Master Thesis
MSc Economic Development and Globalization
Global Value Chain Participation: A driver of regional economic
performance? – The case of European regions
University of Groningen Faculty of Economics and Business
June 2020
Author: Sven Lusti
Student Number: S3621677
Mail address: s.e.lusti@student.rug.nl Supervisor: Prof. Dr. B. Los
2
Abstract
This thesis studies the impact of regional global value chain participation on regional economic performance, measured as the yearly GDP growth and labour productivity. Using the EUREGIO database, a global input-output table with regional details, the global value chain participation index for 245 NUTS2-European regions for the period 2000-2010 is computed. The results, from a fixed effect panel model estimation, show that the immediate effects of an increased participation in global value chains are indeed driving economic performance, more so for initially less developed regions. By extending the observed period, based on a cross-sectional data analysis, the long-run effects on economic performance show that the immediate effects can be reversed, and regions get locked-in into productivity reducing activities.
Keywords: Global Value Chains, Global Value Chain Participation, Input-Output Analysis,
3
Table of Content
Abstract ... 2 Table of Content ... 3 1. Introduction ... 4 2. Literature Review ... 52.1. Regional Economic Performance ... 6
2.1.1. Drivers of Regional Economic Performance ... 6
2.2. What are Global Value Chains? ... 7
2.3. Case studies ... 8
2.4. Governance and Upgrading ... 9
2.5. Macroeconomic Approach to GVC ... 11
2.6. Hypotheses ... 13
3. Methodology ... 14
3.1. Model ... 14
3.2. EUREGIO – Global IO table with regional detail ... 16
3.3.3. Data Limitations ... 18
3.3. Methods ... 18
4. Empirical Analysis ... 22
4.1. Data Summary and NUTS2 RGVC-PI Indices ... 22
4.2. Model selection ... 24
5. Results ... 25
5.1. Discussion and Testing of Hypotheses ... 30
5.2. Limitations ... 32
6. Conclusion ... 33
References ... 35
4
1. Introduction
With the take-over of Škoda by VW in 1991 and the establishment of multiple joint ventures and greenfield investments, the participation in the value chain of the VW-Škoda car manufacturing was enabled to multiple domestic Czech firms. Due to the inclusion in the VW global value chain, companies were forced to upgrade their production to improve both products and processes to comply with the standards set by VW (Pavlínek & Ženka, 2011). The evidence of Czech firms suggests a positive evolution of firm upgrading, with positive spillovers to the regional development, especially the region ‘Mladá Boleslav’, around the manufacturing assembly plant, has benefitted economically (Pavlínek & Janák, 2007). The industrial clustering around the assembly plant made surrounding regions more attractive as they gained a comparative advantage in low-wage labour, attracting further FDI. This showcase of benefits, from participating in an international value chain, is limited, as not all of the gains and value added remained in Czechia, due to the foreign ownership of Škoda leading to a repatriation of profits to the VW headquarters, in Germany (Pavlínek & Ženka, 2011). Nevertheless, it provides solid ground to check whether there is a systematic relationship between participating in global value chains and regional economic development.
The increased level of production fragmentation since the second unbundling, the decrease in communication and coordination costs through better ICT technology (Baldwin, 2015), has changed the organisation and geographical location of production steps towards global value chains (Timmer et al., 2014; Baldwin & Venables, 2010). In this thesis global value chains (GVC), are defined as all value added activities directly and
indirectly involved in the production of a final manufacturing good. These activities include
both goods and services, like the extraction of natural resources, the distribution and marketing, and the final assembly in factories of a manufacturing good. These activities are conducted in different geographical locations and are generally managed by a lead firm (Gereffi et al., 2005). GVCs allow regions or firms to join existing networks and markets more easily, by specialising in a specific value adding activity, without having to build the complete production chain on their own (Baldwin, 2011). Positive correlations between participating in GVCs and economic performance have been found for the Thai car manufacturing industry (Wad, 2009) and countries participating in international trade via GVC in general (Ignatenko et al., 2019). In this thesis the focus lies on a regional analysis, in which regions are defined as the sub-national spatial aggregation level of a
country.
5 Gereffi et al., 2005). Even though the studies in GVC have advanced, there seems to be a lack of regional analysis of GVC and macro-economic evidence of the gains and losses in participating in GVCs (Johnson, 2018). Given the large degree of heterogeneity in terms of productivity and income among EU regions (Becker et al., 2010) as well as cross-regional divergence (Farole et al., 2011), the effect of participating in GVCs on economic performance will be tested on a systematic level for European regions.
By creating a regional GVC-participation index (RGVCPI), defined as the regional value
added in GVC as the share of total regional GDP, the effect of this index on economic
performance indicators will be tested. The indicators of interests are the growth rate of regional GDP, as well as regional labour productivity, measured as regional GDP per hour worked. The RGVCPI makes use of existing techniques in input-output analysis, allowing to break down the regional value added in a given GVC, defined by the geographical location and sector in which the final good is produced (Los et al., 2015; Los et al. 2016). In order to test the research question: is the regional participation in global value chains,
between 2000-2010, driving the economic performance in European regions?, this thesis
makes use of the newly established EUREGIO database. The EUREGIO is a global input-output table with regional detail for 249 EU-25 NUTS2 regions1, containing both data on regional production structures and international trade flows covering 14 industries (Thissen et al., 2018). This unique combination, to the extent of my knowledge, makes it a novelty and allows for a regional analysis of GVC. The period 2000-2010 allows for an interesting period of observation as it has been marked by a fast growth in international trade, largely driven by GVC (Johnson and Noguera, 2012; Baldwin and Lopez-Gonzales, 2013).
The rest of the paper is structured the following way, section 2 provides a discussion about traditional ways of measuring regional economic performance, what GVC analysis contributes and possible channels how regional participation in GVC can affect economic performance. Section 3 will provide the model for the empirical analysis, as well as the data and methods of computing the regional GVC participation indices (RGVCPI). Section 4, will provide descriptive statistics, rough correlation trends and the estimation techniques. The empirical results will be discussed in section 5 and section 6 will provide policy implications as well as concluding remarks for further research in that area.
2. Literature Review
The literature review is split into interrelated subparts, firstly, explaining the traditional drivers of regional economic performance and how the GVC literature helps contributing to the analysis of regional economic performance. Secondly, visualizing the importance of understanding GVCs based on case studies of specific products and value chains. Thirdly, two main theoretical frameworks of GVCs and their implications on performance are introduced: governance and upgrading. Fourthly, advances in the measurement of 1 NUTS 2 correspond to the official sub-national spatial aggregation level of Eurostat defined as “basic regions for the application of regional policies”, more information can be accessed on:
6 GVCs on a macro level are quickly summarized, focusing on the intuition behind them, as they will be further elaborated in the methodology section. The literature review ends with summarizing the advances and gaps in the literature of GVCs and based on the insights the relevant hypotheses for this thesis are derived and discussed.
2.1. Regional Economic Performance
The EU is marked by large regional differences in economic development (Becker & Egger, 2010; Farole et al., 2011), understanding the drivers of regional economic performance is therefore key for policy makers. In this thesis economic performance will be measured by nominal GDP growth, as well as labour productivity. Although GDP is an imperfect measurement for regional well-being, as it excludes a variety of economic activities, it allows for the comparison across a large group of regions and is widely accepted as an objective measure of economic activity (Fraumeni, 2017). Labour productivity, an indicator for how effectively inputs are turned into output, is an important driver of long run economic growth and further allows to make inferences about the competitiveness of a region, as well as the type of activities performed within a GVC. Both indicators are thus focusing on the production capacity and capabilities of regional economies.
2.1.1. Drivers of Regional Economic Performance
Studies focusing on the location decision of multinational enterprises (MNE) have found that the trade-offs between international cost differences, and benefits of co-location are driving the allocations of productions (Baldwin & Venables, 2010). Belderbos et al. (2016) show a trend of regional clustering of firms in order to benefit from agglomeration economies, also known as co-location effects. In their study they analyse where the innovative activities are located by looking at the streams of FDI by MNEs, showing that outsourcing manufacturing activities does not mean that innovative capabilities are also offshored. Their findings go hand in hand with Crescenzi et al. (2013) who extend the OLI framework of MNE’s location decisions, ownership-location-internationalisation, with market potential and sub-national geographical factors, finding that MNEs tend to outsource manufacturing and R&D activities based on regional characteristics like local knowledge assets and the social institutions. Urban economics uses the term agglomeration forces, with Duranton and Puga (2004) showing that industry interactions lead to improved productivity through a process of “sharing, matching and learning”. Increased regional performance is thus dependent on the influx of capital from MNE and regions need to become attractive for MNE’s by the accumulation of human capital and innovative capabilities.
7 human capital accumulation to be endogenous, growth and productivity are driven by investing in education, training and innovation activities to boost productivity and growth (Romer, 1994; Martin & Sunley, 1998; Syverson, 2011). Lastly, Ketterer & Rodríguez-Pose (2018), based on a sub-national analysis of European regions, have shown that regional institutions are key in explaining the economic growth in EU15 regions between 1995-2009. Where institutions were measured via the effectiveness and accountability of governments, the rule of law and the fight against corruption. In their analysis they further showed that the “first-nature” geography, like access to the coast, are only minor drivers for regional development, measured as the growth in regional GDP per capita. In summary, the differences in regional economic performance have been accounted for by regions path dependency, the quality of economic and social institutions, and the participation in exporting (Farole et al., 2011; Porter, 2003; Martin & Sunley, 2006; Winters, 2004). It is therefore important to control for the level of human and physical capital, the technological and innovative capabilities, as well as the level of local or regional institutions. The specific control variables are introduced and further discussed in section 3.
GVC participation could potentially be a driver of regional economic performances in both developed and less developed regions. The rise of GVC eases the access to foreign markets and allows for quicker learning and could therefore favour quicker human capital accumulation and technology spillovers in less developed regions (Baldwin, 2011). While firms, with headquarters in more developed regions, improve their production outputs and costs by offshoring their production to regions with lower wage levels (Baldwin & Lopez-Gonzalez, 2013). It therefore becomes increasingly important to understand GVC. But as official trade statistics are still relying on gross exports measures, new methods and measurements in GVC had to be computed. The intuition behind GVCs and their usefulness in measuring economic performance will be introduced in the following section.
2.2. What are Global Value Chains?
8 extracts raw materials, which it exports to region 2, which uses the materials to produce an intermediate good, which it then exports to region 3, which uses the intermediate good as a component to produce the final good that is then sold domestically or exported.
Figure 1: Abstract and simplified visualization of a global value chain
A further consequence of the newly fragmented international production is that traditional ways of measuring trade based on gross exports are judged as misleading in showing the relative position of a country in trade (Los et al., 2016; Timmer et al., 2013; Johnson and Noguera, 2012). As trade in intermediate goods makes up the majority of international trade (Cattaneo et al., 2013), not accounting for where the value of each stage of production is added, gross exports overstate the value added of each region in the production of a final good. GVCs are therefore defined as: all value added activities
directly and indirectly involved in the production of a final manufacturing good.
However, case studies of specific product types are suitable in visualizing the complexity of GVCs and making their application more understandable.
2.3. Case studies
9 depends on imported intermediate goods and previously assembled components. As China exported the final products worldwide, traditional trade statistics of gross exports have accounted most of the value added to China. This type of case study, by breaking down the location of value added showed that the assembly and exporting of a manufacturing good is not generating a large share of value added and thus the different types of activities can be categorized as low or high value adding. Similarly, the assembly of iPhone’s, done in China’s East Coast, captures a relatively small value added share (Kraemer et al., 2011). Nevertheless, the inclusion of the China’s East Coast region in Apple’s GVC had broader effects as it increased interregional trade in intermediates, reducing the dependency on foreign intermediate inputs. Especially the central region has benefitted from increased demand of intermediate goods (Meng et al., 2012).
The case studies help visualizing two important implications of GVCs on economic performance. Firstly, it shows that there are large differences in the value added, depending on which activity is executed (R&D vs. assembly of a good) and secondly the benefits of participating in the GVC are local but also spread regionally due to the direct and indirect participation. The case studies are limited as data is not perfectly available and more importantly a single product might be misleading and is not representative for a whole industry. In order to make generalised studies a sound theoretical framework for GVCs and their effect on economic performance is needed. The two theoretical concepts governance and upgrading are discussed in the next section.
2.4. Governance and Upgrading
10 Gereffi et al. (2005) created a theoretical framework of the GVC’s governance forms, showing that the type of relationship between the supplier and the buyer is key to understanding another important concept of value chains: learning and upgrading. The intuition behind their framework is that based on a type of governance form, one can make theoretically sound predictive claims how an industry might evolve in case of a shock or when new markets are opening up. Whether goods and services are traded in a market based environment or within a captive relationship has different implications for the upgrading and learning opportunities of firms, industries or regions.
H&S (2001) define four possible ways for firms to upgrade in GVCs, via product, process, functional or interchain upgrading: differentiating the product and moving towards higher value products, optimizing the productivity via better technology, moving towards higher value added functions within the same chain or finally moving into a new chain. H&S (2001) are combining industrial clusters with traditional GVC upgrading theory, based on lead firms governing the chains and imposing standards and quality on suppliers, thus enhancing learning. Their main finding is that the upgrading opportunities for local firms depend on the governance mode of the chains. Pietrobelli and Rabellotti (2011), using the upgrading theory, show that depending on the structure of the chain and the power relations the innovation activities of firms can be enhanced or reduced. In the case of quasi-hierarchical or captive governance forms suppliers are “locked-in” to a specific activity in a chain, as the lead firm or the low number of buyers are interested in improving the product and processes of the suppliers but blocking advances in higher-value activities. This lock-in process is visible in the case of the Sinos valley footwear cluster in Brazil, where local shoe producers gained access to the US shoe market and quickly improved the product and processes but failed to upgrade into higher value functions like design, distribution and marketing. With the entrance of Chinese shoe producers the world prices dropped and coupled with the lack of innovation activities the Sinos valley footwear suppliers were locked-in into the low-value production of shoes by a small number of buyers (H&S, 2002).
To sum up, participating in the global shoe value chain has allowed the Sinos valley region to increase the production, quickly adapt and improve their processes and quality and thus increase overall production and labour productivity levels. This shows the regional benefits of easily being able to participate in a GVC, while also indicating that the volatility of trade forces industries to continuously upgrade in order to remain competitive. A chance that was missed by the Sinos valley, enhanced through the quasi-hierarchical form of governance of this specific value chain, creating a long term negative lock-in effect, in which functional upgrading, towards higher value added stages in the GVC, was not achieved.
11 continuation of manufacturing activities in the Czech republic, has provided upgrading opportunities for the whole manufacturing industry (Pavlínek & Ženka, 2011; Pavlínek & Janák, 2007; Pavlínek, 2012). The Czech suppliers were forced to update their production as quickly as possible in order to continue being a part of the Škoda supply network, or face being driven out by more productive foreign suppliers. The process of upgrading led to a restructuring of the industry directly and indirectly involved in the car manufacturing. In their study Pavlínek & Janák (2007) found that the region ‘Mladá Boleslav’, where the main assembly facility is located, benefited the most as most high value suppliers decided to locate in that area, whereas regions outside of the core benefited from the influx of FDI and relocation of firms due to lower wages and the proximity to the main facility.
A final distinction that should be made is that depending on the type of product produced, the opportunities for learning and upgrading, as well as the governance form changes. Gereffi et al. (2005) predict the type of governance based on the complexity of transactions for a product, its codifiability and the capabilities of the producers. One can assume the degree of product standards is higher for complex products and thus the form of governance is more heavily directed towards a captive relationship, or quasi-hierarchical, in which there is a high degree of technology spillover and thus learning opportunities. Whereas simple products are more often traded in a market governance form, where the standards set are more universal and easier to obtain (H&S, 2001). In order to participate in a GVC for complex goods, like car manufacturing or electronics, lead firms have a higher risk, as the initial investment is higher to set up production processes and to build capabilities in the supply base, making a switch to another supplier more complicated. Whereas simpler products, due to the lower level of complexity and the lower level of supplier capabilities required, allow for an easier switch of supplier. That being said, in the case of complex products, lead firms have no incentive, and possibility, to switch suppliers and thus could lock the suppliers into low value adding activities, like assembling imported intermediate goods and export the final goods. Whereas with simpler products, suppliers could potentially learn and create new capabilities and move into higher value adding activities as design and marketing, as seen in the case of the Asian apparel industry (Gereffi et al., 2005).
In general the economic literature seems to weigh the benefits of GVC participation more heavily than the risks of losing out (Cattaneo et al., 2013). Given the complexity of value chains, the next section shows how the value added activities in a GVCs can be traced on a macroeconomic level.
2.5. Macroeconomic Approach to GVC
12 detail on regional production structures, that represent the fragmented international trade flows within and among industries, and within and across regions in monetary values. A simplified visualization is shown in Figure 2 for a global IO with regional detail. In this case there are two regions and a third entity called rest of the world (RoW), summarizing all other regions outside of region 1 and 2 into one. All three regions consist of two industries. Goods and services are sold from the horizontal axis (rows) to the vertical axis (columns), whereas the vice-versa direction from columns to rows shows the payment of goods received. The stylized version shows the supply and use of intermediate goods and services that industries are trading with each other within and across borders. In Figure 2 the cells Z11 to Z33 are sub-matrices of the total intermediate
trade matrix Z. The sub-matrices contain the monetary value of what is sold from a given industry to another. Whereas the matrix F contains the monetary values of the supply and use of final goods and services. The distinction between intermediate us and final use allows to break down whether goods are used directly and indirectly in the production of final goods. IO table also contain data on primary inputs, such as the monetary value of labour that is used in an industry, or the capital involved in the production. Based on these primary inputs one can derive the total value added in each industry in each region. The final information contained in IO tables is the total output, per industry. Depending on the IO table used, the amount of regions and industries used are higher, allowing to summarize complex data in a relatively simple way.
Figure 2: Visualization of a simplified global IO table with regional detail
One can think of IO analysis as a more generalized version of the case study analysis, seen in section 2.3, for all industries and industry products. The use of IO table, thus, allows to measure what is included in the production of a good in a specific industry and where (which region and industries) the value has been added. This concept of trade in value added was introduced by Johnson and & Noguera (2012). Based on IO table the reality of
Intermediate Use Final Use Total R1 R2 RoW R1 R2 RoW Output
13 trade flows and GVC can therefore be measured correctly as the intermediate goods trade is separated from the trade in final goods and services.
The main direction of measurement of GVC, thus, is going into decomposing and tracking where and by whom value added is generated based on a final demand for a specific good (industry). The exact methods of construction, the limitations and use cases will be further discussed in the methodology section.
2.6. Hypotheses
In summary the literature proposes that GVCs allow less developed regions to more easily engage in international production, and thus favour imminent knowledge and technology transfers and upgrading opportunities. Firms can potentially be locked-in into lower value adding activities and the initial gains from learning will become insignificant. Thus, taking part in GVCs can also have negative effects due to a power asymmetry between the supplier and buyer, as well as higher vulnerability in case of economic downturns that affect the intermediate goods trade more (Ferrantino & Tagilioni, 2014).
Given that also less developed regions in Europe have relatively stable institutions and social capital, a necessary foundation to improve economic performance (Iyer et al., 2005), the potential learning and upgrading effects are assumed to be larger than the lock-in effects. Whereas the developed regions should benefit from the outsourcing mechanisms towards lower-wage regions. The first hypothesis is therefore:
H1: A higher degree of regional overall participation in GVCs is driving regional economic performance: Regional GDP growth and Labour Productivity
As discussed previously, given the large degree of different manufacturing products, the gains from learning and upgrading might differ depending on the GVC a region participates in. The degree of complexity of a final product can therefore have a different effect on economic performance. The second hypothesis is:
H2: The gains from participating in the GVC of complex products are larger compared to participating in simple product GVCs
One can further assume that, given the large degree of heterogeneity amongst European regions, that the effect depends on the level of economic development in a given region. Where the possible efficiency gains and easier access to foreign markets plays a more important role in less developed regions (Baldwin, 2014). Whereas the more developed regions are seeing a decline in actual manufacturing activities (Buckley et al., 2019), the servicification and indirect participation, can have positive effects, but less profound, on economic indicators as well. The third and final hypothesis is therefore:
H3: Less developed regions benefit more from an increased participation in GVCs than more developed regions
14
3. Methodology
This methodology section provides an overview of the model specifications for the regional economic performance regressions, the dataset used for the construction of the regional global-value-chain-participation-index (RGVCPI) and the methods needed to work with IO tables.
3.1. Model
Ultimately, the goal of the thesis is to find a causal relationship between the participation rate of a region and the regional economic performance. The analysis will test the correlation between the RGVCPI and the two regional economic indicators introduced in section 2.1.: yearly regional GDP growth (1) and the level of regional labour productivity (2). Δ𝑦𝑟𝑡 𝑦𝑟𝑡−1 = 𝛼 + 𝛽1𝑅𝐺𝑉𝐶𝑃𝐼𝑟𝑡+ 𝛽2(𝑅𝐺𝑉𝐶𝑃𝐼𝑟𝑡∗ 𝐷𝐸𝑉𝑟𝑡) + 𝛽3𝐷𝐸𝑉𝑟𝑡+ 𝛾𝑋𝑟𝑡+ 𝛿𝑡+ 𝜃𝑟+ 𝜀 (1) 𝑌𝑟𝑡= 𝛼 + 𝛽1𝑅𝐺𝑉𝐶𝑃𝐼𝑟𝑡+ 𝛽2(𝑅𝐺𝑉𝐶𝑃𝐼𝑟𝑡∗ 𝐷𝐸𝑉𝑟𝑡) + 𝛽3𝐷𝐸𝑉𝑟𝑡+ 𝛾𝑋𝑟𝑡+ 𝛿𝑡+ 𝜃𝑟+ 𝜀 (2)
Where r is the identifier for regions (r = 1 … 245) and t is the identifier for the year (t = 2000 … 2010). The right hand side of the equation is based on a constant 𝛼, the regional GVC participation index (𝑅𝐺𝑉𝐶𝑃𝐼𝑟𝑡), an interaction variable controlling for the level of development (𝑅𝐺𝑉𝐶𝑃𝐼𝑟𝑡∗ 𝐷𝐸𝑉), as well as the development dummy (𝐷𝐸𝑉𝑟𝑡 ), a vector of
control variables (𝑋𝑟𝑡), as well as fixed year (𝛿𝑡) and fixed region effects (𝜃𝑟) and 𝜀 is the error term.
15 The data for total hours worked and the rest of the control variables is taken from the Eurostat Regional database2. The data is used to construct a panel dataset with all 249 NUTS2 regions as the first identifier and the years of the period 2000-2010 as the second identifier. As the NUTS2 classification follows a dynamic process the region codes are updated regularly, with the available Eurostat data for the years 2000-2010 formatted in the 2016 classification, whereas the EUREGIO dataset is constructed based on the 2010 classification. In order to harmonize the two data sources, the mismatches in region codes were cross-checked and adjusted based on the NUTS2 classifications documentation.3 Luckily, only a handful of regions have been merged or further split into sub-regions during that period, so that the data validity remained intact, an example of such a merge is given by the region of Brandenburg, that originally was split into West and East but in the 2016 classification was merged into one region, so the data was merged in the Euregio dataset for the computation of the participation index. After the adjustments the panel dataset is covering 245 NUTS2 regions for a period of 11 years, resulting in a total of 2645 observations.
As economic performance is measured, there are quite some effects that need to be controlled for, to avoid omitted variable bias and therefore biased estimators. Due to the restricted availability of regional data, certain variables needed to be approximated or could not be accounted for at all. As introduced in section 2.1.1. the main variables to account for are the level of human capital, the level of investment, the innovative capabilities and technology, and finally the institutional quality.
Table 1 lists the main control variables used in the regressions of the empirical analysis
and their respective measurement units. In order to harmonize the data, all control variables were scaled by total regional population. The level of human capital is proxied by the share of total population having a tertiary education, and by the share of early leavers from education or training aged 18-25. Physical capital, or investment, is proxied by the gross fixed capital formation per capita, measuring the availability of capital in a given region. Technology is proxied by the share of human resources employed in science and technology. The amount of control variables is relatively limited, as there are a large share of missing observations for variables accounting for the level of innovation and technology like public R&D spending, intramural R&D spending, number of patents, including them in the regression analysis would cut the number of observations by almost half. Thus, the regressions are including both year fixed and region fixed effects to account for common shocks and unobserved characteristics that can influence the grGDP and LP. The main concern is that a key driver of economic growth, institutional quality for the years 2000-2010 could not be accounted for, as such indices only exist for the period post-2010. Advances to measure the regional institutional quality have been made by the European Quality of Government Index, and the regional competitiveness indices 2 The Eurostat Database can be accessed on: https://ec.europa.eu/eurostat/web/regions/data/database and the data codes for each variable are listed in Table A6 in the appendix
16 by the European commission combining regional and national institution scores (Annoni et al., 2016). The latter index will be used as a control variable for further robustness checks but cannot be included in the main estimation model.
In order to test whether the level of development of a region influences the gains, or losses, from participating in GVCs four dummy variables were defined, depending on the level of GDP capita and labour productivity of a region in the year 2000, if the values are below the median value regions are considered to be less developed, and if the values are in the lower quartile regions are considered least developed. A region is categorized for the whole observed period as less or least developed, the development dummies, are therefore time invariant. By including the dummy and the interaction effect in the regression, endogeneity can be further reduced as the initial level of GDP and/or labour productivity is accounted for.
Category Variable Name Description
Human Capital EDU_TER % of total pop with tertiary education aged 25-64 LEAVERS % of early leavers from education or training aged 18-25 Physical Capital CAPITAL Gross fixed capital formation in 1000€ per capita Technology HRST % of total pop employed in science and technology
Dummies LESS_DEV Dummy = 1 if the level of GDP per capita in region r is < median of GDP per capita in the year 2000 LEAST_DEV Dummy = 1 if the level of GDP per capita in region r is < lower quartile (p25) of GDP per capita in the year 2000 LESS_PROD Dummy = 1 if the level of LP in region r is < median of LP in the year 2000 LEAST_PROD Dummy = 1 if the level of LP in region r is < lower quartile (p25) of LP in the year 2000 Institutions INST Regional and National Institutional Quality Score between 0 and 100
Table 1: Control variables, dummies and interaction variables used in the regression analysis
3.2. EUREGIO – Global IO table with regional detail
The main dataset used in the construction of the RGVCPI is the relatively new EUREGIO dataset, which maps the intra-regional and international supplier-user linkages for 249 EU-NUTS2 regions4, including 14 industries, for the time period of 2000-2010 (Thissen et al., 2018). In this section a short summary will be provided.
The EUREGIO dataset is the result of merging the regional accounts of the EU25 members with the already existing World Input-Output Database (WIOD)5, as well as interregional trade numbers estimated by the PBL (Netherlands Environmental Assessment Agency), and specific country level data (Thissen et al., 2018). The final goal was to create a global IO table with regional detail, summarizing the trade flows of all industries in all NUTS2 regions, consistent with the WIOD and the national accounts statistics, for which a simplified version was introduced in Figure 2. The WIOD is a project creating world IO tables (WIOT), summarizing the trade flows, in intermediate goods and services and final
4 The complete list of all region names and region codes, for the NUTS2 2010 classification, can be found in the appendix: Table A1
17 goods, between countries’ industries, consistent with national accounts statistics. These WIOTs cover 40 countries, that make up around 85% of the world GDP (Dietzenbacher et al., 2013). In order to make use of the WIOT 2013 release, the data had to be corrected for re-exporting, the exports minus the production, and the exports and imports of countries had to be assigned to destination and source locations. To derive regional values the national accounts and WIOD had to be disaggregated by using the underlying regional supply and use table (Thissen et al., 2018, p.10). The construction of the database further had to fulfil double bookkeeping conditions as for example the exports from region r to region s are equal to the imports of region s from region r and vice versa. Another constraint, based on the double bookkeeping, is that the total of national cells equals to the addition of all regional supply and use cells. By using the regional accounts (Eurostat), the number of sectors have been aggregated into 14 industry classifications, for which both the regional and international trade structure are known. The complete list of the sector codes and industry classifications are presented in Table 2.
Sector Code Industry Classification
S1 Agriculture
S2 Mining, Quarrying and Energy Supply
S3 Food, Beverages and Tobacco
S4 Textiles and Leather
S5 Coke, Refined Petroleum, Nuclear Fuel and Chemicals S6 (S7) Electrical, Optical and Transport Equipment
S8 Other Manufacturing
S9 Construction
S10 Distribution
S11 Hotels and Restaurants
S12 Transport, Storage and Communications S13 Financial Intermediation
S14 Real Estate, Renting and Business Activities
S15 Non-Market Services
Table 2: Industry classification based on 2-digit classification in 14 sectors, EUREGIO Based on Thissen et al. (2018)
18
3.3.3. Data Limitations
The construction of the global IO table with regional details was exposed to a wide variety of constraints and assumptions, in order to get data consistency and follow the accounting relationships. These adjustments create a range of restrictions and implications for the analysis in this thesis. Firstly, due to the limitations of output data on the NUTS2 level the number of industries had to be limited towards an aggregated 14 industries. This aggregation reduces the possibility of a detailed sectoral analysis. Further, the aggregation could potentially lead to strange results due to the combination of industries covering a wide range of products, like the sector 5 covering Coke, Refined Petroleum, Nuclear Fuel and Chemicals. Secondly, the assumptions of homogeneous preferences of consumers throughout all regions in a country, homogeneous government spending across all regions within a country and the same production technology of firms within an industry, need to be taken into account when making use of the dataset for policy recommendations. Third, the IO table does not contain information about the ownership of firms. A German car manufacturer that has its production in the Czech Republic is accounted for as a Czech firm due to the location of the activity. This has a direct implication as this does not allow for an analysis of the type of governance in a GVC and further does not account for where the profits are ultimately accounted for. Fourth and final, the tables are in current prices (€) and are thus no deflated to constant prices or adjusted for purchasing power parities. Thus, the interpretation of the result needs to take this limitation into account.
Nevertheless, this unique dataset is offering novel approaches of measuring regional participation in GVCs. The next section will present the methods used in this thesis to measure GVC-participation.
3.3. Methods
The following section is relatively technical, but hopefully intuitively enough to better understand how IO tables can be used to analyse GVCs.
19 place (Los et al., 2014, p.69). Let us define the value chain as the German car manufacturing as a stylized example. Using the commonly known input-output analysis method by Miller & Blair (2009, Chapter 2.3), we start by taking the gross output, the finished German cars produced, to meet the final demand of all regions for this specific product. This is done by summing the final demand matrix F into a new final demand vector f* over all regions. The new vector f* consists of the gross output of the German car manufacturing industry and zeros for all other final outputs. By selecting only the final demand for this specific industry we obtain the monetary value of the direct effect, showing the direct exports and sales to the domestic market of German cars. In a first round the intermediate deliveries of f* are calculated, which equals the intermediate goods and services required to produce the cars, including example given the windows of the car that imported as they are produced elsewhere and are simply used in the final assembly. This is done by making use of the A matrix, in which the typical element represents intermediate inputs needed to produce one euro worth of gross output (Los et al., 2017). The A matrix is obtained by dividing each element zrs in the Z matrix, the
intermediate input matrix, by the corresponding total output xrs. In matrix notation this
would be 𝐀 = 𝐙 (𝐱̂ )−𝟏. By multiplying Af* the output needed in all regions and industries to produce the intermediate goods are reflected (first round). As this output depends on intermediate inputs as well, by multiplying AAf* shows the effects of the second round, the output produced to produce the intermediate inputs. In our case this could reflect the raw materials needed to produce the glass for the car windows. This is repeated round by round, AAAf* for the third round effect, until the sum x* is reached, which is the total output needed, directly and indirectly, to produce the final goods defined in f*. The vector
x* consists of elements for each industry in all regions. So a typical element of this vector
represents the output, again in monetary values for goods and services, of e.g. the Financial Intermediation sector of the Inner London region to produce all German cars. The process can be summarized in equation (2).
𝐱̂∗ = (𝐟∗+ 𝐀𝐟∗+ 𝐀𝐀𝐟∗+ 𝐀𝐀𝐀𝐟∗+ ⋯ ) (2)
Equation (2) is generally rewritten like equation (3), with (𝐈 − 𝐀)−𝟏 being known as the Leontief Inverse (Miller and Blair, 2009). The Leontief, abbreviated as the L matrix, thus takes all indirect effects into account and quantifies the value directly and indirectly included for the output of one unit of final output (Chen et al., 2017; Dietzenbacher et al., 2013; Johnson and Noguera, 2012).
𝐱̂∗ = (𝐈 − 𝐀)−𝟏 𝐟∗ (3)
20 coefficient vector v’, the four forms of value added in the IO table need to be summarized by adding them up in a row-wise fashion. By dividing each element by the corresponding industry total output, xrs, one receives the value added coefficients, in which the typical
element indicates the additional value added for one € of more output in a specific industry. By multiplying the vector v’ with equation (3), in our case, we receive the total value added worldwide in the German car manufacturing sector as a scalar, a single value. In case we are interested in the value added for a specific region r, the value added coefficient vector v’ can be manipulated by setting all values, with the exception for the industries of region r, to zero, deriving the regional value added vector v’r. In matrix form
the regional value added RVAr (or regional GDP) for a specific final demand could therefore be calculated in the following way (Chen et al., 2018):
RVAr = 𝐯r′(𝐈 − 𝐀)−𝟏𝐟∗ (4)
By manipulating the final demand matrix we can therefore calculate the value added for all 245 regions. Equation (4) is rewritten in equation (5) for a simplified case in which there are three regions, with two industries each. Let’s define the regions as Liguria (Italy), Cataluna (Spain) and Stuttgart (Germany), the abbreviations are (s1, s2 … s6) whereas s1 is industry 1 in region 1, s2 is industry 2 in region 1 until s6 is industry 2 in region 3. The industries are Textiles and Leather, and Electronics, Optical and Transport Equipment. In this stylized example we are interested in the combined value added of the two industries in the region Liguria in the GVC defined by the Electronics, Optical and Transport Equipment industry’s final output in Stuttgart, keeping the example of cars produced in Germany, in this case in the region of Stuttgart. Thus, in the value added coefficient vector only the values, for the Ligurian industries are listed, with the rest of the vector set to zeros. The Leontief-Inverse consists of 36 elements, in which a typical element e.g. ls6s1 states the direct and indirect output provided from industry 2 in region
3 to industry 1 in region 1 for an additional unit of final demand. And finally, the manipulated final output matrix F* has the values for final output only for the Electronics, Optical and Transport Equipment industry in Stuttgart sold to the three regions. As we are interested in the total final output of this industry the final demand matrix is multiplied with a summation-vector consisting of ones, creating a f* vector like in equation 4. The result is the regional value added, a single value, of the region Liguria in Stuttgart’s Electronics, Optical and Transport Equipment value chain.
21 As we want to measure the degree of participation in GVCs of a region, the RVAr is set as a ratio of total regional value added or simply total regional GDP (GDPr). The regional GVC participation index, RGVCPIr, is therefore defined as:
RGVCPI r= RVAr
GDPr (6)
GDPr is computed in equation (7) making use of the same concept as equation (5) but the world total final demand is used. In the equation e is a summation vector equal to the length of the matrix F, in order to sum the world final demand into a single final demand vector.
GDPr = 𝐯′𝐫(𝐈 − 𝐀)−𝟏𝐅𝐞 (7)
As GDPr measures all value added activities in a given region the domestic value added in a given manufacturing GVC must be necessarily smaller. As example given, GDPr measures all services a bank provides including households, governments and firms, whereas the RVA only captures the services of a bank directly involved in manufacturing, like handing out a loan to a car manufacturing company. This approach allows to measure the regional participation in different GVC as an index set between 0 and 1. With 1 being complete participation in GVCs, or all value added in the given region is generated in GVCs, and 0 is equivalent to not participating at all. This measure is an adjustment to the vertical specialisation measure of Johnson and Noguera (2012), who derive the value added in GVCs as a share of gross exports. The proposed participation index here is to be preferred as the measure does not depend on the level of exports and imports of a country, or in other words on the openness to trade of a region. If RVAs were to be divided by gross exports in a region, with a minor share of gross exports of total GDP, the RGVCPI would be extremely large, even though the value added in GVCs is a minor share of the total economy.
22
4. Empirical Analysis
In this section the hypotheses, defined in section 2, will be empirically tested. The section starts with a quick overview of the main data and correlation trends. Afterwards, the main estimation method will be introduced, with the results then critically discussed in section 5 of this paper.
4.1. Data Summary and NUTS2 RGVC-PI Indices
The three RGVCPIs introduced in the previous section are summarized in Table 3. The RGVCPIs were computed for every year between 2000-2010, and will be used as the main explanatory variables in the regressions. The RGVCPI1 will be used to explain the overall impact of GVC participation of a region, as it includes the worldwide production of manufacturing sectors. Optimally, the total manufacturing would have been decomposed into manufacturing sectors for complex and simple products, but as discussed in section 3.3. the aggregation of sectors in the EUREGIO dataset made the desired categorization almost impossible. The Electronics, Optical and Transport Equipment sector therefore represents the complex product GVC, while the Textile and Leather sector represents the simple product GVC. For the remainder of this paper the abbreviation RGVCPI will be used to reference to the three GVCs used for the empirical analysis.
Variable name Value Chain Measured as
RGVCPI1 Total manufacturing GVC Share of total regional GDP between 0 and 1 RGVCPI2 Complex Product GVC Share of total regional GDP between 0 and 1 RGVCPI3 Simple Product GVC Share of total regional GDP between 0 and 1
Table 3: List of the main explanatory variables RGVCPI-1, -2 and -3.
Table 4 provides the summary statistics for the main explanatory variables listed in Table 3. As well as the two dependent variables of interest yearly GDP growth (grGDP) and
23 in both manufacturing and mining industries, explaining the high share of total GDP gained from participating in GVCs (EC, 2020)7. On the other extreme are the regions Attiki and Voreio Aigaio (Greece), as well as the Illes Balears (Spain), which generate below 10% via participation in global manufacturing GVCs. The findings seem valid based on the high dependency of the three regions on tourism and lack of integration in GVC for higher value manufacturing activities (EC, 2020). Contrary to services provided by the banking or insurance sector, tourism is generally not considered to be a direct or indirect part of manufacturing activities. The participation in the complex GVCs is creating above 13% of total GDP in the German regions of Oberpfalz, Mittelfranken, Bremen and Saarland, with Airbus, Ford Werke AG and other big manufacturing companies located in these regions (EC, 2020). On the other hand with less than 1% in RGVCPI2 are, similar to RGVCPI1, Greek regions like Dytiki Ellada, Ipeiors and Voreio Aigaio. For the textile and leather GVC there are two regions generating more than 5% of total GDP in this sector, Norte (Portugal) and the Lombardia (Italy), with Norte still invested heavily in the traditional sectors of clothing and textile, whereas the Lombardia is home to the fashion metropole Milan (EC, 2020). While the lowest shares, below 1%, are registered in the Scottish regions of Highlands and Islands as well as South Western Scotland.8
Observations Mean Median Min Max
grGDP 2695 3.37% 3.46% -21.18% 34.43%
LP 2695 30.28 31.31 3.98 157.92
RGVCPI1 2695 23.40% 23.57% 6.43% 35.90%
RGVCPI2 2695 6.53% 5.94% 0.52% 16.99%
RGVCPI3 2695 1.50% 1.16% 0.30% 9.09%
Table 4: Summary statistics for the dependent and main explanatory variables
The computed growth rates, between 2000 and 2010, for the variables in Table 4 show GDP and labour productivity have, averaged across all regions, grown above 40% in that period, while the three RGVCPI have grown negatively, with RGVCPI1 at around -14%, RGVCPI2 at around -13% and the RGVCPI3 at -37%. The overall growth rates of the dependent and explanatory variables have thus been evolving in opposite directions. A closer look at the data shows that the RGVCPI1-3 have followed a positive growth path as well until 2008, but have seen a major collapse after the financial crisis, which has partially, but as seen not totally recovered. This is in line with studies focusing on the slowdown in international trade caused by the financial crisis (Ferrantino & Faglioni, 2014).
Before defining the model and the estimation method, the pairwise correlation trends between the two dependent and main explanatory variables RGVCPI1-3 are listed, at a significance level of 1%, in Table 5 pooling the data over all years and regions. The
7 The regional information on industries is gained from the Regional Innovation Monitor, provided by the European Commission. Accessible here:
https://ec.europa.eu/growth/tools-databases/regional-innovation-monitor/
24 Pearson correlation coefficients indicates a positive, although small correlation, between all RGVCPIs and grGDP. Whereas the coefficients in the case of LP are negatively correlated with RGVCPI1 and RGVCPI3, while positively correlated with RGVCPI2. Again, the correlation is drawn over the pooled data and needs to be treated carefully, as the correlation if corrected by regional trends could result in a different correlation trend.
grGDP LP RGVCPI1 RGVCPI2 RGVCPI3
grGDP 1 LP -0.1416* 1 RGVCPI1 0.1233* -0.0579* 1 RGVCPI2 0.0576* 0.1165* 0.7768* 1 RGVCPI3 0.1128* -0.3311* 0.1671* -0.1575* 1 * p < 0.01
Table 5: Pairwise correlation, Pearson correlation coefficient, between the dependent and RGVCPI1-3, pooled over all observations and years, with significance level set at 1%
Based on the correlation trends and the final dataset, covering 11 years and 245 regions, the right estimation model had to be selected, the next section provides an overview of the different methods and tests used to derive the correct specifications for the estimation.
4.2. Model selection
25 effects to account for unobserved regional characteristics in the dataset. In a next step the regressions were estimated via random effects model and compared, via Hausman test, with the fixed effects model. The Hausman test was clearly rejected, stating that there is a systematic difference between the coefficients of the two models, and thus the fixed effect estimator is to be preferred. Therefore, unobserved regional characteristics are present, but are not randomly drawn (Hill et al., 2018). The restrictions on the correlation between errors and the explanatory are eased and allow for the error covariance per region to be unequal to zero. Both the set of control variables and fixed year effects were additionally successfully tested for joint significance, with the F-Test was clearly rejected. In an additional step the Fixed Effects model was re-estimated including the lags 1 and 2 of the RGVCPI1, to see whether the fact of participating year t will have an effect on economic performance in year t+1 and t+2 for lag1 and lag2 respectively. Although finding a significant effect for the first lag of the RGVCPI coefficients, given the presence of serially correlated errors and the decreased R-squared and adjusted R-squared make the inclusion of lags for the explanatory variable insignificant. The preferred estimation model, summarized in Table 6, thus becomes the fixed effects model with fixed year and regions effects estimated with robust standard errors. With GDP growth (grGDP) and labour productivity (LP) as dependent variable.
5. Results
26 The overall goodness of fit of the model, is moderate for grGDP, as dependent variable, with the adjusted R-squared at 0.447, translating into around 45% in the variance of grGDP being explained by this regression. In the case of LP, as dependent variable, the adjusted R-squared is at 0.677, explaining a larger share of the variance in LP than for grGDP. Furthermore, the control variables behave as expected as an increase in the capital formation, share of tertiary education, and people employed in science and technology are positively correlated with the dependent variables, and an increase in the share of early leavers from school and training has a negative impact in both cases.
(1) (2) (3) (4) VARIABLES grGDP grGDP LP LP RGVCPI1 0.431*** 8.017 (0.0915) (6.215) RGVCPI2 0.396** -15.06 (0.179) (10.63) RGVCPI3 1.341*** 43.77* (0.324) (24.91) CAPITAL 0.00969*** 0.00991*** 0.645*** 0.639*** (0.00189) (0.00198) (0.167) (0.169) HRST 0.00159 0.00157 0.245*** 0.263*** (0.00110) (0.00114) (0.0889) (0.0854) EDU_TER 0.000176 0.000178 0.0456 0.0335 (0.000767) (0.000774) (0.0602) (0.0609) LEAVERS -0.000809** -0.000955*** -0.0157 -0.0212 (0.000338) (0.000352) (0.0219) (0.0216) Constant -0.173*** -0.117*** 16.08*** 18.47*** (0.0314) (0.0249) (2.852) (2.553)
Fixed Region Effect included YES YES YES YES
Fixed Year Effect included YES YES YES YES
Observations 2,364 2,364 2,364 2,364
Number of Regions 229 229 229 229
Adjusted R-squared 0.448 0.446 0.675 0.677
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Table 6: Fixed Effect Model Estimation Results for GDP growth and labour productivity
27 (1) (2) (3) (4) (5) (6) (7) (8) VARIABLES grGDP grGDP grGDP grGDP LP LP LP LP RGVCPI1 0.263** 0.240** 0.212** 0.228** 19.06** 12.34* 13.71* 9.867 (0.106) (0.0969) (0.102) (0.0961) (7.703) (6.979) (7.938) (7.044) LESS_DEV#RGVCPI1 0.282** -18.48** (0.110) (9.188) LEAST_DEV#RGVCPI1 0.597*** -13.53* (0.181) (7.874) LESS_PROD#RGVCPI1 0.385*** -9.970 (0.112) (9.379) LEAST_PROD#RGVCPI1 0.651*** -5.931 (0.201) (8.456) Constant -0.167*** -0.158*** -0.164*** -0.158*** 15.65*** 15.73*** 15.84*** 15.95*** (0.0323) (0.0342) (0.0328) (0.0351) (2.802) (2.820) (2.822) (2.846) Observations 2,364 2,364 2,364 2,364 2,364 2,364 2,364 2,364 Number of Regions 229 229 229 229 229 229 229 229 Adjusted R-squared 0.450 0.453 0.451 0.454 0.677 0.676 0.676 0.675 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Table 7: Interaction variables for the fixed effect estimation and RGVCPI1
(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES grGDP grGDP grGDP grGDP LP LP LP LP RGVCPI2 0.121 0.125 0.186 0.194 1.229 -6.496 0.598 -6.249 (0.246) (0.211) (0.207) (0.185) (15.81) (14.08) (16.27) (14.37) RGVCPI3 1.134* 0.756* 0.0836 0.373 6.292 3.013 -57.34 -43.12 (0.590) (0.418) (0.474) (0.357) (40.90) (43.06) (55.10) (42.03) LESS_DEV#RGVCPI2 0.501 -25.65 (0.330) (20.98) LESS_DEV#RGVCPI3 0.162 44.54 (0.581) (35.56) LEAST_DEV#RGVCPI2 0.937** -9.586 (0.418) (18.82) LEAST_DEV#RGVCPI3 0.742 54.23 (0.476) (34.31) LESS_PROD#RGVCPI2 0.521 -17.69 (0.316) (21.43) LESS_PROD#RGVCPI3 1.220*** 107.6** (0.440) (48.25) LEAST_PROD#RGVCPI2 0.777* -0.0971 (0.428) (18.96) LEAST_PROD#RGVCPI3 1.075** 103.2*** (0.415) (33.56) Constant -0.112*** -0.101*** -0.107*** -0.0979*** 18.45*** 18.52*** 18.61*** 18.97*** (0.0245) (0.0249) (0.0246) (0.0249) (2.572) (2.656) (2.584) (2.681) Observations 2,364 2,364 2,364 2,364 2,364 2,364 2,364 2,364 Number of Regions 229 229 229 229 229 229 229 229 Adjusted R-squared 0.447 0.449 0.449 0.449 0.677 0.677 0.678 0.679 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Table 8: Interaction variables for the fixed effect estimation and RGVCPI2 & RGVCPI3
28 The results for RGVCPI2 and RGVCPI3 need a more detailed discussion. Focusing on RGVCPI2 first, columns 5-8 in Table 8, the only positive correlation with LP is noted for more developed regions, as less and/or least developed regions, will have a negative marginal effect of an increased participation on LP, no matter the level of RGVCPI2 . In the case of RGVCPI3, regions categorized as less and least developed, based on the level of GDP per capita in the year 2000, gain additionally from an increased participation in the simple GVC, irrespectively of the initial level of RGVCPI3. Whereas the level of RGVCPI3 becomes important if the gains from participation in less and least developed regions, based on the labour productivity level in the year 2000, on LP is analysed. In order for RGVCPI3 to have a positive impact in less developed regions the initial level of RGVCPI3 the interaction term needs to outweigh the negative RGVCPI3 coefficient, if an increase of 0.1 unit is anticipated, equivalent to a 10 percentage point increase in the participation index, the initial level of RGVCPI3 must be higher than 0.053 and 0.042 for less and least developed regions respectively in order to have a positive impact on LP. The results from the fixed effects model estimation are representative of the direct, or short term, implications of an increased GVC participation. But as discussed in section 2.4. the channel through which an increased participation could affect economic performance was defined by learning from lead firms, as well as technology spillovers. While gaining access to cheaper inputs and a larger variety of inputs for developed regions, as well as access to developing markets can certainly have positive effects in the short term, learning and possible lock-in effects are expected to take a long time. Therefore, an alternative to the immediate effect of GVC participation estimation is presented here.
Rather than estimating a panel data model, a cross-sectional model making use of more up to date data for the dependent and control variables between 2010-2016 is used. Offering the chance to include the level of institutions (INST) in a given region as a further control variable, whereas the growth rate of 2010-2016 for the same control variables are used. The main explanatory variables used are the initial RGVCPI level for the year 2000 (INITIAL_RGVCPI), the final RGVCPI level for the year 2010 (FINAL_RGVCPI) as well as the total growth rate over the period 2000-2010 (GROWTH_RGVCPI) for the same GVCs used before. Whereas the dependent variables become the overall growth rate of GDP between 2010-2016 (grGDP2016) as well as the level of labour productivity in 2016 (LP2016).
29 regressions were estimated using a multiple variable OLS estimator with robust errors.
Table 9 summarizes the results for the OLS multiple variable regressions.
(1) (2) (3) (4) (5) (6) VARIABLES grGDP2016 grGDP2016 grGDP2016 LP2016 LP2016 LP2016 INITIAL_RGVCPI1 0.291 -37.55** (0.214) (17.91) INITIAL_RGVCPI2 0.548* 2.808 (0.294) (26.98) INITIAL_RGVCPI3 0.538 0.294 (0.604) (68.68) FINAL_RGVCPI1 -0.221 -59.01*** (0.201) (21.42) FINAL_RGVCPI2 1.292*** 9.598 (0.338) (30.25) FINAL_RGVCPI3 1.240 81.99 (0.786) (111.8) GROWTH_RGVCPI1 -0.0263 -18.63*** (0.0823) (6.755) GROWTH_RGVCPI2 0.0535*** -0.788 (0.0177) (1.188) GROWTH_RGVCPI3 -0.228*** 11.68** (0.0760) (5.392) grCAPITAL 0.226*** 0.225*** 0.250*** 2.904*** 2.949*** 2.685*** (0.0732) (0.0741) (0.0718) (0.444) (0.455) (0.435) grEDU_TER 0.0232 0.0116 -0.00383 -0.174** -0.267*** -0.124* (0.0257) (0.0257) (0.0265) (0.0790) (0.0881) (0.0739) grHRST 0.0149* 0.0195** 0.00465 0.00368*** 0.00364*** 0.00312*** (0.00828) (0.00843) (0.00918) (0.000826) (0.000797) (0.000761) INST 0.00192*** 0.00199*** 0.00171*** 0.316*** 0.348*** 0.273*** (0.000654) (0.000693) (0.000651) (0.0643) (0.0616) (0.0557) Constant -0.147** -0.0828 -0.0320 16.10*** 18.96*** 10.47*** (0.0724) (0.0653) (0.0370) (3.947) (4.668) (2.248) Observations 245 245 245 245 245 245 Adjusted R-squared 0.612 0.622 0.601 0.769 0.785 0.775 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Table 9: Multiple Variables OLS Estimation Results for GDP Growth and Labour Productivity
30 VARIABLES grGDP2016 grGDP2016 LP2016 LP2016 FINAL_RGVCPI1 -0.474* -0.377* -33.12 -2.200 (0.263) (0.209) (29.88) (25.06) FINAL_RGVCPI2 1.078*** 0.934*** 8.951 8.765 (0.396) (0.354) (35.29) (31.70) FINAL_RGVCPI3 2.837*** 1.801** 166.2 108.9 (0.824) (0.876) (116.5) (107.9) LESS_DEV -0.0689 (0.0545) LESS_DEV#FINAL_RGVCPI1 0.361 (0.368) LESS_DEV#FINAL_RGVCPI2 0.651 (0.648) LESS_DEV#FINAL_RGVCPI3 -2.454** (1.037) LEAST_DEV -0.0700 (0.0746) LEAST_DEV#FINAL_RGVCPI1 0.118 (0.490) LEAST_DEV#FINAL_RGVCPI2 1.229 (0.840) LEAST_DEV#FINAL_RGVCPI3 -1.266 (1.415) LESS_PROD 3.156 (5.288) LESS_PROD#FINAL_RGVCPI1 -59.16 (36.46) LESS_PROD#FINAL_RGVCPI2 33.35 (48.96) LESS_PROD#FINAL_RGVCPI3 -98.62 (149.8) LEAST_PROD 1.303 (4.360) LEAST_PROD#FINAL_RGVCPI1 -67.69** (33.62) LEAST_PROD#FINAL_RGVCPI2 36.29 (46.35) LEAST_PROD#FINAL_RGVCPI3 -194.6 (139.7) Constant -0.0668 -0.0663 24.14*** 18.20*** (0.0615) (0.0627) (6.471) (4.163) Observations 238 238 245 245 Adjusted R-squared 0.647 0.642 0.833 0.875 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Table 10: Multiple Variables OLS Estimation Results, including development dummies and interaction variables
5.1. Discussion and Testing of Hypotheses
The results from the second estimation yield some interesting results. Surprisingly, the overall participation in manufacturing GVC seems to have a mostly negative impact in the long run, whereas the immediate or short run effects were mostly positive. Thus, for the period 2000-2010 the first hypothesis “A higher degree of regional overall participation
in GVCs is driving regional economic performance” could be accepted with reginal GVC
31 economic downturns, due to the highly fragmented production and dependency on other intermediate inputs (Ferrantino & Faglioni, 2014). Secondly, an initial gain from accessing GVCs can boost short term labour productivity, through the influx of knowledge and technology, but in the long run, regions could get stuck in these outsourced low productivity activities. Concluding, a higher degree of regional overall participation in GVCs is driving the short term regional economic performance, but the effects are reduced or reversed in the long run.
Decomposing, the general participation into the participation in simple and complex GVCs has shown that the second hypothesis “The gains from participating in the GVC of
complex products are larger compared to participating in simple product GVCs” has to be
rejected. The effects, contrarily to the expectations and over all regions, are significantly stronger for RGVCPI3, when the immediate or short run effects are analysed, with RGVCPI2 even being negatively correlated with labour productivity. Similarly, if the long run estimation model for the best fitted model, using the final RGVCPI levels, is analysed then an increased level of participation in simple product GVCs is driving economic performance more strongly. In section 2.4. the reason to believe for a larger effect in complex value chains was given by the higher level of technology spillovers and the higher level of trust, whereas the potential downside was for firms to become locked-in into a specific activity for a long period of time, reducing the LP and economic growth overtime. Given the results found in the analysis, the latter channel seems to outweigh the gains from increased knowledge transfers if the average effect over all regions is analysed. It can further be concluded that the chances of upgrading and improving the productivity are more likely in the simple product GVCs.
The general conclusion, for the third hypothesis that Less developed regions benefit more
from participating in GVCs than more developed regions is that in terms of GDP growth the
32
5.2. Limitations
There are a few limitations to the extent of significance and validity of the results found. Firstly, given the current data availability on regional trade, the lack of information on FDI and foreign ownership of companies involved in GVCs does not allow for any conclusions on the type of governance as well as to which extend the economic gains are kept within a region. Thus, the repatriation of profits and the location of lead firms and their innovative centres could tell a different story on who is actually gaining and losing from participating in GVCs.
Secondly, further disaggregating the effect on economic performance from regions to industries and firms could provide insights to whether firms are able to upgrade their productions and move into higher value adding activities over time, or whether the lock-in effect takes place. With the given data limitations these effects can only be assumed, given the difference in benefits between less and more developed regions. Especially, in the case of labour productivity it would be interesting to see the effect on the industry level, rather than the regional level, in order to do a sectoral decomposition to see which industries are potentially experiencing a lock-in effect. Thus, with the analysis provided the effect on labour productivity could be explained by an increase or decrease in the LP directly involved in GVCs, or depending on the initial level of RGVCPI, be influenced by industries not involved at all in GVCs. In the future, with more detailed and available data this kind of research could help in deriving more useful policy recommendations, further including industrial employment and the job creation in manufacturing and other sectors. Thirdly, the amount of control variables was relatively limited, thus the chances of omitted variable bias are relatively high, by adjusting for fixed year and region effects and including the time-invariant development dummies and using robust standard errors, the bias in the estimators was certainly limited. In an extra step the dependent variables were replaced by the yearly growth rate in GDP per capita, accepting a lower number of observations, as well as labour productivity measured as GDP per person employed. The results, listed in the appendix Table A5, confirm the trends and coefficient signs found. To further improve the validity of the results and reduce endogeneity, the long difference estimation was conducted, thus controlling for reverse causality. Nevertheless, assuming a potential bias of the RGVCPI estimator seems valid, whereas the direction of the bias, over- or underestimation, cannot be defined definitely.
