Determinants of International Product Fragmentation: Explaining the Recent Decline

(1)

MASTER THESIS

MSc International Economics & Business

Determinants of International Product

Fragmentation: Explaining the Recent Decline

Lieke van de Straat

S2750864

l.m.van.de.straat@student.rug.nl

Supervisor:

Dr. A.A. Erumban

Co-Assessor:

Dr. T. Kohl

Faculty of Economics and Business

University of Groningen

(2)

Abstract:

This paper strives to explain the determinants of product fragmentation and uses these to explain the recent decline in fragmentation. The increasing trade integration in the period after the second unbundling has led to the fragmentation of the production process at multiple stages across many countries. These developments cannot be fully explained by traditional trade theories. This paper will analyse the effect of labour costs, trade costs and ICT on the fragmentation process.

(3)

1. INTRODUCTION

The nature of international trade has shown a significant change in the past decades, mainly caused by globalisation and trade (Gereffi et al., 2005). These developments are fuelled by changes on the global political stage such as the election of President Trump in the United States and the consequences of Brexit. While the nature of trade has changed significantly, international trade integration has been booming, shown by a sharp increase in the last 20 years. Some micro-evidence of this is that there has been a large shift of trade activity in the world from mainly trade among exports in the North to a near even distribution of trade between the North and the South.1_{This trade integration is shown in Figure 1, where in 1990, 27.5 percent}

of total exports originated in developing countries. In the 25 years following 1990, developing countries have increased their proportion of global exports, bringing the international exchange of goods between developed and developing countries to near even in 2015.

Figure 1 - Total Merchandise Exports (in Trillion US$) Source: UNCTAD, Merchandise: Total Trade

This shift in the trade patterns can be explained by using Global Value Chains (GVC). Implementing GVCs allows companies based in developed countries to fragment their production process by outsourcing stages of the production process to developing countries. This leads to a large trade in intermediate products, supported by an OECD study in 2009, which found that 56 percent of trade in goods contain of intermediate products, meaning more than half of the trade flow is of products used in further production (Miroudot et al., 2009). The concept of GVC, as introduced by Porter (1985), describes how all stages of the production process work together to create a final product, yet enables individual parties to make profit

72.5% _70.0% 65.8% 60.4% 53.9% _52.2% 27.5% 30.0% 34.2% 39.6% 46.1% 47.8% 0% 10% 20% 30% 40% 50% 60% 70% 80% 0 1 2 3 4 5 6 7 8 9 10 1990 1995 2000 2005 2010 2015 in T ril lio n

Developed economies Developing economies

(5)

(Tinta, 2017). Fragmentation is determined by splitting up tasks in a production process, where different stages are performed by suppliers in different countries (Timmer et al., 2013). Baldwin (2016) gives three causes for the acceleration of international fragmentation of trade at the beginning of the 21st century, which he coined the ‘second unbundling’.2 Firstly, there was a decrease in trade costs. This can be attributed to the increased availability of cheaper and better transportation, allowing international trade to be more globally dispersed. Secondly, communication across borders improved, creating a smoother production process and, consequently, greater fragmentation worldwide. This lead to a shift from trading in final products, to trading in parts and components. Lastly, Baldwin established the importance of emerging economies in international trade being included in global trade frameworks, such as the World Trade Organisation (WTO) (Erumban & Lundh, 2018). One of the main reason for outsourcing parts of the production process is to profit from lower wages in developing countries. According to Baldwin and Evenett (2013), given that wages are lower in developing countries compared to developed countries, the ICT revolution made the fragmentation of the production process in stages possible, whilst the wage gap between the developing and developed countries (North and South) made it profitable.

Figure 2 - Share of Product Fragmentation in the Manufacturing Sector Source: Own calculation using the WIOD database (see equation 5)

(6)

Figure 2 shows the trend of product fragmentation from 2000 to 2014. Here, the fragmentation increases during the 2003 to 2008 and 2009 to 2012 periods are particularly strongly, with an increase rate around 5 percent. There is a clear fall of almost 3 percent in 2009, which is attributed to the crisis. From 2012 onwards, the percent of product fragmentation has decreased slightly. The recent trends in international trade, characterised by a decline in the foreign share of domestic production, can be seen as a change in the rapidly increasing GVC participation at the beginning of this century. The question is, what has happened to the factors that Baldwin (2016) advocates to have caused the second unbundling and can they be used to explain this recent decline in fragmentation?

The decrease in trade costs previously discussed was mainly caused by the developments in cheaper transportation. While there are still incremental improvements to the transportation systems, the radical innovations of the previous decades are over. The same goes for ICT, even though a larger part of the globe now has access to the internet and there have been some innovations such as cloud computing and video conferencing. These do not render as important as the disruptive ICT technology that caused the second unbundling and which is still widely used by companies (Erumban & Lundh, 2018). Lastly, the improvement of international trade laws has become less momentous, since protectionism is on the rise. This is evident by the recent election of President Trump, who mandated an import tariff of 25 percent on steel and a 10 percent on aluminium (the Economist, 2018). Another example of protectionism is the recent Brexit. In which direction new innovations at the technological level will go is uncertain. On the one hand, it might fragment the production process even further, characterised by Baldwin (2016) as the third unbundling in which services from one country are performed in another country. On the other hand, manufacturing labour may disappear to make way for partly or fully automated production processes which will reduce or eliminate wage costs. In order to explain the recent decline in product fragmentation, we first need to establish an empirical relationship to determine which factors influence fragmentation. When these determinants are identified, we expect to explain the recent decline in fragmentation by looking at these factors and what happened to them during the recent decline in fragmentation. The focus here lies on the European economies.

Following the previous argumentation, the aim of this study is to conduct an analysis based on the following question ‘What are the determinants of product fragmentation and consequently,

(7)

There is little empirical evidence on the determinants of fragmentation, with current literature focussing on the effects of trade costs on fragmentation instead. Moreover, there is little empirical research on which factors are causing the recent decline in fragmentation. Hence, this will be the main focus of this study. An important aspect that is commonly discussed in the fragmentation literature is the impact globalisation has on trade costs and labour costs. However, it is unknown whether these determinants changed in their importance for fragmentation, or changed its direction causing a decline in product fragmentation.

This study analyses the effects of wages, trade cost, specifically tariff and non-tariff barriers, and ICT on fragmentation, using two different models. The results suggest that tariff and non-tariff barriers and wages are most important. Tariff and non-non-tariff barriers are negatively correlated with fragmentation, meaning an increase in barriers to trade causes fragmentation to decline. Wages in particular explain a large part of the decline in fragmentation after the financial crisis, because wages are positively correlated with fragmentation. The wage gap between developing and developed countries is narrowing, meaning it is becoming less ‘cheap’ to outsource production, causing a decline in the fragmentation process. Lastly, ICT has a positive relationship on product fragmentation. ICT assisted the acceleration of fragmentation, however, the nature of technology and its impacts, change with time. ICT is increasingly shifting to the use of robotics, which offsets the previously advantageous ‘cheap labour’ in developing countries, given that full automation allows for reducing the number of employees. So, the arguments Baldwin (2016) uses to explain the second unbundling at the beginning of the 21st century, can also be argued to be influential in the decline of fragmentation 10 years later.

(8)

2. LITERATURE REVIEW

2.1 International Trade Theories

Trade patterns have been changing especially from 1990 onwards, questioning the continuity of the traditional trade theories. According to Baldwin (2016), the second unbundling is characterised by a decrease in trade costs and a decrease in ICT costs, changing the nature of international trade from trade in final products to trade in intermediate products. However, before the occurrence of fragmentation, trade was mainly thought of in terms of Ricardo’s comparative advantage, where trade occurs because of differences in productivity. Or the Hecksher-Ohlin model, which explains trade based on differences in factor endowments (Kierzkowski & Jones, 2000). Due to fragmentation, the applicability of these traditional trade theories of comparative advantage and factor endowments are taken into question (Timmer at al., 2013).

The Ricardian theory of comparative advantage states a country should specialise in the product it can produce relatively better than others, which means produce and export products at low opportunity costs, while importing products involving high opportunity costs. This model is based on final products. Given that final products are not a representative factor anymore, the theory cannot be used as such. Baldwin et al. (2006) argues that there are three reasons comparative advantage trade theory has become less reliable. First of all, the unpredictability due to fragmentation, in which the ‘losers’ and ‘winners’ from globalisation are much harder to predict. Predicting what stage of the production process will lose its competitiveness with the result of production being offshored is difficult, given there is currently no understanding among economists about what exactly holds different production stages together in the first place. This knowledge is necessary in order to predict where it will fragment (Baldwin & Evenett, 2016). Secondly, the fragmentation process is sudden, with previously secure jobs being offshored to developing countries within a short period of time. ‘Situations where a gradual change in underlying conditions causes no visible effect right up to a threshold beyond which a massive reaction such as offshoring occurs’. This tipping point over the threshold cannot be predicted yet (Baldwin & Evenett, 2016). Lastly, individuality is an important factor. The competition is not among sectors anymore, but among individual tasks and intermediate products.

(9)

mainly manufacturing, were offshored, while the high-value added stages, research and development and marketing, were kept at home. The result was that the offshoring stages became commoditised (Baldwin & Evenett, 2013). Baldone et al. (2007) argue that, despite the difficulties, comparative advantage can still be determined using intermediate goods even though the comparison with autarky is pointless, since the intermediate good would not have been produced. However, due to the second unbundling the use of comparative advantage is less valuable, as it does not determine policy reactions resulting from globalisation affecting the economy stage-by-stage (Baldwin & Evenett, 2013).

Another main conventional trade theory that is potentially decreasing in relevance is the Hechsher-Ohlin trade theory. This model is based on factor endowments. A country specialises in the product using its abundant factor endowments, while importing a product produced by the scarce factor endowment. The broad predictions of this theory will still hold in our context. A country will focus on activities in which local value added content is intensively used in their relatively abundant factors (Timmer et al., 2013). An example of this is the way China was used in the beginning of the 21st century. During the boom of fragmentation, ‘cheap labour’ was the country’s abundant factor.

Although the broad theory of comparative advantage and factor endowments still holds, using the theories as a policy guide is less useful, given that stages of production have become the factor influencing globalisation and not sectors. This means that trade policies should focus on specific tasks instead of sectors, while taking into consideration patterns of vertical integration of production (Timmer et al., 2013). None of these fundamental models discussed consider trade in intermediate goods or reasons for the occurrence of fragmentation. An implication of this is that classifying the winners and losers from globalisation, based on either the skill group they belong to or the sector they work in, will be less useful (Baldwin, 2006).

2.2 Determinants of International Trade

2.2.1 Trade Costs, Tariff and Non-Tariff Barriers

(10)

In 2014, the ESCAP World Bank Trade Cost Database estimated that between 0 and 10 percent of the trade costs are tariffs, while 10 to 30 percent are natural trade costs, based on location, history or culture. The remaining portion of the trade costs is due to non-tariff measures, which are mainly determined by business environment, currency fluctuations, and barriers (UNESCAP, 2014; OECD, 2015). Trade costs declined in all major economies between the years 1995 and 2009. This is attributed to negotiations, but mainly to tariffs and border tax measures for regulating imports, which have less of an effect on the GVC (OECD, 2014). It is important to note that non-tariff measures can influence GVC participation, as they have an effect on the efficiency of the operation.

Trade costs have been increasing recently because of the increase in protectionism. The main trade distortions implemented are subsidies, trade defence, tariff increases, localisation requirements and trade finance initiatives (Evenett & Fritz, 2015). Many of the trade distortions are fiscal incentives for exports rather than import restrictions. The European Union is, next to China, the hardest hit by foreign protectionism. Even though trade distortion measures are rising in recent years, Baldwin (2016) argues that these measures will not trigger a sharp increase influencing world trade massively. He reasons this using the example of the global crisis in 2009. At this point in time, the sharpest drop in world trade was experienced since the Second World War. However, this crisis did not trigger a significant shift to protectionism policies. Based on this historical evidence, Baldwin (2016) argues that it is difficult to establish the severity of a world trade disaster required in order to trigger high protectionism policies. Additionally, as the nature of international trade has changed the last few decades and due to globalisation, closing the borders to protect to national economy seems unlikely to work. Yi (2003) argues that the reason a decrease in tariffs has a large impact on fragmentation is because the tariffs accumulate. As mentioned previously, a product crosses many borders before reaching its final stage, meaning a small decrease in tariffs can become large due to its accumulating effect. Moreover, tariff reductions lead to more vertical specialisation, generating nonlinear trade growth (Yi, 2003). Given that trade costs include not only tariff and non-tariff barriers but also transport costs, the same reasoning can be applied. Decreasing transport costs accumulate, however, given that the geographical location of the cities does not change, the changes in the transportation costs determined by distance is highly dependent on the quality of a country’s infrastructure (Bougheas et al., 1999).

(11)

competitive and likely to reduce their consumption of the imported product. Based on the theoretical and empirical literature discussed, we postulate the following hypothesis:

Hypothesis 1: An increase in tariff or non-tariff barriers, by raising the costs of trade will negatively affect product fragmentation.

2.2.2 Labour Costs and Wage Differentials

A main factor in making fragmentation profitable is determined by the outsourcing trend to profit from cheap labour elsewhere in the world. Sourcing internationally to get lower labour costs is a major reason for companies to fragment production. For Europe, China ranks as the third highest outsourcing destination for European companies with 17 percent of the outsourcing locating there, with place one and two respectively EU15 and EU12 (Nielsen, 2017)3. Due to cost efficiency, distinct business functions are moved abroad. The high value-added activities such as R&D and sales are usually kept in the home country. GVCs are heavily reliant on changing labour costs and capital in participating countries. Offshoring or re-shoring is a location decision. Offshoring can be defined as ‘a division of the production processes into separate components that are made by different companies sharing common ownership or not, located in more than one country’ (Radlo, 2016), whereas re-shoring refers to bringing the production process back to the original country. The decision whether or not to offshore production is associated to the trade-off between location advantages identified by Dunning (1980). These are resource seeking advantage, market seeking advantage, efficiency seeking advantage and strategic asset seeking advantage. The major driving force for the offshoring decision is based on efficiency seeking. Cost saving is not the only factor to consider, but there is an increasing tendency to focus on total cost, profitability and customer value creation for locating the manufacturing process (Ellram, 2013). The traditional Stolper-Samuelson model predicts that a country will specialise in its abundant factor, which is in the case of developing economies largely unskilled labour, increasing the demand for, and respectively increasing the relative wages of unskilled labour. In theory, this should reduce wage inequality, though surprisingly wage inequality increased resulting in wage polarisation, where wages for middle-skilled labour decreased, while for low-middle-skilled and high-middle-skilled labour they increased (Reinhold, 2016). This is mainly due to a demand for relatively skilled labour resulting from knowledge and technological externalities of offshoring, increasing the relative wage of skilled labour as well as inequality (Bottini et al., 2007).

3_{EU15 consist of Austria, Belgium, Germany, Denmark, France, Finland, Greece, Ireland, Italy, Luxembourg,} Spain, Sweden and the United Kingdom

(12)

What is clear is that the wages are rising in the main offshoring destination countries. China is a good example of a country used for its cheap labour, where the wages have more than tripled between 1997 and 2007 (Yang et al., 2010). Chinese wages for low skill labour increased by 20 percent a year during this period (Moretti, 2012). A recent report in the Boston Consulting Group Study, GCG (2011) confirms that due to the fast wage growth in China, the manufacturing cost savings of production is around 10 to 15 percent when compared to the US when the labour content is accounted for. At the beginning of the second unbundling, much of the manufacturing production was outsourced to China to benefit from cheap labour. This resulted in significant employment growth and together with rural-urban migration it has caused wages to rise. Low wage labour in China is becoming less common and a middle wage economy is emerging. (Li et al., 2012). While China is flourishing and developing skills in order to attract higher skilled labour, the wages rise with it, triggering the offshore production previously held in China to move to other lower wage countries (De Backer & Miroudot, 2013). This trend is not only happening in China, but many emerging markets follow this path of economic development, increasing labour and raw material costs and taking away the cost benefit of offshoring (Tate et al., 2014). Due to the rising costs in many developing countries, the cost efficiency is lost, which results in many companies re-evaluating their offshoring decision. The choices available to them are either to stay, to re-shore or to move production to another low-cost country.

While argued that the factor endowment theory of Hecksher-Ohlin has become less useful in international trade, the outsourcing trend to ‘cheap labour’ countries can be explained using this theory. If the country where the production of intermediate products is outsourced to has the abundant factor endowment of cheap labour, Hecksher-Ohlin’s theory says this county will specialise in ‘cheap labour’ and export intermediate products that are labour intensive. The Leontief paradox can also be used to argue for outsourcing low-skilled labour production of intermediate products to ‘cheap labour’ countries. This paradox states that ‘a lower capital/labour ratio in exports is lower for a country with a higher capital per worker than in imports’ (Leontief, 1953). This means that a country with low capital per worker, such as low labour costs, has a higher capital/labour ratio in exports than in imports, as they specialise in ‘cheap labour’. This is one of the main driving forces of fragmentation, namely to outsource in order to profit from the cheap labour costs in foreign countries.

(13)

Hypothesis 2: A decrease in the wage gap between home and host country will reduce product fragmentation

2.2.3 Information and Communication Technology

The first unbundling was caused due to the innovation of the steam engine revolutionising the way of travel, not only by steam trains but also through steamships, and thereby sharply reducing the trade costs in the 19th century. This increased international trade. Again, at the end of the 20th-century technological breakthroughs in means of transport changed the way products were moved. This time it was the use of commercial airplanes (Baldwin, 2016). This shows that the evolution of technology has been highly influential for the means of transport and has been important ever since. Not only had the technological progress an effect on the means of transport and on the decrease of transport costs, but it also largely initiated the second unbundling, as long distance communication became possible by the developments in ICT. However, whether this will reach a certain barrier, in which the technological development reaches the point that the cost advantages of cheap labour in outsourcing countries cannot be traded off against the automation process in the home country, will be analysed using the last hypothesis:

Hypothesis 3: An increase in technology and ICT will further accelerate product fragmentation.

The recent ICT revolution lead to greater automation and caused concern in developed countries that computers could take over routine, medium skilled labour causing labour polarisation to occur (Author, 2015). These particular jobs account for the majority of jobs offshored at the beginning of this century. On the one hand, it can be argued that automation keeps jobs in the country because it is cost efficient (De Backer et al., 2016), since cost benefits from automation outweigh the cost benefits of cheap labour, thus making offshoring less viable (Wiesmann et al., 2017). On the other hand, increasing automation does not necessarily create jobs. In many instances automation has allowed firms to heavily reduce their human capital, making firms reconsider their offshoring decision (Tate & Bals, 2017).

2.2.4 Other Factors

(14)

proportional to size, which is measured by GDP, and inversely proportional to the geographic distance between the two trade partners (Chaney, 2013). In other words, according to the gravity model, the bigger the countries involved are, and the closer they are geographically, the higher the trade flow. In this gravity model distance is strictly limited to the geographical distance.

Distance, however, has multiple dimensions, such as for example the cultural distance between two countries. Cultural distance has an effect on trade, since it creates barriers to overcome. Surprisingly, research found that cultural distance reduces bilateral trade, but only beyond a certain threshold level (Lankhuizen & De Groot, 2016) as firms have difficulties to overcome barriers such as lack of understanding and trust (Guiso et al. 2009). However, if the cultural differences between both countries stay below this threshold, it stimulates international trade as cultural differences correspond to differences in comparative advantages, which might outweigh the disadvantages of the actual cultural distance (Lankhuizen & de Groot, 2016). Labour productivity is an important factor in international trade, which influences international trade in a myriad of ways. Fragmentation happened due to firms wanting to take advantage of cheaper labour costs in developing nations. This is determined by the make-or-buy decision, which reflects the trade-off between producing the product in-house or outsourcing it (Williamson 1985b). However, this created the productivity paradox, where in the short-run, outsourcing firms are able to reduce costs. Whereas, in the long-run, firms engaging in outsourcing suffer lower productivity growth than firms that do not engage in outsourcing (Windrum et al., 2009).

All in all, the overall hypothesis tested in the next sections is ‘ICT, trade costs and wages are

the main determinants explaining product fragmentation.’ Including control variables such as

(15)

3. METHODOLOGY

3.1 Defining fragmentation

3.1.1 Vertical Specialisation & Product Fragmentation

The hypotheses will be tested through the use of two separate models, both using a different measurement of fragmentation. First, vertical specialisation will be analysed. Vertical specialisation measures ‘the value of imported inputs used in goods that are exported’ (Hummels et al., 2001). Hummels et al. (2001) specify that vertical specialisation occurs when:

A. A good is produced in two or more sequential stages.

B. More than one country provides value-added to the product in the production process. C. At least one country has to have imported inputs embodied in a stage of the production process and at least part of the outputs of the specific production process has to be exported.

If these stages are all involved in the production process, there is vertical specialisation. In subsets A and B, there is specifically trade in intermediate goods, whereas in C the exported product can be both an intermediate product as well as a final good. The biggest downside of vertical specialisation is that due to data constraints it is not possible to include the value of exports that are embodied in a second country’s export goods (Hummels et al., 2001).

Vertical specialisation measures the total vertical specialisation for a specific country, meaning it is strictly unilateral constraining the possibility for analysis, since the variables such as trade costs, tariffs, non-tariff barriers and distance cannot be reliably included in a unilateral model. Given that this is the case, a second measure is used, referred to in this study as product fragmentation. Product fragmentation measures the domestic value-added embodied in final expenditures abroad (Timmer et al., 2015). This measurement has a few characteristics. First, it is a share and thus a number between zero and one. Second, the value-added is not measured on the ownership of production, but on the geographical location of production. Third, the decomposition is based on values and not on quantities only (Timmer et al., 2015). Product fragmentation specifies the value-added computed in each country, so a bilateral trade analysis is made possible using this measurement as a variable.

3.1.2 Difference in Vertical Specialisation and Product Fragmentation

(16)

value-added is used in exports, while product fragmentation measures how much value-added is used in production.4 Lastly, vertical specialisation is unilateral, meaning its specific for one country, whereas product fragmentation is measured bilaterally, thus the analysis can be focused on bilateral trade between two countries. Hence, the vertical specialisation measure is not on a country to country basis determined, whereas for product fragmentation it is. Figure 3 shows the trend of both vertical specialisation and fragmentation in the manufacturing sector for the sample countries used in this study. For product fragmentation, all the specific bilateral shares of fragmentation are summed up per country per year and taken the average. For vertical specialisation, simply the average is taken. As shown in Figure 3 the trend for product fragmentation and vertical specialisation is exactly the same, only the level is different. This can be explained by the fact that vertical specialisation looks at the fragmentation in exports, whereas product fragmentation focuses on the total fragmentation in production.

Figure 3 - Share of Vertical Specialisation and Product Fragmentation in the Manufacturing Sector. Source: Own calculations used WIOD (see equation 7)

3.2 The Empirical Model

3.2.1 Empirical Model: Vertical Specialisation

Two models will be tested in this study. First, the model with vertical specialisation will be tested on a unilateral basis. Second, the model with product fragmentation will be tested. The following model is used to explain vertical specialisation.

!"_#$ = &_'+ &₎*+,-_.$+ &_/01+2-_.$ + &₃450_.$+ &_#6_#+ 7_.$+ 8 (1)

(17)

In this equation, VS is vertical specialisation, &_' is the constant, labour cost is measured by the *+,-_.$, 01+2-_.$ is trade costs, 450_.$ is ICT cost and &_#6_# are the control variables, namely Labour Productivity, GDP, Landlocked, Human Capital.

There are three options to test the vertical specialisation model, ordinary least square (OLS), random effects and fixed effects. First, the pooled OLS regression is done in order to estimate the effects. However, pooling the different countries without taking individual differences into account might lead to biased coefficients. Moreover, they are assumed to be constant for all countries in all time periods and there is no allowance for individual heterogeneity (Hall et al., 2012). The OLS regression has many limitations, including the restrictive way of acknowledging the panel nature of the data. The fixed effect model can overcome the limitation of the OLS estimation not being able to estimate country-pair specific effects (Zeddies, 2011). The basic intuition behind the fixed effects model is that, in the case that the unobserved variable does not show any changes over time, the effect of the dependent variable must be caused by influences other than the fixed characters. In the fixed effects model, the unobserved heterogeneity of individuals is assumed to be constant over time for every individual. In this model, the control variables distance, landlocked and common language drop out of the equation due to perfect multicollinearity. The data is also tested for heteroscedasticity, which is the case in this model, meaning that we need to include robust standard errors in the model. While fixed effects might be the best option to test the model, another possibility is the random effects model (Zeddies, 2011). The main difference between the random and fixed effects model is whether there are unobserved individual effects embodied elements correlated with the regressors in the model (Greene, 2008, p.183). Random effects assume that the entity’s error term is not correlated with the predictors which allow for time-invariant variables to play a role as explanatory variables (Hill et al., 2011).

The assumption that individual differences are not considered as random disturbances require that the regressors and the unobserved heterogeneity are not correlated. The Hausman-test is applied to identify this. The test compares the coefficient estimates from the random effect model to those from the fixed effects model, in order to check for any correlation between the error component and the regressors in a random effects model. The idea underlying the Hausman test is that both the random effects and fixed effects estimators are consistent if there is no correlation between the error component and the explanatory variables. The Hausman test shows that the coefficients differ between the random effects and the fixed effects model, so the fixed effects model should be used.

(18)

to the data the average trade costs for example of Germany to the USA would be the same as the trade costs of Germany to the Netherlands. Moreover, without bilateral trade, tariff and non-tariff barriers cannot be taken into consideration. In order to solve this shortcoming, a similar model will be provided in the next section based on product fragmentation in which a bilateral trade analysis is possible. The hypotheses will be tested using the model using product fragmentation.

3.2.2 Empirical Model: Product Fragmentation The model for product fragmentation is as follows:

91+,_.:$ = &_'+ &₎*+,-_.:$+ &_/0+1499_.:$+ &₃<=<0+1499_.:$ + &_>450_.:$+ &#₆

.:$# + 7:$+ 7.$ + 8 (2)

All these variables are measured between country i to country j in time t. In this model 91+,_.:$ is product fragmentation, *+,-.:$ is the wage gap, 0+1499.:$ is the tariff, <=<0+1499.:$

is the non-tariff barriers, 450_.:$ is ICT. Moreover, &#₆

.:$# are the control variables, namely

Labour Productivity, Landlocked, GDP, Common Language, Human Capital and Distance.

This model is based on panel data in a bilateral setting from country i to country j in time t. Given that the sample includes data from 25 countries over 15 years means that there are 600 country pairs. Included in the equation are fixed effects on the country-time level in both host and home country. There are three options for the model, the ordinary least square model (OLS), random effects model and fixed effect model. The same argumentation as in the previous model is used, where due to the limitations of OLS, either fixed effects of random effects are used. The Hausman-test indicates, fixed effect has to be used to test this model. Moreover, year fixed effects and robust standard errors are used to test the model.

(19)

Table 1 - Expected Relationship Variable Vertical Specialisation Product Fragmentation Trade − Tariff − − Non-tariff − − Wage + + ICT + + Labour Productivity + + Human Capital + − Landlocked + − GDP + + Common Language + Distance −

(20)

4. DATA

4.1 Measuring Vertical Specialisation

Vertical specialisation (VS) is the imported input content of exports. In other words, the foreign value-added embodied in exports (Hummels et al, 2001). The VS has a value of 0 if county k does not import any intermediate inputs. The higher the VS, the higher the imported input content of exports is. To arrive at the share of VS, the data in the World Input-Output Database (WIOD) is essential. It covers information on the fragmentation trend, in order to enable analysis of the consequences of fragmentation (Timmer et al., 2015). The National Input-Output Tables, used to measure the VS share, are in current prices. The National Input-Input-Output Database (NIOD) covers 28 EU countries together with 15 other major countries in the world for the period from 2000 to 2014. This is exactly the period used for the VS measurement.

Table 2 - Simplified Version Input-Output Table

z11 z21 … zn1 f1 x1 z12 z22 … z n2 f2 x2 … … … … z1n z2n … znn fn xn P1 P2 … Pn P x1 x2 … xn F

The Input-Output database is used to calculate the VS share for the manufacturing sector specifically. The following will be a short explanation of the matrix mathematics used in the Input-Output tables in order to calculate the vertical specialisation.

The @ × @ matrix Z gives the intermediate deliveries matrix, where the element Zij expresses

the money value for the delivery of goods from industry i to industry j. The @ × B F matrix is the final demand matrix, including exports, where the element Fij expresses the money value

for the delivery of goods from industry i to final demand category j. The x vector gives the gross output vector in each sector i. The input coefficient is calculated through + = C/6. In order to quantify the level of production to satisfy the level of final demand F, the Leontief inverse needs to be used (Miller & Blair, 2009), which is calculated as follows:

E = 4 − + F) ₍₃₎

(21)

amount of imports that is required directly and indirectly per unit of exports of good j (I) needs to be calculated. Hereby the Leontief inverse M allows to calculate the value of imported inputs used also indirectly in the production of a unit of exports since it accounts for the multiplier effects that arise. Thus, by multiplying the square matrix Q by M, we arrive at a new square matrix containing coefficients K ∙ E. One unit of exports of good j requires M_#.∙ N_.: units of imports of good i by industry j. Thus, contained in a row vector, the sum over the columns of K ∙ E is calculated. Hence, the total amount of imports required for one unit of export good j is obtained. Having done this, the VS can be calculated by multiplying the import requirements of one export good j by the corresponding export value j and taking the sum over all export goods yields the total amount of imports that is embodied for the whole economy's exports. By dividing this number by the total export value, we obtain the VS for the whole economy. Given that only the VS for the manufacturing sector is necessary, only the rows and columns obtaining information over the manufacturing sector are used in the calculation. In matrix notation, the VS share of manufacturing will be:

!" OℎQRS TU NQ@VUQWXVRY@Z =[\]

^] = V_ES/S# (5)

In which V is a 1 × @ vector of ones, S is the 1 × @ vector of exports and S# is the total country

exports.

(22)

Table 3 - Vertical specialisation share in total manufacturing in European Countries, using the WIOD database Source: own calculation using WIOD database equation 2

Country 2000 2008 2010 2014 Change 2000 - 2008 (%) Change 2008 - 2010 (%) Change 2010 - 2014 (%) Western Europe Luxembourg 0.55 0.63 0.67 0.67 16.10% 6.10% -0.10% Belgium 0.45 0.50 0.56 0.60 12.30% 11.30% 8.20% Netherlands 0.35 0.44 0.56 0.56 26.00% 27.30% -1.40% Ireland 0.47 0.53 0.55 0.52 12.50% 3.20% -4.90% Austria 0.35 0.41 0.43 0.45 18.20% 2.80% 4.50% Portugal 0.37 0.42 0.40 0.45 14.00% -4.20% 11.90% Spain 0.34 0.34 0.39 0.44 -0.70% 15.00% 12.90% France 0.31 0.35 0.36 0.37 12.20% 4.20% 3.60% Germany 0.26 0.31 0.31 0.33 20.70% 0.50% 3.70% Italy 0.23 0.29 0.31 0.32 21.70% 10.00% 0.70% United Kingdom 0.24 0.29 0.34 0.32 22.30% 16.90% -6.00% Eastern Europe Hungary 0.58 0.60 0.61 0.61 2.50% 3.00% -0.10% Slovakia 0.45 0.55 0.57 0.61 23.60% 2.20% 6.80% Bulgaria 0.45 0.55 0.51 0.55 23.70% -6.70% 6.00% Estonia 0.41 0.47 0.50 0.54 15.20% 5.30% 9.40% Lithuania 0.36 0.52 0.50 0.53 41.90% -2.10% 5.30% Czech Republic 0.38 0.46 0.49 0.52 20.70% 5.70% 5.60% Slovenia 0.36 0.43 0.43 0.43 18.60% 0.90% -0.50% Poland 0.31 0.38 0.42 0.41 22.40% 11.90% -1.60% Latvia 0.28 0.38 0.36 0.39 32.80% -5.20% 8.20% Romania 0.32 0.31 0.28 0.35 -3.10% -10.50% 24.80% Croatia 0.30 0.35 0.31 0.34 14.10% -9.30% 6.80%

Major Economic Forces

Japan 0.10 0.22 0.19 0.27 110.00% -12.60% 39.50%

China 0.20 0.27 0.25 0.21 32.60% -6.50% -17.80%

United States 0.14 019 0.17 0.19 31.5% -11.5% 14.6%

Note - This table is ordered from large to small vertical specialisation share per region in 2014.

4.2 Measuring Product Fragmentation

The measurement of product fragmentation follows the calculation used by Timmer et al. (2015) based on the World Input-Output tables. These Input-Output tables cover 40 countries, including all European Union countries together with 13 major economic forces and reaching over the years 1995 to 2014. This dataset provides the value-added content in the final output of manufacturing by countries (Timmer et al., 2015), constructed through closely following the approach used by Timmer et al. (2015).

(23)

unit of output in the manufacturing industry. I is again the identity matrix and (I-A)-1 is known as the Leontief inverse. This Leontief inverse contains the gross output values generated when producing specifically one unit of consumption in the manufacturing industry.5 With all this information, the output levels vector of the manufacturing industry can be calculated, using the following equation:

K = 4 − + F)₅ ₍₆₎

The vector K can then be calculated through the following equation, in which B is a diagonal matrix containing information of the value added to gross output ratios in the manufacturing industry for all the countries in the sample.

a = b 4 − + F)₅ ₍₇₎

In order for the VAX to be computed correctly, C has to refer to the consumption for all the countries j outside of the specific country i.

Table 4 demonstrates an example case, consisting of the percentages of foreign value added in the home country for the years 2000, 2008, 2010 and 2014 and for the specific host country Germany since this is one of the largest trading partners in the sample. For Austria, Luxembourg, Czech Republic and Hungary, manufacturing output contains over 10 percent of German value-added. The United States, Japan and China have the least German value-added embodied in their manufacturing output. This is just a specific example for Germany, however, this data is used for all the countries in the sample.

Table 4 – Product Fragmentation share in total manufacturing in Germany, using the WIOD database Source: own calculation using WIOD database equation 5

Country 2000 2008 2010 2012 Change 2000 - 2008 (%) Change 2008 - 2010 (%) Change 2010 - 2014(%) Western Europe Austria 0.12 0.13 0.12 0.13 14.9% -6.9% 2.1% Luxembourg 0.10 0.12 0.13 0.13 16.5% 10.9% 0.6% Belgium 0.08 0.07 0.07 0.07 -10.2% -4.3% 3.1% Netherlands 0.05 0.05 0.06 0.07 -2.7% 14.4% 16.2% France 0.05 0.05 0.05 0.05 1.7% -3.4% 2.9% Italy 0.03 0.03 0.04 0.04 6.5% 18.7% 13.5% Portugal 0.04 0.05 0.04 0.04 5.1% -16.1% 9.2% Spain 0.04 0.04 0.04 0.04 -16.5% 14.1% 1.4% United Kingdom 0.03 0.03 0.04 0.04 23,00% 18.7% 2.1% Ireland 0.02 0.03 0.03 0.03 21.4% -2.1% -4.6% 5

(24)

Country 2000 2008 2010 2012 Change 2000 - 2008 (%) Change 2008 - 2010 (%) Change 2010 - 2014(%) Eastern Europe Hungary 0.14 0.13 0.11 0.15 -7,00% -12.8% 33.7% Czech Republic 0.12 0.11 0.12 0.13 -3.3% 7.1% 8.9% Slovak Republic 0.11 0.11 0.10 0.12 -5,00% -5.7% 19.4% Poland 0.07 0.07 0.08 0.08 9.3% 4.6% 4.6% Slovenia 0.07 0.07 0.07 0.07 0.5% -4,00% 2,00% Estonia 0.03 0.06 0.06 0.06 64.4% -4.1% 9.5% Bulgaria 0.03 0.03 0.03 0.04 29.6% -6.6% 15.3% Croatia 0.05 0.04 0.04 0.04 -5.9% -13.3% 11.4% Romania 0.03 0.03 0.03 0.04 -19,00% 6.9% 32.6% Lithuania 0.03 0.04 0.03 0.03 15.2% -16.7% -2.1% Latvia 0.04 0.04 0.03 0.03 -12.3% -8.2% 2.1%

Major Economic Forces

China 0.01 0.01 0.01 0.01 74.6% -13.7% -17.2%

Japan 0.00 0.01 0.00 0.01 85.2% -22.5% 28.2%

United States 0.01 0.01 0.01 0.01 30.1% -15.7% 18.4%

Note - This table is ordered from large to small product fragmentation share per region in 2014

4.3 Measuring Trade Costs

Trade costs data is needed in order to be able to test hypothesis one. Trade cost is measured through the ESCAP-World Bank Trade Cost Database, which is ‘a comprehensive all-inclusive measure based on micro-theory and calculated using macroeconomic data, providing an alternative measure of trade facilitation performance’. (Novy, 2012). These trade costs ‘captures all additional costs involved in trading goods bilaterally relative to those involved in trading goods domestically, including international shipping and logistics costs, tariff and non-tariff costs and costs from differences in language, culture and currencies’ (Novy, 2012). To create this database, the World Bank and United Nations ESCAP collaborated in order to create a measurement for comprehensive international trade costs, comprising data for over 180 countries from 1995 to 2015 (UNESCAP, 2016). These trade costs are divided into multiple sectors, including the manufacturing sector, which is used in this case and the data is expressed in current US$. Moreover, the data is enlisted in bilateral trade costs, which causes a problem for the vertical specialisation model, given that the trade costs should be clear per country. In order to estimate these trade costs per country, the total bilateral trade costs of country k are pooled with all countries in the sample by summing up all values and took the average for country k. Given that distance is already included in the trade costs, the model will not be separately adjusted for the distance between the countries.

(25)

of the tariffs imposed by the two partner countries based on the imports of each other. Non-tariff barriers are determined based on the definition of Anderson and van Wincoop (2004), which entails ‘all additional costs other than tariff costs involved in trading goods bilaterally rather than domestically’. Trade costs, tariffs and non-tariff barriers are taken from the ESCAP-World Bank Trade Cost Database.

4.4 Measuring Wage

Measuring labour cost can be done in a myriad of ways, mostly focusing on unit labour costs. This data is necessary in order to test hypothesis 2. These unit labour costs are a broad measure and can be expressed as the cost of labour per unit of output produced (OECD, 2018). However, the unit labour costs are not available for a specific industry. Therefore, the measurement used for labour cost is hourly compensation costs, expressed in US$ and specific for the manufacturing industry. This data is taken from the conference board, out of their International Labour Comparison program (Conference Board, 2018). However, this dataset does not include data from certain countries in the sample, mainly the Eastern Europe countries such as Bulgaria, Croatia, Lithuania, Latvia, Romania and Slovenia. These datasets comprise some missing values, but since it entails all countries, this is not problematic. For these missing data values, the previous data value is used as benchmark, assumed to fill in the gaps. Hourly compensation costs in the specific amount says little about whether the country has ‘cheap’ labour or not, so in order to improve the measure the hourly compensation costs in country i is divided by the sample average, for the vertical specialisation model. While this way of measuring is not ideal as it is still not a perfect measure of cheap or expensive labour, it gives insight in whether a country has a cost advantage in labour or not. For the model based on product fragmentation, the log ratio of hourly compensation costs between country i and country j is computed in order to measure wage bilaterally.

4.5 Measuring ICT

For hypothesis three, ICT will be measured using ICT capital spent per employee. This data is taken from the Total Economy Database from the Conference Board (2017) and is fully balanced. For vertical specialisation, the exact number is used, whereas for product fragmentation the log ratio of ICT capital spent per employee between country i and country j are taken to enable a bilateral analysis.

4.6 Control Variables

(26)

distance between both countries. In the vertical specialisation model, total GDP is used as a control variable, whereas for the product fragmentation model the log ratio between country i and country j is measured. However, in the vertical specialisation model is it impossible to control for distance, whereas in the fragmentation model the distance between the capitals in country i and country j are measured in Kilometres (km). Another major factor used in multiple international trade literature is whether a country is landlocked or not and whether there is a common language, which will also be used in this analysis as a control factor. This data is taken from the CEPII database. The control variable landlocked, is used in both models. It is a dummy variable measuring whether country i is surrounded by only land or lies at the sea. The variable common language, however, can only be used in the product fragmentation model, since it measures whether country i and country j speak the same language. These factors influence international bilateral trade, as a common language between countries facilitates better communication, while a landlocked country has no easy access to a harbour. Moreover, both models are controlled for human capital, which is measured as the percentage of the population that finished tertiary education based on data taken from Barro and Lee (2013). Unfortunately, this database only contains the years 2000, 2005 and 2010. In order to solve for this data shortage, the missing years take up the previous data point measured, in order to not lose many observations. For the vertical specialisation model the specific share is taken, whereas for the product fragmentation model the log ratio between country i and country j is measured Lastly, labour productivity is used as a control variable. Labour productivity can be described as output per unit of labour input (KILM, 2002). The data on labour productivity is taken from the total economy database at the Conference Board (2017). This database contains data on labour productivity, measured as labour productivity per person employed in 2017 US$ (converted to 2017 price level with updated 2011 PPPs). Given that the labour costs are in time units, it is measured as labour productivity per person employed. For the model of vertical specialisation, the exact number is used, as for the model of product fragmentation, the log ratio of the labour productivity between country i and country j are taken.

Table 5 – Control Variables used in both models

Vertical Specialisation Product Fragmentation

GDP GDP

Landlock Landlock

Human Capital Human Capital Labour Productivity Labour Productivity

Distance

(27)

4.7 Sample

The panel data sample used for this analysis is the European Union together with three other economically influential countries, the US, China and Japan, where Finland, Sweden, Norway and Denmark are not taken into consideration. The period of time considered is from 2000 to 2014. This specific time frame is required, since the input-output tables used to construct the dependent variable product fragmentation only includes these years. While this is a short period, it exactly captures the upcoming fragmentation around 2000 and also includes the stagnating period of fragmentation after the financial crisis. Moreover, the dataset has a wide array of countries, ranging from the lower-income eastern Europe countries to the high-income OECD countries, as well as from European Union countries to the Asian countries. This causes the data to have missing values in either the lower-income European countries such as Romania and Estonia, or missing values in the Asian countries such as China or Japan. In order to solve this problem, a proxy is used.

Moreover, the bilateral dataset involves 25 countries, specifically the European Union excluding Finland, Sweden, Norway and Denmark, but including the major economic forces in the world, namely the US, Japan and China. So, all in all the panel dataset includes 25 countries, resulting in 600 country pairs (I and j) for 15 years.

Table 6, shows the descriptive statistics, given that wage, ICT, labour productivity and human capital is measured bilaterally as the log ratio, can explain that the mean in zero, also tariff and non-tariff barriers are measured in logs. Moreover, the bound of minimum and maximum is larger for ICT than for wage, labour productivity and capital.

Table 6 – Descriptive Statistics Product Fragmentation

Mean Std. Dev. Minimum Maximum

(28)

5. RESULTS

In this section, the results of the regressions performed in order to test the three hypotheses will be analysed, based on the model using product fragmentation. First, the preparation of the data will be discussed. Both datasets are tested for normality, multicollinearity and heteroscedasticity. Second, the results of the model based on vertical specialisation will be analysed. Then the hypotheses will be tested based on the model using product fragmentation.

5.1 Data Preparation

The Jarque-Bera test is performed, in order to test for normality. Vertical specialisation is normally distributed c = .09253005 . There is a high Variance Inflation Factor (VIF) value for labour productivity with 10.84, indicating this control variable is correlated. To correct for this, we will model both with and without labour productivity (model 4), in Table 6. Moreover, vertical specialisation is heteroscedastic, so we will use robust standard errors. For product fragmentation, the Jarque-Bera test shows that product fragmentation is also normally distributed c = 4.776Fm3 _{. Based on the VIF test and the correlation test for multicollinearity,}

trade is highly correlated with tariff and especially non-tariff barriers, so trade is excluded from the model. Also wage and labour productivity are correlated based on VIF, so the models will be tested with and without the variable wage. Lastly, the data is tested for heteroscedasticity, in which the dependent variable is heteroscedastic, so we will use robust standard errors. Both the results for the vertical specialisation model and product fragmentation model are interpreted differently since vertical specialisation is based on unilateral data, while product fragmentation indicates the log ratio of country i and country j, bilaterally. Given that there are severe limitations to the model of vertical specialisation, causing the results for trade costs to be unreliable, the hypothesis tests will be based on the product fragmentation model.

5.2 Results Vertical Specialisation

(29)

tests indicate that the results are robust for Western Europe, Eastern Europe and before the financial crisis, however, they do not show any robust results for the Major Economic Forces.

Table 7 – Determinants of Vertical Specialisation

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

Variables Fixed Effect Robust Fixed Effect Robust Fixed Effect Robust Random Effect Fixed Effect Robust Wage -0.0581*** -0.0575*** -0.0567*** -0.0751*** -0.0583 (0.0205) (0.0205) (0.0204) (0.0164) (0.0347) Trade -0.000460*** -0.000450*** -0.000464*** -0.000421*** -0.000364**

(8.27e-05) (8.59e-05) (8.56e-05) (8.27e-05) (0.000171)

ICT 0.00142*** 0.00138*** 0.00123*** 0.000870*** 0.000672***

(0.000107) (0.000141) (0.000153) (0.000152) (0.000238)

Labour Productivity 1.98e-07 -2.53e-07 3.42e-07 4.74e-07

(4.70e-07) (5.04e-07) (4.76e-07) (8.96e-07)

GDP 2.90e-06** 6.38e-07 2.81e-06

(1.22e-06) (8.54e-07) (2.40e-06)

Landlocked 0.133*** (0.0425) Human Capital 0.00545*** 0.00596*** (0.000945) (0.00203) Constant 0.400*** 0.386*** 0.342*** 0.294*** 0.234*** (0.0200) (0.0383) (0.0423) (0.0358) (0.0695) Observations 374 374 374 374 374 Adj. R-squared 0.565 0.563 0.569 0.635 Hausman 4.53 (p = 0.3385) Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

Table 8 – Determinants of Vertical Specialisation, after the Financial Crisis (=2009)

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

Variables Fixed Effect Robust Fixed Effect Robust Fixed Effect Robust Random Effect Fixed Effect Robust Wage -0.0994** -0.112** -0.113** -0.0804*** -0.113*** (0.0469) (0.0467) (0.0468) (0.0304) (0.0272)

Trade 2.38e-05 5.24e-05 5.94e-05 -2.44e-05 5.94e-05

(0.000111) (0.000110) (0.000111) (0.000113) (0.000218)

ICT 0.00166*** 0.000965 0.000876 0.00110* 0.000876

(0.000474) (0.000583) (0.000603) (0.000569) (0.000626)

Labour Productivity 2.31e-06** 3.12e-06* 2.26e-06* 3.12e-06

(1.16e-06) (1.78e-06) (1.35e-06) (2.42e-06)

GDP -1.73e-06 -2.08e-06 -1.73e-06

(2.86e-06) (1.68e-06) (3.95e-06)

Landlocked 0.147*** (0.0517) Human Capital 0.00439 (0.00310) Constant 0.384*** 0.260*** 0.257*** 0.205*** 0.257** (0.0527) (0.0807) (0.0812) (0.0688) (0.105) Observations 125 125 125 125 125 R-squared -0.045 -0.014 -0.021 0.185 Hausman 2.66 (0.4477) Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1

(30)

5.3 Results Product Fragmentation

5.3.1 Results

The hypotheses will be tested using the product fragmentation model. As indicated in the methodology section, both the random effects model and fixed effects are performed, after which the Hausman test statistic estimates to use the fixed effects model. ICT is positive in the random effects model whereas it is negative in the fixed effects model. This is due to the fact that there are different underlying assumptions for the random effects and the fixed effects model. The random effects model has the assumption that the explanatory variables are uncorrelated with the error term. This means that time-invariant variables which are not part of the model might be correlated with the explanatory variables. Hence, an omitted variable bias is found. In the fixed effects model, all time-invariant variables are taken into consideration as part of the constant and not part of the error term.

Table 9 shows eight different models with different underlying assumptions such as the fixed effects model or the random effects model or includes different variables in each model. In the first model, the main independent variables are simply regressed, namely tariff, non-tariff, wage and ICT. The second model adds labour productivity, the third model adds human capital and in the full model GDP is added. The main model is model 5, which includes all variables and is regressed for robust fixed effects. Model 6 and 7 excludes tariff and non-tariff and wage respectively, to account for the high VIF between wage and GDP, and labour productivity and GDP. Lastly, model 8 corrects for the financial crisis year, in which there was a large drop in fragmentation, demonstrated in Figure 3.

First of all, hypothesis one tests whether an increase in tariff or non-tariff barriers can raise the costs of trade and whether it affects product fragmentation negatively. This hypothesis can be confirmed, given that the correlation for as well tariff and non-tariff barriers on product fragmentation are negative and significant at the 1 percent level. For tariff, an increase in tariffs with 1 percent will decrease fragmentation on average by .197 percent ceteris paribus. The same goes for non-tariff barriers, where an increase of 1 percent will lead to a decrease in fragmentation of on average .205 percent ceteris paribus.

(31)

is that it becomes less ‘cheap’ to outsource production, decreasing fragmentation. So, we can confirm hypothesis two.

Lastly, hypothesis 3 tests whether an increase in technology and ICT will further accelerate product fragmentation. The relationship between ICT is negative and significant at the 5 percent level, meaning that a decrease in the difference between ICT in country i and country

j, causes an increase in fragmentation by on average .706 percent ceteris paribus. Therefore,

we reject hypothesis three.

For the control variables, Model 6 shown in Table 9 suggest the relationship between technology and product fragmentation is positive and significant at the 5 percent level. A decrease in the difference between technology in country i and country j with 1 percent causes a decrease in fragmentation of on average .618 percent ceteris paribus. Moreover, while human capital has not a significant effect on product fragmentation, the effect is negative. Lastly, GDP has an insignificant, however, positive effect on fragmentation.

(32)

Table 9 – Determinants of Fragmentation (1) (2) (3) (4) (5) (6) (7) (8) Variables Fixed Effects Robust Fixed Effects Robust Fixed Effect Robust Random Effects Fixed Effects Robust No Tariff No Wage Without the Crisis Year (=2009) Tariff -0.197*** -0.197*** -0.197*** -0.222*** -0.197*** -0.130*** -0.230*** (0.0215) (0.0215) (0.0215) (0.00898) (0.0215) (0.0130) (0.0167) Non-Tariff -0.205*** -0.205*** -0.205*** -0.309*** -0.205*** -0.316*** -0.285*** (0.0385) (0.0375) (0.0376) (0.0165) (0.0371) (0.0400) (0.0369) Wage Gap 1.378*** 1.189*** 1.187*** 1.099*** 1.146*** 0.944*** 1.156*** (0.107) (0.123) (0.122) (0.0530) (0.133) (0.149) (0.141) ICT -0.938*** -0.722** -0.433 0.944*** -0.706** -0.833** -0.438 -0.751** (0.279) (0.323) (0.304) (0.0524) (0.305) (0.399) (0.337) (0.344) Labour Productivity 0.841*** 0.929*** 1.225*** 0.618** 0.583 0.138 0.574* (0.289) (0.288) (0.149) (0.304) (0.434) (0.392) (0.313) Human Capital -0.256* -0.428*** -0.213 -0.128 0.0927 -0.206 (0.141) (0.0608) (0.147) (0.189) (0.128) (0.155) LOG GDP -0.186 0.392 0.680* 2.078*** 0.393 (0.114) (0.330) (0.390) (0.357) (0.348) Landlocked -0.342*** (0.108) Distance -0.356*** (0.0419) Common Language 1.020*** (0.183) Constant 0.0889 0.0889 0.0889 3.213*** 0.0889 -1.071*** -0.266 0.560*** (0.163) (0.159) (0.160) (0.322) (0.157) (0.0189) (0.183) (0.153) Observations 3,336 3,336 3,336 3,336 3,336 4,080 7,644 3,154 Adj. R-squared 0.516 0.524 0.526 0.528 0.486 0.527 0.473

Year Fixed Effects YES YES YES YES YES YES

Hausman 129.07 (p = 0.000) Standard errors in parentheses - *** p<0.01, ** p<0.05, * p<0.1

(33)

Table 10 – Determinants of Fragmentation after the Financial Crisis Year (=2009) (1) (2) (3) (4) (5) (6) (7) (8) Variables Fixed Effects Robust Fixed Effects Robust Fixed Effect Robust Random Effects Fixed Effects Robust No Tariff No Wage Without the Crisis Year (=2009) Tariff -0.179 -0.179 -0.179 0.0517 -0.179 -0.00653 -0.179 (0.124) (0.124) (0.124) (0.0798) (0.122) (0.0133) (0.122) Non-Tariff -0.0842** -0.0842*** -0.0842*** -0.127*** -0.0842*** -0.133*** -0.0842*** (0.0329) (0.0314) (0.0314) (0.0273) (0.0309) (0.0393) (0.0309) Wage Gap 1.385*** 1.428*** 1.428*** 1.137*** 1.430*** 1.280*** 1.430*** (0.161) (0.166) (0.166) (0.107) (0.161) (0.151) (0.161) ICT -0.0475 -0.0951 -0.0951 1.024*** -0.363 -0.663 0.475 -0.363 (0.336) (0.335) (0.335) (0.0704) (0.368) (0.501) (0.344) (0.368) Labour Productivity 1.108** 1.108** 0.115 1.192** 0.198 0.900*** 1.192** (0.475) (0.475) (0.325) (0.468) (0.489) (0.320) (0.468) Human Capital 0.0662 (0.170) LOG GDP -0.287 0.511 0.171 0.964*** 0.511 (0.182) (0.321) (0.464) (0.291) (0.321) Landlocked -0.391*** (0.147) Distance -0.504*** (0.0660) Common Language 0.690*** (0.178) Constant -0.0337 -0.0337 -0.0337 3.846*** -0.0337 -0.737*** -0.723*** -0.0337 (0.153) (0.148) (0.148) (0.484) (0.147) (0.00670) (0.174) (0.147) Observations 780 780 780 780 780 1,360 1,956 780 Adj. R-squared 0.220 0.231 0.231 0.246 0.144 0.105 0.246

Year Fixed Effects YES YES YES YES YES YES YES

Hausman 178.22 (p = 0.000) Standard errors in parentheses - *** p<0.01, ** p<0.05, * p<0.1

(34)

5.3.2 Interpretation of the Results

Since the relationship between tariffs and product fragmentation and non-tariff barriers and product fragmentation is established, we can focus first on what happened in the time period between 2010 to 2014. Table 10 displays the same regression as Table 9, focusing on the time period after the financial crisis, since in that specific time period the decline in fragmentation happened. While tariffs become insignificant, the relationship is still negative. For non-tariff barriers, the relationship becomes slightly smaller but still negative and significant. Given that non-tariff barriers have shown a slight increase in the past years, this might contribute to explaining the decline in fragmentation. Tariffs, on the other hand, have been declining at the beginning of the 21st century, however, these seem to have been stagnating on average after the financial crisis. All in all, tariff and non-tariff barriers have not been increasing enough in order to explain the recent decline in fragmentation, since there are trade agreements in place managing tariff and non-tariff barriers. So, an increase in trade costs might have been a small contributing factor, however, not a main explanatory one. An issue in the analysis is that the tensions at the global political stage that caused an increase in trade barriers happened after the time period investigated in this study. While these increasing barriers have not yet been mandated by 2014, these developments can be used to predict whether fragmentation will decline even further.

Second of all, since the relationship between the wage gap and product fragmentation is as well established, we can examine what happened to the wage gap in the period of declining fragmentation. In the period between 2010 to 2014, wage is still positive and significant. Given that the wage gap has been declining recently due to the increase in wages in the ‘cheap labour’ countries, this is one of the main explanatory variables for the decline in fragmentation.

Figure 4 – ICT Differential (Calculated as the average log ratio between country i and country j)

While this wage gap data in Figure 4 shows that since the year 2000, the wage gap has been decreasing. This would indicate that even in the years of booming fragmentation the wage gap has been decreasing. This can be explained through the unpredictability of fragmentation. The

Determinants of International Product Fragmentation: Explaining the Recent Decline

MASTER THESIS

MSc International Economics & Business