have influenced earlier research. Further, future research could investigate whether EIAs mainly affect horizontal- or vertical FDI. Finally, future research could test if the effect of EIAs is heterogeneous between, regions and income groups, which were out the scope of this study.

References

Ahmad, N., & Ribarsky, J. (2014). Trade in Value Added, Jobs and Investment. In 33rd General Conference of the International Association for Research in Income and Wealth, Rotterdam, Netherlands.

Aitken, N. D. (1973). The effect of the EEC and EFTA on European trade: A temporal cross-section analysis. The American Economic Review, 881-892.

Anderson, J. E., & Van Wincoop, E. (2003). Gravity with Gravitas: A Solution to the Border Puzzle, American Economic Review 93.

Antràs, P., & Staiger, R. W. (2008). Offshoring and the role of trade agreements (No. w14285). National Bureau of Economic Research.

Antràs, P., & Staiger, R. W. (2012). Offshoring and the role of trade agreements . The American Economic Review, 3140–3183

Baier, S. L., & Bergstrand, J. H. (2004). Economic determinants of free trade agreements. Journal of international Economics, 64(1), 29-63.

(32)

32 Baldwin, R., & Lopez‐Gonzalez, J. (2014). Supply‐chain Trade: A Portrait of Global Patterns

and Several Testable Hypotheses. The World Economy.

Bergstrand, J. H. (1985). The gravity equation in international trade: some microeconomic foundations and empirical evidence. The review of economics and statistics, 474-481. Blinder, A. S. (2009). How many US jobs might be offshorable?. World Economics, 10(2),

41.

Cardamone, P., & Scoppola, M. (2012). The impact of EU preferential trade agreements on foreign direct investment. The World Economy, 35(11), 1473-1501.

Dedrick, J., Kraemer, K. L., & Linden, G. (2009). Who profits from innovation in global value chains?: a study of the iPod and notebook PCs. Industrial and Corporate Change, dtp032.

Dietzenbacher, E., Los, B., Stehrer, R., Timmer, M., & De Vries, G. (2013). The construction of world input–output tables in the WIOD project. Economic Systems Research, 25(1), 71-98.

Feenstra, R. C., Markusen, J. R., & Rose, A. K. (2001). Using the gravity equation to

differentiate among alternative theories of trade. Canadian journal of economics, 430-447.

Frankel, J., Stein, E., & Wei, S. J. (1995). Trading blocs and the Americas: The natural, the unnatural, and the super-natural. Journal of development economics, 47(1), 61-95. Frankel, J., & Rose, A. (2002). An estimate of the effect of common currencies on trade and

income. Quarterly Journal of Economics, 437-466.

Ghosh, S., & Yamarik, S. (2004). Are regional trading arrangements trade creating?: An application of extreme bounds analysis. Journal of International Economics, 63(2), 369-395.

Gómez, E., & Milgram, J. (2010). Are estimation techniques neutral to estimate gravity equations? An application to the impact of EMU on third countries’ exports. mimeo. Gugler, P. (2015). World Investment Report 2014: Investing in the SDGs: An Action Plan. Head, K., & Mayer, T. (2013). Gravity equations: Workhorse, toolkit, and cookbook.

Hummels, D., Ishii, J., & Yi, K. M. (2001). The nature and growth of vertical specialization in world trade. Journal of international Economics, 54(1), 75-96.

Johnson, R. C. (2014). Five facts about value-added exports and implications for

macroeconomics and trade research. The Journal of Economic Perspectives, 119-142. Kohl, T. (2014). Do we really know that trade agreements increase trade?. Review of World

Economics, 150(3), 443-469.

Kohl, T., Brakman, S., & Garretsen, H. (2015). Do trade agreements stimulate international trade differently? Evidence from 296 trade agreements.

Krishna, P., Mansfield, E. D., & Mathis, J. H. (2012). World Trade Report 2011. The WTO and Preferential Trade Agreements: From Co-Existence to Coherence by World Trade Organization Geneva: World Trade Organization, 2011. World Trade Review, 11(02), 327-339.

Lawrence, R. Z. (1996). Regionalism, multilateralism, and deeper integration. Brookings Institution Press.

Los, B., Timmer, M. P., & Vries, G. J. (2015). How global are global value chains? a new approach to measure international fragmentation. Journal of Regional Science, 55(1), 66-92

Mayer, T., & Zignago, S. (2011). Notes on CEPII’s distances measures: The GeoDist database.

Do trade agreements stimulate participation in global value chains?