August 2014 Academic year 2013/2014
Supervisor: Dr. D. J. M. Veestraeten Second reader: Drs. N. J. Leefmans
Master’s Thesis
Faculty of Economics and Business
Estimating the Effect of Relative Specialization on
Divergence in Economic Growth within the European
Union and Euro-area
A Panel Data Approach
Maarten Tellegen
6383424
A Master’s Thesis Submitted in Partial Fulfilment of the Requirements for the Degree of Master of Science in Economics
Contents
1 Introduction ... 2
2 Convergence and Specialization within the Optimum Currency Area Framework ... 4
2.1 Historical Overview ... 4
2.2 The New Theory of Optimum Currency Areas ... 6
2.3 The Endogenous Nature of Optimum Currency Area Criteria ... 8
2.4 The Likelihood of Formation of Clusters within the EU... 10
3 Empirical Strategy ... 14
3. 1 Measuring Business Cycles ... 15
3.2 Measuring Relative Specialization ... 15
3.3 Measuring Synchronisation ... 16
3.4 Specifying Additional Explanatory Variables ... 17
3.5 Limitations of the model ... 18
4 Related Results from the Literature ... 20
5 Sample and Data ... 22
6 Empirical Results ... 24
6.1 Analysis of the Development of the Specialization Index over Time ... 24
6.2 The development of the Specialization Index and the EU ... 26
6.3 The Development of the Specialization Index and the Euro-area ... 29
6.4 Regression Results ... 30
6.5 Validity Issues ... 33
6.6 Interpretation and Discussion of the Results ... 34
7 Summary ... 36 8 Literature ... 37 Appendix I ... 39 Appendix II ... 40 Appendix III ... 40 1
1 Introduction
A considerable part of legislation within the European Union (EU) is aimed at convergence in macro-economic policy among individual Member States. The Stability and Growth Pact (SGP) is a rule-based framework for the coordination of national fiscal policies in the Economic and Monetary Union (EMU). It requires Member States to submit Stability or Convergence Programs (SCPs) to the European Commission (EC), which are at the basis of the multilateral surveillance of budgetary positions and the coordination of economic policies. Non-euro area members present convergence programs while euro area members present stability programs. The basis of the SCPs is formed by two rules; the budget-deficit of individual member states must not exceed 3% of GDP and public debt must not exceed 60% of GDP. In case either of the two thresholds is exceeded a member should diminish sufficiently towards the threshold. Following the sovereign debt crisis, economic and fiscal governance rules are fundamentally strengthened. The SGP is now imbedded in the Six-Pack (came into force in December 2011), which strengthened both the preventive and the corrective arm of the Pact. In addition, the Fiscal Compact - which came into force on 1st of January 2013 - requires Member States to enshrine the country-specific medium-term objectives in national binding law, preferably of constitutional nature (European Commission, 2013)
The EU currently consists of 28 Member States. Of these 28 Members States 17 are member of the Euro-area as well. Members of the Euro-area have a common currency, the Euro. Having a common currency entails conducting common monetary policy, which in the Euro-area is done by the European Central Bank. The consequence of the centralization of monetary policy is that internal (between members of the Euro-area) exchange rates no longer can be adjusted. Moreover, the external exchange rate is equal for all members of the Euro-area. Members of the EU that have not adopted the euro are expected to participate for at least two years in the European Exchange Rate Mechanism II (ERM II). ERM II puts limitations on exchange rate fluctuations; the currencies are allowed to fluctuate as much as 15% from their assigned value.
From above it follows that a member of the EU is in two ways constrained in conducting autonomous macro-economic policy:
1. Room to manoeuvre with fiscal policy is restrained within clear boundaries;
2. For members of the EU that are not a member of the Euro-area yet but aim at becoming a member, monetary policy is restricted. For members of the Euro-area, the possibility of conducting autonomous monetary policy is completely absent
In addition, EU-members have, through the establishment of the internal market completely abandoned autonomous trade policy. These constrains severely limit a member of the EU to respond independently to economic shocks. This can be problematic in case economic shocks are not equal (either in direction or time) for all members; that is asymmetrical. From this potential problem follows the desirability of convergence in economic growth rates between members of the EU.
As stated above, members of the EU form a single market. A consequence is that internal trade barriers no longer exist. This could in theory advance the formation of specialised regional economic clusters within members of the EU, where a regional economic cluster means the geographic concentration of economic activity. In case this formation of specialized clusters actually takes place, the consequence can be that the economic diversity of a member of the EU decreases; members become specialized. If this specialization takes place for multiple members of the EU and those members specialize in different fields, the economic diversity between members’ increases.
If the economic diversity between members of the EU, as a consequence of the formation of specialized clusters, increases, the risk of shocks with asymmetric effects increases as well. This in turns threatens the viability of the EU and EMU. This paper examines the latter danger. In particular, this paper addresses two main questions:
1. Is there growing specialization within the EU and the Euro-area?
2. What is the effect of this specialization on the convergence in economic growth between members of the EU and Euro-area?
This paper is constructed as follows; chapter 2 examines the literature in order to formulate a theoretical hypothesis for the two main questions of this paper. In chapter 3 we develop the empirical strategy that is used in this paper to study specialization and convergence in economic growth. Chapter 4 briefly discusses relevant findings of other empirical studies. Chapter 5 describes the data and sample used in this paper. Chapter 6 describes and interprets the empirical results. Chapter 7 concludes this paper with a brief summary of the findings.
2 Convergence and Specialization within the Optimum Currency Area
Framework
The debate on the desirability of monetary integration has a long and rich history. Through time this debate has become a field of research of its own within the field of international economics, and is general addressed as the discussion on Optimum Currency Areas (OCA). A currency area is a domain within which exchange rates are fixed. One can distinguish further between a currency area and a monetary union; a currency area has irrevocably fixed exchange rates, capital mobility and the members of the currency area conduct common monetary policy. The Euro-area is called a monetary union, which is the stronger form of a currency area. The fixed exchange rate is replaced by a common currency and instead of conducting coordinated monetary policy, monetary policy is centralized at the European Central Bank. For a country, the optimum currency area/monetary union is realised within that group of countries for which the positive difference between benefits and costs of monetary integration is maximised (Mavromatis, 2013).
2.1 Historical Overview
The natural starting point when discussing OCA is the influential paper A Theory of Optimum
Currency Areas by Mundell (1961). Mundell sketches an OCA by systematically examining
hypothetical configurations of countries or regions and their exchange rate regimes. Consider two entities that are initially in full employment and balance-of-payment equilibrium. Money wages and prices cannot be adjusted in the short run without causing unemployment, and monetary authorities act to prevent inflation. What happens if demand shifts from entity A to entity B? If the entities are two countries with a fixed exchange rate, the shift in demand causes unemployment in A and inflationary pressure in B. If B would allow prices to rice, the terms of trade of A would improve and thereby would relieve it of some of the burden of adjustment. The priority of B is however preventing inflation. B tightens credit restrictions, and A’s output and employment will decline further. Now suppose that the entities are regions within a closed economy with a common currency, and suppose that the national government pursues full-employment. A shift of demand from region A to B, again, causes unemployment in A and inflationary pressure in B. Since full-employment has priority, the government increases the money supply; inflation rises in B, improving the terms of trade of A and thereby correcting unemployment in A. Therefore, a fixed exchange range or common
currency between countries or regions creates an inflation-unemployment trade-off when confronted with asymmetric demand shocks.
In the country example, depreciation by country A in response to the shift in demand from A to B would correct the external imbalance. This correction through the exchange rate would mitigate unemployment in country A while at the same time restrain inflation in country B (Mundell, 1961). Does this imply that countries should always have a flexible exchange rate between each other in order to circumvent the inflation-unemployment trade-off? The answer is no. Consider a two country world in which both countries consist of two regions; West and East. The countries have a separate currency. Moreover, West produces cars in both countries and East produces lumber products, again in both countries. What is the effect of a shift in demand from cars to lumber products? The effect within each country is the same as the region example above; unemployment pressure in West, which can be circumvented by rising inflation in the East. Key is that this trade-off arises in both countries simultaneous; flexibility of the exchange rate between the two countries would not serve to correct the balance-of-payments situation between the regions, which is the essential problem (Mundell, 1961). The above analysis leads Mundell to conclude that the optimum currency area is the region. What then defines a region? A region that is best suited to have a single currency is a region that has (high) internal factor mobility and (low) external factor immobility (Mundell, 1961). This implies that an optimal currency area could consist of multiple countries as long as there is factor mobility between them, but at the same time, even nations could be unfit to have one currency in case they have low internal factor mobility.
McKinnon (1963) discusses the benefit for a (small) open country to engage in a currency area. Open is defined as a country that has a high tradable to non-tradable ratio. If a country is sufficiently small, the money prices of the tradable goods in terms of the outside currency are not influenced by domestic exchange rates or domestic choices. Or, the terms of trades will be immune to domestic economic policy. If exportables and importables make up a large percentage of the goods domestically consumed, flexible exchange rates become both more damaging to internal price level stability and less effective as a control device for external balance. The former because exchange rate movements translate directly into domestic price changes, due to lack of influence on world prices. The latter follows than from the former since prices for importables and exportables are equally affected. In the extreme case of a completely open country, e.a. the country only consumes tradables, the external balance will 5
not be affected at all (McKinnon, 1963). From this, it follows that the higher the extent of trade between countries the more suited they are to form a currency area.
The similarity of shocks and cycles is a pre-condition of an optimum currency area but is itself the result of the economic design of a country as well. Kenen (1969) discussed the conditions under which asynchronous macroeconomic shocks across countries would become less likely. He suggested that the higher product diversification in countries is that potentially form a currency area, the lower the risk of asymmetric shocks between them. The more output is diversified within a country the lower is the impact of a sector specific shock on total output. The law of large numbers then dictates that sector specific shocks will be offset between countries that form a currency area (Kenen, 1969; McKinnon, 2001; Horvath & Komarek, 2002). It could be that countries form a currency area while not having synchronised business cycles. It is also possible that factor mobility is low between those countries. In this case, a high degree of fiscal integration could dampen the (negative) effect of asymmetric shocks. Centralizing fiscal policy allows for a system of budgetary transfers from countries that are favourably affected to the countries that are negatively hit (Bordo et al., 2011; Mavromatis, 2013).
2.2 The New Theory of Optimum Currency Areas
Mongelli (2002) addresses the work by Mundell, McKinnon and Kenen as the pioneering phase. A problem with the OCA properties formulated above is that they are difficult to measure unambiguously and to evaluate against each other. Mongelli provides a detailed literature overview regarding the theory of OCA in which he furthermore distinguishes between the reconciliation phase (1970s), the re-assessment phase (1980s and early 1990s) and the empirical phase (last 20 years). During the reconciliation phase, the OCA properties started to be analysed and weighted with one another to gauge their relative importance. The analysis of the cost and benefits of joining a currency area acquired more structure (Mongelli, 2002).
After the pioneering phase and the reconciliation phase, development of OCA theory lost some momentum. On one side European monetary integration slowed down, on the other side the analytical framework behind the OCA theory weakened (Mongelli, 2002). Indeed, the early view on monetary policy was that there is a clear trade-of between inflation and unemployment. However, in the 1970s and early 1980s some countries experienced rising unemployment as well as rising inflation. This led to the conclusion that expected inflation 6
should be taken into account, and that short-term changes have no impact on real variables. This suggests that monetary policy has no real effect. Hence, the cost of losing monetary policy, by entering a currency area or monetary union, as an instrument for macro-economic policy is smaller than suggested by for example Mundell (1961). Along the same line of reasoning, the effect of fiscal policy is debated. For example, expansionary fiscal policy through a reduction in taxes should lead to an appreciation of the exchange rate. However, if consumers are forward looking, they expect that the tax reduction is followed by a future tax increase to compensate for the tax reduction. Instead of consuming more, they save in order to pay the future tax. As a result, there will not be an increase in consumption and hence no exchange rate appreciation. These insights undermine the assumed effectiveness of fiscal policy measures as a tool to manipulate the exchange rate, and thereby reduce the presumed loss of limiting fiscal policy measures when joining a currency area (Mongelli, 2002; Mavromatis, 2013). In the second half of the 1980s interest in monetary integration regained momentum. Members of the European Union re-launched the idea of a European Economic and Monetary Union and were confronted with concerns regarding the timing and modalities of implementing a currency union once the political decision to create one had been taken. The revisions of the analytical framework behind the “old” OCA theory led to a “new” OCA theory. The theoretical and empirical advancements resulted in a reassessment of the main benefits and costs form monetary unification and the balance of judgement shifted in favour of currency areas (Mongelli, 2002).
Theoretical innovations, advancements in econometric techniques, and the renewed interest towards European economic and monetary integration led to a rise in the number of empirical studies on the diverse OCA properties in the past 20 years. The aim of this empirical phase is to operationalize the OCA theory. This is done by assessing why specific groups of countries may form an OCA by analysing and comparing OCA properties using several econometric techniques. A large share of the empirical studies investigating OCA properties are backward looking (Mongelli, 2002). This is partly because the OCA criteria discussed so fare are formulated as preconditions; countries should satisfy them prior to forming a currency area. However, as pointed out by Frankel and Rose (1998), the OCA criteria are largely endogenous. Taking the synchronization of business cycles as an example, Kenen (1969) stated that the higher the synchronization of business cycles the lower the cost of forming a currency union. However, synchronization could occur after countries become a currency
union. More strongly, this synchronization can occur precisely because the countries form a currency area.
2.3 The Endogenous Nature of Optimum Currency Area Criteria
Figure 2.1 sketches the possible relations between the effects of joining a currency area and the symmetry of output fluctuations discussed in this chapter. Coe and Helpman (1995) estimate the effect of Research and Development (R&D) on Total Factor Productivity (TFP). They assume that in a world with international trade in goods and services, foreign direct investment, and an international exchange of information and dissemination of knowledge, a country’s productivity depends not only on its own R&D but also on the R&D efforts of its trade partners.
Figure 2.3.1 Relations between effects of joining a currency area and the symmetry of output fluctuations
Source: Kalemli-Ozcan et al. (2001)
More similar supply (knowledge spillovers) More similar policy Less trade barriers More capital market integration More demand spillovers and trade Output fluctuations asymmetry More industrial specialization and trade
-+
-a c b d e 8Some of their estimates suggest that foreign R&D capital stocks have stronger effects on domestic productivity the larger the share of domestic imports in GDP. This let them to conclude that economies which are more open, extract larger productivity benefits from foreign R&D than less open economies (Coe and Helpman, 1995). Since TFP is an important determent of economic growth, the more knowledge spillovers (in the form of R&D) take place between countries the more synchronised output fluctuations become. If, as a consequence of forming a currency area, trade integration between members increases, productivity shocks may spread more rapid, hence raising the covariance of the growth rates of the members (Frankel and Rose (1998), p. 1014. Relation a in figure 2.3.1).
The emphasis within the EU on convergence in fiscal policy (the deficit and debt requirements) could lead to convergence in output growth as well. This relation is formalized and tested empirical by Darvas et al. (2005). They point out that the theoretical relation between fiscal policy convergence and business cycle synchronization is ambiguous. In case of country-specific shocks, countries can react with countercyclical fiscal measures (the automatic stabilizers). In this case, diverging fiscal policy leads to more converged business cycles. The opposite can occur as well, countries can respond to country-specific shocks with procyclical fiscal measures, in which case the divergence in business cycles is amplified. However, fiscal policy is not only a response mechanism, but can be a source of shocks as well. Darvas et al. find, using a panel of 21 OECD countries observed over 40 years, that countries with similar budget positions tend to have business cycles that fluctuate more closely. They attribute this to the observation that fiscal divergence tends to occur when a country runs a substantially and persistently higher budget deficit than other countries, and simultaneously creates fiscal shocks. They find that irresponsible fiscal policy (a persistently high deficit) coincides with idiosyncratic (fiscal) instability. Stimulus to close the budget deficit, for example the deficit and debt requirements within the EU, supresses the fiscal shocks. So convergence in fiscal policy increases convergence in business cycles (Darvas et al., 2005. Relation b in figure 2.3.1).
As can be seen in figure 2.3.1 the effect of the removal of trade-barriers on the convergence in business cycles is ambiguous; two opposite paradigms with different implications have been put forward (Mongelli, 2002). Frankel and Rose (1998) estimate the relation between the intensity of trade links and the correlation of business cycles using a panel of 21 industrialized countries observed over thirty years. Although they leave the potential outcome open when forming their hypothesis they do express a strong expectation towards a positive relation. 9
They mention a number of channels that effect this relation in a positive way. The removal of trade barriers intensifies trade relations. This will increase aggregate demand-spillovers since an increase of public or private spending in one country tends to raise demand for both foreign and domestic output (Relation c in figure 2.3.1). The effect on the supply side of intensifying trade relations on the symmetry of output fluctuations is less straightforward. Frankel and Rose (1998) acknowledge that more trade liberalization may lead to specialization. However, they highlight that intra industry trade accounts for a major share of international trade. They state that if most trade remains intra-industry specialization will be low (Frankel & Rose, 1998). On the other side of the paradigm is the already mentioned specialization effect of the formation of a currency area.
2.4 The Likelihood of Formation of Clusters within the EU
Why are specialized clusters likely to arise and/or increase in the EU? Krugman addresses this question in his paper Lessons of Massachusetts for EMU (1993). It is clear that the integration of markets leads to more regional specialization when it comes to goods that are highly suited to be produced in that particular region; for example producing oranges in Spain using the sun instead of producing them in Netherlands using greenhouses. Krugman deepens this by stating that a reduction in the transaction costs between two regions, whether these costs take the form of transportation expenses, tariffs, or disparities in regulation, will make it more likely that any given degree of external economies will be sufficient to lead to geographical concentration of an industry (Krugman (1993), p. 244).
Consider a two-country model and an industry that can operate in either of the two countries. Demand for the product produced by the industry exists in both countries and the industry is assumed to be perfectly competitive, but subject to location specific external economies. Figure 2.4.1 (next page) outlines the situation prior to the removing of trade barriers. Both countries are self-sufficient, with country i producing OQ at cost c and country j producing
QO* at cost c*. So country i has a cost advantage of c-c*, which may either be due to a larger
home market, or because of some factor cost advantage. This cost advantage, however, is not sufficient to lead to concentration of the industry. As long as the transaction costs exceed c-c* a geographically dispersed industry is a stable outcome. However, if the transaction costs drop below c-c*, because of the removal of trade barriers for example, a snowballing cost advantage for country i arises leading to a geographically concentrated industry (Krugman (1993) p. 244-245. Relation d in figure 2.3.1).
Figure 2.4.1Two country model prior to the removal of trade barriers
Source: Krugman (1993)
This model leaves room for debate. One potential flaw is addressed by Krugman himself. External economies are often market size effects; is it still necessary to be in the same country to have access to the forward and backward linkage effects when markets become more integrated? Although the answer is no, Krugman argues that integration will still lead to greater geographical concentration of industries. Consider an industry that faces demand from two countries, where demand is equally divided between the two countries. Moreover, firms operating in this industry are symmetric, meaning that they all deliver to both markets, and sell a fraction of output, 𝜇𝜇, as intermediate goods to other firms in the industry. If firms can only have one plant, the choice of were to settle is easy; settling near the other firms saves the cost of transporting fraction 𝜇𝜇. However, what if a firm is allowed to open an additional plant at a fixed cost? Opening a plant in the foreign country saves the cost of transporting the fraction sold in the foreign country. This is an equilibrium as long as the fixed cost does not exceed the transportation cost. Remarkable about this equilibrium is that its likelihood is decreasing when transportation costs decrease.
The theoretical model of Krugman focusses on external economies of scale. The effect of the removal of trade barriers on specialization through internal economies of scale is less straightforward. The expansion of the market size through the removal of trade barriers allows firms to exploit economies of scale, leading to big specialized firms. This increases
inter-Cost C C* C C* c* c O Q O* 11
industry trade However, the increase of market size also increases demand for different varieties allowing countries to specialize in producing one or more varieties. This leads to more intra-industry trade. Although countries become more specialized on the variety level, they do not on the sector level.
Casella (1993) agrees with the fundamental economic proposition that specialization in response to trade is the main source of gains from free exchange. However, she calls into question the identification of the relevant economic region by asking what a region is and to what extent we can identify an economic region with a given political border. Economic regions may be much smaller, so that a country contains many economic regions. Or economic regions may overlap political borders, in which case regional specialization does not coincide with national specialization (Casella, 1993). Closely related is the critique by Frankel (1999) who performs a thought experiment with implications of the model by Krugman; if a region is diversified enough to form a currency area with a other region that itself is diversified enough, the combination of this two regions is even more diversified and should form itself a currency union with another region. This process has the corner solution of a single currency for the entire world. If however the region is not diversified enough to form a currency area itself, it should break op into smaller currency units. But these currency units will be even less diversified, and should break up itself. This process has the corner solution of 5 billion currencies. However, his thought experiment seems to neglect two things. First, the model by Krugman is dynamic. It presumes an increase in specialization after the formation of a currency area, precisely because of the removal of trade barriers due to the formation of the currency area. This would break the chain of events leading to the ever-growing currency area. Second, the increase of specialization is considered a potential problem because it could lead to asymmetric business cycles. Asymmetric business cycles in turn are considered a cost within a currency area because countries have lost the ability to perform, for instance, autonomous monetary policy. This is less relevant at the country level since countries can use fiscal policy to address business cycle asymmetry within their borders. This ability will reduce the necessity to split up in ever smaller currency areas (i.e. smaller than the country level) while at the same time preventing countries from joining a currency area of multiple countries (for as far fiscal policy is not integrated).
Kalemli-Ozcan et al. (2003) demonstrate empirically that more insurance possibilities against output fluctuations are associated with higher specialization. The main mechanism for spreading risk among regions and countries is geographical diversification of income sources 12
achieved via capital markets. In order to formalize the relation between insurance and specialization, Kalemli-Ozcant et al. (2003) present a simple model where production technology exhibits increasing returns to scale. A group of regions of equal size all produce the same consumption good. Labour supply is assumed to be inelastic and there are no fixed costs. A region can chose any of several ex ante identical technologies which exhibit increasing returns to scale. This creates a trade-off between increasing returns in production and gains from diversification across technologies. In the case of perfect interregional income insurance each region will fully specialize and moreover each country will specialize in a different technology in order to fully profit from diversification (Kalemli-Ozcan, Sørensen, & Yosha, 2003. Relation e in figure 2.3.1).
This chapter has discussed the theoretical framework of the optimum currency area. The early phase of the OCA theory reveals that there are costs and benefits to forming a currency area. This phase of the theoretical debate focusses on the characteristics of countries prior to the formation of a currency area that either increase or decrease the potential benefits and costs of the formation of a currency area. Especially relevant for this paper is the observation that the costs of forming a currency area is high when output fluctuations are asymmetric between potential members of the currency are. Through time, the debate has evolved and started to focus on the endogenous nature of OCA criteria. This chapter has discussed a number of effects on the divergence of output fluctuations, that arise as a consequence of forming a currency area. Some effects enlarge the probability of asymmetric output fluctuations while others decrease this probability. The rest of this paper focusses on one specific channel, namely the effect of specialization on the asymmetry of output fluctuations. In this chapter, we have discussed the probability of the likelihood of specialization as a consequence of the removal of trade barriers. This is especially relevant for this paper since it broadens the perspective from a currency area to trade unions, like the EU. In the following chapters, we will examine the relation between specialization and the asymmetry of output fluctuations.
3 Empirical Strategy
The main interest of this paper is the effect of specialization on the asymmetry in business cycles between countries. This section will discuss the base model that will be used to estimate the effect of specialization on the symmetry of business cycle fluctuations. Frankel and Rose (1998) express the per capita GDP growth process of countries i and j as:
1. ∆𝛾𝛾𝑡𝑡𝑖𝑖 = � 𝛼𝛼𝑠𝑠𝑖𝑖𝑢𝑢𝑠𝑠,𝑡𝑡+ 𝑣𝑣𝑡𝑡𝑖𝑖 + 𝑔𝑔𝑖𝑖 𝑠𝑠
2. ∆𝛾𝛾𝑡𝑡𝑗𝑗 = � 𝛼𝛼𝑠𝑠𝑗𝑗𝑢𝑢𝑠𝑠,𝑡𝑡+ 𝑣𝑣𝑡𝑡𝑗𝑗 + 𝑔𝑔𝑗𝑗 𝑠𝑠
Where variables ∆𝛾𝛾𝑡𝑡𝑖𝑖 and ∆𝛾𝛾𝑡𝑡𝑗𝑗 are the growth rates of GDP per capita of countries i and j. Variable 𝑢𝑢𝑠𝑠,𝑡𝑡 represents a time t shock that is specific to sector s, but equal for countries i and
j. Variables 𝑣𝑣𝑡𝑡𝑖𝑖 and 𝑣𝑣𝑡𝑡𝑗𝑗 represent a shock at time t that is specific to country i or j but are
common to all sectors in those specific countries and variables 𝑔𝑔𝑖𝑖 and 𝑔𝑔𝑗𝑗 are the trend rates of output for countries i and j. 𝛼𝛼𝑠𝑠𝑖𝑖 and 𝛼𝛼𝑠𝑠𝑗𝑗 represent the weight of sector s in total output of countries i and j.
From this simple representation follows immediately the effect of specialisation on the correlation in GDP fluctuations. If country i specialises in sector 1 relatively to country j, the weight of this sector in total output, measured by 𝛼𝛼1𝑖𝑖, in country i also increases relatively to country j. In that case, a shock to sector 1 will affect countries output stronger than country j. So specialisation will reduce the correlation in GDP fluctuations.
In order to implement the above representation in a testable model three things need to be done:
1. Fluctuations in the business cycle need to be isolated and made measureable; 2. Specialisation needs to be made measurable;
3. The model has to control for factors other than specialisation that influence movements of the business cycle.
3. 1 Measuring Business Cycles
When measuring business cycles a distinction has to be made between classical business cycles and deviation cycles. Classical business cycles are defined in terms of absolute expansions and contractions of economic activity. However, since business cycles in general move around a (upward) trend most classical business cycles are not stationary. This is why most research focusses on deviation cycles, i.e. the deviation of economic activity from a trend. The most straightforward way of decomposing output into trend and cycle is by calculating first differences, as will be the technique used in this paper (Haan et al., 2008). A limitation of using first differences is that it does remove a trend from time-series, but potentially at the cost of a shift in the peaks and troughs of the differenced series and a larger volatility, as argued in a paper by Baxter and King (1999). In order to circumvent this limitation a variety of filters exist, like the Hodrick-Prescott (HP) filter. Haan et al. (2008) provide an overview and discussion of different filter techniques. This paper uses a different strategy.
In order to measure fluctuations of the business cycle for a country the following formula will be used:
3. 𝛾𝛾𝑡𝑡𝑖𝑖 − 𝛾𝛾𝑡𝑡−1𝑖𝑖
Where 𝛾𝛾𝑡𝑡𝑖𝑖 is the natural logarithm of GDP per capita at time t for country i. Natural logarithms are used since the difference of the log of a variable yields growth rates. Hence, this paper focusses on the divergence in growth rates of per capita GDP.
3.2 Measuring Relative Specialization
A specialization index is computed annually (for every country) for the relevant sample years. This index comes from a paper by Kalemli-Ozcan et al. (2001)
4. SPEC𝑡𝑡𝑖𝑖 = � �GVA𝑖𝑖 𝑠𝑠 GVA𝑖𝑖 − 1 𝐽𝐽 − 1 � GVA𝑗𝑗𝑠𝑠 GVA𝑗𝑗 𝑗𝑗≠𝑖𝑖 � 2 𝑆𝑆 𝑠𝑠=1 15
where GVA𝑖𝑖𝑠𝑠 is the Gross Value Added (GVA) of sector s in country i¸ GVA𝑖𝑖 is the total GVA of this country, S is the number of sectors, and J is the number of countries. SPEC𝑡𝑡𝑖𝑖 measures the extent to which country i differs from the other countries in terms of sector composition at time t. This is done by summing the squared distance between sector shares of country i and the average sector shares in the countries other than country i (Kalemli-Ozcan et al., 2001). 3.3 Measuring Synchronisation
Now that we established a method to measure fluctuations of business cycles of individual countries and relative specialization, we need to examine whether these business cycles move together across countries and estimate the influence of specialization on this co-movement. Most studies examine some sort of correlation of business cycles between countries over time. In turn this correlation is regressed on the explanatory variable of interest. For example Frankel and Rose (1998) calculate for each pair of countries in their dataset the correlation coefficient of economic activity over a given span of time. This immediately reveals a limitation of this method; a 2-dimensional observation, activity per country at a point in time, is transformed to a one-dimensional variable. Frankel and Rose resolve this problem by dividing their sample into four time spans. This way the time dimension is recovered. However, their dataset covers 34 years. The common currency (and thus the removal of exchange rate uncertainty) is of recent origin such that it pays to focus on the more recent period. As a consequence, however, the correlation analysis is not feasible, due to the more limited time span of the sample (16 years versus 34 years). We therefore opt for an alternative way, namely examining the effect of specialization on growth rates.
The point of interest is the explanatory power of relative specialisation regarding the difference between countries in fluctuations of the business cycle. Instead of linking the business cycle of each country by calculating the pair wise correlations we conclude that the fluctuations of the business cycle of each country move around the mean of this fluctuation of the complete set of countries included in the sample. This leads to the following model:
5. �𝛾𝛾𝑡𝑡𝑖𝑖− 𝛾𝛾𝑡𝑡−1𝑖𝑖 � − (𝛾𝛾̅𝑡𝑡− 𝛾𝛾̅𝑡𝑡−1) = 𝛼𝛼 + 𝛽𝛽1SPEC𝑡𝑡𝑖𝑖+ 𝛽𝛽2�𝑍𝑍1,𝑡𝑡𝑖𝑖 − 𝑍𝑍������ + ⋯ + 𝛽𝛽1,𝑡𝑡 𝑥𝑥�𝑍𝑍𝑥𝑥,𝑡𝑡𝑖𝑖 − 𝑍𝑍������ + 𝜀𝜀𝑥𝑥,𝑡𝑡 𝑡𝑡𝑖𝑖
In this model the left-hand side represents the difference between the growth rate of GDP of country i at time t,�𝛾𝛾𝑡𝑡𝑖𝑖 − 𝛾𝛾𝑡𝑡−1𝑖𝑖 �, and the average growth rate of GDP of the total set of countries at time t, (𝛾𝛾̅𝑡𝑡− 𝛾𝛾̅𝑡𝑡−1). This difference is regressed on the difference between the relative specialization index of country i at time t and the average specialization index of the total set of countries at time t. In order to address omitted variable bias a number of additional explanatory, 𝑍𝑍1,𝑡𝑡𝑖𝑖 … 𝑍𝑍𝑥𝑥,𝑡𝑡𝑖𝑖 , will be included (these variables will be defined below). In the above equation the country average of the additional explanatory variables is subtracted as well. In equation 5, all the averages (𝛾𝛾̅𝑡𝑡− 𝛾𝛾̅𝑡𝑡−1 and 𝑍𝑍����� … 𝑍𝑍1,𝑡𝑡 �����) are equal for all countries at a 𝑥𝑥,𝑡𝑡 given time t and therefore only depend on time t. This property allows including the averages in a time specific intercept. Rewriting the equation above gives:
6. �𝛾𝛾𝑡𝑡𝑖𝑖 − 𝛾𝛾𝑡𝑡−1𝑖𝑖 � = 𝛼𝛼𝑡𝑡+ 𝛽𝛽1SPEC𝑡𝑡𝑖𝑖 + 𝛽𝛽2𝑍𝑍𝑡𝑡𝑖𝑖+ ⋯ + 𝛽𝛽𝑥𝑥𝑍𝑍𝑥𝑥,𝑡𝑡𝑖𝑖 + 𝜀𝜀𝑡𝑡𝑖𝑖
Where 𝛼𝛼𝑡𝑡 is the time specific intercept.
3.4 Specifying Additional Explanatory Variables
There are typical other drivers of growth. In order to identify these we start with the Neoclassical Growth Model. The Solow-Swan Growth Model derives growth on the basis of technology, labour and physical capital. Output growth is a function of growth of the input variables in the production function. Mankiw et al. (1992) augment the Solow Growth Model, while making a distinction between human and physical capital. They regress, using OLS, the growth of output per worker on the log of base output per worker, investment divided by GDP, the growth rate of the population, and a proxy for human capital. Depreciation and the growth rate of technology are included, but are assumed to be equal for all countries and constant over time.
The central assumption of Mankiw et al. (1992) that allows them to use OLS is that difference of the base level of technology between countries is not correlated with savings and population growth rates. Islam (1995) debates this assumption by arguing that the level of technology is not only defined in the narrow sense of production technology. It also includes resource endowments, institutions, etc. and it is therefore not convincing to argue that savings and population growth rates are not correlated with country specific technology shocks. He 17
suggests using a panel data method. This entails including country-specific dummy’s that can control for the difference in technology levels between countries. Islam assumes constant and equal depreciation and the growth rates as well.
Bloom et al. (2004) follow a different approach. They state that technology represents the efficiency with which inputs are used, that is, total factor productivity (TFP). They incorporate TFP by splitting it up in a country specific level, a worldwide technological frontier and a deviation term. Taking first differences and substituting out the deviation term allows them to estimate a model that incorporates a technology term.
Moving from a specialized country towards a more diversified country or vice versa requires capital investments. Growth of the labour force allows employing people in new sectors and therefore could encourage diversification. An increase in human capital or technology allows a country to develop sectors that require more developed labour or technology inputs, and thereby allows diversifying the economy. Given the significance, as found in the papers described above, of these explanatory variables in explaining economic growth these variables need to be controlled for in order to control for omitted variable bias. Inclusion in equation 6 gives:
7. �𝛾𝛾𝑡𝑡𝑖𝑖 − 𝛾𝛾𝑡𝑡−1𝑖𝑖 � = 𝛼𝛼𝑡𝑡+ 𝛿𝛿𝑖𝑖+ 𝛽𝛽1SPEC𝑡𝑡𝑖𝑖 + 𝛽𝛽2∆𝑘𝑘𝑖𝑖𝑡𝑡+ 𝛽𝛽3∆𝑙𝑙𝑡𝑡𝑖𝑖 + 𝛽𝛽3∆ℎ𝑡𝑡𝑖𝑖 + 𝛽𝛽4∆𝑇𝑇𝑇𝑇𝑇𝑇𝑡𝑡𝑖𝑖 + 𝜀𝜀𝑡𝑡𝑖𝑖
Where ∆𝑘𝑘𝑡𝑡𝑖𝑖 is the growth rate of capital, ∆𝑙𝑙𝑡𝑡𝑖𝑖 is the growth rate of labour, ∆ℎ𝑡𝑡𝑖𝑖 is the growth rate of human capital, and ∆𝑇𝑇𝑇𝑇𝑇𝑇𝑡𝑡𝑖𝑖 is the growth rate of total factor productivity. Growth rates are measured by taking the first difference of the log of the variables. A country specific intercept, 𝛿𝛿𝑖𝑖 is included to control for country-specific properties that are not covered by the control variables. It is reasonable to assume that for example geographic location, national government policies or weather conditions influence economic growth. For as far as we are dealing with time-invariant properties, adding country-specific intercepts is sufficient to control for omitted variable bias caused by these properties if not controlled for.
3.5 Limitations of the model
The effect of specialization on economic growth depends on the direction of the sector specific shock. A negative sector specific shock has a negative effect on the relative growth 18
rate of a country that is specialized in that specific sector, whereas a positive sector specific shock has a positive effect on the relative growth rate of a country that is specialized in that specific sector. This entails that if positive and negative shocks to a given sector are equally distributed with a zero mean, their effect on growth running through specialization cancels out. Moreover, the specialization index is constructed using squared deviations. As a consequence of the squaring, deviations are always positive. An important limitation of the constructing the model as is done in this paper is that it is possible that relative specialization will appear to have no effect at all. This is a typical problem that arises when analysing complex systems, but unfortunately inevitable.
4 Related Results from the Literature
As stated in chapter 2, Frankel and Rose (1998) have estimated the effect of bilateral trade intensity on business cycle synchronization. They find within a panel of 21 industrial countries, observed between 1959 and 1993, that bilateral trade intensity has a strong and positive effect on business cycle synchronization. A problem of their approach is that they estimate the net effect of trade instead of distinguishing between inter- and intra-industry trade (Gruben, Koo & Millis, 2002). Gruben et al. estimate the effect of bilateral trade intensity on business cycle synchronization, using the same sample as Frankel and Rose, but distinguish between inter- and intra-industry trade. Their results on inter-industry trade, which can be seen as a measurement of specialization, are mixed. The effect of intra-industry is significant, but about half as strong as found by Frankel and Rose.
Otto et al. (2001) develop an empirical model to estimate the influence of a large set of explanatory variables on the bilateral correlation of GDP growth. Using data for 17 OECD countries observed quarterly between 1960 and 2000 they examine amongst other factors the role of industry structure differences. They find a significant negative relation between differences in industry structures and bilateral growth correlation. Kalemli-Ozcan et al. (2001) regress a utility based asymmetry index for output fluctuations on the relative specialization index that is used in this paper as well. The utility based asymmetry index is constructed using a model of risk sharing among countries inhabited by representative agents. The index is then calculated by evaluating the welfare that each country would obtain if it were constrained to consume its own GDP and evaluating the welfare that each country would obtain if output where pooled across the entire OECD. They use a sample of the US states and 11 OECD countries and find a significant negative relation between the specialization of the production structure and the correlation of output fluctuations. Referring to their earlier work on the positive relation between capital market integration and regional specialization (Kalemli-Ozcan et al., 20031. Discussed in chapter 2) they conclude that the findings support the idea that higher capital market integration leads, through the effect on specialization, to less symmetric fluctuations (Kalemli-Ozcan et al., 2001). Imbs (2004) states that the interactions between trade openness, financial integration, specialization, and business cycle synchronization are more complex; he uses a system of simultaneous equations to estimate the
1 The later date is due to the fact that the paper was first published in 1999 as an working paper. We used the
publication in the American Economic Review (2003).
20
interactions within a sample of 24 countries observed quarterly in the 1980s and 1990s. His main findings are that: 1. Specialization patterns have a sizable effect on business cycles. Most of this effect is independent of trade or financial policy, but directly reflects differences in GDP per capita; 2. Economic regions with strong financial links are significantly more synchronized, even though they also tend to be more specialized; 3. The overall effect of trade on business cycles synchronization is strong, but a sizable portion is found to actually work through intra-industry trade (Imbs, 2004).
Amiti (1999) examines the development of specialization in 10 EU countries in light of the ongoing process of dismantling trade barriers between members of the EU. He finds that there was a significant increase in specialization between 1968 and 1990 for Belgium, Denmark, Germany, Greece, Italy, and The Netherlands; no significant change for Portugal; and a significant fall for France, Spain and the UK. Although not all countries experienced an significant increase from 1968 till 1990, all countries did between 1980 and 1990. A plausible explanation, according to Amati, for the fall in specialization during the overall period is that before joining the EU these countries had high trade barriers for industries for which they did not have a comparative advantage. If the trade barriers a removed as a result of joining the EU competitive pressure arises to expand industry in which each country has a competitive advantage. This could explain why late joiners like Spain, Portugal, and the UK saw an overall decline, but an upward trend starting in the late 1970s and early 1980s (Amiti, 1999).
5 Sample and Data
This paper constructs a panel of 24 countries observed every year for the period 1995-2012. The 24 countries are all the countries that where member of the EU in 2012 with the exclusion of Cyprus, Malta and Luxembourg. These countries are outliers in most of the variables used and/or are too small. The period is chosen because of great volatility driven by political and economic events prior to 1995 in the Central and Eastern Europe Countries, and limitations in data availability past 2011.
Data for output growth, capital growth, labour force growth, human capital growth and total factor productivity (TFP) growth are obtained from the Penn World Tables. GDP is measured as output-side real GDP at chained PPPs (in mil. 2005US$). This way of measuring GDP is chosen because it is best suited to compare relative productivity across countries over time (see Feenstra et al. (2013) for a technical explanation). Capital and TFP are at current PPSs (in mil. 2005US$). GDP per Capita and Capital per Capita is obtained by dividing GDP and Capital by the population in millions, which is also obtained from Penn World Tables. The index of human capital per person is based on years of schooling and returns to education. Labour is measured as the number of people in the labour market in millions.
Data for the specialization index are obtained from Eurostat. Eurostat provides national accounts aggregates that are split out by the gross value added of ten industry branches. These branches are:
1. Agriculture, forestry and fishing; 2. Industry (except construction); 3. Construction;
4. Wholesale and retail trade, transport, accommodation and food service activities; 5. Information and communication;
6. Financial and insurance activities; 7. Real estate activities;
8. Professional, scientific and technical activities; administrative and support service activities;
9. Public administration, defence, education, human health and social work activities;
10. Arts, entertainment and recreation; other service activities; activities of household and extra-territorial organizations and bodies.
This paper will construct the specialization index defined in chapter 3, using these ten industry branches.
6 Empirical Results
The way the baseline-model is formulated in equation 7 it estimates the effect of the degree of relative specialization on the symmetry of economic growth within the sample. It does not, however, reveal whether countries within the sample indeed have become more specialized. In order to address the central hypothesis we therefore need to answer the two questions that were stated in the introduction, namely:
1. Is there growing specialization within the EU and Euro-area?
2. What is the effect of this specialization on the convergence in economic growth between members of the EU and Euro-area?
The first three sections of this chapter address the first question. The subject of the question, namely the development of a single variable over time, makes it suitable for a graphical analysis. The other sections focus on the second question, using the more formal methods that are described in chapter 3.
6.1 Analysis of the Development of the Specialization Index over Time
Table 6.1.1 (next page) displays the values for the specialization index, as formulated in equation 4 in section 3.2, for the 24 countries included in the sample multiplied by 1000. Displayed are the observations for 1995, 2004, 2007 and 2011 and the difference between 2011 and the other three years2. These years are selected because 1995 and 2011 are the first and the last year included in the dataset, and some of the countries in the dataset joined the EU in 2004 or 20073. The bold observations indicate the most relevant period, namely the period in which the countries considered where a member of the EU. Parentheses indicate a negative value.
If a country is completely specialized in a sector, and all the other countries are completely specialized in one other sector, the specialization index will be two (and the value in the table 6.1.1 would be 2000) for that specific country. Or the specialization index of all countries would be zero if all countries have the same sector shares for all sectors. Focussing on 1995, table 6.1.1 shows that Romania, Bulgaria and Greece where very specialized compared to the other countries in that year. This is mainly due to a relative high sector share for Agriculture,
2 See table A.I in the appendix for all the observations of the specialization index. 3
See table A.II in the appendix for the history of EU and Euro-area enlargement.
24
forestry and fishing in total value added (19% for Romania, 15% for Bulgaria, and 10% for
Greece) compared to the average share for that sector in the total sample in 1995 (6%).
Table 6.1.1 Development of the Specialization Index over Time
Country 1995 2004 2007 2011 Δ('11-'95) Δ('11-'04) Δ('11-'07) EU member in 1995 Ireland 4.04 8.37 7.15 12.88 8.84 4.52 5.73 France 9.89 11.37 14.37 15.07 5.18 3.70 0.70 United Kingdom 4.43 8.13 9.20 8.95 4.53 0.83 (0.24) Belgium 4.69 5.27 5.85 8.44 3.76 3.18 2.60 Portugal 2.39 4.40 4.99 4.26 1.86 (0.15) (0.74) Greece 16.35 21.18 15.13 17.92 1.58 (3.26) 2.79 Netherlands 4.60 6.15 4.64 6.09 1.49 (0.06) 1.45 Germany 7.38 8.05 9.61 8.79 1.41 0.74 (0.82) Spain 5.02 7.13 7.29 5.64 0.62 (1.49) (1.65) Italy 2.35 2.31 1.98 2.95 0.60 0.64 0.98 Finland 3.95 4.51 5.70 4.24 0.29 (0.27) (1.46) Denmark 7.11 5.43 4.60 7.18 0.06 1.75 2.57 Sweden 5.87 6.50 6.25 4.75 (1.11) (1.75) (1.50) Austria 2.69 0.85 0.72 0.80 (1.89) (0.04) 0.08 EU member in 2004 Lithuania 6.92 8.94 9.25 18.00 11.07 9.06 8.75 Poland 6.39 4.05 4.75 8.19 1.80 4.15 3.45 Czech Republic 5.95 10.11 12.65 10.65 4.70 0.54 (2.00) Hungary 2.74 4.42 4.28 4.00 1.25 (0.42) (0.28) Slovakia 6.68 8.77 11.05 7.64 0.96 (1.12) (3.41) Latvia 2.43 13.73 13.35 12.56 10.12 (1.17) (0.79) Estonia 3.40 3.53 3.87 2.19 (1.21) (1.35) (1.69) Slovenia 2.34 4.85 3.20 1.27 (1.08) (3.59) (1.93) EU member in 2007 Romania 42.61 21.80 11.73 29.46 (13.14) 7.66 17.73 Bulgaria 26.06 7.56 4.97 6.65 (19.40) (0.90) 1.69
Chart 6.1.1 (next page) shows the development of the average and the standard deviation (SD) of the specialization index over time for the whole sample. A number of things stand out from this chart. First of all, between the start and the end of the sample period considered, specialization has on average increased from 7.8 in 1995 to 8.8 in 2011. During this period the SD has decreased, indicating that the countries are more closely distributed around a higher average relative specialization in 2011 than in 1995. The increase in the relative specialization index was, however, not steady. The sharp increase in the average between 1995 and 1997 is largely due to Latvia, who saw a large increase in the share of Wholesale and retail trade,
transport, accommodation and food service activities, Czech Republic (a large increase in Construction and Industry), Lithuania (large increase in for Agriculture, forestry and fishing)
and Bulgaria (large increase in Wholesale and retail trade, transport, accommodation and 25
food service activities and Professional, scientific and technical activities; administrative and support service activities).
6.2 The development of the Specialization Index and the EU
Before continuing, it is important to point out that the specialization index measures the relative specialization, where relative refers to all the other countries taking in to account when calculating the index. This implies that one has to be very careful when drawing conclusions regarding the influence of being a member of the EU or being a member of the Euro-Area, especially since countries joined either the EU or Euro-area at different points in time, or in case of the Euro-Area not at all. Take the fourteen countries that were EU member for the full period between 1995 and 2011 as an example for now. A number of possibilities arise when one wants to analyse whether these members of the EU have experienced increased specialization.
The difference between 1995 and 2011 in the specialization index (column 5 in table 6.1.1) shows that 12 out of the 14 countries indeed experienced an increase in specialization. Chart 6.2.1 visualises this by displaying again the average and SD of the specialization index over time, but now only for those 14 members. The specialization index itself is, however, constructed using the full set of 24 countries; so chart 6.2.1 tells us that these 14 EU members have become more specialized relative to all 24 countries, of which 10 where no EU member during the full period considered. This is relevant information; the period prior 2004 tells us that during their membership these 14 EU members have become more specialized taking a large subset of countries into account. The period starting in 2007 (grey box in chart area)
7,76 8,55 9,29 7,55 6,78 6,34 7,13 7,09 7,20 7,81 7,29 7,56 7,36 7,05 7,28 7,84 8,69 8,88 9,68 12,75 7,25 4,82 3,64 5,03 4,95 4,53 5,01 3,96 4,02 3,92 3,95 4,08 5,38 6,37 0,00 2,00 4,00 6,00 8,00 10,00 12,00 14,00 16,00 18,00 20,00 0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00 9,00 10,00 1995 1997 1999 2001 2003 2005 2007 2009 2011 S tandar d D ev iat ion A v er age
Chart 6.1.1 SPEC index for all countries
Average SPEC index Standard Deviation SPEC index
shows that this increased specialisation continued for these 14 countries when the full set of countries where member of the EU.
Since the theory predicts an increase in specialization between countries that remove trade barriers between them, an interesting question is whether relative specialization took place within this subset of countries that had a removal of trade barriers for the full period considered. Chart 6.2.2 displays the development of the average and the SD of the specialization index over time for the 14 countries considered in isolation. That is, the specialization index is calculated using only those 14 countries.
Chart 6.2.2 shows that increased specialization has taken place during the period 1995-2004 between the countries that where member of the EU during that period. After 2004 the relative specialization between these countries stabilized.
5,77 5,70 6,11 5,78 5,58 5,70 6,43 6,77 6,93 7,12 6,56 6,87 6,96 6,63 6,95 7,10 7,71 3,56 3,90 4,58 3,90 3,39 3,43 3,92 4,36 4,64 4,66 3,42 3,54 3,93 3,92 3,50 3,79 4,62 0,00 2,00 4,00 6,00 8,00 10,00 12,00 14,00 16,00 18,00 20,00 0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00 9,00 10,00 1995 1997 1999 2001 2003 2005 2007 2009 2011 S tandar d D ev iat ion A v er age
Chart 6.2.1 SPEC index for all countries that where EU member in 1995
Average SPEC index Standard Deviation SPEC index
4,59 4,45 4,61 4,83 4,94 5,10 5,67 6,13 6,13 5,99 5,36 5,63 5,82 5,50 5,76 5,76 6,04 3,61 3,81 4,34 3,96 3,89 4,06 4,86 5,49 5,43 5,04 3,44 3,46 3,72 3,64 3,61 3,65 4,50 0,00 2,00 4,00 6,00 8,00 10,00 12,00 14,00 16,00 18,00 20,00 0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00 9,00 10,00 1995 1997 1999 2001 2003 2005 2007 2009 2011 S tandar d D ev iat ion A v er age
Chart 6.2.2 SPEC index for all countries that where EU member in 1995 SPEC index calculated using only those countries
Average SPEC index Standard Deviation SPEC index
Charts 6.2.3 and 6.2.4 repeat the same exercise, but now for the eight countries that became EU member in 2004. Chart 6.2.3 displays the mean and SD over time for the specialization index calculated using all 24 countries. Chart 6.2.4 does the same, but now the specialization index is calculated using only the subset of eight countries that became EU member in 2004. The transition between the dotted and the solid lines indicates the point of entry in to the EU.
The graphs show that the eight countries have experienced increased specialization between 1995 and 2011, both relative to the complete sample as relative to each other. After their accession to the EU in 2004 these increases continued, peaking in 2006. Table 6.1.1 shows that the majority of the eight countries experienced a decline in specialization relative to the complete set of countries. However, due to a large increase in specialization of Lithuania and Poland the group on average experienced a moderate increase, as can be seen from chart
4,61 6,03 5,00 5,23 5,62 5,59 5,73 5,47 6,12 7,30 7,95 8,21 7,80 7,06 6,77 7,19 8,06 1,92 2,25 2,31 1,46 1,52 1,69 2,48 2,81 2,77 3,41 4,41 4,57 3,96 3,94 4,05 4,61 5,28 0,00 2,00 4,00 6,00 8,00 10,00 12,00 14,00 16,00 18,00 20,00 0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00 9,00 10,00 1995 1997 1999 2001 2003 2005 2007 2009 2011 S tandar d D ev iat ion A v er age
Chart 6.2.3 SPEC index for all countries that became EU member in 2004
Average SPEC index Standard Deviation SPEC index
5,76 6,31 6,79 6,38 5,62 5,30 5,11 5,39 3,42 4,94 3,97 4,52 5,14 5,05 4,98 4,80 5,08 5,76 4,39 5,34 5,51 4,60 3,67 3,14 2,90 3,09 1,33 1,75 2,00 2,08 1,97 2,11 2,48 2,48 3,61 4,39 0 2 4 6 8 10 12 14 16 18 20 0 1 2 3 4 5 6 7 8 9 10 1995 1997 1999 2001 2003 2005 2007 2009 2011 S tandar d D ev iat ion A v er age
Chart 6.2.4 SPEC index for all countries that became EU member in 2004 SPEC index calculated using only those countries
Average SPEC index Standard Deviation SPEC index
6.2.3. Chart 6.2.4 shows that the eight countries experienced a moderate decline in specialization relative to each other during their membership of the EU.
Chart 6.2.5 shows the development of the mean of the specialization index over time for the two countries that joined the EU in 2007, Romania and Bulgaria. Since this are only two countries the specialization index is not calculated separately for this countries. For the same reason the SD is not calculated.
As stated earlier both countries started with a high relative specialization. This increased until 1997, but decreased to more moderate levels until 2007. In 2007 the average specialization index of these countries was 8.35 vs. 7.36 on average for the total sample (chart 6.1.1). After joining the EU in 2007 both countries became relatively more specialized. This effect was much stronger for Romania, who experienced an increase of 17.73 between 2007 and 2011 (table 6.1.1), making it the most specialized country in 2011 which it was in 1995 as well. 6.3 The Development of the Specialization Index and the Euro-area
Charts 6.3.1 and 6.3.2 (next page) display the development over time of the average and SD of the specialization index for the 10 countries that adopted the Euro in 1999. In chart 6.3.1 the specialization index is calculated using all 24 countries. For chart 6.3.2 only the 10 countries are used when calculating the specialization index. The transition between the dotted and the solid lines indicates the point of adoption of the Euro for these countries.
34,33 38,57 48,68 29,27 19,86 13,87 17,69 15,75 13,43 14,68 9,85 9,84 8,35 9,92 11,62 15,59 18,06 0,00 10,00 20,00 30,00 40,00 50,00 60,00 1995 1997 1999 2001 2003 2005 2007 2009 2011 A v er age
Chart 6.2.5 SPEC index for all countries that became EU member in 2007
Average SPEC index
The graphs show that the 10 countries have experienced increased specialization between 1995 and 2011, both relative to the complete sample as relative to each other. This increase in specialization was especially strong after their Euro-area membership started in 1999.
6.4 Regression Results
In order to adress the question “What is the effect of this specialization on the convergence in
economic growth between members of the EU and Euro-area?” this part of the chapter
estimates the effect of the degree of relative specialization on economic growth. We first estimate the baseline model as specified in chapter 3:
�𝛾𝛾𝑡𝑡𝑖𝑖− 𝛾𝛾𝑡𝑡−1𝑖𝑖 � = 𝛼𝛼𝑡𝑡+ 𝛿𝛿𝑖𝑖 + 𝛽𝛽1SPEC𝑡𝑡𝑖𝑖 + 𝛽𝛽2∆𝑘𝑘𝑡𝑡𝑖𝑖 + 𝛽𝛽3∆𝑙𝑙𝑡𝑡𝑖𝑖 + 𝛽𝛽3∆ℎ𝑡𝑡𝑖𝑖 + 𝜀𝜀𝑡𝑡𝑖𝑖
We first need to determine whether to use a fixed effect model or random effects model. The Hausman test rejects the random effects model in favour of the fixed effects model
4,70 4,51 4,72 4,57 4,74 4,94 5,67 6,02 5,73 5,84 5,92 6,37 6,23 5,92 6,27 6,54 6,92 2,24 2,78 3,00 2,52 2,66 2,50 3,30 3,89 2,94 2,92 3,09 3,51 3,64 3,54 3,50 3,75 4,21 0,00 2,00 4,00 6,00 8,00 10,00 12,00 14,00 16,00 18,00 20,00 0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00 9,00 10,00 1995 1997 1999 2001 2003 2005 2007 2009 2011 S tandar d D ev iat ion A v er age
Chart 6.3.1 SPEC index for all countries that became Euro-area Member in 1999
Average SPEC index Standard Deviation SPEC index
3,94 3,64 3,68 4,10 4,42 4,65 5,32 5,84 5,34 5,24 5,34 5,80 5,79 5,48 6,06 6,10 6,36 1,93 1,73 1,80 1,99 2,35 2,35 3,38 4,23 2,93 2,59 2,74 3,07 3,10 3,03 3,65 3,51 4,10 0,00 2,00 4,00 6,00 8,00 10,00 12,00 14,00 16,00 18,00 20,00 0,00 1,00 2,00 3,00 4,00 5,00 6,00 7,00 8,00 9,00 10,00 1995 1997 1999 2001 2003 2005 2007 2009 2011 S tandar d D ev iat ion A v er age
Chart 6.3.2 SPEC index for all countries that became Euro-area Member in 1999
SPEC index calculated using only those countries
Average SPEC index Standard Deviation SPEC index
(Prob.>chi2 = 0.000). The results of the regression of the base line model can be found in table 6.4.1.
Table 6.4.1 Baseline Model
Model uses state and time fixed effects regression. Dependent variable is growth rate of GDP per capita.
Right-hand side variables 1
Specialization index -1.525**
(0.59)
Capital Growth Rate 0.156***
(0.05)
Labour Force Growth Rate 0.545***
(0.10)
Human Capital Growth Rate -0.136
(0.33)
R2 0.48
N 384
* p<0.1; ** p<0.05; *** p<0.01 Terms between parentheses are standard errors
Both state and time fixed effects where included but the corresponding coefficients are not showed in the results. In order to prevent perfect multicollinearity n-1 time dummies where included. Using an F-statistic, we can reject the joint hypothesis that all time specific effects are equal to zero. In order to control for heteroskedasticity robust standard errors have been used.
The coefficient on the specialization index indicates that higher specialization leads to downward divergence in economic growth; a high score on the specialization index is associated with lower economic growth. This result is significant at the 5% level. The coefficients on the control variables have the expected sign and are significant (at the 1% level), except for the coefficient on human capital. This sign is negative where we would expect a positive sign based on economic theory. This result is however not significant (the p value is 0.679)4.
As already mentioned while analysing the development of the specialization index over time,, members of the EU and the Euro-area joined at different points in time within the period
4 We intended to control for TFP using TFP as measured by Pen World Table. However, the way TFP is
constructed in Pen World Table removes al the significance from the results (see appendix III, the correlation between TFP growth and GDP per capita growth is very high (.7239) Instead we follow Islam (1995) and assume that TFP growth is not country specific and can therefore be controlled for by time-specific intercepts.
31
considered in the sample. In order to incorporate this fact in the regression model we add two interaction terms to the baseline model.
�𝛾𝛾𝑡𝑡𝑖𝑖− 𝛾𝛾𝑡𝑡−1𝑖𝑖 � = 𝛼𝛼𝑡𝑡+ 𝛿𝛿𝑖𝑖 + 𝛽𝛽1SPEC𝑡𝑡𝑖𝑖 + 𝛽𝛽2∆𝑘𝑘𝑡𝑡𝑖𝑖 + 𝛽𝛽3∆𝑙𝑙𝑡𝑡𝑖𝑖 + 𝛽𝛽3∆ℎ𝑡𝑡𝑖𝑖 + 𝜇𝜇1∗ 𝐷𝐷𝑢𝑢𝐷𝐷 𝐸𝐸𝐸𝐸𝑡𝑡𝑖𝑖 ∗ SPEC𝑡𝑡𝑖𝑖 + 𝜇𝜇2
∗ 𝐷𝐷𝑢𝑢𝐷𝐷 𝐸𝐸𝑢𝑢𝐸𝐸𝐸𝐸𝑡𝑡𝑖𝑖 ∗ SPEC𝑡𝑡𝑖𝑖 + 𝜀𝜀𝑡𝑡𝑖𝑖
Where 𝐷𝐷𝑢𝑢𝐷𝐷 𝐸𝐸𝐸𝐸𝑡𝑡𝑖𝑖 equals one if country i is EU member in year t and 𝐷𝐷𝑢𝑢𝐷𝐷 𝐸𝐸𝑢𝑢𝐸𝐸𝐸𝐸𝑡𝑡𝑖𝑖 equals one if country i is Euro-area member in year t
Table 6.4.2 Baseline Model with interaction terms
Model uses state and time fixed effects regression. Dependent variable is growth rate of GDP per capita.
Right-hand side variables 1
Specialization index -1.756***
(0.40)
Capital Growth Rate 0.150***
(0.05)
Labour Force Growth Rate 0.525***
(0.09)
Human Capital Growth Rate -0.125
(0.38)
SPEC * Dum EU 0.93
(0.89)
SPEC * Dum Euro 0.15 (0.74)
R2 0.48
N 384
* p<0.1; ** p<0.05; *** p<0.01 Terms between parentheses are standard errors.
The interpretation of the interaction terms is the additional effect of specialization on economic growth for members of either the EU or Euro-area. The combination of both the coefficient on specialization and the coefficient on the interaction terms yields the total effect of specialization on growth for members of the EU or Euro-area. Please note that a number of countries are member of the EU but not of the Euro-area in which case only the EU-dummy interaction is relevant. Other countries are member of both the EU and the Euro-area, in which case both interaction terms are relevant. Table 6.3.3 (next page) presents the results of the sum of the combinations. As can be seen in table 6.4.2 the interactions terms themselves are not significant. In order to test the null hypotheses whether the sums are equal to zero Wald tests have been executed. The results of these tests can be found in table 6.4.3 as well.
