The Determinants of Service Exports
A Gravity Model Analysis Using Service Value-Added Data
University of Groningen
Faculty of Economics and Business
Research Paper for MSc Economic Development and Globalization
Name Student: Rik Sulman Student ID number: S3801284
Student email: h.sulman@student.rug.nl Date: 16-06-2020
Abstract
Key words: Gravity Model, Value-Added, Services Export
1
1. Introduction
Over the past decades global value chains have witnessed a further fragmentation of tasks. Services have proven to play a pivotal part in coordinating production chains. With the increased fragmentation Bohn et al. (2019) found that the export of value-added in services has grown faster than the export of value-added in goods. Services can be seen as glue that link the fragmented value chain together (Low, 2013). While not all services are directly linked to manufactured goods, in recent publications the growing importance of services in international trade of manufactured goods have been established (Drake-Brockman &, Stephenson, 2012). However, due to a lack of reliable data the role of services in these international production chains remain less understood (Low, 2013).
From the first phase of an international production chain until the last phase, all different phases add value. The sum of all value-added is called the total value-added (Qiu, 2014). This paper looks at the determinants of value-added exports of services. Services are less easy to be commoditized, but their intangible nature allows services to be exported without crossing physical borders (Low, 2013). Therefore it is hypothesized that geographical distance is less important for export in value-added services as it is for value-added goods. Administrative is hypothesized to be more important, since services are not easy to be commoditized and harder to ensure quality and enforce contracts. Cultural distance is expected to also be more important because in order to exchange the service, interaction between people is often needed.
This paper is structured as follows. Section 2 explains the growth in trade in services and its relation to manufacturing goods, through existing literature to conclude in thee hypothesizes. Section 3 describes the methodology and the data sources. Section 4 applies the approach described in section 3, discusses the results and answer the hypothesizes. Section 5 concludes, discusses some implications and offers suggestions for future research.
2. Literature Review
2
Fernandez-Stark (2011) as “the full range of activities that firms and workers perform to bring a product from its conception to end use and beyond” (Gereffi & Fernandez-Stark, 2011, p. 4). These value chains are made global due to the fact that these activities are spread over multiple countries, and sometimes even over multiple continents (Bernard et al., 2007; Morschett et al., 2010). GVCs are the phenomenon where production is broken into activities and tasks carried out in different countries. To coordinate these dispersed value chains, infrastructure services as telecom, internet, and more, are used (Baldwin 2013). Low (2013) refers to these services as the glue that holds the GVCs together.
This chapter will first explain the unbundling of trade. Thereafter recent empirical findings of the impact of GVCs will be looked into and discuss the intertwined relation between goods and services. Special attention will be given to the value-added across the production chain, and its change over time. Thereafter the paper will address determinants of trade found in empirical research. To conclude with the hypothesizes of this paper.
2.1. The Unbundling of Trade
Before GVCs were viable, production and consumption were tied together. Baldwin (2016) describes globalization as a progressive reversal of the forcible bundling of production with consumption. There are three costs that constrain the limit of separation between production and consumption; transportation costs, coordination costs, and the costs of moving people. Two advances in connective technologies have driven the unbundling of production and consumption forwards over the two periods in time; advances in transportation during the first unbundling, and transmission technologies in the second unbundling (Baldwin 2013, 2016).
2.1.1. The First Unbundling
Poor transportation technology was one of the reasons production and consumption have been bundled for many years. Steam technology began to change this in the 19th century;
3
The first unbundling resulted in factories clustering in industrial districts. Labor migration boomed and cities grew in number and size. While transportation costs got cheaper, the coordination costs and the cost of moving people remained less or not reduced (Baldwin, 2016). Economies of scale made it more profitable for industries to cluster, however today’s GVCs were not yet profitable due to the high coordination costs. Although there are exemptions, industries seemed to have clustered mainly in todays developed countries (Baldwin, 2013). Innovation, increased scale production and specialization gave these industrializing countries a cost-advantage over other countries’ industries, and resulted in a large divergence in international income (Pritchett, 1997).
2.1.2. The Second Unbundling
The second unbundling started in the mid-1980s. The combination of the wage differences together with the development of transmission technologies, such as the internet have revolutionized and further stimulated international trade (Bloom et al., 2012; Carballo et al., 2016). The innovation of ICT made it possible to coordinate complex tasks at distance, and therefore made it possible to further split up the GVCs (Baldwin, 2013).
While total trade-flows in intermediates kept growing, trade data from 2007 pointed out that the global supply chain was still very regional (Lopez González, 2012). Baldwin (2013) explain this empiric finding, by introducing regional hub-and-spoke dependencies of factory economies and headquarter economies. Whereby, headquarter economies are defined as economies whose export contain relatively little imported intermediaries, and factory economies as whose exports contain a large share of imported intermediates.
Baldwin (2008, 2013) points to three main factory regions. First factory region Baldwin (2008) points out is Asia whereby Japan is traditionally seen as the headquarter economy, however in recent years China is rapidly developing itself as headquarter economy. Second factory region is North America, whereby the USA is the headquarter economy. Third factory region is Europe, whereby Germany is the headquarter economy. To coordinate the dispersed production between headquarter and factory economies, infrastructure services as, but not limited to, telecom and internet are used (Baldwin 2013).
4
in the spatial coordination of the dispersed tasks of the GVCs (Baldwin, 2013; Timmer et al., 2019). Low (2013) refers to these services as the glue that holds GVCs together. Which is in line with recent findings that shows that services are often directly or indirectly embodied in exports (Drake-Brockman &, Stephenson, 2012; Heuser and Mattoo, 2017).
2.2. Manufacturing and Services Intertwined
Multinational Enterprises (MNEs) have had played a big part in the first and second unbundling and the increased trade in intermediate products. A large portion of international trade is in fact intrafirm trade. International intrafirm trade is trade between MNEs’ subsidiaries located in different countries (Bernard et al., 2010).
Nordas and Rouzet (2015) calculated that 75% of the Economic Co-operation and Development (OECD) countries’ gross domestic product (GDP) and around 80% of its employment, is accounted for by services. However, the precise role and impact of services in international trade and GVCs remain less clear, Low (2013) explains this is mainly caused by the lack of reliable data due to the intangible and heterogeneity nature of services, and the absence of a fully developed and commonly agreed product nomenclature. Drake-Brockman and Stephenson (2012) writes that traditional statistical measurement techniques overlook the value of services when these are embodied in other traded, and expect that embodied services alone account for an average growth of 25% as proportion of global merchandise exports.
International trade in services can be divided into two groups; firstly, services embedded in material substances. Secondly, capital or goods that absorb services and transform them in some way after they have crossed an international border (Grubel, 1987).
The first group consist of products produced by the service industry, that are intangible and can vary per experience. Therefore, these products cannot be easily commoditized and are often traded while bundled with goods produced by MNEs (Drake-Brockman & Stephenson, 2012). Few examples are: after sales service, transportation, training as well as the growing computer software industry (Qiu, 2014).
5
internet and satellites can be a substitute for physically crossing borders, and allows for digitally crossing of international borders. Few examples are: packaging, shelling, wholesale activities, or assembly services (Qiu, 2014).
2.2.1. Value-added in the Global Value Chain
As the name suggests, from the first phase of a GVC until the last phase, all phases add value. The sum of all value-added in the GVC is called the total value-added (Qiu, 2014). Not all phases add the same amount of value. After the second unbundling a clear change in the pattern of value-added can be seen in the GVCs.
Some countries have specialized in high-value adding and skill-intensive services as design, research and development, sales and marketing. While other countries have specialized in the commoditized and relative labor-intensive intermediate manufacturing stages. In recent years researchers have tried to determine the cause of the specialization in services (Bladwin, 2013). Through the use of the function specialization the relation between countries’ GDP and its function in the GVCs are found (Timmer et al., 2019).
During the second unbundling a large part of the labor-intensive production moved to countries with a comparative advantage in labor, while the high-value-added phases of the value chain seemed to have stayed behind. This empirical phenomenon is captured in the smile-curve, as can be seen in Figure 1.
Figure 1. The smile curve.
6
The steeper the curve in the smile-curve, the more unequal the distribution of value-added along the development process. In the case of the deepening smile-curve, this indicates that the manufacturing stages have lost relative share of value-added, while the initial and final stages have gained relative share. One explanation for this phenomenon can be that manufacturing tasks can often be relatively easy offshored and executed in countries where labor is cheap. Baldwin (2013) further explains that due to relative market power and mobile technology the smile has deepen since the 1980’s. Labor abundant countries are often eager to attract such tasks (Baldwin, 2013; Timmer et al., 2019).
Countries and its industries specialize in particular stages of the GVCs according to their comparative advantages, and are linked through trade in intermediate products (Timmer et al., 2019). Skill-intensive tasks are often more difficult to offshore and are kept at home where firms have relative high market power. Baldwin (2013), conclude that offshored tasks have become commoditized, and the onshore tasks have not. Important to note is that not all services are high-value-added and skill-intensive, like not all manufacturing work is labor-intensive.
2.2.2. Attraction of High-Value Adding Industries
7
On average MNEs often have their headquarters in developed countries where the governments are stable, contract enforcement is good, there is less governmental corruption, and it is easier to access capital (Spencer & Gomez, 2011). Empirical evidence also identified that countries where governmental stability is high, contracts are protected, and the availability of capital is stable, seem to be specialize in the early and final stages of the GVC’s, there where the relative share of the total value-added is high (Alfaro & Chauvin, 2017). In recent work Bohn et al. (2019) found that the importance of services has become greater in interregional trade between 2000 and 2014. In their findings Bohn et al. (2019) notices difference in trends between regions; trade in services have grown in relative size to trade in manufacturing products in all areas, however substantially faster in the OECD countries. Services have a growing impact on bilateral trade-flows for OECD countries, therefore it is important to understand the determinants of trade in services. So have Arnold et al. (2016) found that liberalization in the service industry could result in stimulated productivity in the manufacturing sector. Additionally, reduced trade restrictions for services complimented by domestic policies can lead countries to gain a comparative advantage in producing goods that depend on services, which are relative more often complex goods (Van der Marel, 2016). Bohn et al. (2019) argues that improved access to services raises competitiveness and productivity in the manufacturing industries and contribute to economic growth. Therefore they advocate too for countries to lift service barriers. Fully understanding the determinants will let each country develop its own policies to optimize their economic future.
2.3. Gravity and the Determinants of Trade
Distance has been proven to be an important determinant in trade, whereby the gravity model has been often used as base model for analyzing the determinants of bilateral trade since its introduction (Tinbergen, 1962). The gravity model is developed to estimate bilateral trade-flows based on economic size and distance between two countries or regions. With the right econometric applications the model have been witnessed as an empirical success in that it accurately predicts bilateral trade-flows (Head & Mayer, 2014).
8
multiple counting and the registration of the export to the correct industry (Kelly, 2012). The gravity model can be expressed as
𝑋𝑖𝑗 = 𝑀𝑖𝛽1𝑀𝑗𝛽2𝑑𝑗𝑖𝛽3, (1) where 𝑋𝑖𝑗 is the volume of trade between countries 𝑖 and 𝑗. 𝑀𝑖 is the mass of the country of origin, 𝑀𝑖 is the mass of the country of destination, and 𝑑𝑗𝑖 is the bilateral accessibility between country 𝑗 and country 𝑖. Therefore, 𝛽1 is the potential to generate trade-flows, 𝛽2
the potential to attract trade-flows, and 𝛽3 reflect the distance decay factor in trade.
Double or multiple counting can occur in a dataset when a producer purchase inputs to process and export them. In exporting countries, the value-added is equal to the value paid to the factors of production. Double counting happens when the exports include both the added value by processing and the costs of importing the intermediates or raw resources. When official trade statistics are measured in gross terms, they can double count the value of intermediate goods that cross international borders multiple times. Head and Mayer (2014) explain that when bilateral trade is not registered to its proper origin or final destination, the results of the gravity model can be biased.
2.3.1. Distance as Determinant
Like the Newton’s gravity equation, the basis of the economic gravity equation includes two masses and a distance measure. Countries’ GDP is often used as the two masses and geographical distance has proven a statistically significant determinant that negatively affect the size in bilateral trade (Ghemawat, 2001). The determinants found in literature are categorized in this paper in four distances following Ghemawat (2011); geographical distance, administrative distance, economic distance and cultural distance.
9
than just how far two countries are from each other: a country’s physical size, within-country distances to borders, access to ocean, topography, shared border, and even time zones can all be of influence (De Santis et al., 2015; Ghemawat, 2001).
Administrative distance include measures of governmental and institutional quality. Increased administrative distance can increase the (perceived) risk of trading, and can result in hold-up problems and diminishing trade-flows. Additionally, historical and political associations between countries such as colonial links, are proven to have a profoundly effect on bilateral trade patterns in some papers (Ghemawat, 2011).
Economic distances that can affect size in bilateral trade can include are membership of economic agreements, targets of economic sanctions, consumer wealth and income (Santos Silva and Tenreyro, 2006).
Culture can be defined as the collection of beliefs, values, and social norms that shape the behavior of individuals and organizations (Ghemawat, 2011; Kaufmann et al., 2004; North, 1990). This dimension can be difficult to quantify, and therefore existing databases are not without controversy.
2.4. Hypothesis
The second unbundling was characterized by further unbundling of production with consumption and the vertical specialization of countries in the GVCs (Lopez González, 2012; Baldwin, 2016). In and after this unbundling countries have taken different paths to develop their economies, some have joined GVCs while others did not. Some countries have specialized in manufacturing intermediates, while other countries focus on the early or final stages of the manufacturing process (Low, 2013). Service trade-flows are closely connected with the manufacturing flows, and their share in value-added created in total trade-flows have greatly increased across the world. The increase in share for OECD countries in the period between 2000 and 2014 has been the largest (Bohn et al., 2019).
10
between services and manufactured goods is difficult to entangle, however, is essential in order to make the right policy decisions. Understanding the difference between determinants of exports and services and goods can help countries find the optimal way of developing their economies.
This paper will add to the empiric search for the determinants of trade in services, by using the final demand perspective. This will be done through the use of the gravity model. Whereby the paper will focus on geographical, administrative, economic and cultural distances. Current literature result in the following hypotheses:
Hypothesis 1:
• Geographical distances is less important for value-added trade-flows in services than it is for trade-flows in manufactured goods.
International trade in services does not require to move an physical object across borders. Services can moved across border via internet or by the use of satellites. Bohn et al. (2019) not only found that trade in services has increased in interregional trade, but also that services travel further than manufactured goods. Since a large part of services are embedded in the trade in goods, geographical distance is expected still be significant for trade in services, but the impact smaller.
Hypothesis 2:
• Administrative Distance is more important for value-added trade-flows in services than it is for trade-flows in manufactured goods.
11 Hypothesis 3:
• Cultural Distance is more important for value-added trade-flows in services than it is for trade-flows in manufactured goods.
Goods are more easily commoditized than services. Services are intangible and the received experience can be different each time, for it often involves interaction between at least one people and/or software. For this reason it is hypothesized that cultural distance will have a bigger impact on trade in services as it has on trade in goods.
3. Methodology
This chapter will look at the basis of the gravity model and the different problems that can occur, and will discuss the solutions possible that fits the research in this paper. Thereafter, it will discuss the commonly used variables related to the hypothesizes of this paper and which variables will be used in this paper.
3.1. Gravity Model
International trade-flows can be predicted with an analogy to Newton’s law of universal gravitation. This gravity model hypothesizes that the gravitational force between countries’ mass is directly linked to the amount of trade and inversely proportional to their geographical distance (Burger et al., 2009). While the gravity model is successful in the explanation of aggregated trade, for the analysis of sector or product specific trade this traditional outline is too general. Therefore, the theoretical and econometric backing of the gravity model have developed into a structural form, including fixed effects for the exporter and the importer (Anderson & Wincoop, 2003). Hereby the mass often is calculated by the countries’ GDP (Head & Mayer, 2014).
12 𝑋𝑖𝑗 = 𝑌𝑖 Φ⏟𝑖 𝑀𝑖 𝑋𝑗 Ω⏟𝑗 𝑀𝑗 𝑑𝑗𝑖, (2)
Where 𝑌𝑖 = ∑ 𝑋𝑗 𝑖𝑗 is the value of production, 𝑋𝑗 = ∑ 𝑋𝑖 𝑖𝑗 is the value of the importer’s
expenditure on all source countries. Φ𝑖 and Ω𝑗 are multilateral resistance terms. Multilateral resistance terms refers to the barriers which both countries i and j face in trading with all their trading partners, domestic or internal trade including. These multilateral resistance terms can defined as
Φ𝑖 = [∑ (𝛽𝑗Ω𝑗 𝑡𝑗𝑖) 1−𝜗 𝑗 ] 1 1−𝜗 and Ω𝑗 = [∑ (𝛽𝑖Φ𝑖 𝑡𝑖𝑗) 1−𝜗 𝑖 ] 1 1−𝜗 , (3) where 𝛽𝑗1−𝜗 denotes the share of country 𝑗 in country 𝑖′𝑠 consumption. Trade resistance is assumed to be symmetric so that 𝑡𝑖𝑗 = 𝑡𝑗𝑖. 𝜗 is the elasticity of substitution between all goods. Φ𝑖 and Ω𝑗 are the multilateral resistance terms, and are also considered the be consumer price indices for country 𝑖 and country 𝑗, for each is the function of the country’s full set of barriers to trade between country i and j (Adam & Cobham, 2007; Anderson & Van Wincoop, 2003).
Equation 2 allows for a more complete calculation of impacts of trade costs changes, because the multilateral resistance terms can be solved for a given set of trade costs. Structural gravity can be used to estimate trade-flows on aggregate or industry level (Head & Mayer, 2014).
3.2. Underlying Model Difficulties
The gravity model is widely used, and different regression models have been developed to deal with several issues each with their own specific restrictions and assumptions. To identify the regression model best for this research the expected difficulties must be uncovered. Expected is that when econometric and methodologic issues arise, they will bias the estimated results. The expected issues for this research are the following:
13
Santos Silva and Tenreyro (2006) and Head and Mayer (2014) use different regression models to test certain situations to find the most efficient one. Both papers compared Poisson pseudo-maximum likelihood (PPML), Gamma pseudo-maximum likelihood (GPML), and OLS. How well the models deal with four issues will be discussed in the following sub-chapters, along with other methods how this paper can deal with these issues.
Following the advice of Head and Mayer (2014), which are in line with the WTO, worded by Yotov et al. (2016) and the conclusion of and Santos Silva and Tenreyro (2006), this paper will make use of the PPML model. In the following sub-chapters all issues will be discussed and explained why the use of the PPML model is justified.
3.2.1. Zero Valued Trade-flows
A log-normal model based on equation 1 cannot deal well with zero-valued trade-flows; one of the reasons, the logarithm of zero is undefined. Two ways to deal with this issue while using the log-normal model is to omit all zero-valued flows or arbitrarily insert a small positive number to all zero-valued trade-flows to ensure that the logarithm is well defined (Yotov et al., 2016). However, both options lead to their own biases. Removing all zero-valued trade-flows leads to a selection bias, while arbitrarily inserting positive numbers also lead to distorted results (Santos Silva & Tenreyro, 2006; Head & Mayer, 2014). The use of PPML is justified, for can deal with datasets containing zero-valued trade-flows, and its estimations remain efficient (Yotov et al., 2016).
3.2.2. Multilateral Trade Resistance
Bilateral trade between any country pair is affected by both trading partners’ interaction with the rest of the world. Multilateral trade resistance captures the third-country effect on trade between the country pair (Magerman et al., 2013).
14 3.2.3. Endogeneity of Trade Policy
Briefly mentioned earlier, it may be possible for two countries to be unable to trade with each other, while the gravity model in equation 2 would imply that based on their characteristics they should, but are not able to by an existing trade sanctions. These unobserved bilateral trade costs can lead to endogeneity in the dataset (Yotov et al., 2016).
This paper will make use of fixed effects of the country of origin and destination enables to control for unobservable characteristics that affect trade-flows, that vary over time for each country, Agnosteva et al. (2014) show that fixed pair-effects concerning trade agreements or sanctions, are a proper proxy for accounting for these hidden trade costs. PPML regression makes it possible to include year, country and country-pair fixed effects.
3.2.4. Heteroskedasticity
Head and Mayer (2014) have conducted a series of tests to see which model is least affected by heteroskedasticity in the gravity model. Head and Mayer (2014) uses the variance to mean ratio (VMR) to see whether difference of the estimated results and its expected values, are dispersed or clustered compared to a standard statistical model. The VMR is a normalized measure, and can be used to gain insight into the nature distribution of the data. The ratio is equal to zero in the case of a not dispersed, and constant random variable. Between zero and one the distribution is under dispersed, while above one the distribution is over dispersed. The ratio is equal to one in a Poisson distribution (Head & Mayer, 2014; Yotov et al., 2016). Following the advice of Santos Silva and Tenreyro (2006) and Head and Mayer (2014), this paper will use PPML regression, instead of the OLS model, to avoid significant biases in the estimations.
3.3. Regression Model
Combining the previous mentioned literature, the core of equation 2, and the goal of this paper the following lognormal equation has been created:
𝑻𝑖𝑗 = 𝛽0+ 𝛽1𝒅𝑗𝑖 + 𝛽2𝜸𝑖 + 𝛽3𝝋𝑗+ 𝛽4𝝎𝑗𝑖 + 𝜺𝑗𝑖, (4)
where 𝑇𝑖𝑗 is the bilateral value-added export between countries 𝑖 and 𝑗, 𝛽1 reflect the
15
of destination, 𝝎𝑖𝑗 is the pair-fixed effect of the country of origin and the country of
destination, and 𝜺𝑖𝑗 is a random error term which is assumed to be independent and
identically distributed. Furthermore, it is noteworthy to mention that this paper use bilateral exports as dependent variable, such that each country pair yields two observations, with each country either as exporter or importer. The data used for the bilateral value-added export 𝑻𝑖𝑗 will be explained in chapter 3.4, and the variables and data used for the distance vector
𝒅𝑖𝑗and the fixed effects 𝛾𝑖, 𝜑𝑗 and 𝝎𝑖𝑗 will be discussed in chapter 3.5.
The distance vector in equation 4 contains distance variables linked to the distances mentioned in previous chapter; geographical distance, administrative distance, economic distance and cultural distance. The fixed effects can either have a positive impact on the trade-flows when these effect close the distance between the two countries, or a negative effect when it increases the distance between them (Santos Silva & Tenreyro, 2006; Lankhuizen & Thissen, 2019).
3.4. Required Data
To be able to estimate the determinants of value-added trade in services, detailed data containing the bilateral trade-flows of services is needed. This dataset is preferably split up into different industries, so it will be possible to perform a detailed analysis. Furthermore, this data needs to be corrected for double or multi-counting.
3.4.1. World Input Output Tables
To avoid double or multiple counting Bohn et al. (2019) argue, that focusing on the value-added statistics instead of on the gross export statistics will help in making sure that inputs passing through multiple countries within GVCs are not double counted. Manufacturing industries depend heavily on domestic and foreign service industries, gross export statistics report often the directly traded services, but not the domestic services that can be embodied in the export of manufactured goods (Bohn, et al., 2019). For the same reason, Los et al. (2015) and Timmer et al. (2016) used value-added data in their analysis of the fragmentation of the GVC’s and international trade respectively.
16
WIOTs document trade-flows of industry intermediate and final outputs within an economy (Timmer et al., 2015). WIOTs enable to estimate trade interdependencies in global production, and the separation of trade between industries and countries (Bohn et al., 2019). Detailed WIOT tables are needed to gain a balanced and complete yearly overview of the trade-flow between each countries’ industries, and to be able to calculate the value-added trade exports.
This paper will look at the bilateral trade in services, which looks at which country exports how much trade for whom? Whereby the paper’s starting-point is the standardized WIOT in Table 1 with n countries, each with s industries.
In the text of this paper matrices will be indicated by being shown in bold capital letters, vectors in bold lower case letters, and scalars in italicized letters. A circumflex is used to indicate a diagonal matrix, and an apostrophe for transposition.
The ns x ns matrix D of intermediate products, the ns x n matrix F of final use, and the ns output vector x, are given by:
Table 1. WIOT with n countries
Intermediate use (S columns per country)
Final use (C columns per country)
Total 1 ⋯ n 1 ⋯ n S industries, country 1 ⋮ S industries, country n 𝑫11 ⋯ 𝑫1𝑛 ⋮ ⋱ ⋮ 𝑫𝑛1 ⋯ 𝑫𝑛𝑛 𝑭11 ⋯ 𝑭1𝑛 ⋮ ⋱ ⋮ 𝑭𝑛1 ⋯ 𝑭𝑛𝑛 𝒙1 ⋮ 𝒙𝑛 Value-added (𝒘1)′ ⋯ (𝒘𝑛)′ Output (𝒙1)′ ⋯ (𝒙𝑛)′
The intermediate matrix 𝐃𝑟𝑛, element 𝑑𝑖𝑗𝑟𝑛 gives the value of intermediate deliveries from industry i from country r to industry j in country n. The same logic can be applied for the final use matrix F, and the vector x. These values are given in money value. Input coefficients give the input per unit of receiving industry’s output, and can be calculated by 𝐀 = 𝐃𝐱̂−1, resulting in a ns x ns matrix A. In the same train of thought, the value-added coefficients can be calculated by 𝐯′= 𝐰𝒙̂−1, resulting in a ns vector containing the value-added per unit of its
17 𝐃 = [ 𝐃11 ⋯ 𝐃1⋯ ⋯ 𝐃1𝑛 ⋮ 𝐃⋯1 ⋮ ⋱ ⋮ ⋱ ⋯ 𝐃⋯⋯ ⋯ ⋱ ⋮ ⋱ ⋮ 𝐃⋯𝑛 ⋮ 𝐃𝑛1 ⋯ 𝐃𝒏⋯ ⋯ 𝐃𝑛𝑛] , 𝐱 = [ 𝐱1 ⋮ 𝐱⋯ ⋮ 𝐱𝑛] 𝐅 = [ 𝐅11 ⋯ 𝐅1⋯ ⋯ 𝐅1𝑛 ⋮ 𝐅⋯1 ⋮ ⋱ ⋮ ⋱ ⋯ 𝐅⋯⋯ ⋯ ⋱ ⋮ ⋱ ⋮ 𝐅⋯𝑛 ⋮ 𝐅𝑛1 ⋯ 𝐅𝒏⋯ ⋯ 𝐅𝑛𝑛] , 𝐰 = [ 𝐰1 ⋮ 𝐰⋯ ⋮ 𝐰𝑛]
To be able to calculate how much trade from country r is in the final use of country n, the Leontief inverse is needed. In this paper the Leontief inverse matrix is a ns x ns matrix 𝐋 = (𝐈 − 𝐀)−1, and given by:
𝐋 = [ 𝐋11 ⋯ 𝐋1⋯ ⋯ 𝐋1𝑛 ⋮ 𝐋⋯1 ⋮ ⋱ ⋮ ⋱ ⋯ 𝐋⋯⋯ ⋯ ⋱ ⋮ ⋱ ⋮ 𝐋⋯𝑛 ⋮ 𝐋𝑛1 ⋯ 𝐋𝒏⋯ ⋯ 𝐋𝑛𝑛]
Now for any country’s final use vector, the output in each country and how much value-added in involved can be calculated. Take for example the final use in country r, which can be indicated by 𝐟𝑡𝑟 (with t – 1, …, n). The production needed in country s that is needed to satisfy these final use is ∑𝑛𝑡=1𝐋𝑠𝑡𝐟𝑡𝑟. When premultiplying this production in country s with the value-added coefficients results in the vector ∑𝑛𝑡=1𝐯̂𝑠𝐋𝑠𝑡𝐟𝑡𝑟. Concluding, the total value-added generated in country s that is embodied in the final use in country s is 𝑉𝐴𝑋𝑠𝑟 = ∑𝑛 (𝐯𝑠)′𝐋𝑠𝑡𝐟𝑡𝑟
𝑡=1 . In other words, the value-added exports from country s to country r is 𝑉𝐴𝑋𝑠𝑟
(Bohn et al., 2019). This calculation enables to calculate the value-added trade-flows between countries.
3.4.1. Re-exports in Trade-flow Dataset
18
A potential problem that can occur using trade data is double counting, which is illustrated in Figure 2(a). The WIOD corrects these re-exports by leaving out the import flows of country B, as illustrated in Figure 2(b), eliminating the problem of double counting (Timmer et al., 2015). However, according to Lankhuizen and Thissen (2019) the correction of the WIOD according to Figure 2(b) is incorrect. While the total volumes of trade are being corrected, the re-exports are not reassigned to the right countries. Therefore, Lankhuizen and Thissen (2019) propose the correction shown in Figure 2(c).
Figure 2. Double counting, WIOD corrected, & Lankhuizen and Thissen (2019) correction
Note. Adapted from “The implications of re-exports for gravity equation estimation, Nafta and Brexit,” by R. Lankhuizen & M. Thissen, 2019, Spatial Economic Analysis, 14(4), p. 390. Copyright 2019 by the R. Lankhuizen & M. Thissen
Re-exports are goods that are imported and subsequently exported without having undergone any significant industrial processing (Lloyd & Sandilands, 1985). Correcting for re-exports is relevant for this research for two reasons. Firstly, to avoid biased estimations. When re-exports are registered to the incorrect country, it may blur bilateral trade patterns. Consequentially, when bilateral trade patterns are incorrect, estimations from the gravity model may be biased (Lankhuizen & Thissen, 2019). Secondly, it is important for a policy perspective. Re-exports may lead to misidentification of main trade-partners, and can lead to wrongly targeted trade policies (Lankhuizen & Thissen, 2019).
19 3.4.2. Trade-flow Correction
In order to correct the WIOD, Lankhuizen and Thissen (2019) advise a three step method to reassign the export to the correct country, as shown in Figure 2(c). Assuming that re-exports have similar country patterns as regular trade, the first step is to calculate the probabilities of the origin of imports and exports:
𝑷𝑰𝑝𝑖𝑗 = 𝑻𝑝 𝑖𝑗 ∑ 𝑻𝑖′ 𝑝𝑖′𝑗 (5) 𝑷𝑬𝑝𝑖𝑗 = 𝑻𝑝 𝑖𝑗 ∑ 𝑻𝑗′ 𝑝𝑖𝑗′ (6)
Whereby 𝑷𝑰𝑝𝑖𝑗 is the probabilities of the country of origin i of imports and 𝑷𝑬𝑝𝑖𝑗 is the probabilities of the destination j of exports. 𝑻𝑝𝑖𝑗 are the imports of product p by country j
coming from country i, excluding trade margins. The probabilities are needed to calculate the re-export table 𝑹𝑬𝑶𝑫𝑖,𝑞,𝑗,𝑝, describing the exports of product p coming from country i,
re-exported by country q and with final destination country j, in three steps;
𝑹𝑬𝑶𝑝𝑖𝑞 = 𝑷𝑰𝑝𝑖𝑞𝒓𝒆𝑝𝑞+ 𝒆̂𝑝𝑖𝑞 (7) 𝑹𝑬𝑫𝑝𝑞𝑗 = 𝑷𝑬𝑝𝑞𝑗𝒓𝒆𝑝𝑞+ 𝒆̂′𝑝 𝑖𝑞 (8) 𝑹𝑬𝑶𝑫𝑝𝑖𝑞𝑗 =𝑹𝑬𝑶𝑝 𝑖𝑞 𝒓𝒆𝑝𝑞 𝑹𝑬𝑫𝑝 𝑞𝑗 + 𝒆̂𝑝𝑖𝑞𝑗 (9)
Whereby 𝑹𝑬𝑶𝑝𝑖𝑞 describes the origin of the re-exports and 𝑹𝑬𝑫𝑝 𝑞𝑗
the destination of re-exports. 𝒓𝒆𝑝𝑞 are the re-exports by country q and product p, excluding trade margins. Equation
9 specifies that the origin and destination of re-exports are determined endogenously (Lankhuizen & Thissen, 2019).
In the third and final step, 𝑹𝑬𝑶𝑫𝑝𝑖𝑞𝑗 is used to determine the trade matrix 𝑻𝑹𝑬𝑝𝑖𝑗: 𝑻𝑹𝑬𝑝𝑖𝑗 = 𝑻𝑝𝑖𝑗 − 𝑹𝑬𝑫𝑝𝑞𝑗+ ∑ 𝑹𝑬𝑶𝑫𝑞 𝑝𝑖𝑞𝑗+ 𝒕̂𝑖,𝑗,𝑝
𝑎𝑑𝑗
(10)
20
that origin (Lankhuizen & Thissen, 2019). Together with the error terms of equation 7 to 9, the adjustment term 𝒕̂𝑖,𝑗,𝑝𝑎𝑑𝑗 are included in the minimized function Z:
𝒁 = ∑𝑖,𝑞,𝑝(𝒆̂𝑝𝑖𝑞)2+ ∑𝑞,𝑗,𝑝(𝒆̂′𝑝 𝑖𝑞)2+ ∑𝑖,𝑞,𝑗,𝑝(𝒆̂𝑖𝑞𝑗𝑝 )2+ ∑𝑖,𝑗,𝑝(𝒕̂𝑖,𝑗,𝑝𝑎𝑑𝑗)2 (11) Whereby the following constraints need to be met:
∑ 𝑹𝑬𝑶𝑞 𝑝𝑖𝑞 ≥ 𝒆𝒙𝑝𝑖 (12) ∑ 𝑹𝑬𝑫𝑞 𝑝𝑞𝑗≥ 𝒊𝒎𝑝𝑗 (13) 𝒆𝒙𝑝𝑖 ≥ ∑ 𝑻𝑹𝑬 𝑝 𝑖𝑗 𝑗 (14) 𝒊𝒎𝑝𝑗 ≥ ∑ 𝑻𝑹𝑬𝑖 𝑝𝑖𝑗 (15) 3.5. Variables
21
Table 2. Descriptive statistics of non-dummy variables
Variable Obs Mean Std. Dev. Min Max
1 Log. Distance 25284 7.987041 1.098922 4.087008 9.825874 2 Administrative Distance 25284 2.042187 2.411567 0.0029 15.0917 3 Cultural Distance Hofstede 7728 1.631437 1.017421 0.0687 4.976 4 Cultural Distance WVS 4998 1.552278 1.09835 0. 0404 8.1233 5 Log GDP o / d 25284 12.9952 1.708753 9.01824 16.6375 6 Corrected Total Exports Value-added (TEVA) 25284 2044.066 6754.417 0.07 168920.8 7 Corrected TEVA Services 25284 892.2169 2645.46 0.04 49080.15 8 Corrected TEVA Excluding Services 25284 1151.849 4303.805 0.02 125331 Note. Variable 1: Geographical Distance. Variable 2: Administrative Distance. Variables 3 & 4: Cultural Distance. Variable 5: Economic Distance. Variables 6 – 8: Dependent Variables.
Table 3. Descriptive statistics of dummy variables
Variable Obs. Mean Std. Dev.
9 Common Border 25284 0.060 0.237
10 Landlocked o / d 25284 0.139 0.347
11 Colony o / d 25284 0.004 0.066
12 Common Language 25284 0.217 0.412
13 Sanctions Threatened on Trade 23479 0.008 0.091
14 Sanctions Imposed on Trade 23479 0.017 0.127
15 Agreement PTA Goods 25284 0.556 0.497
16 Agreement PTA Services 25284 0.442 0.497
17 Custom Union (CU) 25284 0.353 0.478
18 Economic Integration Agreement (EIA) 25284 0.446 0.497
19 Free Trade Agreement (FTA) 25284 0.507 0.500
20 Partial Scope Agreement (PSA) 25284 0.024 0.154
21 Member EU o / d 25284 0.561 0.496
22 Joint Member EU 25284 0.322 0.467
23 Member WTO o / d 25284 0.977 0.151
24 Joint Member WTO 25284 0.954 0.210
25 Member GATT o / d 25284 0.814 0.389
26 Joint Member GATT 25284 0.659 0.474
22 3.5.1. Geographical Distance
Geographical distance includes the distance between countries, whereby often the straight distance between the countries’ capitals is used, and has often a negative statistically significant influence on bilateral trade-flows; when the distance between countries increases, the bilateral trade-flow declines (Head & Mayer, 2014; Lankhuizen & Thissen, 2019). However, the impact of distance does not need to be constant, so can the first kilometer have a different sized coefficient as for example the 6000th kilometer. There are two ways to
address this; taking the logarithm of the distance, or using adjacency dummies for countries which are connected by a border (Yotov et al., 2016). Additional geographical distance can occur when countries without access to a port and are locked in by other countries. Being landlocked have been found to be in a disadvantage when it comes to trade (Yotov et al., 2016).
In this paper geographical distance will be measured as the straight distance between countries, using their capitals as its center of gravity. The distance will be calculated by using the geographical coordinates provided by the CIA’s World Factbook (2020). Additionally, the CIA’s World Factbook (2020) provides data on which countries border each other, and if countries are landlocked. Data on shared borders and being landlocked will be used to generate dummy variables.
Country size, time zones and distance to border have not been included into this dataset. They are excluded because the WIOD dataset only contain production information on country level, and does not provide the exact location of production, exports and imports. Therefore, the between the countries’ capitals is used as proxy for the geographical distance between countries in kilometers. For example, several Chinese provinces are known for their specific expertise and exports (Jarreau & Pncet, 2010). However, this paper will use the Chinese capital as center of export and import. Some countries are located in multiple time zones, and are diverse in country size and distance to border, as can be seen in Figure 3. A full list of all countries included can be found in Appendix B. Using additional variables that all hinge on the proxy of geographical distance, can magnify the bias of the proxy.
23
Figure 3. Visualization of the Countries in the Dataset.
Note. Countries included in the dataset are highlighted in white, countries not included in the dataset are left out in blue.
3.5.2. Administrative Distance
The concept of administrative quality is difficult to quantify and therefore proxies are used. To measure administrative distance by De Groot et al., (2004) and Lankhuizen and Thissen (2019) used proxies that included rule of law, governmental accountability, quality of regulation, control of governmental corruption, government effectiveness and political stability. De Groot et al., (2004) found that administrative quality has a significant, positive and substantial impact on bilateral trade-flows. However, looking at the distance between administrative quality between importers and exporters have mixed results, in some papers they have been found statistically significant, while in others they have not (Lankhuizen & Thissen, 2019). Colonial ties is another cultural variable that can regularly be seen in gravity models. For countries who share colonial ties have a long history of bilateral trade, and sometimes share the same language (Ghemawat, 2011).
24
Unfortunately, for data on Rule of Law, Accountability, Quality of regulation, Control of Corruption, Government Effectiveness and Political Stability from the Worldbank (2019) are missing for 2001. The Administrative Distance in Table 2 is measured by using the data on the six dimensions of governance, and calculate the index developed by Kogut and Singh (1988):
𝐼𝐷 𝑖𝑗 = 1 6∑ (𝐼 𝑘𝑖−𝐼𝑘𝑗) 2 𝑉 𝑘 6 𝑘=1 (16) Where 𝐼
𝑘𝑖 indicate the ith country score on the kth dimension, and 𝑉𝑘 is the variance of this dimension across all countries.
The colony dummy in Table 3 for the destination country is 1 when it was ever a colony of the origin country. The other way around, the colony dummy for the origin country is 1 when it was ever a colony of the destination country.
3.5.3. Economic Distance
As mentioned, GDP is used as mass in the gravity model, and can be used as variable for both the exporters’ GDP or importers’ GDP. While often statistically significant, the size of the coefficient varies (Head & Mayer, 2014). Large countries, in terms of GDP, attract more trade. However, another way to attract trade can be through membership of trade organizations or sign trade agreements. Membership can reduce the MNEs’ risk to invest (Head & Mayer, 2014).
Economic data on the countries’ GDP in international dollars at purchasing power parity will be retrieved from the Worldbank (2019a). Whereby data on the membership of trade agreements will be retrieved from the Worldbank (2019b). Data on the membership of countries will be used to create dummy variables for both the exporter and importer of the services. In addition information on economic sanctions will be added from the United States International Trade Commission (Gurevich & Herman, 2018).
25
Regarding the economic variables in Table 3, 2014 data for sanctions threatened on trade and sanctions imposed on trade is missing. Agreement PTA is a dummy variable recording if the country pair is in at least one active preferential trade agreement, and is further split into Goods and Services. Furthermore, there are dummy recording if the country pair is in an agreement concerning a custom union (CU), economic integration agreement (EIA), free trade agreement (FTA) or partial scope agreement (PSA). Membership of the European Union (EU), World Trade Organization (WTO) and General Agreement on Tariffs and Trade (GATT) are split up into three dummies each, one if both countries are joint members of the trade organization, and the other two if the country of origin or destination is member.
3.5.4. Cultural Distance
A shared language can reduce the perceived distance between people. While not always found statistically significant, shared language is frequently included as a control variable (Ghemawat, 2011; Head & Mayer, 2019). Two databases used in this paper for measuring cultural distance are Hofstede and the World Value Survey (WVS).
The WVS is created, whereby for each time period of the WVS suggestions for questions are solicited by social scientists from all over the world. The questionnaires are distributed in the countries’ own language and the results can be found unadjusted and unaggregated (De Santis et al., 2015). Cyrus (2015) found similar results when looking at the effect of cultural distance for bilateral trade-flows for EU-member, when using the WVS as basis of cultural differences. Using Hofstede’s cultural dimensions Lankhuizen and de Groot (2016) found a non-linear significant effect, and suggest that cultural differences reduce trade only at higher levels of cultural distance.
26
et al., 2014). Cultural difference between countries based on Hofstede’s and WVS’s dimensions in Table 2 are calculated similarly as equation 16.
4. Determinants of Trade
Examining the patterns in the total exports of value-added (TEVA) trade-flows between 2000 and 2014 of the dataset, figure 3 visually shows the monetary and relative patterns. The figure is divided in three panels the distribution of the corrected TEVA for countries in de European Union. All three panels are split into two graphs; a graph indicating the TEVA trade-flows in current United States Dollars (USD) for the total dataset, when the exporter is member of the EU and the importer is not a member of the EU, when the exporter is not a member of the EU and the importer is a member of the EU, and when both the importer and exporter are members of the EU. Secondly, the right of that panel the percentages are taken of the flows when either the importer or exporter or both are members of the EU, of the TEVA trade-flows of the dataset. Figure 3 is split into three panels, showing the division for the TEVA trade-flows, TEVA trade-flows for services and the TEVA trade-flows excluding services.
Looking at Figure 3 two points stand out; firstly, the EU exports almost triple the amount of value-added in goods and services as it imports. More than half of the EU’s value-added in services is exported outside the EU, which have grown from 51.31% in 2000 to 63.97% in 2014. This is in line with the results of Bohn et al. (2019), who found that intraregional trade has become more important for trade in services. Around 40% of the EU’s value-added in services is exported within the EU, and the amount of value-added in services that has been imported by EU members have dropped from 24.04% in 2000 to 18.19% in 2014.
27 Figure 3. Trade-flows Between 2000 - 2014
Note. All patterns contain 5% intervals estimations. 4.1. Determinants of Trade Fixed Effects
28
Table 4. Country-Pair, Year and Trade Agreements Fixed Effects
(1) (2) (3) (4) (5) (6) Log Distance -0.7483*** -0.7294*** -0.7473*** -0.7270*** -0.5800*** -0.6641*** (0.0328) (0.0402) (0.0329) (0.0402) (0.0384) (0.0450) Log GDP Origin 0.8227*** 0.8260*** 0.8218*** 0.8202*** 0.8475*** 0.8149*** (0.0270) (0.0263) (0.0287) (0.0268) (0.0258) (0.0249) Log GDP Destination 0.8198*** 0.9342*** 0.7961*** 0.9281*** 0.8211*** 0.9257*** (0.0284) (0.0283) (0.0291) (0.0285) (0.0308) (0.0245) Common Language 0.3553*** 0.3361*** 0.3546*** 0.3387*** 0.3778*** 0.2510*** (0.0895) (0.0917) (0.0910) (0.0919) (0.0906) (0.0866) (Former) Colony of Destination 0.1292 -0.0026 0.1022 -0.0042 0.2099 0.1149 (0.3771) (0.3180) (0.3769) (0.3193) (0.3954) (0.3107) (Former) Colony of Origin 0.5579* 0.3536 0.5174 0.3506 0.6107* 0.4953 (0.3258) (0.3842) (0.3196) (0.3867) (0.3478) (0.3777) Common Border 0.1046 0.3249** 0.0942 0.3294** 0.0901 0.1871 (0.1287) (0.1439) (0.1284) (0.1442) (0.1138) (0.1250) Landlocked Origin 0.0188 -0.2041** 0.0113 -0.2155** 0.0003 -0.1592* (0.0932) (0.0849) (0.0938) (0.0859) (0.0898) (0.0833) Landlocked Destination 0.1070 0.0562 0.0511 0.0437 0.0323 0.1058 (0.0953) (0.1002) (0.0951) (0.1003) (0.0929) (0.0951) Constant -10.1290*** -11.7780*** -9.7273*** -11.6601*** -11.9722*** -12.1667*** (0.5032) (0.6067) (0.5206) (0.6131) (0.6209) (0.7048) Country-Pair Fixed Effects
Yes Yes Yes Yes Yes Yes
Year Fixed Effects No No Yes Yes Yes Yes
Trade Agreements No No No No Yes Yes
Observations 25,284 25,284 25,284 25,284 23,479 23,479
R-squared 0.7159 0.7684 0.7186 0.7710 0.7605 0.8238
Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Trade agreement effects includes the dummies: sanctions threatened on trade, sanctions imposed on trade, agreement PTA goods, agreement PTA services, CU, EIA, FTA and PSA. The log of distance coefficient stays negative and highly significant for the TEVA for only services and excluding services. However, the coefficient in model 5 is smaller than the coefficient in model 6, indicating that distance has a less pronounced effect on the TEVA for services. In contrast, the coefficient of a common language between importer and exporter is larger in model 5 as it is in model 6. In Appendix C the full table can be found, whereby all control variables’ coefficients can be seen.
29
Table 5. Trade Organization Membership Fixed Effects
(1) (2) (3) (4) Log Distance -0.6212*** -0.6221*** -0.6697*** -0.6705*** (0.0342) (0.0347) (0.0421) (0.0435) Log GDP Origin 0.8690*** 0.8688*** 0.7765*** 0.7760*** (0.0241) (0.0243) (0.0285) (0.0284) Log GDP Destination 0.8495*** 0.8492*** 0.9237*** 0.9233*** (0.0267) (0.0267) (0.0251) (0.0253) Common Language 0.3153*** 0.3096*** 0.2167** 0.2112** (0.0845) (0.0842) (0.0979) (0.0979) (Former) Colony of Destination 0.1820 0.1849 -0.0174 -0.0152 (0.4392) (0.4393) (0.2936) (0.2930) (Former) Colony of Origin 0.5207 0.5239 0.6284** 0.6333**
(0.4082) (0.4085) (0.3161) (0.3155) Common Border 0.1541 0.1495 0.1983 0.1931 (0.0964) (0.0971) (0.1306) (0.1326) Landlocked Origin -0.0045 -0.0066 -0.2146** -0.2180** (0.0924) (0.0926) (0.0900) (0.0902) Landlocked Destination 0.0325 0.0303 0.1081 0.1060 (0.0893) (0.0892) (0.0969) (0.0969) Member EU origin 0.2991*** 0.2887*** -0.3839*** -0.3990*** (0.0947) (0.0966) (0.1137) (0.1168) Member EU Destination 0.2484*** 0.2382*** -0.0474 -0.0592 (0.0858) (0.0877) (0.0977) (0.0995) Joint Membership EU 0.4786** 0.4843* (0.2221) (0.2606)
Member WTO origin -0.6472*** -0.7308*** -0.1690 -0.4490*
(0.1005) (0.2301) (0.1241) (0.2294) Member WTO Destination 0.8659*** 0.7816*** 0.7603*** 0.4840**
(0.1148) (0.2380) (0.1597) (0.2234)
Joint Membership WTO 0.0844 0.2823
(0.2627) (0.2787)
Member GATT origin 0.4521*** 0.3538* 0.0650 0.0127
(0.0977) (0.2097) (0.1128) (0.2523)
Member GATT Destination 0.1812* 0.0834 0.0863 0.0356
(0.1024) (0.1814) (0.1290) (0.2249)
Joint Membership GATT 0.1173 0.0680
(0.2116) (0.2676)
Constant -13.3692*** -13.1795*** -12.2508*** -11.9093***
(0.5692) (0.6579) (0.6969) (0.7776)
Country-Pair Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
Trade Agreements Yes Yes Yes Yes
Observations 23,479 23,479 23,479 23,479
R-squared 0.8056 0.8057 0.8279 0.8283
Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
4.1.1. Hypothesis 1: Geographical Distance
30
a cultural difference variable based on the WVS’ country scores, and model 5 and 6 include a cultural difference variable based on the dimensions of Hofstede.
Table 6. Administrative and Cultural Distance
(1) (2) (3) (4) (5) (6) Log Distance -0.6000*** -0.6671*** -0.5458*** -0.6620*** -0.5747*** -0.6330*** (0.0351) (0.0434) (0.0605) (0.0647) (0.0416) (0.0548) Administrative Distance -0.0505*** -0.0080 -0.0868*** -0.0241 -0.0639*** -0.0160 (0.0140) (0.0150) (0.0247) (0.0238) (0.0199) (0.0199) Cultural Distance WVS 0.0173 0.0139 (0.0279) (0.0383) Cultural Distance Hofstede 0.0401** -0.0065 (0.0156) (0.0204) Log GDP Origin 0.8686*** 0.7766*** 0.8742*** 0.7449*** 0.8859*** 0.7775*** (0.0235) (0.0285) (0.0358) (0.0366) (0.0317) (0.0333) Log GDP Destination 0.8502*** 0.9235*** 0.7658*** 0.8557*** 0.8365*** 0.9357*** (0.0258) (0.0254) (0.0360) (0.0374) (0.0331) (0.0325) Common Language 0.3444*** 0.2180** 0.2838** 0.0527 0.2348 -0.0213 (0.0811) (0.0934) (0.1280) (0.1130) (0.1446) (0.1125) (Former) Colony of Destination 0.2001 -0.0208 -0.4163 -0.1917 0.6566*** 0.3263 (0.3691) (0.2839) (0.3448) (0.5090) (0.2115) (0.2116) (Former) Colony of Origin 0.5357 0.6290** 0.6034* 0.8078*** 0.3871 0.6582** (0.3324) (0.3025) (0.3541) (0.2834) (0.3643) (0.2565) Common Border 0.1142 0.1853 0.3020* 0.3582** 0.2596** 0.3731*** (0.1013) (0.1311) (0.1549) (0.1755) (0.1015) (0.1296) Landlocked Origin 0.0077 -0.2144** 0.0746 -0.3709*** -0.1486 -0.3758*** (0.0902) (0.0902) (0.1346) (0.1218) (0.0944) (0.1235) Landlocked Destination 0.0481 0.1091 -0.1248 -0.2495*** -0.1332 -0.0414 (0.0865) (0.0971) (0.1384) (0.0947) (0.1053) (0.1019) Constant -13.3514*** -11.9469*** -13.1372*** -13.2200*** -13.9486*** -12.7848*** (0.6617) (0.7814) (0.8744) (0.8279) (0.7466) (0.9587) Country-Pair Fixed Effects
Yes Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes Yes
Trade Agreements Yes Yes Yes Yes Yes Yes
Trade Organization Membership
Yes Yes Yes Yes Yes Yes
Observations 25,284 25,284 4,646 4,646 6,944 6,944
R-squared 0.8075 0.8264 0.7566 0.7988 0.8463 0.8940
Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
31 4.1.2. Hypothesis 2: Administrative Distance
Hypothesis 2 hypothesized that administrative distance would be of greater importance for the TEVA for services relative to the TEVA excluding services. As hypothesized, administrative distance has a profound, although limited, negative effect on the TEVA of services, as can be seen in Table 6. An increase in administrative distance predicts a decrease in export of value-added in services. Even when controlling for cultural distance in model 3 till 6, administrative distance remains negative and highly statistically significant. In contrast, in the models for TEVA excluding services administrative distance is not statistically significant, and therefore confirms the hypothesis.
4.1.3. Hypothesis 3: Cultural Distance
Hypothesis 3 hypothesized that cultural distance would be of greater importance for TEVA of services. Like Lankhuizen and de Groot (2016), this paper found an statistically significant variable in model 5 and 6. Following the hypothesis, the coefficient was expected to be negative, however, model 5 suggests a positive relation between cultural difference and TEVA for services. Lankhuizen and de Groot’s (2016) research also resulted in a positive coefficient, suggest that cultural differences reduce trade only at higher levels of cultural distance. Furthermore, cultural difference measured in model 3 and 4 does not result in statistically significant estimations for the WVS cultural difference coefficient.
Possible explanations of the unexpected sign and insignificant coefficients are: The loss of data and potential bias in the models containing cultural distance variables; only 18.38% of the bilateral trade-flows found a match in the WVS dataset, and 24.46% found a match in the dataset of Hofstede. Additionally, a selection bias can also result in unexpected sign and insignificant coefficients, the dataset does not include any African nor Middle-Eastern countries, nor has it an equal spread across the world for the countries included in the dataset, as can be seen in figure 3.
4.1.
Impact of Correction for Re-exports32
Table 7. Impact of Re-export Correction on Determinants of TEVA for services
(1) (2) (3) (4) (5) (6) Log Distance -0.6068*** -0.6000*** -0.5626*** -0.5458*** -0.5828*** -0.5747*** (0.0353) (0.0351) (0.0582) (0.0605) (0.0413) (0.0416) Administrative Distance -0.0519*** -0.0505*** -0.0861*** -0.0868*** -0.0658*** -0.0639*** (0.0140) (0.0140) (0.0246) (0.0247) (0.0200) (0.0199) Cultural Distance WVS 0.0190 0.0173 (0.0273) (0.0279) Cultural Distance Hofstede 0.0437*** 0.0401** (0.0169) (0.0156) Log GDP Origin 0.8737*** 0.8686*** 0.8737*** 0.8742*** 0.8919*** 0.8859*** (0.0236) (0.0235) (0.0236) (0.0358) (0.0319) (0.0317) Log GDP Destination 0.8350*** 0.8502*** 0.8350*** 0.7658*** 0.8271*** 0.8365*** (0.0254) (0.0258) (0.0254) (0.0360) (0.0326) (0.0331) Common Language 0.3305*** 0.3444*** 0.3305*** 0.2838** 0.2157 0.2348 (0.0805) (0.0811) (0.0805) (0.1280) (0.1440) (0.1446) (Former) Colony of Destination 0.1714 0.2001 0.1714 -0.4163 0.6224*** 0.6566*** (0.3674) (0.3691) (0.3674) (0.3448) (0.2032) (0.2115) (Former) Colony of Origin 0.5019 0.5357 0.5019 0.6034* 0.3654 0.3871 (0.3218) (0.3324) (0.3218) (0.3541) (0.3504) (0.3643) Common Border 0.1070 0.1142 0.1070 0.3020* 0.2548*** 0.2596** (0.0990) (0.1013) (0.0990) (0.1549) (0.0972) (0.1015) Landlocked Origin 0.0094 0.0077 0.0094 0.0746 -0.1453 -0.1486 (0.0893) (0.0902) (0.0893) (0.1346) (0.0945) (0.0944) Landlocked Destination 0.0042 0.0481 0.0042 -0.1248 -0.1702 -0.1332 (0.0850) (0.0865) (0.0850) (0.1384) (0.1037) (0.1053) Constant -13.1800*** -13.3514*** -13.0145*** -13.1372*** -13.8228*** -13.9486*** (0.6498) (0.6617) (0.8570) (0.8744) (0.7334) (0.7466) Country-Pair Fixed Effects
Yes Yes Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes Yes Yes
Trade Agreements Yes Yes Yes Yes Yes Yes
Trade Organization Membership
Yes Yes Yes Yes Yes Yes
Observations 23,479 23,479 4,646 4,646 6,944 6,944
R-squared 0.8075 0.8075 0.7622 0.7566 0.8454 0.8940
Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
Table 7 indicate that correcting for re-exports have little effect on the determinants of TEVA of services. While the theory expects the effects to be limited due to the their intangible nature, the effect is smaller than anticipated.
4.2. Administrative Distance Before and After 2009
In Figure 3 a drop in TEVA flows can be witnessed after 2009. This drop is a result of the global financial crisis that started with collapse of the subprime mortgage market in the United States of America in 2007. After the undoing of Lehmann Brothers in 2008, the crisis led to a serious recession around the world (Hein & Detzer, 2015).
33
the regions around Germany, like parts of Poland, Slovakia and Czech Republic. Crescenzi, Luca and Milio conclude that peripheral regions, like most of Spain, Greece and Ireland, have been hit harder than the factory economies. Table 8 shows how this is reflected in the determinants of TEVA for services in model 1 and 3, and in the determinants of TEVA excluding services in model 2 and 4. Model 1 and 2 show the determinants for 2000-2009 and model 3 and 4 for 2010-2014.
Table 8. Determinants Before and After 2009
(1) (2) (3) (4) Log Distance -0.5719*** -0.6633*** -0.6244*** -0.6593*** (0.0417) (0.0459) (0.0347) (0.0478) Administrative Distance -0.0608*** -0.0155 -0.0369** 0.0008 (0.0132) (0.0151) (0.0158) (0.0163) Log GDP Origin 0.8931*** 0.8068*** 0.8474*** 0.7466*** (0.0231) (0.0269) (0.0276) (0.0335) Log GDP Destination 0.8704*** 0.9465*** 0.8347*** 0.9058*** (0.0264) (0.0259) (0.0273) (0.0271) Common Language 0.3452*** 0.1956** 0.3582*** 0.2700** (0.0768) (0.0839) (0.0935) (0.1161) (Former) Colony of Destination 0.3054 0.0314 0.0533 -0.0855 (0.3448) (0.2644) (0.3939) (0.3153) (Former) Colony of Origin 0.5880* 0.7435*** 0.4648 0.4383
(0.3066) (0.2762) (0.3655) (0.3276) Common Border 0.1354 0.2098* 0.0801 0.1418 (0.0971) (0.1220) (0.1169) (0.1480) Landlocked Origin 0.0604 -0.1954** -0.0263 -0.2119** (0.0849) (0.0908) (0.1023) (0.0986) Landlocked Destination 0.0851 0.1420 0.0350 0.0905 (0.0877) (0.1009) (0.0896) (0.0957) Constant -14.2458*** -12.5997*** -13.1272*** -11.5147*** (0.6404) (0.7357) (0.7529) (0.8653)
Country-Pair Fixed Effects Yes Yes Yes Yes
Year Fixed Effects Yes Yes Yes Yes
Trade Agreements Yes Yes Yes Yes
Trade Organization Membership Yes Yes Yes Yes
Observations 16,254 16,254 7,225 7,225
R-squared 0.8290 0.8439 0.7783 0.8095
Note: Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1
34
become more determinant. Overall the models have lost explanatory value, shown in the smaller value of R-squared, suggesting that there is may be an excluded variable that has grown in importance after the crisis hit.
5. Conclusion
By using the gravity model, this paper has tried to add to the empiric search for the determinants of trade in services. This paper has looked at the value-added and final demand perspective. Whereby the paper has focused on three hypothesizes concerning geographical, administrative, and cultural distances.
Geographical distance as determinant still matters for the export of value-added of services. There is some evidence that geographical distance is of lesser importance for the export of value-added of service than it is for other value-added exports, however, geographical distance remained one of the largest determinants for the export of value-added of service. A possible explanation can be that this paper did not separate the exports of embedded services from separate services. Since the export of embedded services are heavily dependent on other goods, this might result in an overestimation of the effect of geographical distance. Administrative distance have been proven to be of an important determinant for the export of value-added of service. This paper provide statistical evidence that administrative distance is more important for the export of value-added of service than it is for other value-added exports, however the size of its corresponding coefficient remained limited.
Cultural distance has been difficult to determine. Using the dataset of the WVS the determinants were statistically insignificant, and using the Hofstede dataset the results were statistically significant but the wrong sign. This can be the result that cultural differences reduce trade only at higher levels of cultural distance (Lankhuizen & de Groot, 2016). A possible explanation can be found when looked at the countries included in the paper’s dataset. Since the spread of countries in the dataset is not equally spread, this might result in a selection bias.
35
determinants remained large and statistically significant, indicating that embedded services might influencing the estimations, the impact of the correction was less than expected. Looking at the difference between determinants before and after the financial crisis in 2009, another surprising results came to light; administrative distance has lost strength. Geographical distance and a common language have become more important, indicating that international trade have become more regionalized after 2009.
36
6. References
Acemoglu, D., Johnson, S., & Robinson, J. (2002). Reversal of fortune: Geography and
institutions in the making of the modern world income distribution. The Quarterly Journal of Economics, 117(4), 1231-1294.
Adam, C. & Cobham, D. (2007). Modelling multilateral trade resistance in a gravity model with exchange rate regimes. Centre for Dynamic Macroeconomic Analysis, CDMA Conference Paper Series.
Alfaro, L., Chauvin, J., (2017). Foreign Direct Investment, Finance, and Economic
Development. Chapter for the Encyclopedia of International Economics and Global Trade, Forthcoming. Retrieved from: https://ssrn.com/abstract=2908440
Agnosteva, D., Anderson, J., Yotov, Y., & National Bureau of Economic Research.
(2014). Intra-national trade costs : Measurement and aggregation (Nber working paper series, no. 19872). Cambridge, Mass.: National Bureau of Economic Research. (2014). Anderson, J., & Van Wincoop, E. (2003). Gravity with gravitas: A solution to the border
puzzle. American Economic Review, 93(1), 170-192. Retrieved from:
http://doi.org/10.1257/000282803321455214
Arnold, J. M., Javorcik, B. S., Lipscomb, M., & Mattoo, A. (2016). Services reform and manufacturing performance: Evidence from India. Economic Journal, 126(590), 1–39. Retrieved from: https://doi.org/10.1111/ecoj.12206
Baccini, L., & Dür, A. (2018). Global Value Chains and Product Differentiation: Changing the Politics of Trade. Global Policy 9(2), 49-57. Retrieved from:
http://dx.doi.org/10.1111/1758-5899.12614
Baldwin, R. (2008). Managing the Noodle Bowl: The Fragility of East Asian Regionalism. Singapore Economic Review, 53(3), 449-478. Retrieved from:
37
Baldwin, R. (2013). Global Supply Chains: Why They Emerged, Why They Matter, and Where they are Going. In Elms, D. K., & Low, P. (Ed.), Global value chains in a changing world (pp. 13-59). Geneva: WTO Publications.
Baldwin, R. (2016). The great convergence. Cumberland: Harvard University Press. (2016). Baldwin, R., Taglioni, D., & National Bureau of Economic Research. (2011). Gravity chains: Estimating bilateral trade-flows when parts and components trade is important (Nber working paper series, no. w16672). Cambridge, Mass.: National Bureau of Economic Research. (2011). Retrieved from: https://doi.org/10.3386/w16672
Baskerville, R. F. (2003). Hofstede never studied culture. Accounting, Organizations and Society, 28(1), 1–14. https://www.doi.org/10.1016/S0361-3682(01)00048-4
Bernard, A. B., Jensen, J.B., Redding, & S. J., Schott, P. K. (2007). Firms in International Trade. Journal of Economic Perspectives, 21(3), 105-130
Bernard, A., Jensen, J., Redding, S., & Schott, P. (2010). Intrafirm trade and product contractibility. The American Economic Review, 100(2), 444-448.
Bernhofen, D., El-Sahli, Z., & Kneller, R. (2016). Estimating the effects of the container revolution on world trade. Journal of International Economics, 98, 36-50. Retrieved from:
http://dx.doi.org/10.1016/j.jinteco.2015.09.001
Bloom, N., Sadun, R., & Van Reenen, J. (2012). Americans Do IT Better: US Multinationals and the Productivity Miracle. American Economic Review, 102(1), 167–201. Retrieved from: http://dx.doi.org/10.1257/aer.102.1.167
Bohn, T. (2019). Global trade in services, jobs, and incomes. [Groningen]: University of Groningen, SOM research school. Retrieved from:
https://doi.org/10.33612/diss.104863895
