What Are the Driving Factors Fuelling Global Value Chain Localisation in Advanced Countries: A Country and Industry Level Analysis

(1)

What Are the Driving Factors Fuelling Global Value Chain Localisation in

Advanced Countries: A Country and Industry Level Analysis

University of Groningen

Faculty of Economics and Business

Master Thesis International Economics and Business

Name Student: Mike Jimmink

Student ID number: S2975459

Student E-mail: M.R.Jimmink@student.rug.nl

Date Thesis: 18-06-2019

Name Supervisor: Prof. Dr. H.H. van Ark

(2)

ii ABSTRACT

This paper aims to identify the main drivers that influence industries their decision to localise production in advanced countries. This localisation trend started in 2011 and has since caught attention of researchers. By using fixed and random effects regressions on a large panel data set consisting of 14 countries and 18 industries over the period 2000-2014, I find that an increase in China’s wages has played an increasing role in localisation while technology does not. In contrast, as localised industries decrease in productivity it is likely that technology is substituted for labour, causing a negative relationship between localisation and technology.

(3)

iii

Global value chains (GVC) have been a topic of interest since they started to develop in the 1960s. Researchers have been closely watching the developments of GVC and have been analysing the effects which they have on economies. Moreover, politicians often blame GVC for a loss of jobs in advanced countries, making it a topic in which the public is interested. However, the most recent developments show manufacturing industries return their production sites to advanced countries, and with it labour jobs. It is unclear what specifically drives this return of manufacturing. Even so, advanced countries are welcoming the return of manufacturing labour jobs (Irwin-Hunt & Shehadi, 2019).

The aim of this paper is to identify the main drivers for the localisation of manufacturing industries. The paper is inspired by a report of The Conference Board (De Vries, Maselli, Erumban, Lundh, & Ozyildirim, 2019). In there, the authors showed that there is a recent trend of localisation among manufacturing industries in Western Europe and North America in the period of 2011-2014. Moreover, they show the effects which this has had on, among others things, productivity and wages within their respective industry. This paper takes the analysis one step further by looking at which factors are driving this decision to localise. It aims to provide more insight in how advanced countries can let their manufacturing industries grow as to create more employment opportunities. Currently, no research is done on the drivers of localisation at the industry or country level using regression analysis. Most existing literature uses firm-level surveys to explain why firms localize (Bailey & De Propris, 2014; Ellram, 2013; Dachs, Kinkel, Jäger, & Palčič, 2019). This paper will aim to provide an answer to the following research question:

Which factors are the main drivers of the recent localisation trend in manufacturing sectors?

(6)

2 researched. Finally, results show that there are large differences between drivers of localisation, both at the industry and the country level.

The remainder of this paper is organized as follows. Section 2 will provide the reader with an overview of existing research on this subject. On top of this, it will lay the foundation of the hypotheses, which are presented in section 3. Section 4 will then go on by describing the methodology used in this paper as well as the data used. Next, section 5 will discuss the diagnostic checks as well as providing the reader with the main results and its discussion. At last, section 6 includes the conclusion as well as a summary of the caveats of this paper.

2. Literature Review

The literature review will mainly be used to explain the current state of knowledge on localisation to inform the research question and hypotheses for this paper. Many firm-level surveys are done based on reshoring or the localisation of manufacturing which suggest a large set of drivers to focus on. However, only a small part of the literature focuses on the industry level, which will be used in this report. Nonetheless, some of the drivers, although hard to quantify at the industry level, will still be discussed in short to give a good idea as to why firms and industries localise.

2.1. Defining Localisation

The literature has yet to agree on a clear definition for the ongoing process of firms re-locating their manufacturing processes back into their own country. A few definitions which are being used are: reshoring (De Backer et al., 2016), back shoring (Wu and Zhang, 2014), near-shoring (Fratocchi et al. 2016), and localisation (De Vries, Maselli, Erumban, Lundh, & Ozyildirim, 2019). As this report builds on the report written by De Vries et al. (2019), the term localisation will be used to define the activity of firms increasingly using locally sourced inputs for the assembly of a final product. The focus is on the country of completion of the final product, and not on the ownership of the company producing the final product. For clarification, De Vries et al. (2019) provide the reader with two examples: “A German car company opening up a plant in the United States for the production of cars benefitting from United States third-party suppliers will attribute to the U.S. going local. Moreover, a German-based plant starting to use more inputs from German third-party suppliers counts towards the German automotive sector going more local” (p. 7).

(7)

3 as going global. Finally, all results in between are categorized as neutral. They have opted to use the time period of 2011-2014 due to interference of the global recession. By starting at 2011, they make sure that the noise in the data caused by the global recession has faded.

2.2. Discussion of the Localisation Trend

In recent decades global value chains have been a much studied subject. This section briefly explains how and why value chains have become global in the past and aims to provide a starting point for the discussion on localisation.

2.2.1. Value chains in the past. The reason as to why value chains have become more “global” in the last two centuries is the fall in transport and communication costs, and the increase of technology (Baldwin, 2006). Researchers often distinguish between two waves of globalization. The first wave of globalization happened in two parts, and was caused by a decrease in transport costs around the 1850s and 1960s (Baldwin, 2006). The effect of this was that the production and consumption became unbundled, and could happen at different places. Later, in the 1980s and the 1990s when communication costs started falling with inventions such as the internet, it became more attractive for firms to separate production stages (Baldwin, 2006). According to Fukao, Ishito and Ito (2003), Japanese industries were one of the first to unbundle their production stages in the mid-1980s and send parts of it abroad. The reason for this was the large wage discrepancy between China and Japan, as well as the close geographical proximity. Because of this, Japanese firms could easily cut down on production costs and increase their profits and competitiveness. A similar situation unfolded during this time period with American firms using Mexican maquiladora’s for the production of their goods (Feenstra and Hanson 1997). As a result value chains became more global.

In his latest book, Richard Baldwin (2019) has identified a third wave of globalization. In here, he discusses something which he calls “telemigration”. Currently the level of technology in the field of telecommunications is growing rapidly with inventions such as the 5G mobile network and improved machine translation. Baldwin (2019) argues that this can cause a new wave of globalization, where workers do not have to move in order to work abroad. Instead, due to the increased telecommunication technologies they can compete in an ever increasing global labour market.

2.2.2. Recent global value chain trends. Today, GVC are still an intensively researched topic. However, the discussion has shifted in different directions. The focus is not only on why industries would split up their entire value chain or the benefits of this, it also focusses on reasons as to why there are some industries which show signs of reshoring and localisation in America and Europe. This section will discuss some of the literature which identifies recent trends in localisation and the reasons for the changes compared to the previous period.

(8)

4 employed in its global value chain. This value-added approach has as advantage that it avoids the double accounting problem often present in GVC analysis (Koopman, Wang, & Wei, 2014). Timmer et al. (2014) distinguish multiple trends in their research. First of all, the share of foreign value added in total value added increased for 85% of the product chains in the time period 1995 till 2008, indicating that international fragmentation is an ongoing process. The second trend found is that of an increasing importance of capital and high skilled labour as a share of the total value added. Moreover, it was found that between 1995 and 2008, the share of value added from manufacturing chains has decreased from 74% to 56% in high-income countries. Another paper with similar findings is that of Timmer, Dietzenbacher, Los, Stehrer, and De Vries (2015), which shows that foreign value added shares in the automotive manufacturing industry are still growing up until 2011. Moreover, Timmer et al. (2015) report that the global financial crisis did not have any significant lasting effects on the pace of international fragmentation between 2008 and 2011. However, they note that to see more structural effects data for a longer period is necessary. Additionally, a different paper from Los, Timmer, and De Vries (2015) indicate that the foreign value added shares in 2011 are considerably higher than in 1995 by analysing a dataset including 40 countries and 14 manufacturing product groups.

In contrast to these two papers, De Vries et al. (2019) report a trend since 2011 untill 2014 towards localisation in most manufacturing industries within a set of European and North American countries. It is important to note that the paper by de Vries et al. focusses on the time period 2011-2014, whereas the two previously mentioned papers focus on time periods prior to 2008 or up until 2011. They show the most recent trend towards localisation by using the WIOD database and looking at domestic value added shares for a set of 18 different industries. This is important, as it becomes clear that there are significant differences between the industries. For example, manufacturing sectors which are most prone towards going local are the metal and automotive sectors, as well as the clothing sector and the rubber and plastic sector (De Vries et al.). In contrast other sectors such as furniture and leisure, computer and electronics, and printing and media still tend to become more global.

(9)

5 The paper of De Backer et al. (2016) puts recent research into perspective. They provide an extensive overview of the current literature on the reshoring trend as well as its potential effects on economies. They suggest that reshoring news is prone to bias as corporations widely announce this as good news, whereas offshoring is not mentioned in great detail by firms. Due to this, the reshoring trend can be overestimated. They also touch upon an important problem for GVC reshoring analysis. To be specific, there is a lack of data above the plant level. Much of the current empirical research is being done by using data from surveys at the firm level. An example of this is the work of Kinkel (2014), who concludes that reshoring is often a short-term correction of previous decisions in the past five years. However, empirical research done at industry levels is often more accurate due to the absence of selection biases (De Backer et al.) De Backer et al. also show that back-shoring is not a dominating effect by looking at the share of imports for a set of developed countries. In case of strong localisation, imports into a domestic market should be declining. Although the trend has slowed down, in most countries the share of imports in domestic demand are still increasing.

In the end, most papers agree there is some kind of localisation trend going on. However, the important question is how much emphasis should be put on this. Some papers like that of Sirkin, Zinser, and Hohner (2011); Wu and Zhang (2014); and De Vries et al., (2019) argue that this is an important upcoming trend. At the same time, other papers like: Dachs et al. (2019); Bailey and De Propris, (2014); Los, et al., (2015); and Timmer et al., (2014), suggest that it is blown out of proportion and that the world is still becoming more global. There is a significant gap in the literature regarding changes in localisation at the industry level. Much of the research is done based on limited firm-level surveys. Broader research done at the industry level like that of De Vries et al. could help to paint a clearer picture of the ongoing processes in the manufacturing sector.

2.3. Global Value Chain Theories

As GVC have been a large field of interest, there are many theories which aim to explain and predict GVC behaviour. Some of these theories will be explored in this section and will later on be used to develop a set of potential drivers for the localisation phenomenon. When it comes to the reason as to why industries tend to locate in certain countries, it all comes down to the search for a competitive advantage. Because of this, the next theories will explain different ways in which competitive advantages can be gained.

(10)

6 as well as the lower need for asset specific investments. Due to lower asset specific investments from one party, opportunism of the other firm becomes less likely (Gereffi et al.). However, even in the case of high asset specific investments, the production of complex products can still be outsourced. This is due to the effects of repeat transactions, firm reputation, and social norms, which can all be used to reduce the chance of opportunism (Gereffi et al.) TCE predict that, ceteris paribus, firms will always move from high cost to low cost regions (Ellram, Tate, & Petersen, 2013). This means that for example, countries with worse intellectual property rights, which increase the chance of opportunism, are less attractive as countries of production. Literature has argued that changes in transaction costs and risks in the host-country have driven back-shoring (Dachs et al. (2019); Ellram et al., (2013); Fratocchi et al., (2016)). Due to asset specific investments, especially sectors with high levels of technology are affected by these changes (Gereffi et al., 2005). Important to note is that TCE looks at the decision to do something in-house or via a contractor, which not necessarily means to do something local or to off-shore. One could still buy product parts (via a contractor) but do it locally, or keep production in-house but have it off-shored. So, although the location decision and TCE are closely linked by reductions in transaction costs when dealing with firms in one’s own region (McIvor, 2013), they are not entirely the same.

2.3.2. Resource based view. The resource based view (RBV) aims to explain firm

performance by looking at available resources (Wernerfelt, 1984). Essentially, the RBV is used by firms to discover where there competitive advantages lie via a theoretical framework. For the topic of localisation of manufacturing industries, the RBV is of interest as it can help explain why industries shift towards different locations. Through organisational activities, firms are looking to enhance their competitive advantages (McIvor, 2013). Thus, firms will compete over the locations that with the most resources to give them an advantage over their competition (Peteraf, 1993). In the case of localisation, manufacturing sites are moving towards advanced countries that have become more attractive. This can have several reasons, such as an increase in the costs of production in the original country, or a shift in the priority between efficiency seeking and market seeking. An example of this is given in Bailey and De Propris (2014). They mention in their study on manufacturing reshoring in the United Kingdom that it could be that firms put more priority on market seeking motives and less on resource seeking motives, and thus are relocating to advanced countries.

2.4. Linking the Theories to Drivers

(11)

7 technology or a skilled labour force. Below I will first focus on what factors could push manufacturing industries away from countries.

One of the first push factors that come to mind are the increasing labour costs in China. Williamson (2008) state that following transaction cost economics, firms will always move from high cost to low cost regions. However, Bailey and De Propris (2014) state that reshoring in the United Kingdom could be due to market seeking motives. If that is the leading driver, then the increase in Chinese wages will not have much of an effect. A second push factor is increasing transportation costs as described by Behar and Venables (2010). They find that distance matters for trade costs. In terms of localisation, in most cases sourcing intermediate products locally will decrease distance and thus transport costs. Via this way, increasing transport costs can act as a push factor away from foreign sourcing. Following transaction cost economics, Gereffi et al. (2005) discuss another set of push factors. These concern transaction costs such as increased lead time, opportunism, contract costs and having less control on the quality of products. With China having the largest manufacturing value added in the world in 2017 (The World Bank, 2019), as well as being criticized concerning intellectual property theft (Griswold, Boudreaux, 2019), transaction costs economics predicts that manufacturing industries will relocate away from China (Vanchan, Mulhall, & Bryson, 2018).

Important pull factors described by the literature are decreases in production cost due to digitalization of manufacturing in advanced countries (De Backer et al., 2016), and increased production flexibility (Dachs et al., 2019). Increasingly, manufacturing producers want to locate closer to their market. One of the reasons for this is that consumers are demanding more customized products (Dachs et al.). To deliver these, the lead time cannot be too long and thus it is beneficial to localise. Moreover, De Backer et al. emphasize the importance of developments such as sensors, machine-to-machine communication, and artificial intelligence. All of these are technological developments, gradually pushing manufacturing from becoming labour intensive towards becoming capital intensive. At the same time, Dachs et al. have found that the largest reshoring trends are happening in technological complex manufacturing industries. Following transaction cost economics, manufacturing industries will locate in countries in which it is cheapest to produce. Thus, if capital becomes more productive than labour, this might be an explanation for localisation.

3. Hypotheses

(12)

8 cost difference, continuing to act as a pull-factor for local sourcing in manufacturing industries. The hypotheses are to be tested for a set of manufacturing industries which are used in the report of de Vries et al. (2019).

H1: Manufacturing industries in advanced countries are increasingly sourcing locally due to increasing Chinese wages.

H2: Manufacturing industries in advanced countries are increasingly sourcing locally due to increases in production technology.

4. Methodology

The aforementioned hypotheses will be tested by using regression analysis on a series of relationships. The main variables to be included in the econometric model will be outlined in this section. It is important to note that I test the hypotheses both on the country level, as well as the industry level. For these two regressions the same variables will be used on different levels in panel data format. The time period for this study ranges from 2000 until 2014 and considers the following countries: Austria, Belgium, Germany, Denmark, Spain, Finland, France, United Kingdom, Greece, Italy, Netherlands, Portugal, Sweden, and the United States. Moreover, a set of 18 manufacturing industries is used which can be found in Appendix A.

I will run the analysis explained below over the entire time period 2000-2014. Preferably, the analysis is ran over two separate time periods, 2000-2007 and 2011-2014, to see whether the importance of drivers has changed and to exclude the years of global economic recession. However, the latter time period has proven to be too short with a range of only 4 years. Instead, to check for any differences in the pace of localisation I add a time-period dummy for the period 2011-2014 to the regression analysis. I am using the period 2011-2014 as the data shows that the localisation trend was not as strong pre- 2011. De Vries et al. (2019) defined a country as going local when the domestic value added share of 2014 is larger than in 2011 plus 0.02. Using this definition, only 2 out of 16 countries go local in the 2010-2014 period. In contrast, in the period 2011-2014 9 out of 16 countries show a localisation trend.

4.1. Country-level Model

(13)

9 Figure 1

Going Local or Going Global? Period 2000-2007

Notes: Countries shown in light blue are showing a localisation trend. Countries shown in red have continued towards globalization. Data source: WIOD 2016

Figure 2

Going Local or Going Global? Period 2011-2014

(14)

10 It becomes clear from Figures 1 and 2 that there is a significant difference between the two periods. In the first period, only Canada and Spain showed a localisation trend in their manufacturing sector. In the latter time period this changed dramatically, with 9 out of 15 countries showing a trend towards localisation. To discover any relationships, the following model specification will be used:

(1) LVASi,c,t = β0 + β1(WRi,c,t) + β2(WRi.c,t#TRENDt) + β3(Techc,t) + β4(LSt) + δ(Crisist) + εi,c,t

Where LVAS is the change in local value added share in total value added for industry i, country c at time t. WR is the wage ratio of the host country relative to Chinese wages and WR # TREND is an interaction variable between the wage ratio and the period dummy variable trend, which takes a value of one for the period 2011-2014. Tech is the sum of the investment in economic competencies, software database and R&D, taken as a ratio to GDP. LS is a control variable for the level of labour skill in Chinese manufacturing. Moreover, crisis is an intercept dummy variable which obtains the value of one for the years 2008 until 2010 and ε is the error term.

Chinese labour skills are used to correct for an increase in Chinese labour wages due to productivity increases. The inclusion of this variable originates from the fact that a growing level of labour skill can indicate a movement towards the higher value added part of the value chain. In this case, the manufacturing sectors can be drawn towards China and thus produce a larger share of value added. Because of this, I expect that an increase in Chinese labour skills will have a negative effect on the localisation trend in advanced countries.

4.2. Industry-level Model

The second model focusses on the 18 different manufacturing industries as shown in Appendix A. This analysis is important as there can be large within sector differences, as shown in De Vries et al. (2019) where almost 95% of fabricated metals manufacturing showed to be localising, whereas only 9% of other transport equipment has seen a shift towards localisation in 2016. Due to these differences between industries, I run a large set of panel data regressions over different subsets of industries. First of all, I will provide regression results over all separate industries. This is possible due to a time period of 15 years, combined with the second dimension of countries. This results in 14 observations per year per industry, resulting in a total of 210 observations per industry.

(15)

11 Table 1

Consumer Industry Grouping Industry:

Industry Code Manufacture of food products, beverages and tobacco

products C10-C12

Manufacture of textiles, wearing apparel and leather products C13-C15 Printing and reproduction of recorded media C18 Manufacture of computer, electronic and optical products C26 Manufacture of motor vehicles, trailers and semi-trailers C29 Manufacture of furniture; other manufacturing C31_C32

The industry specific model is specified as follows:

(2) LVASi,c,t = β0 + β1(WRi,c,t) + β2(WRi,c,t # TRENDt) + β3(Techct) + β4(LSi,t) + εi,c,t

In here, LVAS is the change in local value added share in total value added for industry i and country c at time t. WR is the wage ratio of the host country relative to Chinese wages and WR # TREND is an interaction variable used to measure the effect of the wage ratio in the period 2011-2014. Tech is the sum of the value of economic competencies, software database and R&D expenditure taken as a ratio to GDP. LS is a control variable for changes in the Chinese labour skills in industry i. Moreover, ε is the error term.

4.3. Data and Variable Specification

The following section focusses on the variables used in the models and their data sources. Moreover, I highlight the expectation of this research for every variable.

The dependent variable of this study is the nominal local value added share (LVAS). It is originating from the WIOD and is transformed to its current form by De Vries et al. (2019). The LVAS is the share of domestic value added in total value added per industry country combination, given annually. An increase in this percentage means that the industry country combination has used a larger share of intermediate products from its own country.

(16)

12 workers will earn a higher wage. If this is the case in China, this might decrease the effect that Chinese labour wages might have on localisation in advanced countries due to a smaller or no change in the cost/productivity ratio. For this reason it is importance to include Chinese labour skill in the model. If I neglect this, an endogeneity problem may appear due to omitted variables. When Chinese labour skill is omitted, the regression error will be correlated with the wage ratio variable, making it endogenous (Hill, Griffiths, & Carter, P. 407, 2014). Following hypothesis 1, I expect that there is a positive relationship between changes in wages and localisation.

The second explanatory variable is the level of technology of the home country and is used to test the second hypothesis. The data used for this variable is originating from the INTAN INVEST database (version 2019). This is a harmonized dataset based on expenditure based measurements of the value of intangible assets (Corrado, Haskel, Jona-Lasinio, & Iommi, 2016). The data is given as current, annual, sector data per country. This means that it provides data on the manufacturing sector as a whole, but not on an industry specific level. The variable taken from this dataset to use as a proxy for the level of technology is the sum of investment in economic competencies, computer software and databases, and R&D, as a ratio of GDP. This is further explained in Appendix B. The INTAN INVEST database gives all of these in nominal values of national currency. To be able to create the ratio, using GDP in USD, I convert the INTAN INVEST data to USD using the same exchange rates as above. I do this to make it easier to spot measurement errors in the level data. However, due to this a small measurement error is possible due to exchange rate differences for the separate parts of the ratio. I expect that the level of technology for a country will have positive effects on localisation. Important to note that although the value is given for all manufacturing sectors together, it is not for corresponding sub-sectors of manufacturing. Thus, for year t and country c it enters with the same value for all industries. This is a clear limitation to this study, as the level of technology is different for every sector. However, data on the same industry classification has not been found.

The final variable is a control variable and is defined as Chinese Labour Skill. This variable is incorporated into the model to control for endogeneity. It is likely that the change in wages is caused by a change in the skill-level of Chinese labourers. At the same time, an increase in labour skill can act as a pull factor for the manufacturing location decision. Excluding this variable could lead to invalid regression results. If this variable is not incorporated in the model, the least square estimator of the effect of an increase in wages is likely to be positively biased and inconsistent. This means that the coefficient will become larger than it actually is, and that this bias will not disappear even in the case of large samples (Hill et al, p. 407, 2014). I expect that an increase in Chinese labour skill will decrease the LVAS in advanced countries. More information about this variable can be found in Appendix C

(17)

13 dummy variable to interact with the explanatory variables to measure if the coefficients of these variables differ for this time period.

5. Diagnostic Checks and Discussion

In this section, the main analysis and statistical tests will be done using the earlier described data and variables. First of all, I put the data in a long panel data format. Afterwards, I grouped the country and industry codes together using the “egen” command in Stata. This group is then used together with the year variable to set the data to panel data. Important to note is that for the variable on Chinese wages, there is no data for the industry “Repair and installation of machinery and equipment”, and thus this industry is dropped.

5.1. Data Description

Below in Table 2, the summary statistics are presented. It becomes clear that the data set is balanced and complete, with no observations missing. Moreover, the mean LVAS over the entire period is 68.95%. When comparing maxima and minima with the means, the LVAS share minimum of 8.5% catches the eye. This low value is the LVAS of the Netherlands in the industry “Manufacture of coke and refined petroleum products (C19)” for the year 2012. Further inspection is done via a scatterplot of LVAS for industry C19 over the entire time period. Figure 3 confirms the reason why De Vries et al. (2019) have dropped this industry in their research. They argue that the volatility within this industry is too high. In 2012, the highest point is at 66.7, where the lowest percentage is only 8.5. Because of this high volatility, the industry C19 is dropped from the research leaving the dataset with 3570 observations.

Table 2

Descriptive Statistics

Variable Obs Mean Std.Dev. Min Max

LVAS 3780 68.957 13.774 8.499 93.381 Wage Ratio 3780 15.62 9.187 .785 71.214 Software 3780 .002 .002 0 .009 Competencies 3780 .008 .003 .003 .021 R&D 3780 .011 .008 0 .034 Technology 3780 .021 .012 .005 .06 Labour Skill 3780 .03 .019 .003 .095

(18)

14 Figure 3

Local Value Added Share of Industry C19 Plotted over Time

Next, I take a look at the correlation matrix in Table 3 to get a better feel for the data. The wage ratio has a positive correlation with LVAS of 0.1932. As an increase in the wage ratio means that the wage gap increases, this is rather unexpected. This means that increases in China’s wage rate are negatively associated with the LVAS in advanced countries, which is the opposite of what is hypothesised. Moreover, all three measures of technology have a negative correlation with the LVAS. Intuitively, this does not seem right. When an industry reaches higher level of technology, it becomes easier to produce certain products. However, an opposite train of thought should also be considered. De Vries et al. (2019) find that industries that go local create more jobs, but decrease in productivity. One explanation for the decrease in productivity is that capital gets substituted by labour. Moreover, the report of De Vries et al. finds that most localisation takes place in industries that are labour intensive. A negative sign for the level of technology indicates that indeed, capital is being substituted when industries localise.

(19)

15 used within industries, one single number is not specific enough to catch all changes in the level of technology. Table 3 Matrix of correlations Variables (1) (2) (3) (4) (1) LVAS 1.000 (2) Wage Ratio 0.255 1.000 (3) Technology -0.093 0.161 1.000 (4) Labour Skill -0.419 -0.315 0.033 1.000 Note: LVAS stands for local value added share.

In the next section I graph the variables on the industry- and country level. The reason for this is that it becomes clear that there are large between industry and country differences. In Figure 4 below, I graph a trend line of the LVAS over the time periods 2000-2010 and 2011-2014 per industry. By doing this, the overall trend during these periods become clear. Here, it can clearly be seen that something has changed in the downward trend. For most industries, the LVAS has stagnated or started to increase since 2011. However, some industries also show barely any change. An example of such a sector is C30, “other transport equipment”, which is in line with the findings of De Vries et al. (2019).

Furthermore, Appendix D2 shows a Figure which describes the LVAS over 2011-2014 for industry C24 per country. It becomes apparent that Denmark, the United Kingdom, and the Netherlands are among the largest contributors for the increase in LVAS. According to the United States Energy Information Administration, this industry is one of the largest industrial energy users. Consequently, many of the countries increasing in LVAS have low electricity cost prices, except for the United Kingdom which are relatively high (Eurostat, 2019). The corresponding Figure can be found in Appendix D1. This could indicate that besides labour costs and the level of technology, energy costs are an important factor for localisation as well. However, due to data limitations this could not be implemented into the model.

Next up, I take a look at the change in the wage ratio per industry over the years. The results of this can be found in Figure 5. The change of the wage ratio over time is much as expected, with a steady decrease over the years for all industries. Interesting that most industries experience an increase in the wage ratio in or around the year 2004. This is especially clear in C31_C32 “Manufacture of furniture; other manufacturing”.

(20)

16 Figure 4

Trend Line of the LVAS per Industry over Time

When looking at country and industry specific graphs for the three technology variables it becomes clear that Finland and Sweden have the highest level of technology, ranging between the 4% and 6%, in contrast to most countries such as the U.S. that hover around the 2%. Important to note is that most countries, although slowly, do increase this ratio over time. For the Chinese labour skill variable no surprises are found, and the values over time are increasing for all industries.

(21)

17 Figure 5

Wage Ratio per Industry over 2000-2014

However, some industries do not show much of a correlation. These are C21, C26, and C27: “Manufacture of basic pharmaceutical products and pharmaceutical preparations, Manufacture of computer, electronic and optical products, and Manufacture of electrical equipment” respectively. According to the European Competitiveness and Sustainable Industrial Policy Consortium (ECSIP, 2013), this group of sectors has become increasingly capital intensive and reliant on technology leading up to the year 2013. This can explain why changes in the wage ratio do not have much of an effect on the LVAS in advanced countries.

(22)

18 Figure 6

Scatterplot of LVAS on the WR – Years 2011-2014 per Industry

As the correlation matrix has indicated, the correlation between the level of technology and the LVAS is negative. This holds when considering the industry level. Only C20 and C26 have slight positive correlations. However, the trend line is close to being flat for all industries. This could indicate that the level of technology does not have a large impact on the decision for industries to localize. The Figure that explains this relationship can be found in Appendix D4.

More interesting to look at is the results of the scatterplot between the Chinese skill level and the wage ratio as presented in Figure 7. For most industries there is a large negative correlation, indicating that as the wage ratio decreases, the level of skill increases simultaneously. This signals that including the level of skill for Chinese manufacturing workers in the model might help to reduce any omitted variable bias for the wage ratio variable.

5.2. Statistical Checks

(23)

19 Figure 7

Scatterplot of the Chinese Manufacturing Labour Skill and the Wage Ratio

5.2.3. Multicollinearity. First, I check the model for presence of multicollinearity. I do this by running a standard regression with the variables as specified in equation 1 and 2 while including all four potential technology variables one by one. To double check, I run a second test. I estimate a fixed effects regression by including a dummy for each year in an ordinary least squares regression. The VIF values are presented in Appendix E1-E5. The reported VIF values are all around one, and thus indicate that multicollinearity is not a problem present in the model. When retrieving the VIF values for the second test, again no multicollinearity is found. The only VIF values that could indicate multicollinearity are considering the dummy variable crisis and the interaction variable between the wage ratio and the time trend dummy. However, this is not surprising as a dummy for each year is included in the second regression.

(24)

20 run a regression on their lags and test whether or not the coefficient of the lagged variables are equal to -.5. I find that this is not the case and the null hypothesis of no present serial correlation is rejected with a P-value of 0.000. To control for both heteroscedasticity and serial-correlation problems, I use the cluster option in all of my models.

5.2.5. Fixed effects or random effects. Using the models specified in (1) and (2), I preform a test of overidentifying restrictions to help decide whether the fixed effects or random effects specification would most suit my data. I choose to do this test over a standard Hausman test following Wooldridge (p.291, 2002) who advices a test that is able to account for heteroscedasticity and serial correlation in case these are present. Using the command xtoverid in Stata, I am able to run a specification test using clustered standard errors. The results for this test can be found in Appendix F1. The test reports a P-value of 0.000, and thus rejects the null-hypothesis that the random effects model is the preferred type. For this reason, I will use the fixed effects specification for my models. One of the advantages that the fixed effects model has is that it accounts for unobserved time fixed effects and can take account of individual-specific heterogeneity (Hill et al., p.548, 2014).

5.3. Country-level Analysis

In this section, I will present the findings for the country-level analysis. Regression results will be presented on a per country level. I have run the full regression, including a time period dummy for 2011-2014, as well as a crisis dummy for the period 2008-2010. Moreover, an interaction variable for the period 2011-2014 and the wage ratio is added. When testing which technology specification is preferred, no large differences were noticed on the country level. Due to this reason, I will use the combined technology variable of all three technology measures in further regressions. To achieve the best comparability per country, I will display the preferred specifications in two separate tables. The results of the main specifications can be found below in Table 4 and 5. The full range of specifications can be found in Appendix G.

The first thing that catches the eye in Table 4 is that the wage ratio is only significant in three countries. These are the Netherlands and Portugal at the 10% level, and Greece at the 1% level of significance. Interestingly enough, the coefficient is positive, rather than the

hypothesized negative coefficient. With a coefficient of 0.388, it would mean that a 1% increase in the Greece-China wage ratio will increase the LVAS in Greece by 0.388%. In Table 5 I include an interaction variable between the wage ratio and a trend variable which is 1 for the period 2011-2014. Now, Belgium also reports a significant and positive coefficient for the wage ratio. However, Spain does not anymore. The interaction variable is significant for 8 out of 14 countries at the 5% and 1% level.

(25)

21 Table 4

(26)

22 Table 5

(27)

23 Hypothesis 2 states that manufacturing industries in advanced countries are increasingly sourcing locally due to increases in production technology. The data suggests that the level of technology is indeed, slowly, increasing for most countries in the sample. Moreover, the results in the regression analysis vary. In Table 4, 7 out of 14 countries show that the level of technology is a significant predictor of the LVAS of the home country. However, the sign varies showing positive signs for the United Kingdom and the United States. When excluding the general 2011-2014 trend variable, and including the wage ratio # trend interaction variable, the same seven countries report a significant p-value without any sign changes. Because of this, I conclude that the variable is robust to changes in the model. The country-level model rejects hypothesis 2. In contrast, an increase in the level of technology is associated with decreases in the LVAS of advanced countries. The reason for this could be that industries that go local are the most labour intensive industries. Moreover, it indicates that capital gets substituted by labour. A country which invests less in technology will see an increase in the LVAS.

Furthermore, the effect of the Chinese labour skill variable seems to differ in importance per country. In the first model, it is significant for 11 out of 14 countries all of which report a negative sign. In Table 5, 12 out of 14 countries show a significant P-value and a negative coefficient for the Chinese labour skill variable. Over the two models, the majority of countries show a significant and negative coefficient as expected. There are differences in the size of the coefficient. According to Table 4 a one percentage point increase in Chinese labour skill will decrease Germany her LVAS by 2.8%, whereas this is 3.2% in Table 5. The skill level of Chinese labourers seems to have a significant negative effect on the LVAS of advanced countries. With coefficients ranging between -0.464 and -3.242 this is not an effect that should be ignored.

Moreover, I include the crisis dummy in Table 4 and Table 5, as well as the period trend in Table 4. In Table 4, the crisis dummy is significant for 11 out of 14 countries and reports a negative coefficient for 10 of those. In Table 5, only 6 out of 14 countries report a significant P-value for the crisis dummy variable, of which one with a positive coefficient. This indicates that the effect of the crisis differs per country and should be controlled for.

When looking at the trend variable in Table 4, 10 countries show that this time period has a significant influence on the LVAS share via variables not incorporated in the model. All of these report a negative coefficient. This intercept indicator variable indicates that for most countries, the LVAS is lower for the years 2011 until 2014.

(28)

24 have access to data ranging from 2000 until 2014, which gives a time period of five years where the trend is present. The model could be improved in robustness by using data for a longer time period, when this becomes available. Assuming the model is consistent, this should give results which are closer to the true values.

Figure 8

Trend Line of LVAS over 2000-2010 and 2011-2014

5.4. Industry-level Analysis

(29)

25 variable is insignificant for C21, C23, C26, C30 and C31_C32. These five industries all have relatively high capital intensity, and thus it is not surprising that these show a non-significant P-value. Again, the negative coefficient of WR*TREND is larger than the positive wage ratio coefficient for all industries if significant, except for C25, “Manufacture of fabricated metal products, except machinery and equipment”, a capital intensive industry. Combining the effect of the two variables, a decrease in the wage ratio leads to a decrease in the LVAS of industries pre-2011, whereas the same change leads to a positive change in the LVAS post 2010. Following this logic, I conclude that the wage ratio has reached a tipping point after 2010, which fuels localisation of industries in advanced countries. This is also robust to changes in the model. The interaction variable remains significant and negative for most specifications.

Again, the technology variable shows results that reject hypothesis 2. It appears that an increase in the level of technology is negatively influencing the LVAS in 6 out of 18 industries. When including a trend interaction for the level of technology, the results remain largely insignificant (see Appendix H6). Moreover, the industries for which it is significant, C17 at the 5% level, and C25 as well as C27 at the 10% level, the overall effect including the technology coefficient is still negative. For example, the interaction variable for C17 reports a coefficient of 0.8589 and for the technology variable negative 2.972. Due to this, I conclude that the increase in production technology in advanced country does not have a positive effect on the LVAS. Rather, the regression results show that there is in fact a negative relationship between the level of production technology and the LVAS for some industries. For most industries, changes in the level of technology have no significant effect. However, important to note is that for the period 2011-2014 this effect is significantly smaller.

The Chinese labour skill variable is consistent throughout this analysis and the previous analysis. It reports a negative and significant coefficient for 15 out of 17 industries. Moreover, it is robust to changes in the model, as can be seen in Appendix J. The model provides evidence for the fact that increases in Chinese labour skill act as a pull-factor away from advanced countries. However, as the data does not include the LVAS of China, I cannot conclude, but only assume that the products are then sourced from Chinese industries. This would be the logical explanation.

(30)

26 Table 6

(31)

27 An explanation for the insignificance of the technology variable is the lack of industry-level data for this variable. By taking the entire industry sector industry-level of technology, the model fails to anticipate that every technology uses different technologies. Thus, a technological breakthrough in the manufacturing of food products is not related to production capabilities in the basic metals sector. Moreover, the technology variable shows a low variance, being 0.206 at its maximum, and 0.013 at its lowest. This is shown in Table 7. A low variance in a predictor variable can greatly lower the precision of the estimates (Kennedy, 2005). Because of this, the level of technology as measured in this paper might not be a good predictor of LVAS. This explanation also holds for the country-level model.

Table 7

Descriptive statistics - variance by Country

technology AUT .206 BEL .116 DEU .093 DNK .154 ESP .016 FIN .53 FRA .104 GBR .047 GRC .019 ITA .03 NLD .037 PRT .013 SWE .16 USA .016

5.4. Industry Sub-group Analysis

In this section I provide the regression results with a consumer product group dummy. This dummy is one if the industry is considered a primarily consumer good focused industry, and zero otherwise. Important to note is that, as this variable is time-invariant, I cannot use the FE regression. Because of this, I have to choose between a simple pooled OLS estimation and the random effects (RE) estimator. I do this by running a Lagrange multiplier test as suggested in Hill et al. (p.554, 2014). This test rejects the null-hypothesis with a P-value of 0.000, and thus the RE model is preferred over a pooled OLS estimation. However, it is important to consider that the earlier I noted that the FE model is preferred over the RE model. This means that it is likely that the RE estimation is inconsistent due to a correlation between the error term and other explanatory variables (Hill et al., p. 558, 2014). Because of this, the results should be interpreted with caution. However, this will not lead to the model reporting the wrong signs.

(32)

28 analyses, there are no changes in the sings of coefficients. The wage ratio again enters positively for the entire time period. However, it becomes negative when interacted with the years in which the localisation trend is observed. Additionally, the coefficient of the interaction variable is smaller than the coefficient of the corresponding wage ratio. Thus, for the years 2011-2014, an increase in Chinese wages holding domestic wages constant, will cause an increase of the LVAS in advanced countries. Moreover, Technology is again negatively associated with the LVAS, as well as Chinese labour skill. The crisis dummy is insignificant for the consumer focused industries, yet significant for the investment and intermediary industry group.

Table 8

Random Effects Regression Results – Whole Sample

(33)

29 argued. However, this variable is not robust to changes in the model as in the second and third regression, the variable is insignificant.

One has to keep in mind that the manufacturing industries used in this research entail a wide range of activities. Because of this, it is not possible to make a hundred percent accurate group of consumer product focused industries. With this in mind the model above does not give an exclusive conclusion, but does show that there is a good possibility of a link between localisation and the degree of which an industry focusses on consumers.

6. Discussion & Caveats

In this paper I have tried to answer the following research question “What are the drivers behind the recent trend of manufacturing localisation in advanced countries?” I have investigated this by using different panel data models, both on the country and industry level data over the period 2000-2014. By running the panel data over 14 countries and 17 industries, I have found mixed results. First of all, the decrease in the wage ratio between specific advanced countries and China has led to an increase in localisation for the time period 2011 until 2014. However, during the 2000-2010 periods this effect is insignificant. This finding is in line with hypothesis 1.

In contrast, hypothesis 2 is rejected. The results in this paper indicate that if there is a relationship, it is a negative relationship between the level of technology and the LVAS. However, this negative relationship can be explained by considering that industries that go local often see a decrease in productivity. This indicates that capital is substituted for labour, which causes a negative relationship. This result is dependant both on industry and country, with insignificant findings in most industries. At the country level, this variable is robust to changes in the model. However, results should be interpreted with caution, due to a lack of industry-specific data as well as a low variance within the variable. Additionally, by including an intercept indicator variable, I conclude that consumer focused industries show a higher degree of localisation.

This study included dummy variables for the recession years 2008 until 2010, and a time period dummy for the years 2011-2014, in which the localisation trend is observed. The recession years are found to negatively influence the LVAS of manufacturing industries, whereas the trend does not seem to be significant overall. Lastly, this report finds that the level of Chinese labour skill is a significant predictor of the LVAS in advanced countries. As the skill of Chinese labourer’s increases, the LVAS of advanced countries will decrease. This is most likely due to manufacturing industries increasingly sourcing from China. However, this report has not studied where industries source from, if not from the country in which it is located. The gap that this report leaves regarding this might be of interest to study. An example research question could be: To what extent does the increase in skill of Chinese labourers attract more manufacturing production?

(34)

30 available up to 2014, and going back until 2000. Additionally, because the trend is still at its early stage and is relatively weak when comparing it to the prior trend of globalisation, regression results will likely be biased. However, assuming the models used are consistent and the localisation trend continues, the analysis should be re-ran when more data is available to retrieve results closer to its true value. Moreover, the consumer dummy variable cannot be run with the fixed effects model as it does not change over time. Although the fixed effects model is the preferred choice, the random effects model does still give the correct signs. Lastly, literature indicates other variables that can influence the localisation decision of manufacturing industries, such as the level of intellectual property rights in the current host country and the energy costs in a country. However, due the difficulty of measuring some of these variables they are not included. If in the future these variables can be measured they should be incorporated into the model.

(35)

31 References

De Backer, K. et al. (2016), “Reshoring: Myth or Reality?”, OECD Science, Technology and Industry Policy Papers, No. 27, OECD Publishing, Paris.

Bailey, D., & De Propris, L. (2014). Manufacturing reshoring and its limits: the UK automotive case. Cambridge Journal of Regions, Economy and Society, 7(3), 379-395.

Baldwin, R. E. (2006). Globalisation: the great unbundling (s). Economic Council of Finland. Baldwin, R. (2019). The Globotics Upheaval: Globalization, Robotics, and the Future of Work.

Oxford University Press.

Baum, C. F. (2001). Residual diagnostics for cross-section time series regression models. The Stata Journal, 1(1), 101-104.

Behar, A., & Venables, A. J. (2011). Transport costs and international trade. Handbook of transport economics, 97-115.

Corrado, C., Haskel, J., Jona-Lasinio, C., & Iommi, M. (2016). Intangible investment in the EU and US before and since the Great Recession and its contribution to productivity growth.

EIB Working Papers,No. 2016/08.

Dachs, B., Kinkel, S., Jäger, A., & Palčič, I. (2019). Backshoring of production activities in European manufacturing. Journal of Purchasing and Supply Management.

De Vries, K., Maselli, I., Erumban, A., Lundh, E., and Ozyildirim, A. (2019) Going Local: Changing global value chains and the impact on revenue and jobs. The Conference Board, Research report 1688.

Drukker, D. M. (2003). Testing for serial correlation in linear panel-data models. The stata journal, 3(2), 168-177.

Ellram, L. M., Tate, W. L., & Petersen, K. J. (2013). Offshoring and reshoring: an update on the manufacturing location decision. Journal of Supply Chain Management, 49(2), 14-22. Eurostat, Energy Statistics – Prices of Natural Gas and Electricity (2019) Electricity prices for

non-household consumers. Retrieved from:

https://ec.europa.eu/eurostat/web/energy/data/database

Feenstra, R. C., & Hanson, G. H. (1997). Foreign direct investment and relative wages: Evidence from Mexico's maquiladoras. Journal of international economics, 42(3-4), 371-393.

Fukao, K., Ishido, H., & Ito, K. (2003). Vertical intra-industry trade and foreign direct investment in East Asia. Journal of the Japanese and international economies, 17(4), 468-506.

Fratocchi, L., Ancarani, A., Barbieri, P., Di Mauro, C., Nassimbeni, G., Sartor, M., et al. (2016). Motivations of manufacturing reshoring: an interpretative framework. International Journal of Physical Distribution & Logistics Management, 46(2), 98-127.

Gereffi, G., Humphrey, J., & Sturgeon, T. (2005). The governance of global value chains. Review of international political economy, 12(1), 78-104.

Griswold, D., & Boudreaux, D. J. (2019). How the United States Should Respond to China’s

(36)

32 https://www.dentonisd.org/cms/lib/tx21000245/centricity/Domain/82/APA%206th%20File s/APAelectronicsources.pdf

Hill, R. C., Griffiths, W. E., & Lim, G. C. (2012). Principles of econometrics (4th ed.). Asia: John Wiley & Sons.

Irwin-Hunt, A. & Shehadi, S. (2019, April 18). On the move: manufacturing’s return to the developed world. FDI Intelligence.

Kennedy, P. E. (2005). Oh no! I got the wrong sign! What should I do?. The Journal of Economic Education, 36(1), 77-92.

Kinkel, S. (2014). Future and impact of backshoring—Some conclusions from 15 years of research on German practices. Journal of Purchasing and Supply Management, 20(1), 63-65.

Koopman, R., Wang, Z., & Wei, S. J. (2014). Tracing value-added and double counting in gross exports. American Economic Review, 104(2), 459-94.

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

McIvor, R. (2013). Understanding the manufacturing location decision: the case for the transaction cost and capability perspectives. Journal of Supply Chain Management, 49(2), 23-26.

Peteraf, M. A. (1993). The cornerstones of competitive advantage: a resource‐based view. Strategic management journal, 14(3), 179-191.

Reinhard, M., Strobel, T., & Vieweg, H. G. (2013). Study on the competitiveness of the electrical and electronic engineering industry. ECSIP.

Sirkin, H. L., Zinser, M., & Hohner, D. (2011). Made in America, again: Why manufacturing will return to the US (pp. 3-5). Boston: Boston Consulting Group.

The World Bank, World Bank national accounts data, and OECD National Accounts data files (2019). Manufacturing, value added (current US$). Retrieved from:

https://data.worldbank.org/indicator/NV.IND.MANF.CD?locations=CN-US

Timmer, M. P., Dietzenbacher, E., Los, B., Stehrer, R. and de Vries, G. J. (2015), "An Illustrated User Guide to the World Input–Output Database: the Case of Global Automotive Production", Review of International Economics., 23: 575–605

Timmer, M. P., Erumban, A. A., Los, B., Stehrer, R., & De Vries, G. J. (2014). Slicing up global value chains. Journal of economic perspectives, 28(2), 99-118.

Vanchan, V., Mulhall, R., & Bryson, J. (2018). Repatriation or reshoring of manufacturing to the US and UK: dynamics and global production networks or from here to there and back again. Growth and Change, 49(1), 97-121.

Wernerfelt, B. (1984). A resource‐based view of the firm. Strategic management journal, 5(2), 171-180.

(37)

33 Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data. MIT press. Wu, X., & Zhang, F. (2014). Home or overseas? An analysis of sourcing strategies under

competition. Management Science, 60(5), 1223-1240.

(38)

34 Appendix A – Industry Classification

Table A1

Industry Classification

Industry

Industry Code Manufacture of food products, beverages and tobacco products C10-C12 Manufacture of textiles, wearing apparel and leather products C13-C15 Manufacture of wood and of products of wood and cork, except furniture;

manufacture of articles of straw and plaiting materials C16

Manufacture of paper and paper products C17

Printing and reproduction of recorded media C18

Manufacture of coke and refined petroleum products C19

Manufacture of chemicals and chemical products C20

Manufacture of basic pharmaceutical products and pharmaceutical preparations C21

Manufacture of rubber and plastic products C22

Manufacture of other non-metallic mineral products C23

Manufacture of basic metals C24

Manufacture of fabricated metal products, except machinery and equipment C25 Manufacture of computer, electronic and optical products C26

Manufacture of electrical equipment C27

Manufacture of machinery and equipment n.e.c. C28

Manufacture of motor vehicles, trailers and semi-trailers C29

Manufacture of other transport equipment C30

Manufacture of furniture; other manufacturing C31_C32

Repair and installation of machinery and equipment C33

(39)

35 Appendix B – Technological Change

The level of technology is influenced by multiple variables. To reduce noise and get a value closer to the true value of the technological level in a countries manufacturing industry I use the sum of three separate measures. Moreover, Corrado et al. (2016) report that these three variables are the major components of intangible assets. In the estimations, I have used these three variables separately as well as combined. Separately, the variables are not robust to changes in the model. For this reason, I have chosen to use the variables taken together in one.

First of all, I use the amount of investment in computer software and databases, as well as the investment in R&D. To increase the level of technology, investment in computer software as well as investment in research and development is necessary. Moreover, I include the level of economic competencies. Corrado et al. (2016) report that they include this variable in their database as it captures knowledge assets that firms use to run their business. It includes for example investments on training of personnel.

(40)

36 Appendix C – Chinese Labour Skill

(41)

37 Appendix D – Additional Figures

Appendix D1

Price of Electricity per Country – 2014 Prices in Euro

Source: Eurostat (2019), Electricity prices components for non-household consumers - annual data. Consumption less than 20 MWh - band IA

Appendix D2

Time Trend of Local Value Added Share – Years 2011-2014 per Country

(42)

38 Appendix D3

Cross-correlation of Local Value Added Share over Wage Ratio – Years 2000-2010 per Industry

Appendix D4

(43)

39 Appendix D5

(44)

40 Appendix E – VIF Values

Appendix E1 VIF Values VIF 1/VIF Trend 1.64 0.612 Wage Ratio Crisis Labour Skill Technology 1.54 1.27 1.27 1.14 0.647 0.786 0.878 0.895 Mean VIF 1.112 .

VIF Larger than 10 = sign of multicollinearity

(45)

41 Appendix E5 VIF Values VIF 1/VIF Wage Ratio 1.69 0.591 Software Labour Skill Crisis Wage Rate # Trend Year 2001 1.15 1.20 5.24 4.81 1.88 0.866 0.833 0.191 0.208 0.530 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Mean VIF 1.88 1.89 1.89 1.89 1.90 1.92 1.90 1.90 1.88 4.04 3.53 3.44 3.33 2.53 0.531 0.529 0.529 0.528 0.527 0.520 0.526 0.527 0.533 0.247 0.283 0.290 0.301 .

(46)

42 Appendix F –Test of Overidentifying Restrictions

Appendix F1

Test for Overidentifying Restrictions

_Coefficient

chi2 (238) 31551.77

Prob>chi2 ₀

(47)

43 Appendix G – Fixed Effects Regression Results by Country

Appendix G1

Fixed Effects Regression Results - by Country

Appendix G2

(48)

44 Appendix G3

(49)

45 Appendix G4

(50)

46 Appendix G5

(51)

47 Appendix H – Regression Results by Industry

Appendix H1

Fixed EffectsRegression Results - by Industry

Appendix H2

(52)

48 Appendix H3

Fixed Effects Regression Results - by Industry

Appendix H4

(53)

49 Appendix H5

(54)

50 Appendix H6

What Are the Driving Factors Fuelling Global Value Chain Localisation in Advanced Countries: A Country and Industry Level Analysis

What Are the Driving Factors Fuelling Global Value Chain Localisation in

Advanced Countries: A Country and Industry Level Analysis

Table of Contents