Master Thesis in the framework of the
Double-Degree Program:
M.A. International Economics
(Georg-August Universität Göttingen)
M.Sc. International Economics and Business
(Rijksuniversiteit Groningen)
Lennart C. Kaplan
July 30, 2014
Supervisor: Tristan Kohl, Ph.D., University of Groningen
Developmental prospects for CEECs in the EU's internal market
of goods and services - A Value Added perspective
Lennart C. Kaplan
July 30, 2014
Abstract
The structure of international trade has changed markedly due to the processes of increased fragmenta-tion since the late 1980s and trends towards economic integrafragmenta-tion. Producfragmenta-tion processes are increasingly split across borders. In this new framework it becomes more and more important to examine, which amount of domestic Value Added (VA) is embodied in nal products. Furthermore, in order to assess the competitiveness of a respective country the composition of the VA needs to be analyzed. Based on a new comprehensive database, allowing to account for VA ows and its factorial composition across countries, the implications of economic integration are reassessed. This is namely done for the example of the Eastern Enlargement of the European Union (EU) by ten Central and Eastern European Coun-tries (CEEC) in 2004. In order to obtain reliable estimates of accession eects, theoretically founded speci cations of the well-known gravity model of trade are applied. The results are surprising: While CEEC's VA exports seem to be stimulated by EU accession, industrial upgrading processes in intra-EU trade seem to be slowed down vis-à-vis CEEC's exports to third countries. Therefore regional integration might disproportionally favor VA by low-skill tasks.
Acknowledgements
List of Abbreviations
AvW - Anderson and van Wincoop BB - Baier and Bergstrand
BRIC - Brazil, Russia, India and China CAP - Common Agricultural Policy
CEEC - Central and Eastern European Countries
CEEC-2 - The 2007's European Union entrants: Bulgaria and Romania
CEEC-10 - The 2004's European Union entrants: The Czech Republic, Cyprus, Estonia, Hungary, Latvia, Lithuania, Malta, Poland, Slovakia, Slovenia
CEFTA - Central European Free Trade Agreement
CEPII - Centre d'Etudes Prospectives et d'Informations Internationales CIA - Central Intelligence Agency
EAs - Europe Agreements ed./eds. - Editor/Editors
e.g. - exempl gr ati a (for example) EIA - Economic Integration Agreement EMU - European Monetary Union et al. - et alii (and others) EU - European Union
EU-15 - The European Union members in the pre-Eastern Enlargement period/up to 2004 EU-27 - The present members of the EU excluding Croatia (EU-15 + CEEC-10 + CEEC-2) FGLS Feasible Generalized Least Squares
GVC - Global Value Chain
GVCI - Global Value Chain Incomes H - Hypothesis
ICT - Information and Communication Technology
IDE-JETRO - Institute of Development Economies of the Japan External Trade Organization IMF - International Monetary Fund
IOT - Input-Output Table LabHS - High-Skilled Labor LabMS - Medium-Skilled Labor LabLS - Low-Skilled Labor Mio. - Million
N - Number of Observations NA - Not Available
OECD - Organisation for Economic Co-operation and Development OLS - Ordinary Least Squares
p.c. - per capita
Ph.D. - Doctor of Philosophy
PPML - Poisson Pseudo-Maximum Likelihood p-value - probability value
R² Coe cient of Determination R&D - Research and Development RoW - Rest of the World
UN - United Nations
UNCTAD - United Nations Conference on Trade and Development US - United States
VA - Value Added
VAX - Value Added in Exports VC - Value Chain
VIF - Variance In ation Factor vs - versus
Contents
1 Introduction 7
2 Theory 9
2.1 Economic Fragmentation . . . 9
2.2 Upgrading in a fragmented world economy . . . 10
2.3 Integration theory . . . 12
2.4 Integration design of the EU . . . 13
3 Methodology and Data 16 3.1 The Gravity Model of Trade . . . 16
3.1.1 Multilateral Resistance . . . 17
3.1.2 Accounting for endogeneity . . . 18
3.2 Data . . . 21
3.2.1 Data Sources . . . 21
4 Data Analysis 23 4.1 Descriptive Analysis . . . 23
4.2 Empirical Analysis . . . 26
4.2.1 Distance's diering eect on gross and VA exports . . . 26
4.2.2 First dierences accession eect estimates . . . 28
4.2.3 Digging deeper Accession eects on sectorial VAX . . . 31
4.2.4 Is it trade with new or old members that promotes CEECs' VA exports? . . . 33
4.2.5 Are integration eects diering between 2004 and 2007 entrants? . . . 35
4.3 Robustness Analysis . . . 37
5 Discussion 40 6 Limitations and prospects for future research 44 6.1 Limitations . . . 44
6.2 Avenues for future research . . . 46
7 Conclusion and policy implications 47
List of Figures
1 Per capita income gap of EU-15 and CEEC-10 in ¿ . . . 143 VA exports from CEEC-10 to EU-27 . . . 24
4 Composition of VA exports from CEEC-10 to EU-27 . . . 25
5 Composition of VA exports from CEEC-10 to non-EU states . . . 41
6 Schematic World Input Output Table . . . 58
7 Domestic nal demand fb for world nal production and domestic nal production for world nal demand in a schematic IO-table . . . 59
8 Distribution of Residuals for total VA exports in the AvW-type model . . . 73
List of Tables
1 List of Countries contained in the WIOD . . . 222 Distance and its eect on VA and gross exports (AvW-speci cation) . . . 27
3 First Dierences - Baseline Estimates . . . 29
4 First Dierences - Manufacturing . . . 32
5 First Dierences - Services . . . 33
6 First Dierences: Splitting the accession eect up . . . 34
7 First Dierences - 2004 vs 2007 entrants . . . 36
8 First Dierences - Baseline with Leads . . . 38
9 Number of signi cant lead terms in the dierent speci cations . . . 38
10 Summary of previous speci cations . . . 40
11 Descriptive Statistics . . . 60
12 Accounting for In ation . . . 62
13 Distance with country-time xed eects . . . 62
14 Multicollinearity Test . . . 63
15 Wooldridge Test for Autocorrelation . . . 63
16 Baseline Estimates - Fixed Eects . . . 64
17 Baseline - Leads . . . 65
18 Manufacturing - Leads . . . 66
19 Services - Leads . . . 66
20 Splitting the accession coe cient up - Leads . . . 67
21 2004 vs 2007 entrants - Leads . . . 68
22 Sensitivity Analysis - First Dierencing - GVCI . . . 69
23 Sensitivity Analysis - First Dierencing - Eurozone . . . 70
24 Sensitivity Analysis - Fixed Eects - Dynamic Panel . . . 72
1 Introduction
Global economic integration is neither a new phenomenon nor as global as perceived. Already during the 19th century a spatial separation of production processes and consumption took place, interrupted by the two World Wars. New however, is the increasing fragmentation of production steps and tasks across countries since the mid-1980s (Baldwin, 2006).
Due to this fundamental change, it becomes less and less meaningful to use gross trade numbers as the level of economic analysis. In the new framework rather the Value Added (VA) by domestic labor and physical capital, embodied in gross exports, needs to be examined (Baldwin and Taglioni, 2011; Daudin, 2009).
An example is the geographical dispersed production chain of the Porsche Cayenne. Its engine and high-end interior and exterior components are manufactured in Germany. These components are assembled into the car in Bratislava, Slovakia, whereas the nal production steps and inspection are carried out again in Germany in order to deliver Porsche Cayennes for German nal demand. In a traditional trade perspective, gross trade statistics, would tell that Germany exported engines and high-end car components to Slovakia, while Slovakia is exporting nearly nished cars including engines and high-end components (Dudenhöer, 2005). However, from a VA perspective the story is rather dierent. Germany is not exporting any VA for Slovakian nal demand, while Slovakia is only exporting its assembly services and manufacturing products on a medium-skill level. This nding makes it so important to consider trade not on a gross, but rather on a VA level.
The nature of fragmentation in the Porsche example points also to another feature of the 21th century world economy. Driven by the intensive emergence of regional Economic Integration Agreements (EIA), Value Chains (VCs) seem to be increasingly organized regionally (Baldwin and Lopez-Gonzalez, 2013). A pioneer of the EIAs is the EU, which already emerged in the 1950s and constitutes with its nowadays 28 member states the largest single market in the world economy. In 2004 the EU faced its most comprehensive enlargement, when ten Central and Eastern European Countries (CEECs) acceded to the EU. Hand in hand with the accession went expectations of the CEECs to promote growth due to oshoring and outsourcing from the further developed EU-15 countries and their rms (Marin, 2006). A fragmented world economy oers countries the opportunity to participate in international VCs, rather than building their own from scratch. In this way technical knowledge can get accumulated, which provides opportunities to upgrade capabilities of domestic producers. Firms can then consequently move to higher VA activities, promoting the growth of GDP, which is by de nition the sum of all VA activities in a country.
H1: The EU membership of the CEEC-10 led to more trade between the entrants and other member states. The EU accession led to signi cant reductions of impediments to trade, among the 10 CEECs and the EU incumbents. Classical trade theory suggests that liberalization is linked to an increase in trade, especially due to the perceived comparative advantage of CEECs in labor intensive goods (Baldwin, 1995).
H2: The EU membership of the CEEC-10 led to more VA Exports between the entrants and other member states.
Classic economic theory suggests that a reduction of trade barriers leads to an increase in trade. It is possible that the increase in gross trade is attributable either to an increase of domestic VA activities or an increasing use of foreign inputs (Dedrick, Kraemer&Linden, 2010). For this reason it is of great importance to compare the gross with the VA trade perspective.
H3a: The increased VA in trade due to EU members' nal demand embodies a shift to more advanced production factors (e.g. high-skilled labor and capital).
The largest share of VA is captured by technology intensive activities located in pre- and post-production of goods and services (Baldwin and Evenett, 2012). These activities mainly involve highly skilled workers and larger amounts of capital. This paper seeks to answer whether the accession induced a stronger contribution of these factors to CEECs' exports for EU members' nal demand. This is done to assess consequently if industrial upgrading is achieved among CEECs.
H3b: The increased VA in trade is induced by outsourcing of low-skilled routine tasks to the CEECs from other EU member states, leading consequently to industrial downgrading.
Contrasting, it is also imaginable that the EU accession of CEECs led to increased exports for other mem-bers' nal demand, which are mainly based on the use of low-skill labor from the CEECs. In this scenario VA exports might increase quantitatively due to an increased export of low-skill tasks. However, no qualitative upgrading in Global Value Chains (GVCs) and VA per worker would have taken place. This would leave the CEEC economies stuck in low VA activities or even lead to downgrading. The growth prospects in this scenario would be comparatively lower.
2 Theory
2.1 Economic Fragmentation
After post-war recovery, in the 1960s, a second wave of globalization gained momentum. It was driven by a widespread industrialization and signi cant drops in trade cost, causing increasing trade across nations in intermediate goods. As a result the domestic VA share in gross exports seems to have dropped by 10-15 percentage points in the last four decades (Johnson, 2014). This phenomenon was discussed in the trade literature and subsumed under the headline of economic fragmentation (Feenstra and Hanson, 1996; Jones and Kierzkowski, 1990; Krugman and Venables, 1995).
New however, is the quality of trade integration, which was induced by the revolutions in information and telecommunication technology (ICT). The broad availability of the internet, computers and the mobile-phone opened further avenues of global trade integration. While formerly production processes were carried out in one speci c location, as face to face contact between employees was needed to coordinate complex tasks, ICT facilitated the coordination of tasks across borders. Recent theory by Baldwin (2006) and Grossman and Rossi-Hansberg (2006) implies that in the new paradigm not only value-chains but speci c production steps are unbundled. Therefore, it is not complete goods and services that are traded internationally anymore, but it is rather tasks, jobs and skills. This nding is associated with severe alterations of international labor markets: Formerly the production was allocated to the company or the country, which was able to most e ciently perform the average of all tasks involved in the production of the good. Therefore, competitive pressure from developing nations for developed states was perceived primarily from low-skilled sectors, while the developed world had an edge in high-skilled sectors as a whole (Feenstra and Hanson, 1996; Krugman and Venables, 1995). In the new paradigm speci c tasks can get coordinated across borders. Therefore, although developing countries are not most e cient for the whole production process, individual jobs in high-skill branches might be challenged by developing countries' workers (Baldwin, 2006). Nevertheless, this allocation to most e cient rms and individuals increases the average labor productivity and enhances e ciency (Grossman and Rossi-Hansberg, 2006).
to assess the competitiveness of a country (Tre er and Zhu, 2010; De Vries, Foster-McGregor&Stehrer, 2012). This just recently became possible, due to an increased establishment of comprehensive multi-country Input-Output Tables (IOTs). These global IOTs are now widely used in international trade analysis and facilitate fruitful research.
A substantial part of fragmentation research covers the conceptualization of VA measures. Johnson&Noguera (2012a) proposed in this regard the measure of Value Added Exports (VAX), which will be also used for the empirical analysis in this paper. VAX exhibits the VA of a source country that is embodied in the nal demand of a partner and is therefore a footprint measure. Based on this measure it was found in a gravity model of trade that VA ows are less repressed by physical distance than gross trade (Johnson&Noguera, 2012b). With data from IOTs these aggregate measures can be easily decomposed into its factorial shares. For instance, Stehrer, Foster&de Vries (2012) found that according to theoretical predictions advanced countries would be exporters of high-skilled labor. In contrast especially medium-skilled manufacturing jobs would be oshored (Foster, Stehrer, Timmer&de Vries, 2012). As of today not only low-skill, but increasingly medium-skill jobs seem to be sourced from abroad, industrial upgrading might be assumed to be happening for the catching-up countries.
In this context this papers seek to combine the work done by Johnson&Noguera (2012b) and the studies on factorial decomposition. The former authors applied a gravity-model for a large number of countries and dierent EIAs and examined the eects of distance and trade integration on VAX on an aggregate level. This paper focuses on a more in-depth examination of a speci c example in the framework of a gravity model, namely the EU's Eastern enlargement 2004. In this regard not only VAX on an aggregate level will be examined, but its factorial composition in dierent sectors in order to analyze, whether industrial upgrading took place. For this purpose, the following section provides a more in-depth de nition of industrial upgrading in a fragmented world economy.
2.2 Upgrading in a fragmented world economy
Sustainable economic growth and the explanation for income dierences across countries has been always one of the main objectives of economic science. In economic literature it has been acknowledged that international trade is an important growth promoter (Frankel and Romer, 1999). Proponents of a developmental paradigm that is only based on bare accumulation of physical and human capital are becoming increasingly challenged by the view that the assimilation to advanced technologies via the engagement into foreign trade contributes signi cantly to a country's growth perspectives. In the business literature Porter (1990) stressed that countries can achieve competitiveness via the implementation of new production processes and technologies. This technological progress should be most preferably accomplished via innovation, which makes competitiveness rather hard to achieve for countries that are far away from the technological frontier.
value chain on their own anymore. Nevertheless, in a fragmented world economy it becomes easier for other countries to engage in the production of advanced goods on a global scale as well. Therefore, countries and their rms need to gradually improve their productive capabilities and avoid being stuck in low VA activities in order to achieve international competitiveness - upgrading is thereby not an automatic process. The case of the CEEC-10 leads to the question whether the acceding countries were capable to make use of their skill structure or whether they became suppliers of low VA intermediate inputs for other EU economies.
In order to answer if the CEECs succeeded in upgrading it is important to have a clear-cut de nition of upgrading. Humphrey and Schmitz (2002) de ne upgrading as the implementation of new production processes or products in regional clusters and distinguish four types: (i) process upgrading, (ii) product upgrading, (iii) functional upgrading or (iv) inter-sectoral upgrading. In this context (i) process upgrading describes the switch to more e cient production processes, e.g. due to technological change or reorganization. While (ii) product upgrading, exhibits the introduction of products with a higher consumer value, (iii) functional upgrading is associated to a skill increase in the production structure, e.g. switching from basic assembly to the production of basic input materials. Finally (iv) inter-sectoral upgrading takes place in rms, which use knowledge from speci c sectors in a dierent higher value adding sector, e.g. using production processes learned in the domain of computers for the production of tablet devices. An important determinant of upgrading prospects remains the form of GVC governance set by the lead rms. The governance mode is mainly based on the complexity of the tasks carried out, the codi ability of the knowledge and the industry-speci c capabilities of the supplier (Humphrey and Schmitz, 2002). Last but not least, the willingness of the lead rm to accept suppliers' upgrading is of much importance, as upgrading suppliers might be future competitors for the lead rm. Especially due to the complexity of the upgrading and GVC governance structures, upgrading processes were so far mostly described in a case study setting (Gere and Fernandez-Stark, 2011; Ivarsson and Alvstam, 2011; Schmitz, 2006). With regard to the CEECs recent literature provides industry-speci c evidence that upgrading took place for example in the Czech call center industry as well as in Polish and Slovakian furniture production (Ahmed, 2013; Kaplinsky, Readman&Memedovic, 2006).
2.3 Integration theory
In an attempt to resolve the ambiguity in de nitions of economic integration, Balassa (1961) describes economic integration as abolishing discrimination between economic units belonging to dierent national states (and) [...] the absence of various forms of discrimination between national economies. This reduction of discrimination consisting of taris and non-tari trade barriers (quotas, local content requirements etc.) is expected to increase trade and specialization in absolute cost advantages following classic economic theory. Ranging from a simple free trade agreement (FTA), over a customs union (FTA plus a common external tari) until an economic union (a customs union plus a common market for production factors) trade integration can achieve diering degrees. Depending on the intensity of integration dierent economic outcomes are attainable.
In assessing the outcomes of an FTA, the starting point of integration analysis is the internal opening up and the external closure of the contracting parties with regard to the trade in goods. (Similar assump-tions can be made for services. Yet, in more narrow bounds, as services might be partly non-tradable.) Following basic economic theory of comparative cost advantage, the decrease of trade impediments, as an act towards free trade, should increase trade ows. Therefore, also the welfare-enhancing eects of trade would be increased. Following Viner (1950) these eects were labeled the trade creation eects of a customs union. The economic bene ts of trade creation are usually subsumed under the terms (i) allocation and (ii) accumulation eects. The rst term (i) allocation eects of trade describes the before mentioned changes towards the comparative cost advantages of countries, as ineective domestic production is substituted by more eective foreign production. When trade barriers are reduced within the customs union, the partner country's rms gain improved access to the domestic market, becoming able to make use of their compar-ative cost advantage, which is now not distorted by taris anymore. The eect is increased competition in the domestic market as only the most competitive foreign rms select themselves into internationalization modes. This forces ine cient domestic producers out of the market and releases their productive factors (Altimonte, Aquilante&Ottaviano, 2011). Consequently, these production factors can be used for more pro-ductive courses, for which the home market faces comparative advantages. This reallocation to more e cient domestic and foreign producers leads to economies of scale and increases pro ts of the eective producers as now foreign and domestic consumers are served. Additionally, consumers as well as producers are expected to face a wider product and intermediate good variety for lower prices (Baldwin, Francois&Portes, 1997).
In addition in the long term (ii) accumulation eects are expected to increase welfare even further, in terms of fostering economic agglomerations and competition. Especially the latter increases the incentives to invest in human capital and R&D (Baldwin and Wyplosz, 2012). These accumulation eects towards more advanced production factors (high-skilled labor and capital intensity) will be of interest with regard to the CEECs.
customs union (Viner, 1950). Let us assume the perspective of one small country 1, which faces imports of one single good from another small country 2 as well as from one big country 3 (assumed to be the Rest of the World). In this example the producers in the Rest of the World (RoW) are assumed to produce the imported good more e ciently and can thus provide the good cheaper. In the initial situation country 1 poses taris on imports from both countries, but only imports goods from country 3 as it is the most e cient producer. Now both countries 1 and 2 select into a customs union: Assuming that the value of the tari is higher than the cost advantage gap of country 2 and 3, country 1 will switch its imports from country 3 towards country 2. Trade is diverted into the customs union, away from the most productive producer in the world market. Although the initial situation with taris was not optimal, the establishment of the customs union leads to a further diversion of trade into the customs union by reducing the import share of the RoW. Furthermore tari incomes, which are in general assumed to ow back to the consumer via the redistribution of governmental returns, are reduced. After accounting for taris the consumers pay now higher prices for goods, linked to a deterioration of the bilateral terms of trade. Despite these theoretical implications, it has to be noticed that countries are both exporters and importers. This means that in an extended analysis, rms of the importing country could in turn export more to country 2, even if they were not the most price-competitive suppliers. This would lead to more exports and potential bene ts for domestic rms. Consequently the common perception is, that positive welfare eects (at least for the participating customs union members) prevail (Balassa, 1961). EIAs might even increase in speci c cases the world welfare, but never reach a pareto-optimum vis-à-vis absolute free trade (Lipsey, 1957; Balassa, 1961). Therefore, bilateral and regional trade integration, in contrast to world free trade, is only seen as the second best solution.
Nevertheless, almost 350 bilateral and multilateral trade agreements have been implemented in the post-WWII world (Kohl, 2012). This is due to the fact that a global consensus about world trade policy is hardly imaginable, as a result of the seemingly endless WTO trade rounds. Consensus seems to be only possible, if trade integration is also accompanied by political integration, which is an outstanding attribute of the EU (Baldwin and Wyplosz, 2012; Pomfret, 2007). The speci c characteristics of the economic integration process of the EU will be subsequently described in more detail.
2.4 Integration design of the EU
Irrespective of its current debt crisis and ongoing discussions on how far the integration of European states should go, the EU is often considered as a role model of regional integration, especially in the developing world (Pomfret, 2007). This is driven by the fact that the European integration process has advanced both in economic and political aspects.
In this respect already the foundational Treaty of Rome expressed the desire to ensure the development of the [...] prosperity (European Economic Community, 1957) for its member states. Prosperity enhancing integration was speci cally desired as a means to an end of a peaceful Europe.
and capital and exceeds in this way the scope of a customs union. The internal market is supported by an European competition policy and a currency union, which however does not include all CEECs to this date. More important is that the European economic integration process is complemented by political integration. The EU has its own executive (European Commission), legislative (European Parliament, Council of the European Union) and judicial branch (Court of Justice of the EU) and important decisions are increasingly made in Brussels, Strasbourg and Luxembourg. This deep political cooperation enhances also economic cooperation, which would not be possible to such a degree in a loose customs union (Baldwin et al., 1997). Moreover, the EU budget, although often object of profound criticism with regard to wasteful use of resources, is contributing substantially to European cooperation. For instance, due to its utilization for structural cohesion policies and the promotion of rural development (European Commission, 2014). With the accession of ten CEECs in 2004 the EU faced both a historical chance and challenge: On the one hand the incumbents were able to win the CEECs for the European idea after the end of the Soviet hegemony and build the foundation for future stability. On the other hand the union also faced its largest expansion in terms of countries, enlarging the EU from 15 to 25 members and by circa 75 million citizens. Additionally, the developmental gap between entrants and incumbents, depicted in Figure 1, was partly tremendous. While in 2004 the CEEC-10 exhibited an average GDP p.c. of circa 12,500¿, EU-15 inhabitants earned with 24,500¿ nearly twice as much. Also marked income dierences existed among entrants, reaching from close to 10,100¿ p.c. in Latvia, to 19,600¿ in Cyprus.
Figure 1: Per capita income gap of EU-15 and CEEC-10 in ¿
Above that the EU-15 consisted mainly of developed economies with a functioning institutional system, whereas the CEECs were characterized by heterogeneous institutions and a middle-income status. For this reason the EU accession 2004 was associated with diering economic expectations on both sides: The opinion towards the EU accession was rather positive in the CEECs due to the expected gains from participation in the single market. Additional points were the eligibility for nancial support in the framework of the Common Agricultural Policy (CAP) and structural policy, as well as political support for the transformation process via the implementation of the EU's legislatory body. In this regard it was expected that CEECs would be stabilized politically by the integration prospect, providing stimuli for investment. The rationale is that before the EU accession the CEECs were in a state of transformation, accompanied by unpredictable changes in in ation, taxes and subsidies. This led ultimately to a higher investment risk premium. Due to the stabilizing function of the EU entry the risk premium was expected to decrease, leading to lower interest rates and in eect a higher capital accumulation. This is especially important as the lack of capital is perceived as a bottleneck for CEECs' growth, because the countries are rather labor abundant, while lacking capital (Baldwin et al., 1997). In contrast, although some EU-15 countries like Germany were having intensi ed trading relations with the CEECs already in the 1990s, skepticism prevailed among the old member states. Migration pressure was expected in the Northern EU-states, whereas the Southern states perceived the CEECs mainly as competitors with regard to structural funding and exports. Additionally, in a political economy context the incumbent countries feared that the new entrants would use their newly gained political power in order to enforce their speci c interests to the detriment of the EU-15. This nally resulted in changes in the political framework of the EU (Treaties of Nice & Lisbon), as well as long lasting transition periods for CEECs' free migration (Baldwin et al., 1997). It can be conjectured that these pre-enlargement reforms limited the bene cial eects of integration. Moreover, it needs to be accounted for the fact that liberalization had already taken place in the framework of the Europe Agreements during the 1990s, lowering additional bene ts of full membership.
Notwithstanding, was it expectable that the entry gave further impetus to CEEC-10's growth for following reasons: Full EU membership equals a further reduction of trade frictions in terms of abolished border controls and product standard harmonization, which formerly increased trade costs by circa 10 percent without producing bene ts for the importing countries. Furthermore industrial upgrading prospects might have bene ted from the capital in ows into the CEECs due to stabilization in combination with integration into production networks of the EU-15.
UK and to a lesser extent Italy would source and re-import goods and increasingly services from the factory economies of the CEECs, especially Poland and the Czech Republic Baldwin and Lopez-Gonzalez (2013). Based on the same data Los, Timmer&de Vries (2014) depict a more nuanced picture: In their study the factory economies Poland and the Czech Republic show a clear trend towards globalization of VA ows between 1995 and 2009. In contrast Hungary and Latvia seem to have become increasingly oriented towards their region. Nonetheless, for both groups of countries regional VAX ows increased. Yet, for the former -rather globalizing group of countries - increases in intra-regional VAX ows were surpassed by even stronger increases concerning inter-regional trade.
Nevertheless, even if regional eects are surpassed by quantitatively stronger eects of global trade inte-gration, this paper wants to examine if the 2004 accession has enhanced CEECs' VA exports qualitatively. Especially, due to the deepness of European integration, a qualitative dierence in VAX is assumable, regarding industrial upgrading prospects. In order to shed more light on this question an analysis of the changes in factorial composition of VAX after the EU accession will be carried out. For this purpose the well-known gravity model of trade will be applied, building on a new comprehensive dataset allowing for decomposition of VA ows the WIOD. Subsequently, both methodology and data will be explained in more detail.
3 Methodology and Data
For the empirical analysis of the accession eects a gravity model of trade was chosen. It convinces due to its exibility to diverse modi cations and its low demands in terms of inputs on trade policy measures (e.g. change in taris and non-tari barriers). The intuition behind the gravity model of trade, its advantages, shortcomings and possible solutions for the latter are outlined subsequently.
3.1 The Gravity Model of Trade
The gravity model dates back to a seminal work by Tinbergen (1962), who as a trained physicist applied Newton's law of universal gravitation in the framework of trade. The model builds on following basic equation:
Eij = a0∗ Yia1∗ Y a2 j ∗ D a3 ij ∗ G a4 ij (1)
where the Eijstands for the respective ow between two countries i and j. Yiand Yjsignify the respective
gross domestic products of the trade partners, Dij describes the bilateral distance and Gij stands for the
gravitational constant, explaining all other relevant bilateral characteristics. The exponents a1−4 signify
that the explanatory power of the independent variables diers, while a0 is a constant, which scales the
well as time varying determinants of trade. The most prominent among them are common language, coastal access, colonial history and trade policies (e.g. EIAs). For the latter partly ambiguous outcomes were found (see for example with regard to the eects of WTO membership Rose (2004a) and Kohl (Forthcomming)). The gravitational constant Gij is actually not a constant term. It depends on the bilateral relation examined,
varies over time and more importantly it is correlated to unobserved trade policy variables, leading to omitted variable bias - a topic covered later in this section (De Benedictis and Taglioni, 2011).
In order to make the gravity model computable in the standard OLS-regression design, the basic model is often log-linearized, yielding:
lnEij = lna0+ a1lnYi+ a2lnYj+ a3lnDij+ a4lnGij+ eij (2)
Despite of its good data t (R² ~ 0.7) Tinbergen's gravity model, originating from physics, was heavily criticized for its non-theoretical, intuitive justi cation (Deardor, 1995). However, this seems not to apply for micro-economic foundations. They were thoroughly developed by Anderson (1979) and Bergstrand (1985) or more recently by van Bergeijk and Brakman (2010) and Baldwin and Taglioni (2006).1 Further
theoret-ical contribution regarding the applications of gravity models in the new paradigm of a fragmented world economy was provided by Costinot and Rodríguez-Clare (2013). Although micro-economically well founded, the basic gravity model has important theoretical shortcomings, which induce estimation biases. Important contributions to the gravity model have been made since 2000 regarding the remoteness to trade and the endogeneity of trade agreements. These extensions, which are taken into consideration in the data analysis of this paper, are outlined brie y in the following paragraphs.
3.1.1 Multilateral Resistance
Already Anderson (1979) and Bergstrand (1985) indicated that above bilateral trade costs, the trade between two partners will be determined also by the average barrier to all other international partner countries. Still, trade economists ignored this nding for several years. This explains for example a border eect estimation of a 2200% dierence between US interstate trade and trade between US states and Canadian provinces by McCallum (1995). It was not possible to account for the relative average trading costs, labeled 'multilateral resistance' or remoteness, until Anderson and van Wincoop (2003) (AvW) introduced a new modeling approach. They formulate a basic gravity model:
xij = yiyj yw ( tij PiPj )1−σv (3)
where xij is the trade ow and yi and yj are GDPs of countries i, j. They are divided by the world GDP
yw. The following term describes the relative trade resistance of the two countries (assuming symmetric trade
barriers), where tij is the bilateral resistance and Pi and Pj constitute the price indexes of the respective
countries. Theoretically the bilateral resistance factor tij is unobservable, but might be estimated via
ob-servables. Examples for these observables are geographical distance, common borders, adjacency or linguistic identity. Econometrically it is more viable to include these unobservables via country-speci c xed eects as they can be estimated easily in an OLS setting. However, estimates with the monadic xed eects can be less e cient (than using proxies) and the speci cation error might be correlated with the observable determi-nates of trade costs (Anderson and van Wincoop, 2003). Notwithstanding, Head and Mayer (2013) underline the importance of introducing country xed eects, as they control for further unobservables (like being a regional export hub, which can fundamentally bias gross export numbers). Additionally, if country-speci c xed eects are used, it is arguable that time xed eects should also be added to the regression model in order to account for unobservable changes in the world economy (Anderson and van Wincoop, 2003; Head and Mayer, 2013).
3.1.2 Accounting for endogeneity
Baier and Bergstrand (2007) and Baier, Bergstrand and Vidal (2007) point out that further unobservables should be considered in order to avoid endogeneity bias of EIA estimates. This involves following issues: (i) The intensity of trade integration in the framework of EIAs diers signi cantly. However, until recently EIAs were mostly measured via a dichotomous 0-1 variable (measurement errors). (ii) The decision to select into an EIA might be endogenous, because countries selecting into a common EIA would share similar characteristics that are not accounted for in the gravity equation (omitted variables). (iii) Finally, dependent and indepen-dent variables could be mutually depenindepen-dent, leading to reversed causality. The depenindepen-dent trade variable might be causally linked to higher GDP levels and a higher propensity to select into an EIA (simultaneity bias).
With regard to the (i) measurement error problem, it should be noticed that trade agreements dier signi cantly in their intensity. While preferential trade agreements can be considered as the rst building block, integration can advance even into political integration, leading to a harmonization of regulation and policies like in the EU. If now all EIAs are displayed as a dichotomous variable, a signi cant estimation bias might emerge, which overestimates the eects of basic EIAs and underestimates the eects of deeper integration. This problem can be addressed by using continuous variables, displaying the progress of trade integration in the form of an index and was applied for example by Kohl, Brakman and Garretsen (2013). This paper focuses mainly on one single EIA, namely the EU. For this reason, it is assumed - though dierences in integration intensity are in place e.g. for migration - that trade liberalization via an EU membership is similar and can thus be exhibited by a dummy variable.
countries can barely be conceptualized. Nevertheless, particularly the latter could induce a large bias for following reasons:
Assume high regulatory barriers in a speci c bilateral trade relation, e.g. the rigid competition policy of post-Communist CEECs. If now a country selects into an EIA, it is imaginable, that the EIA induces in addition to tari reduction also a convergence of national regulations, e.g. in terms of the EU's regulatory body. This convergence enhances the prospects of trade due to reduced trade frictions. Policy makers anticipate this and for this reason the probability to select into an EIA is positively in uenced. Therefore, the mutual high regulatory dierences are both positively correlated with the selection into an EIA and the magnitude of the EIA eect. Its inclusion would be then negatively correlated to the error term, as a larger share of the variance would be explained. In contrast, the exclusion of an internal regulatory variable, leads to a larger error term and to an underestimation of the EIA eect omitted variable bias is induced (Baier et al., 2007).
Baier and Bergstrand (2007) suggest that the EIA-selection-decision of trade policy makers is based on the levels of trade ows and not on recent changes. Therefore, the selection would take place on a cross-sectional level. With panel data this assumption makes it possible to mitigate the selection-bias via an estimation model based on dyadic and country-time xed eects. Although mitigating potential endogeneity bias, dyadic xed eects are perfectly collinear to the time-invariant variables in the gravity model. These time-invariant characteristics include geographical distance and bilateral properties like a common colonial history. Due to the collinearity, these variables have to be excluded from the estimation. Beyond xed eects, the rst-dierencing approach proposed by Baier and Bergstrand (2007) can help to mitigate the endogeneity issues, outlined in this paragraph. In the framework of this approach rst dierences of all dependent and independent variables are taken. This leads to an exclusion of the time-invariant country-pair characteristics, whereas the time-diering country characteristics can be measured via country-time xed eects in the model. The methodology seems to oer two additional bene ts. (i) Firstly it can be assumed that trade ows do follow unit-root processes, thus are non-stationary. Fixed-eects however, are similar to a dierencing around the mean in terms of the xed eects acting as a level. In this regard they assume stationarity of trade ows. Thus, if xed eects are used in long panels, spurious-regression issues could arise. In contrast rst dierencing induces a stationarity of the variables as they depict the dierences from the previous period. In eect the rst dierenced data t the model better. This reduces concerns of spurious regressions. (ii) Secondly, in panels with a long time-series component problems might especially arise if the error of the xed eects model is correlated over time. In eect bias due to serial correlation would arise, which might be mitigated via the use of rst dierences (Baier and Bergstrand, 2007; Wooldridge, 2002). Both rst dierences and the previously proposed xed eects will be applied in the empirical part of this paper.2
Another issue arises in terms of (iii) reversed causality or respectively the ambiguous causation between
2 As it is also a question of this paper to examine the time-invariant diering eects of distance on VA vis-à-vis gross exports,
dependent and independent variables (Baier and Bergstrand, 2007). The basic gravity model - outlined in Section 3.1 - assumes that bilateral trade is explained among other factors by the GDP and the existence of trade agreements between the partners. There seems to be evidence that the latter two independent variables are partly explained by the size of bilateral trade ows (dependent variable).
Firstly, following classical trade theory free commerce is enhancing economic growth by specialization on country-speci c advantages. Thus, countries with higher trade ows might have on average also a higher GDP. This makes the interpretation of causation ambiguous. Frankel (1997) examined this issue in an instrumental variable (IV) approach. GDP mass variables were instrumented in a gravity setting with past income levels. Frankel's results pointed out that the endogeneity of GDP and gross exports causes little bias. Nevertheless, to avoid a dependent variable misspeci cation in the new paradigm of a fragmented world economy, VAX instead of gross trade data are used for most speci cations of this paper (Baldwin and Taglioni, 2011). In this framework endogeneity issues can become more severe. This is due to the fact that GDP depicts the sum of VA produced for domestic and foreign use in a speci c country. Regressing GDP on VAX resembles therefore a circular reasoning. Country-time xed eects proposed by Baier and Bergstrand (2007) can account for this, as they capture the GDP component completely. This acts as a further reason why this speci cation is preferred for most of the paper's estimations.
Secondly, the selection into an EIA might be driven by the initial amount of bilateral trade ows. The reasoning is that countries tend to have an interest in liberalizing trade with countries that are already important partners, rather than with countries, with which no cooperation is established. For this reason it is suggested that selection into an EIA is driven by initial trade ows, leading to a further endogeneity bias in the basic equation. To control for this issue in a subsequent section of this paper, pre-integration trade ows will be regressed on the future EIA-dummy (Baier et al., 2007; Frankel, 1997; Magee, 2008). This follows the intuition that future trade liberalization cannot have eects on pre-integration trade ows, whereas the latter might have an impact on the selection into an EIA. This endogeneity test will then reveal if the trade ows between the EU-15 and CEECs, in e.g. the year 2000, had an impact on the decision to select into the EU by 2004.
In this manner it is also possible to suggest that future EIA dummies depict theoretically founded an-ticipatory eects, which might not be linked to endogeneity. This can be imagined especially, if exporters anticipate that a trade agreement comes into place and deepen links to the future partner country. Regard-ing the 2004 EU accession round another reasonRegard-ing for anticipation exists: Countries with a prospect of EU membership were granted preferential terms in the framework of the Europe Agreements (EAs), reducing trade barrier in the pre-accession phase. Statistically signi cant lead variables might also be attributable to this context.
in order to capture the full eect of selection into trade agreements (Anderson and Yotov, 2011; Rose, 2004b). This will be done in the framework of this paper, based on the time-series data of the World Input Output Database (WIOD). The database will be, among the other data sources used, subsequently explained.
3.2 Data
3.2.1 Data Sources
In order to assess bilateral commerce via the previously described gravity model, data on trade as well as on country and country-pair characteristics are needed. For the speci c purpose of this paper the VA trade data of the dependent variable were derived from the WIOD. It is a comprehensive source, oering data on the input-output structure of 40 countries from 1995 to 2009. The included countries are depicted in Table 1.3
Its 40 economies account for circa 85% of world GDP, making it truly an international input-output table (Timmer, 2012). The Rest of the World (RoW) acts as a residual for the missing 15% of world GDP.
In this regard it needs to be mentioned that further international input-output tables are oered by other organizations as well. The most prominent examples are the IOTs by: (1) the OECD, (2) a joint OECD-WTO project (TiVA), (3) the Institute of Development Economies of the Japan External Trade Organization (IDE-JETRO) and the (4) Global Trade Analysis Project (GTAP). These databases dier slightly in annual coverage and the scope of included countries. In this regard (1) OECD and (2) TiVA project provide data on a larger set of countries, but lack a consecutive time-series. This is also a characteristic of (3) the IDE-JETRO data, which additionally includes only East-Asian economies, making it not applicable for the research question of this paper. Finally, (4) the GTAP database comprises more countries, but mainly after 2000 (Johnson, 2014). This is problematic in the context that the estimations of the EU accession eect in this paper should be based on a su cient pre-enlargement period. In this regard the profound statistical basis of the WIOD, the countries included and especially the European focus of the database in combination with its annually time coverage, make it particularly applicable for the research question investigated.
Data on VA can be attributed to sectors and its contributing factors via the WIOD's socio-economic satellite accounts. Concerning this, the WIOD distinguishes between physical capital and three labor types, which are characterized by their educational attainment. Using the International Standard Classi cation of Education, the editors of the WIOD dierentiate between high-skill, medium-skill and low-skill labor. In this context low-skilled labor depicts all workers, which have attained at the maximum lower secondary education. Medium-skill labor comprises educational attainments including post-secondary, but non-tertiary education. Finally, the class of highly skilled workers describes the tertiary educated workforce, including holders of a Ph.D. (Timmer, 2012). Based on the data it is possible to calculate measures related to the research question, like for example the VA from medium-skilled workers in Lithuanian service industries, which is embodied in Lithuanian exports to Finland. The measures which will be used for the dependent variables
3 Full use was made of the 15 years included in the WIOD. Sensitivity analysis revealed that the included time and
Table 1: List of Countries contained in the WIOD
European Union Asia and Paci c North America Latin America
Austria Germany Netherlands Australia Canada Brazil
Belgium Greece Poland China USA Mexico
Bulgaria Hungary Portugal India
Cyprus Ireland Romania Indonesia
Czech Republic Italy Slovakia Japan
Denmark Latvia Slovenia Korea
Estonia Lithuania Spain Russia
Finland Luxembourg Sweden Taiwan
France Malta United Kingdom Turkey
in the gravity models of the empirical examination are namely (i) VA exports (VAX) and (ii) Global Value Chain Incomes (GVCI). As previously outlined, the (i) VAX concept was developed by Johnson&Noguera (2012a) and describes the foreign VA footprint of domestic nal consumption e.g. how much Hungarian VA is embodied in Dutch consumption. Another measure is the (ii) GVCI, which was proposed by Timmer et al. (2013) and will be used to control for the robustness of the ndings based on VAX. GVCI, as a footprint measure of production, exhibits the foreign VA that is embodied in domestic nal production. Although having an overlap, (i) VAX is mainly consumed in the speci c country of interest, while (ii) GVCI might be re-exported. Both measures account for indirect inputs via the so-called Leontief inverse.
In this context it is worthwhile to mention that a recent contribution by Wang, Wei&Zhu (2013) suggests that forward-linked measures might be biased due to double counting. VAX and GVCI are focused on the nal demand or respectively nal consumption and are thus forward-linked measures that could induce bias for speci c applications. Nonetheless, Leontief's insights of forward based measures are applicable, if foreign ultimate absorption is considered (Wang et al., 2013). This is speci cally the case in the subsequent analysis for VAX, as it depicts the CEEC's VA that is embodied in other EU countries' nal demand. For GVCI the notion is not that clear, as only parts of the VA are absorbed in the partner country, whereas other fractions are re-exported in the partner country's nal products (Wang et al., 2013). For this reason in subsequent estimations the VAX measure will be used. GVCI measures will be used for a robustness analysis, provided in the appendix. A more detailed description of the two measures and how to derive them from the WIOD data can also be found in the appendix.
entry into the EU and European Monetary Union (EMU) were compiled from the web page of the European Comission's Directorate General for Economic and Financial Aairs (2014). Data on whether a country is landlocked were retrieved from the CIA's World Factbook (2014). Finally, data on weighted geographic distance, this is bilateral distance weighted by the geographic distance between populous centers in the country, data on contiguity, common o cial languages and historical colonial relationships are based on the CEPII's GEODIST dataset by Mayer and Zignago (2011). An issue arose in this regard with the distance of the WIOD's RoW residual. As the RoW comprises some 150 countries it is hard to attribute a distinct measure of physical distance, colonial relationships or contiguity to it. These characteristics are nonetheless needed for the construction of a gravity dataset, as distance is one of the main explanatory variables. For this reason the RoW was not included in the subsequent data-analysis. This should be kept in mind as a caveat, when interpreting subsequent results. In the following section some descriptive statistics of the gross and VA export data will be given, motivating the more in-depth empirical analysis.
4 Data Analysis
4.1 Descriptive Analysis
Figure 2: Gross exports from CEEC-10 to EU-27
Source: Own calculations based on data on gross exports by UNCTAD. Figure 3: VA exports from CEEC-10 to EU-27
Source: Own calculations based on WIOD.
CEEC-10 countries to the other EU members from 1995 to 2009. It becomes visible that the trend of the VAX curve is basically in line with the gross exports curve the rst take-o after 2002/03, a further increase two to three years after accession and the eects of the global nancial crisis. However, it is noticeable that the trend of the VAX curve is not as steep as for gross ows. This is also re ected in the relatively lower increase of VAX vis-à-vis gross exports in the 1995-2009 period, which was with approximately +300% still noticeably high. While on the one hand VAX increased with a slower pace than gross exports in the observed 15-year period, they were on the other hand also more resilient towards the economic downturn in 2008/09 and dropped only by 14%. However, this is not generalizable for VAX in the whole world economy, as VAX of CEEC-10 to non-members dropped like gross exports approximately by one quarter. This gives the notion that integration in the EU might have been a stabilizing feature for VAX during the crisis. The EU's stabilizing function during the nancial crisis remains as an interesting question for further research. Nonetheless, the focus is returned to the 2004 accession's eect on CEEC-10's VAX by subsequently examining the composition of VAX.
Figure 4: Composition of VA exports from CEEC-10 to EU-27
Source: Own calculations based on WIOD.
this reason, the notion of industrial upgrading in CEEC-10's exports to other EU members is given. This is nevertheless only a rst impression based on a descriptive data analysis. In order to assess if industrial upgrading is in fact linked to the CEEC-10's entry into the EU or is attributable to a general transformation in the new member states, a more in-depth analysis is needed. This should be subsequently done via the application of a theoretically founded gravity model. The underlying idea is to test via an EU membership dummy if the accession set further impetus on CEECs' intra-EU gross exports, VAX and industrial upgrading vis-à-vis CEECs' trade with non-members.4
4.2 Empirical Analysis
4.2.1 Distance's diering eect on gross and VA exports
In a rst step of the empirical analysis the question of distance's eect on VA ows should be reexamined. Like previously outlined Johnson and Noguera (2012b) found that VA exports travel further than gross trade. In this paragraph it should be tested whether this notion can also be con rmed for the WIOD dataset. Going beyond Johnson&Noguera's (2012b) study, it is sought to examine if the distance eect diers for VA exports in manufacturing and services sectors. In both speci cations proposed by Baier and Bergstrand (2007) distance as a country-pair speci c characteristic - would be cancelled out: (i) Using country-time and country-pair xed eects, physical distance would be perfectly collinear to latter dyadic xed eects. (ii) Taking rst dierences of all variables, the time-invariant distance between two countries would be canceled out similarly. Therefore, in order to estimate the eects of distance the Anderson and van Wincoop (2003) speci cation is used. Following equation describes the estimated model:
ln(yij,t) = α0+ β0ln(GDPi,t) + β1ln(GDPj,t) − β2ln(dij) + β3Cij+ β4M Ri+ β5M Rj+ β6F Et+ εij,t (4)
where yij,t describes the respective trade ow beteween country i and j in period t. α0 is a constant,
GDPi,t and GDPj,t describe the economic masses of the trade partners and dij the geographical distance.
Cij contains further time-invariant bilateral characteristics, including the number of countries in the dyad
that are landlocked, as well as dummies for common o cial languages, colonial relationships and whether the countries share a common border. MRi and MRj describe xed eects that proxy the unobservable
multilateral resistance terms introduced by Anderson and van Wincoop, whereas F Et exhibits the time
xed eects. Finally, εij,t depicts the pair- and time-speci c error term.5 Robust standard errors, clustered
around the cross-section of country-pairs, were used to mitigate potential bias due to serial correlation and heteroskedasticity. Table 2 displays the results for this estimation with (1) the log of exports, (2) the log of
4 An overview of the statistical properties of the underlying dataset can be found in the appendix.
5 The Anderson&van Wincoop (2003) approach is more susceptible to endogeneity concerns vis-à-vis the Baier&Bergstrand
Value Added exports and its disaggregation for (3) the manufacturing and (4) services sector respectively.6,7
In a rst step the impacts of distance on exports and VAX on an aggregate level are compared. It becomes Table 2: Distance and its eect on VA and gross exports (AvW-speci cation)
(1) (2) (3) (4)
Exports VAX aggregate VAX manufacturing VAX services
ln(GDPi) 0.797∗∗∗ 0.756∗∗∗ 0.774∗∗∗ 0.776∗∗∗ (0.000) (0.000) (0.000) (0.000) ln(GDPj) 0.909∗∗∗ 0.925∗∗∗ 0.973∗∗∗ 0.897∗∗∗ (0.000) (0.000) (0.000) (0.000) Landlocked(One) -1.730∗∗∗ -1.146∗∗∗ -1.170∗∗∗ -1.158∗∗∗ (0.000) (0.000) (0.000) (0.000) Landlocked(Both) -3.201∗∗∗ -2.144∗∗∗ -2.204∗∗∗ -2.128∗∗∗ (0.000) (0.000) (0.000) (0.000) ln(Distanceij) -1.423∗∗∗ -0.877∗∗∗ -0.971∗∗∗ -0.808∗∗∗ (0.000) (0.000) (0.000) (0.000) Contiguityij 0.379∗∗ 0.349∗∗∗ 0.341∗∗∗ 0.341∗∗∗ (0.003) (0.000) (0.001) (0.000) Colonyij 0.346∗ 0.225∗ 0.175 0.258∗∗ (0.019) (0.021) (0.112) (0.007) Common_languageij 0.142 0.107 0.126 0.103 (0.230) (0.199) (0.153) (0.229) Constant -18.91∗∗∗ -29.47∗∗∗ -31.82∗∗∗ -30.42∗∗∗ (0.000) (0.000) (0.000) (0.000) N 23340 23400 23399 23400 R2 0.893 0.948 0.952 0.940 p-values in parentheses ∗p < 0.05,∗∗p < 0.01,∗∗∗p < 0.001
visible that the eects of distance are larger for gross exports than for VAX. The repressive eects of physical
6 Estimates are based on trade ows and GDP data, which were de ated with country-speci c de ators from the World Bank
in order to avoid bias caused by inappropriate de ation with US-CPI de ators. The estimates of the xed eects are not reported in order to save space. A test on multicollinearity of the independent variables is available in the appendix as well.
7 It has to be kept in mind that VAX is by de nition a part of the GDP. Due to this fact the basic Anderson&van Wincoop
distance and being landlocked is markedly more negatively correlated for gross exports. In a same manner the enhancing eects of contiguity and colonial history exhibit at least a slightly larger impact on gross exports. This nding is in line with the research by Johnson and Noguera (2012b). Also intuitively this makes sense, as an exporting country and more speci cally its rms earn on average more from trading one unit of pure domestic VAX than one unit of gross exports, which might embody large shares of foreign inputs. Therefore, the willingness to cover potential transport costs would be on average larger for the exporter of pure VAX.
Zooming in on the sectorial level of VAX regarding the estimates in Column (3) and (4), the notion that VAX are less aected by distance than gross exports is again con rmed. Furthermore, comparing the regression results for both sectors, it is visible that (3) VAX in manufacturing seem to be more sensitive to the repressive eects of distance regarding the landlocked and physical distance coe cients. An explanation might be that industrial goods in Column (3) are more aected by transport costs, related to physical distance. In contrast, services can be delivered via the channels of modern ICT and thus are not that contingent on transport cost. Further on, the coe cient for colonial history is signi cant and positive in the speci cation with services as dependent variable, while insigni cant for manufacturing. In this regard it could be assumed that colonial history aects VAX in services due to legal similarities, lowering adaption costs for exported services. Sumarizing the results, the notion of a more repressive eect of the various forms of distance on gross exports vis-à-vis VAX was con rmed.
After this short review on the empirical evidence concerning the young topic of the relationship of VAX and distance, it should be turned towards the main research question of this paper: The eects of EU acces-sion on the CEECs.
4.2.2 First dierences accession eect estimates
In this section the main research hypotheses of this paper, outlined in the introduction, should be examined. These are namely whether an increase in gross exports (H1), increased VAX (H2) and industrial upgrad-ing/downgrading (H3a/b) were induced by the CEEC-10's EU accession. The accession eect is modeled in the gravity setting via the dummy variable eu04eu, which equals one if the exporting country is a 2004 EU entrant trading with an EU member and is zero in all other cases.
As it was outlined previously, the Anderson and van Wincoop (2003) approach of country and time xed eects might be susceptible to endogeneity bias regarding EIA-eects. Therefore, the modeling techniques of Baier and Bergstrand (2007) are used subsequently in order to derive estimates for the EU accession rounds, mitigating endogeneity concerns. It was stated before that the e ciency of xed eects estimators might be adversely aected by serial correlation. A test on serial correlation, which is provided in the appendix, reveals that serial correlation is indeed an issue in the xed eects speci cation. First dierencing seems to mitigate serial correlation issues. For this reason the rst dierencing technique is used subsequently to examine the research hypotheses.8
For the speci c case examined in this paper, the rst dierences model can be described as follows: yij,t− yij,t−1 = β0(eu04euij,t− eu04euij,t−1) + β1(M Ri,t− M Ri,t−1) + β2(M Rj,t− M Rj,t−1) + vij,t−(t−1) (5)
where yij,t−yij,t−1and eu04euij,t−eu04euij,t−1exhibit the rst dierences of the dependent trade variable
(e.g. VA or gross exports) and the rst dierences of the 2004 accession coe cient. MRi,t− M Ri,t−1 and
M Rj,t− M Rj,t−1 are the respective country-time xed eects for country i and country j. Latter can
be interpreted as the rst dierences or changes of multilateral resistance terms. Finally, vij,t−(t−1) =
eij,t− eij,t−1 depicts white noise (Baier and Bergstrand, 2007).9
As stated, the rst dierencing of all variables leads the time-invariant characteristics to drop out. This includes distance, colonial relationships, contiguity, as well as the dyadic xed eects. Nonetheless, country-time xed eects (being country-time-variant) remain in the model (Baier and Bergstrand, 2007). They avoid bias by accounting for e.g. price changes at the country level. This is especially important as many of the CEECs, among them Romania, Lithuania, Latvia, Hungary, Estonia and Bulgaria, faced tremendous in ationary pressure in their transformation phase.10 Like in the previous estimation cluster robust standard errors are
used.11
Table 3: First Dierences - Baseline Estimates
(1) (2) (3) (4) (5) (6)
Gross_Exports VAX_Total VAX_Capital VAX_LabHS VAX_LabMS VAX_LabLS Panel (i) eu04euij,t−(t−1) 0.00924 0.0809∗∗ 0.0526 0.0653∗ 0.0878∗∗ 0.131∗∗∗ (0.871) (0.002) (0.080) (0.012) (0.001) (0.000) N 21771 21840 21816 21840 21840 21840 R2 0.222 0.507 0.509 0.521 0.507 0.530 Panel (ii) eu04euij,t−(t−1) 0.00924 0.0809∗∗ 0.0526 0.0653∗ 0.0878∗∗ 0.131∗∗∗ (0.871) (0.002) (0.080) (0.012) (0.001) (0.000) eu04euij,t−(t−6) 0.00369 0.0944∗∗∗ 0.114∗∗∗ 0.0694∗ 0.0873∗∗ 0.0978∗∗ (0.950) (0.000) (0.000) (0.012) (0.001) (0.002) N 21771 21840 21816 21840 21840 21840 R2 0.222 0.507 0.510 0.521 0.507 0.530 p-values in parentheses ∗p < 0.05,∗∗p < 0.01,∗∗∗p < 0.001
9 The rst dierences of the accession eect are subsequently depicted as eu04eu
ij,t−(t−1).
10 In Appendix Table 12 it is shown that the approach of Baier and Bergstrand (2007) is more robust, if it is accounted for
de ation. Including the country-time xed eects helps to avoid the bronze medal error of using wrong de ators (Baldwin and Taglioni, 2007).
11 First-dierencing estimates make use of Stata's reg2hdfe command, which help to reduce computation times signi cantly
Table 3 shows the results of the model both excluding and including phasing-in eects. In latter esti-mations ve year lags of the rst dierence of the EU accession coe cients - eu04euij,t−5− eu04euij,t−6
- are added to equation (5).12 Turning the attention in a rst step to the results of the basic estimations
without lags in Panel (i), it becomes visible that four coe cients are signi cant. Signi cant coe cients can be found namely in the estimations with VAX and its labor contributions as dependent variable (Column (2) & (4)-(6)). VA exports in Column (2) seem to be positively aected, while gross exports in Column (1) remain stable. Furthermore no signi cant increase of VAX by capital linked to EU accession is found in Column (3). This implies a decrease in the capital-labor ratio as VAX increase for all labor-skill levels. Above that the results suggest that the percentage increase of low-skilled labor in Column (6) is twice as high as the increase for high-skilled labor in Column (4). A decrease in the labor-skill ratio is therefore implied.
If now the results of the regressions including a phasing-in eect of accession in Panel (ii) are analyzed, it becomes visible that an inclusion of the lags increases the added-up accession impact markedly.13 The
initial level-eects of accession are not aected by the inclusion of lags with regard to signi cance and size. Furthermore now even all regressions with a VAX related dependent variable show a signi cant increase due to the entry into the EU 2004, while the gross exports seem to be not changed signi cantly. VAX added by all production factors increases namely by 18%.
Investigating the contributions of speci c production factors in more detail, the inclusion of lags seems to support previous results of the basic speci cation. The eect on VAX by capital in Column (3) of Panel (ii) is with an increase by 12% smaller than the eect on aggregate VAX. This implies a decreasing capital-labor ratios in CEEC-10's VA exports to the EU. Moreover, the increase of VAX by low-skilled labor in Column (6) of Panel (ii) is the largest accession eect with circa 24%. This impact surpasses the increases of VAX by high- and medium-skill labor, which exhibited 14% and 18%. Therefore, the labor-skill ratio in VA exports is actually decreasing. This indicates moderate downgrading trends combined with increasing VAX due to the EU entry.14
This is rather a surprising nding, as Timmer et al. (2013), using the same data-set, found mainly an increase in high- and medium-skilled GVC workers for CEEC-10 countries between 1995 and 2008. Addi-tionally, the authors located the largest increases in services sectors. For this reason more in-depth analysis should be carried out in the subsequent section, making use of the sectorial data provided by the WIOD.
12 The rst dierences of the ve year lagged accession coe cient are subsequently depicted as eu04eu
ij,t−(t−6).
13 If lags are calculated with statistical software, a part of the observations cannot be computed due to a lack of previous
observations. In order to avoid unnecessary reduction of sample size the lagged values of EU accession were manually adjusted. When non-adjusting, the sample-size was reduced by 7,800 observations. Same applies for lead values, which are used later on in this paper. A more detailed explanation on the adjustment procedure is provided in association with the do- le on Table 3, which can be found in the appendix.
14 Estimates of percentage changes are for all estimations referring to the summation of baseline and theoretically motivated
4.2.3 Digging deeper Accession eects on sectorial VAX
The WIOD oers detailed Input-Output structures for 35 sectors in all 40 countries. In order to elucidate the accession eects and their implication on structural change in VAX, the economy is divided into the secondary and tertiary sector. More speci cally the data is split up in two sub-aggregates of the (i) fourteen manufacturing and the (ii) twenty service sectors in the World Input Output Tables (WIOTs).15
Accession eects on manufacturing sectors
In a rst step the estimates for the manufacturing sectors are examined in Table 4. VAX seem to be the more meaningful dimension in the new paradigm of the second unbundling vis-à-vis gross exports. For this reason, the sectorial analysis will mainly focus on the comprehensive data on VA, which is oered by the WIOD. Examining the basic eects of the EU accession dummy eu04euij,t−(t−1)in Panel (i), all coe cients
are signi cant and positive. Therefore, VAX on an aggregate level as well as by all factors is increasing. Considering the capital-labor ratio, the accession eect on VAX by capital in Column (2) can be identi ed as the lowest. This consequently implies a slightly negative accession eect on the capital-labor ratio in CEEC-10's manufacturing VA exports. Concerning the labor-skill ratio, the estimates of the EU entry's eect are surprisingly similar for the dierent skill levels. This is also con rmed if it is accounted for phasing-in via the ve year lag of the membership coe cient eu04euij,t−(t−6)(see Panel (ii)). All factors show increases of VAX
due to integration into the EU. Again it is capital that shows the smallest coe cient, while the eects on labor seems to be equally distributed across skill-sectors. This is especially interesting as the manufacturing sector is generally perceived as capital intensive. The CEEC-10's however, exhibit a decrease in the capital-labor ratio after accession. In this context the competitive advantage of the CEEC-10's factories might be not attributable to capital, but to its supply of relatively cheap labor comprising all skill levels. While manufacturing's VAX by capital increased by 17%, VAX by manufacturing workers increased on all skill levels by 22-23%. Interestingly there is no discrimination by skill-levels, pointing to the large low-cost labor pool of the CEEC-10, comprising all skill-levels (Marin, 2006). Taken together, the accession eect seems to have induced an increase of manufacturing's VAX by circa 21%.
Accession eects on services sectors
A slightly dierent picture is derived for the services sector. Table 5 reveals that in the basic regression in Panel (i) only the accession coe cient is statistically signi cant in the regression with VAX by medium-skilled labor as the dependent variable. While the EU entry seems not to have promoted higher VA exports in CEEC-10's service industries, there seems to be a positive and signi cant eect for the contribution of medium-skilled workers in Column (4). Also the positive eect for low-skilled labor in Column (5) is with a p-value of 0.051 close to statistical signi cance. However, the previous panel-estimations showed that parts of the accession eects are missed, if phasing-in of the EIA is not considered. This is underlined by the regressions including the lagged eects in Panel (ii). Adding the lagged coe cients to the mostly insigni cant
