Assessing Turkey’s Opportunity Costs of Remaining in
a Customs Union rather than becoming a full-fledged
EU Member.
University of Groningen
Master Thesis Presented to the Faculty of Economics and Business
in Partial Fulfillment of the Requirements for the Degree
Master of Science (M.Sc.)
Supervisor: Dr. Abdul A. Erumban Co-assessor: Prof. Dr. Marcel P. Timmer
June 9, 2016 Andr´e Assenmacher a.assenmacher@student.rug.nl
Abstract
Welfare gains and economic implications from integration in general, and from EU acces-sion in particular, have received a lot of attention in the literature. Economic integration is arguably considered to be mutually beneficial for both parties. However, we argue that when integration is premature, the gains from pursuing integration are not fully achieved and the opportunity costs of not integrating fully are rather high. In this paper we evaluate the case of Turkey, because Turkey joined a customs union with the EU in 1996, but, as of yet, has not managed to become a full-fledged EU member. We compare Turkey’s GDP per capita increases from joining the customs union to the hypothetical gains that Turkey could have realized if it had become a member of the EU. To create a synthetic Turkey, which has not joined a customs union with the EU, the synthetic control method is employed. In a subsequent step, the differences in GDP levels and from it the resulting growth differentials are determined. After that, the GDP per capita gains from EU membership as estimated by Campos et al. (2014) are applied to the case of Turkey controlling for country and economic integration characteristics. The findings suggest that both joining a customs union and joining the EU (would have) brought eventually substantial and significant GDP per capita increases and additional growth for Turkey. However, we argue that major increases for Turkey failed to appear, because it did not continue to pursue integration, i.e. become a member of the EU. We attribute this find-ing mainly to the lack of institutional harmonization that Turkey would have received through the EU assessment process.
Table of Contents
List of Abbreviations IV
List of Tables and Figures V
1 Introduction 1
2 Literature Review 4
2.1 Integration in the European Union . . . 5
2.2 Turkey between Customs Union and EU Membership . . . 8
2.3 A Brief History of Turkey-EU Relations . . . 9
2.4 Hypotheses . . . 12
3 Turkey’s Gains from a Customs Union with the EU 13 3.1 Methodology: Time Series Regressions . . . 13
3.2 Methodology: Synthetic Control Groups . . . 14
3.3 Data for Synthetic Control Groups . . . 17
3.3.1 Dataset for Synth A and Synth B . . . 17
3.3.2 Dataset for Synth A-C* . . . 18
3.4 Results . . . 19
3.4.1 Regression Results . . . 19
3.4.2 Synthetic Control Group Results . . . 20
3.5 Robustness and Limitations . . . 24
4 Gains from EU Membership 28 4.1 Methodology: Turkey as an EU Member . . . 28
4.2 Empirical Results . . . 30
4.3 Econometric Results . . . 31
4.4 Robustness and Limitations . . . 32
5 Discussion 34 5.1 Conclusion . . . 37
References VI
List of Abbreviations
ASEAN Association of Southeast Asian Nations. CU Customs Union.
EU European Union, referring to the current 28 member states, if not indicated differ-ently.
EU10 European Union, Member States that joined the EU on 1 May 2004 (CZ, EE, CY, LT, LV, HU, MT, PL, SI and SK).
EU15 European Union, 15 Member States before 1 May 2004 (BE, DK, DE, EL, ES, FR, IE, IT, LU, NL, AT, PT, FI, SE and UK).
FDI Foreign Direct Investment. GEM General Equilibrium Model. G20 Group of Twenty major economies.
NAFTA North American Free Trade Agreement. NATO North Atlantic Treaty Organization.
OECD Organisation for Economic Co-operation and Development. PPP Purchasing Power Parities.
PWT Penn World Tables. TFP Total Factor Productivity. TPP Trans Pacific Partnership.
List of Tables
1 Channels through which integration affects GDP growth. . . 4
2 OLS regressions for Turkey with dependent variable GDP growth. . . 20
3 Actual and synth a/synth b with average variable values for Turkey. . . 21
4 Actual and synth a-c* with average variable values for Turkey. . . 22
5 Estimated GDP per capita gains from EU accession. . . 29
6 Descriptive statistics for regression variables. . . 30
7 Pairwise correlation coefficients for explanatory variables. . . 31
8 OLS regressions with dependent variable gap (see Table 5 for a detailed description of the variable). . . 33
9 Summary of the results: Turkey’s GDP gains in different scenarios. . . 34
10 EU Enlargement and GDP growth for pre- and post-accession. . . XI 11 Variables for synth a and synth b. . . XII 12 Variables for synth a*, synth b* and synth c*. . . XII 13 Variables for counterfactual: Turkey as EU member. . . XII 14 Weights of the synthetic control groups. . . XIII
List of Figures
1 Development of Turkish exports of goods and services. . . 102 Development of Turkish unemployment and GDP growth. . . 11
3 Comparing growth rates between synth c* and actual Turkey. . . 23
4 Synthetic control groups for Turkey. . . 24
5 Robustness check with broader set of variables: synth c*. . . 25
1
Introduction
A common rationale is that economic integration and cooperation have proven to be suc-cessful, when it comes to promoting economic growth and sustaining welfare. Examples for this include the Association of Southeast Asian Nations (ASEAN), the North Amer-ican Free Trade Agreement (NAFTA) and the EU. Moreover, their effects on income and welfare have all been discussed extensively in the literature (e.g. see Kurus (1993); Hufbauer (2005); Baldwin et al. (2006)). More recently, the public awareness has partly become critical of economic integration and resulted, for example, in rallies against trade agreements such as the Transatlantic Trade and Investment Partnership (TTIP)1 and the Trans Pacific Partnership (TPP)2. In addition, also the EU – a unique institution for po-litical and economic cooperation – is facing headwind in nowadays; in part, due to TTIP and autonomy movements (European Union, 2015).
In general, economic integration aims at ruling out discrimination of other economic agents within an area. Balassa (2013, p.7) explains that integration refers in this context to the abolishment of ”trade-and-payments restrictions and increased state intervention and is designed to mitigate cyclical fluctuations and to increase the growth of national income.” This shows that economic integration is mainly about increasing GDP per capita and countering volatile GDP developments in order to enhance welfare.
Economic integration can have several different forms depending on its degree, such as a free trade area, a customs union, a common market, and total integration ultimately. The latter refers to the legally binding merger of monetary, fiscal, social, and counter-cyclical policies (Balassa, 2013). In the case of the European Union (EU) this process has neither been completed, nor is total integration the overall goal (European Union, 2015). First of all, it is important to define the above-mentioned concepts and stress their differences. A free trade area is a form of economic cooperation, in which tariff and non-tariff barriers across countries are abolished or gradually lowered. In principle, a customs union is a free trade area but additionally countries within a customs union have agreed upon common external tariffs applying to non-members. Therefore, it can be considered as a deeper level of integration, because successful negotiations require an intensification of cooperation beforehand (Reinert et al., 2009). The case of the European Union is more complicated, because the EU – as stated above – is not fully integrated but rather combines partly characteristics of an economic union such as a common market and aligned policies, macro-economic and regulatory cooperation and ultimately a customs union (OECD, 1999). In order to join the European union a potential candidate needs to fulfill the so-called Copenhagen criteria, which include among other things, having democratic institutions enforcing the rule of law and human rights, but also a functioning market economy. The application process can be sketched as follows. In the beginning,
the current EU members need to negotiate a consensus. Subsequently, the EU agrees on a time frame with the potential member state, in which it has to comply with and implement all rules and regulations of the EU, the so-called acquis communautaire. If and only if this process has been completed successfully, the candidate will be granted accessed to the EU (European Commission, 2015).
The EU – and its predecessors like the European Economic Community (EEC) – is an organization, in which economic integration has been taking place from 1958 onwards by enlarging the number of member states on one hand and by increasing the level of integration within the Union on the other hand. One of the first goals of the EU was to foster economic cooperation. That is why a transition was set in motion towards a common single market, while this transition is still ongoing nowadays (European Union, 2015). Therefore, the EU is offering a natural experiment for analyzing the actual outcomes of economic integration.
It has been the EU’s rationale to proceed with enlargement in order to reunify Europe after World War II and it is seen by the EU Commission as a mutually beneficial process for the joining country as well as for the members of the EU (European Commission, 2010). Since further enlargement is inherently limited, on one hand geographically but on the other hand also politically and culturally, the EU is currently seeking for a novel rationale. This can be attributed to the fact that in terms of enlargement only a limited number of countries are available within the region qualifying for EU accession.
While enlargement and integration have been politically without questioning, ex-post evaluations of actual economic benefits for member states are rare (Campos et al., 2014). Therefore, it is of interest to quantify the GDP per capita gains from EU membership ex-post and weigh them against other political forms of economic cooperation, namely forming a customs union. Indeed, all countries that became an EU member had joined a free trade agreement before such as the EU10 countries (Lejour et al., 2001).
Consequently, one might reasonably ask: what are the advantages of deeper levels of integration such as the EU over lower levels such as a customs union? In order to answer this question, the case of Turkey is assessed in particular, because Turkey aimed for EU membership but remained in the status of solely forming a customs union.
In this regard, Esfahani (2003) argues that solely joining a customs union has been costly for Turkey, because it exposed Turkey to more competition, while not helping Turkish institutions to grow stronger as it would have been the case if integration would have been pursued further.
Taking this as a starting point, the objective of this paper is to assess how large the opportunity costs have been for Turkey to remain in the preliminary status of forming solely a customs union relatively to a counterfactual EU membership. These costs are formulated as the difference of GDP per capita levels that can be attributed to actually forming a customs union and the theoretical GDP per capita levels that would have been realized as full EU member.
Using synthetic control groups, we are able to create a synthetic Turkey not being affected by a customs union to calculate these opportunity costs. The GDP per capita levels of synthetic Turkey are compared to Turkey’s actual levels. After having derived the gains from a customs union, we compare these gains to the counterfactual gains for Turkey from EU membership. That is, applying the gains from other countries having joined the EU to Turkey controlling for country characteristics and the channels affecting growth in the context of integration (Campos et al., 2014). We find that Turkey forming a customs union with the EU has resulted in additional GDP growth ranging from 0.53 to 1.42 percentage points on average in the ten-year period after 1995. Moreover, our results suggest that on top of these gains, EU membership would have resulted in another 0.53 percentage points of additional growth in the ten-year period after the accession. The sim-ilarity of the numbers is a coincidence. Moreover, our results do not take dynamic effects such as institutional harmonization into account that might affect the results positively or negatively. However, these are beyond the scope of this paper.
2
Literature Review
Economic effects from integration can be quite different and depend on how integration is pursued in particular. They can be either negative or positive, whereas Balassa (2013) denotes that the overall goal of integration should always be to increase welfare.
If integration is understood as the elimination of mobility frictions for capital, labor, goods, and services, welfare gains might be realized. Contrarily, if the process of in-tegration is understood as national economic policies or government planning, it could also hamper welfare improvements and induce beggar-thy-neighbor effects, in which one country benefits at the expense of another country (Balassa, 2013).
Further, the process of integration determines the channels through which economic welfare could be realized. One of the main channels explaining welfare increases is the reallocation of resources, i.e. trade (Balassa, 2013). For example, Behrens et al. (2007) evaluate the effect of the trade channel on economic growth by modeling two sectors and various countries. Their simulations show that the benefits outweigh the costs of economic integration, but only if transport costs are low. This is an important finding, because it underlines that different countries in different phases of development do not necessarily benefit from integration. For example, a country with higher transport costs due to technological differences or limited market access would rather face large costs from integration than benefits under this simulation. This is also in line with Eaton and Kortum (2002), who stress the differences in gains from trade due to technology and geography.
Table 1 – Channels through which integration affects GDP growth.
authors channels details direction
Baldwin et al. (1995) • Trade effects – Abolishon of trade barriers, costs + ∗ • Single market effects – Increases in efficiency, competition + ∗ • Factor movements – Capital flows + ∗ – Migration +/−
Acemoglu et al. (2005) • Institutions – Promoting long-term growth + and Lejour et al. (2004)
Breuss (2001) • Costs – Transfers and harmonization −
Notes: Effects denoted with ∗ are also present in a customs union, whereas the other effects are at work in deeper levels of integration such as the EU or economic unions.
of labor changes not only the supply of labor, but also increases the market potential by creating more demand.
Baldwin et al. (1995) summarize effects from integration contributing positively to growth in three categories: trade effects referring to increased trade due to the decrease of tariffs and other trade costs, single market effects concerning changes in productivity and competition, as well as factor movements referring to flows of capital and labor migration.
Moreover, Acemoglu et al. (2005) describe the importance of growth promoting insti-tutions for economic development. Therefore, it can be argued that especially developing countries not only benefit from integration through the established integration channels, but also gain in the long-run from institutional harmonization (Lejour et al., 2004).
In addition, Breuss (2001) considers a fourth channel being negatively associated with GDP, namely the cost of enlargement referring to transfers from rich to poor countries. On the contrary, the costs of joining the EU for membership candidates have been rather small. Despite the fact that the EU has been designed for relatively rich democracies, poorer countries have benefitted from EU accession (Baldwin et al., 1997). Table 1 sum-marizes the potential channels of integration and their effects on welfare.
2.1
Integration in the European Union
Incidentally, it has already been acknowledged that the EU is a unique case of economic integration, in which single countries may have benefited from the channels as pointed out in the previous section. The EU can be considered as one of the largest supra-national bodies uniting political and economic cooperation of different sovereigns under a single roof (European Union, 2015). Therefore, the EU is an example of an economic and political union without being fully integrated, because the majority of competences remain with the national sovereigns. Nevertheless, the EU is working towards a single common market in order to enhance welfare. This common market is based on the principle of four freedoms or more specifically: the free movement of goods, capital, services, and people (European Commission, 2016b).
There is an extensive set of literature dealing with ex-ante estimations of enhancing welfare or in other words on the economic gains from EU membership (Badinger, 2005; Baldwin et al., 1997; Breuss, 2001; Kohler, 2014; Lejour et al., 2001; Barry, 2004; Grassini et al., 2001; Heijdra et al., 2002; Kohler, 2004; Keuschnigg and Kohler, 2002; Maliszewska, 2004; Read and Bradley, 2001). The European Commission (2006) emphasizes that all major studies find ex-ante significant GDP increases from enlargement for new EU mem-bers and mostly also for old EU memmem-bers.
enlargement to 56 percent after and is thus showing the so-called export success story (European Commission, 2009, Ch. 2).
In order to evaluate potential gains from EU membership, most approaches employ a General Equilibrium Model (GEM) (Baldwin et al., 1997; Breuss, 2001; Kohler, 2014; Lejour et al., 2001; Grassini et al., 2001; Heijdra et al., 2002; Kohler, 2004; Keuschnigg and Kohler, 2002; Maliszewska, 2004). Usually, this kind of approach models supply and demand under strong assumptions, such as utility maximizing agents and with the help of commonly known, but generalized, production functions. A GEM offers an approximation of the expected GDP (growth) performance. However, the studies deviate in their outcome substantially due to their different assumptions beforehand. Additionally, it needs to be stated that all these simulations might be mitigated in the long-run. Nevertheless, consistent with the EU’s reports, all models find significant GDP increases that can be attributed to EU membership (Badinger, 2005; Baldwin et al., 1997; Breuss, 2001; Kohler, 2014; Lejour et al., 2001).
Among others, Baldwin et al. (1997) and Kohler (2014) argue that permanent and temporary growth effects of enlargement are mainly attributed to major increases in effi-ciency and investment, since the risk for the latter has become smaller due to institutional harmonization (Badinger, 2005). The main driver behind boosting growth is the fact that EU integration has lowered the risk premium for investment in the EU10 countries sig-nificantly. In one simulation, the resulting gain is considered to be about 18.8 percent of additional GDP for the EU10 until steady state (Baldwin et al., 1997). The net effects of EU accession can be decomposed even further, as suggested by Breuss (2001): into trade-and single market effects as well as into gains from factor movements (capital trade-and labor) and the cost of enlargement, respectively. According to Breuss (2001) the single market effect seems to be the leading driver and simulations show that this effect ultimately leads to positive additional GDP for Hungary and Poland (8-9 percent), the Czech Republic (5-6 percent) and for the EU15 (0.5 percent) after the 2004 enlargement. If the modest gains in terms of GDP for the EU15 countries are considered, Austria, Germany, and Italy benefit the most, while in Denmark, Portugal and Spain losses occur. Generally speaking, the findings of Breuss (2001, p. 15) can be seen as evidence for the idea that enlargement does not lead to an evenly distributed welfare increase or a necessary welfare increase at all, but rather suggest enlargements as a ”shock with asymmetric outcome[s]” for the affected countries.
8 percent of additional GDP income until their simulation reaches steady state, but is varying across industries (Lejour et al., 2001).
The European Commission (2010, p. 1) officially states that ”[e]nlargement is a mu-tually beneficial process for the EU and its new members” and concludes that after the 2004 enlargement the new member states benefited with rapid growth rates. Nevertheless, it is unclear whether these growth rates can be solely attributed to EU membership or whether they reflect the general economic development of the European economic area.
However, there are several countries, which have been experiencing higher GDP growth rates on average after the EU accession than before. Table 10 in the Appendix reports the accession dates of all EU members and additionally the averaged GDP growth rates of the ten-year period before and after the accession. It can be seen that Portugal, Spain, Finland and Sweden, but also several EU10 countries, namely Czech Republic, Estonia, Latvia, Lithuania, Slovakia, Bulgaria and Romania have on average experienced higher growth rates after joining the EU than before. To summarize the findings so far, 21 countries have joined the seven founding members between 1973 to 2013, of which eleven experienced higher growth rates post EU accession than before. Consequently, we cannot conclude that EU accession has led to higher growth per se.
On the other hand, the reports of the EU Commission and the Directorate-General for Economic and Financial Affairs about economic achievements of an enlarged EU (Eu-ropean Commission, 2006, Ch. 2.3) emphasize that with integration into the EU comes higher growth in general. But at the same time, they also acknowledge that even the countries with higher growth rates do not experience exceptional high growth, when com-pared to other countries’ growth performance such as emerging economies. Nevertheless, it still remains unclear whether these growth rates reflect the state and development of the world economy or whether they have been realized due to EU accession. Surprisingly, the success of enlargement seems to be only justified by positive GDP growth data of new member states and by ex-ante forecasts and simulations.
In this line of reasoning, it is problematic to argue based on forecasts and simulations only without having validated these findings with real world data ex-post. Moreover, it is important to acknowledge that the enlargement of 2004 was accompanied by a liber-alization of trade, which already had begun after the breakdown of the Soviet Union. By the early 2000s almost 85 percent of trade within the EU15 and EU10 was covered in trade agreements. However, the macroeconomic impact of the EU10 joining the EU is considered to be modest, since the new members add only 9 percent of GDP3 to the
economic strength of the whole Union (European Commission, 2006, Ch. 4).
In contrast to earlier research, one theoretical problem of an ex-post evaluation is that researchers are facing a missing counterfactual problem as every country can only be observed with or without EU membership. Campos et al. (2014) circumvent this identification problem by employing a recently developed approach, in which a synthetic
control country is created. By this, they are able to quantify gains from EU membership ex-post. The methodology makes it possible to create a synthetic country out of a weighted recombination of other countries which are not part of the EU. The idea is to come up with weights based on covariates prior to the treatment of EU accession to reproduce a country’s GDP trend correctly. Campos et al. (2014) follow Abadie and Gardeazabal’s (2003) approach, who used the same methodology to estimate the cost of political conflict. This method has been applied to all members joining the EU until the 2004 enlargement and finds positive gains for all countries, except for Greece. The latter is considered to be an exceptional case, in which early losses can be attributed to weak domestic industries in Greece. Nevertheless, the results vary heavily across countries. For example, the gains for Finland seem to be rather small, whereas the EU10 countries are gaining significantly from EU membership (Campos et al., 2014).
2.2
Turkey between Customs Union and EU Membership
Following from the discussion so far, Turkey would be expected to benefit from EU ac-cession through the channels being summarized in Table 1. Nevertheless, forming solely a customs union can be understood as a lower level of integration, in which still imped-iments to trade, factor flows and only limited access to the EU market exist (Balassa, 2013). However, a customs union corresponds to a deeper level of integrations than a free trade agreement, because the same external tariffs apply, on which need to be agreed on and negotiated about in the first place. Thus, larger welfare gains can be realized (Facchini et al., 2013). Considering trade specifically in the case of Turkey, it must be acknowledged that not even all products are covered in the customs union agreement with the EU (European Commission, 2016d).
Mercenier and Yeldan (1997) conclude ex-ante that a customs union, which is not followed by further integration, might be rather costly and eventually impose negative welfare effects on Turkey eventually. They attribute this to the Turkish economy at the time of forming the customs union, because it used to be highly state-controlled and not competitive enough to cope with imported products from the EU (Mercenier and Yeldan, 1997).
On the contrary, Harrison et al. (1997) find ex-ante positive effects for forming a customs union with the EU and mainly hold the improved access to third country markets accountable for this effect. In their GEM simulation, a customs union yields additional GDP of 1–1.5 percent for Turkey. But they also acknowledge possible negative effects such as a budget shortage and the importance of finding alternative income, since tariffs will not contribute as extensively as before to the Turkish budget anymore.
realized due to an institutional harmonization yielding additional 5.6 percent of GDP until their simulated economy has reached steady state. Yet, the effect from the shared common internal market would be rather small with 0.8 percent and the effects from migration remain unclear. These findings stress the integration channel of institutional harmonization, which might positively affect GDP due to full EU membership. However, this finding is based on ex-ante simulations and might be mitigated in the long-run.
2.3
A Brief History of Turkey-EU Relations
Besides the economic discussion of a theoretical EU accession and Turkey’s benefits from the EU customs union, Turkey’s relations with the EU have been a matter of discussion in the EU since the very beginning, which is discussed in detail below. However, rela-tions between Turkey and the EU’s predecessor started in 1959, whereas the cornerstone for the customs union has been laid in the Ankara Agreement in 1967 including a grad-ual convergence of tariffs and a gradgrad-ual intensification of cooperation. Finally, in 1996 the customs union came into force (European Commission, 2016a). The next step was Turkey’s attempt to join the EU by applying for membership in 1987. It took another ten years until the EU accepted this application and finally, in 2005 negotiations over EU membership started only to be suspended in 2006 (European Commission, 2016a; Schimmelfennig, 2008).
To be able to understand the relations between the EU and Turkey, one needs to consider the historical background and especially the dispute of Turkey and Greece, when it comes to the island of Cyprus. Cyprus had emerged out of the Ottoman Empire and was subsequently leased to the British Empire (1878-1960). Afterwards Turkey as well as Greece claimed the island to be their own territory. In 1974 this conflict resulted in a de-facto division of the island into the Republic of Cyprus – being a member of the EU since 2004 – and a smaller, Turkish part being only recognized by Turkey. The Republic of Cyprus is recognized not only by the EU, but also by the UN as a sovereign. On the contrary, Turkey refuses to recognize the Republic of Cyprus as sovereign, but supports the autonomous Turkish region among other things with military backup (Hill, 2010). That is the reason why Greece and other countries blocked Turkish EU membership attempts. When Cyprus joined the EU in 2004, Turkey in turn refrained from expanding the customs union treaty with the EU to Cyprus. According to Schimmelfennig (2008), the main reason for that is a conflict of national identity that kept Turkey from both extending the EU customs union to Cyprus and recognizing the new EU member as sovereign. As a consequence, this caused the EU to put the negotiations on hold.
On one hand this can be seen as the historical reason, why Turkey refused to enlarge the customs union and on the other hand, it also helps to understand why Greece and other member states have blocked Turkish membership attempts for decades.
have resulted in resistance from the member states for a long time. Assuming current population growth to remain stable, Turkey would not only have the largest votes within 20 years in the Council of the EU, but would also contribute with a population of 80 million people – of which the majority is Muslim – extensively to the EU population being mainly Christian (Flam, 2004). Furthermore, the EU and its members claim that Turkey does not respect human rights in a way the EU does and suppresses minorities and the opposition (Flam, 2004).
More recently, the EU is open for debating Turkish EU accession again in order to resolve the current refugee crisis caused by the conflict in Syria, in which Turkey shall play the role of the EU’s gatekeeper. That is why good relations to Turkey have been restored including negotiations about visa-free travel to the EU for Turkish citizens and other areas, where negotiations were stuck. However, it is neither clear if this attempt of co-operation is permanent or rather short-lived, because some EU members already declared to be willing to block this deal, to which Turkey responded that they are not willing to cooperate with the EU without a visa deal (European Commission, 2016c).
Figure 1 – Development of Turkish exports of goods and services.
10
20
30
40
exports of goods and services (percentage of GDP)
1980 1985 1990 1995 2000 2005 2010 2015
Year
Turkey European Union
Source: World Bank (2016b), whereas European Union refers to the EU28 World Bank aggregates.
a convergence of exports as percentage of GDP towards EU levels has been taking place. Yet, Turkey’s imports and exports account for 60 percent of the country’s GDP and Turkey has had a negative trade balance for the last thirty years with only a few exceptions. Comparing these figures to the ones of the EU, where trade contributes by 81 percent to GDP on average, this suggests that Turkey is not an open economy per se (World Bank, 2016b).
According to World Bank (2016a) Turkey is the 17th largest economy in the world and belongs to the group of middle income countries. Moreover, Turkey is a member of the OECD, the G20 and the NATO. Turkey has been making progress in terms of increasing welfare and fighting poverty. From 2002 to 2012 it was possible to reduce extreme poverty from 13 to 4.5 percent of the Turkish population and since the financial crisis in 2008 6.3 million new jobs were created (World Bank, 2016a). Figure 2 supports this finding and clearly shows that more recently Turkey has been able to sustain unemployment levels well below 10 percent and is thus doing better on average than the EU.
Figure 2 – Development of Turkish unemployment and GDP growth.
6
8
10
12
14
percentage of total labor force 1980 1985 1990 1995 2000 2005 2010 2015 Year A. UNEMPLOYMENT −5 0 5 10 percentage change 1980 1985 1990 1995 2000 2005 2010 2015 Year B. GDP GROWTH
Turkey European Union
Source: World Bank (2016b), whereas European Union refers to the EU28 World Bank aggregates.
arising from the conflict in Syria, the domestic conflict with the Kurds and the question, whether the government has the capabilities to handle corruption effectively. This is ultimately resulting in less trust as well as hampered private demand and by this affecting economic growth.
Turning explicitly to the EU customs union, real GDP growth at market prices ten years prior to 1996 was on average 4.7 percent and after 1995 on average 4.6 percent (World Bank, 2016b).4 Consequently, it can be said that from GDP growth as such it is
not possible to obtain any effects, which the customs union may had on GDP growth.
2.4
Hypotheses
Following the reasoning so far, we hypothesize that Turkey – joining solely a customs union with the EU – might have only limited access to the channels through which integration has a positive impact on GDP. Therefore, Turkey might only be able to exploit partial additional growth benefits, when comparing the opportunities in a customs union to the opportunities of enhancing welfare from EU membership. Nevertheless, after having formed a customs union with the EU, Turkey is expected to realize additional GDP growth and thus, higher GDP per capita levels, albeit less than in the theoretical case of full EU membership. Hence, we can state a first hypothesis that will be addressed in the empirical analysis in the next section.
Hypothesis 1: After joining a customs union with the EU, Turkey is expected to expe-rience additional GDP growth.
However, economic integration is mainly driven by positive trade effects, factor flows and in particular capital flows as well as efficiency increases due to a common single market. On the contrary, the other factor flow – labor migration – either plays a limited role or is not examined in particular (Lejour et al., 2001). Generally, it can be summarized that EU accession is associated with higher GDP growth and results thus in higher GDP per capita levels, because the benefits outweigh the costs as stated above. Therefore, we formulate the following hypothesis:
Hypothesis 2: In the hypothetical case of being an EU member, Turkey would realize a GDP growth, which is larger than when being only in a customs union.
4A similar picture emerges, when instead of market prices PPPs are taken into account (The Conference
3
Turkey’s Gains from a Customs Union with the EU
First of all, it is our objective to test for Hypothesis 1 and to quantify, whether forming a customs union has had a positive impact on Turkish GDP per capita levels or not, whereas Hypothesis 2 is tested in section 4, in which also the related methodology is explained.
To further investigate the effects of an EU customs union on GDP growth, we employ a time-series regression to gather first evidence. However, Turkish GDP growth needs to be related to the global economic environment. This is important for answering the question whether additional growth can be solely attributed to forming a customs union or whether other omitted global developments and country-specific unobservable characteristics cause the GDP growth patterns. Moreover, the volatile Turkish GDP growth makes it hardly possible to isolate the effect of a customs union with the help of a simple regression. Additionally, it is crucial to determine the extent of the effect, which a customs union has had on Turkish GDP.
That is the reason why we employ in a second step so-called synthetic control groups for a case study, in which Turkey in a customs union is compared to a synthetic Turkey with the same characteristics except for the fact that it is not forming a customs union. Thereby, the gains from a customs union are not only evaluated from a different per-spective, but are also corrected for omitted variable biases (Abadie and Gardeazabal, 2003).
3.1
Methodology: Time Series Regressions
In order to capture the effects from the customs union, Turkish GDP growth is regressed on potential determinants of economic growth. An additional indicator variable for the customs union being equal to unity from 1996 onwards and otherwise zero, is added to capture the effects a customs union might have had on GDP growth.
YYY = XXX β + DDD γ + ε,
in which YYY is a 41 × 1 vector with Turkish GDP growth from 1970 to 2010, DDD is a matrix with several dummy variables among others the customs union dummy and XXX is a 34 × i matrix with the explanatory variables. i refers to the number of explanatory variables in the different models, which are explained below.
In a subsequent step, model 2 builds upon the work of Barro (1996) and Mankiw et al. (1990), in which a broader set of variables is used to capture the effects of human capital on GDP growth. Therefore, secondary and tertiary school enrollment (both as percentage, (Barro and Lee, 2013)) as well as life expectancy (logarithm of life expectancy at birth, (World Bank, 2016b)) are added as explanatory variables. In addition, the initial income level (logarithm of GDP per capita converted to 2014 price level with updated 2011 PPPs, (The Conference Board, 2016)) as well as trade (exports and imports as share of GDP), inflation (annual change of consumer price index, both (World Bank, 2016b)) and government consumption as percentage of GDP (calculated as share from the absolute values in current national prices, (UN Statistics Division, 2015)) are added as explanatory variables. Lastly, model 2 is completed by a democracy index combining the democracy scores of ten independent indices5 (Pemstein et al., 2010).
Model 3 is specified to test for the robustness of the results and to control for For-eign Direct Invesment (FDI, net inflow as percentage of GDP, World Bank (2016b)). In contrast to model 2, this model does not take government consumption into account, but rather focusses on the structure of the whole economy in the form of capital formation, agriculture share and manufacturing share of GDP (calculated as share from the absolute values in current national prices, UN Statistics Division (2015)) are taken into account.
Lastly, model 4 differs only in the sense that it contains dummy variables for each decade in order to control for time fixed effects. The results are discussed and reported in section 3.4.1.
3.2
Methodology: Synthetic Control Groups
In an ideal world it would be possible to compare Turkey forming a customs union to a country with exactly the same characteristics as Turkey has, except for the fact that it is not forming a customs union. On one hand, regressions allow to control for certain characteristics in order to isolate the effect of a customs union on GDP, but on the other hand unintentionally omitted variables will lead to biased estimates. Therefore, this paper creates a synthetic Turkey from a number of similar countries based on a set of covariates
5The unified democracy index of Pemstein et al. (2010) combines ten different indices such as the Freedom
to compare it to actual Turkey and subsequently determine the difference in GDP per capita levels between those two. Since it is the idea that the two countries do not differ, except for fact that actual Turkey is forming a customs union, the differences in the GDP per capita levels can be solely attributed to the EU customs union.
An important aspect of such a comparison would be to identify a country or a set of countries that can be combined to create a synthetic Turkey. In general, it will be misleading to chose a comparison country randomly or to select a pool of countries by chance, because their GDP per capita development is biased by the global economic development and individual country characteristics (Abadie and Gardeazabal, 2003). To tackle this issue, the methodology of synthetic control groups is employed, which creates a synthetic country consisting of the weighted average of other countries. These other countries may not be affected by forming an EU customs union. The idea is that before 1996, the year in which Turkey joined the customs union, the weighted average of the other countries’ GDP levels should be matching with Turkish GDP per capita levels as exactly as possible. The differences that occur from 1996 onwards can then laregly be attributed to the fact that Turkey formed a customs union, whereas synthetic Turkey (the weighted countries) does not. Yet, there are limits to this approach, which are discussed later on in section 3.5.
The main challenge is that the differences in GDP per capita levels prior to 1996 between actual and synthetic Turkey must be as small as possible. Consequently, optimal weights must be chosen in such a way that they are minimizing this difference. The definite methodological approach employed to achieve this is described subsequently (Abadie and Gardeazabal, 2003; Abadie et al., 2010, 2015; Campos et al., 2014).
We follow Abadie et al.’s (2015) approach and consider a balanced panel of J + 1 countries, in which j = 1 is Turkey. That means that j = 1 is the country being affected by forming a customs union with the EU, whereas the other countries in the dataset j = 2 until j = J + 1 may not engage in a customs union with the EU. Borrowing from Abadie and Gardeazabal (2003), the remaining countries are called donor pool, from which one can derive a synthetic control group with the help of individual weights.
It is necessary to assume having a balanced panel, in which all country observations are present for the same years t = 1, . . . , T . Secondly, a positive number of pre- (Tpre) and
post-customs union (Tpost) country observations with T = Tpre+Tpostare needed. Another
important assumption is the clear distinction for Turkey (j = 1) between the period in which a customs union is in force (Tpre+ 1 . . . T ) and the period prior to 1996 (1, . . . , Tpre),
in which there is no customs union. It is necessary that a potential effect from the customs union on GDP per capita levels is present only from the date the customs union came into force onwards and that there are no anticipation effects (Abadie et al., 2015). These assumptions and the possible limitations following from them are discussed later on.
The question is, how are these weights chosen and which countries should be taken into account. As already mentioned above, the countries from the donor pool are selected in such a way that they reflect the Turkish GDP development prior to 1996 (during Tpre) as
precisely as possible and the weights W need to fulfill the side condition of minimizing the difference between synthetic and actual Turkey during that time. To come up with optimal weights, X1 defines a k × 1 vector, which embodies the values of characteristics
for Turkey (j = 1) before engaging in a customs union, whereas X0 describes the same
variables but contrarily for the countries in the donor pool. We formulate the differences of the variables in these vectors as X1− X0W before the enforcement of a customs union
with the EU and define the optimal weights as W∗ minimizing the difference between X1− X0 (Abadie et al., 2015).
We follow Abadie and Gardeazabal (2003) and Abadie et al. (2010) in order to solve for the optimal weights. These weights are derived from variables capturing the Turkish economy and its structure and include the economy’s output composition, indicators for human capital and initial income levels. In this regard, X1m is referring to the value of
the m-th variable for Turkey, whereas X0mconsists of the same variable for the countries
in the donor pool. Afterwards W∗ can be obtained by minimizing:
k
X
k=1
vm(X1m− X0mW )2, (1)
where vm is a vector of weights referring to the importance of the variables in X0 and
X1. To implement this model and to calculate vm, the computation process6 follows
thereby a number of optimization routines. First of all, a number of regressions will be run determining the explanatory power of the variables in vm for explaining GDP per
capita. Following from the results the variables are weighted in such a way that the explanatory power of the variable is reflected in its weight (vm) in equation (1). However,
there are various ways of choosing vm, which will not be discussed any further such as
cross-validation or even manual selection (Abadie et al., 2015). After having determined vm, the derivation of equation (1) is calculated with respect to W yielding the optimal
weights w∗. This will result in w∗ = 0 for some countries, which means that they are not contributing to synthetic Turkey, because only countries for which w∗ > 0 are central to form synthetic Turkey.
Finally, the synthetic control estimator is defined by Abadie et al. (2015) as:
Y1t− J +1
X
j=2
wj∗Yjt, (2)
6In this paper, the method is implemented by using the statistical software package STATA with the
synth-module and the option nested. Thus, a regression is used to find the optimal weights in vm by
where Y1t is the actual GDP per capita of Turkey in year t with t spanning from Tpre to
Tpost. In the same manner, Yjt is the GDP per capita for the countries j = 2 . . . J + 1 in
the donor pool at time t. The second term of equation (2) refers to synthetic Turkey’s GDP per capita and is a combination of GDP per capita levels of the countries in the donor pool weighted by w∗.
To conclude with, X1 and X0 in equation (1) are only observed prior to 1996 and are
used as independent predictors to derive the weights. Then, the weights are used to solve equation (2) and eventually synthetic Turkey can be created (Abadie et al., 2015).
3.3
Data for Synthetic Control Groups
It is important to pay respect to the two main assumptions underlying the estimation of synthetic control groups. First, the countries in the donor pool may not form a customs union with the EU. Since being a member of the European Union includes also being a member of the EU’s customs union, the countries in the donor pool are not supposed to be EU members either. Second, variables may not include characteristics from the period prior to 1996 that anticipate possible GDP per capita effects (Campos et al., 2014). Therefore, it is important to chose a set of variables in X0 and X1 that is explaining
GDP per capita development approximately rather than anticipating the development. Abadie et al. (2015) use trade openness, inflation, industry share of GDP, schooling and the investment rate to estimate the economic impact of the German reunification on GDP. Consistent with this choice, Campos et al. (2014) take a similar set of variables into account to estimate gains from EU membership. In this paper, we collect four different datasets from two different sources. The synthetic control groups synth a and synth b are estimated from the first dataset, whereas the synthetic control groups synth a-c* are estimated from the second dataset.
3.3.1 Dataset for Synth A and Synth B
First of all, we will follow Campos et al.’s (2014) approach in estimating a synthetic control group for Turkey being not in a customs union to make the case comparable to the benefits from EU membership as estimated by Campos et al. (2014). Therefore, capital formation, population growth, agriculture and industry share of GDP, secondary and tertiary school enrollment as well as initial GDP per capita levels are taken into account. Table 11 in the Appendix displays all the variables and the corresponding sources that have been used for estimating synth a-b.
Hence, all observations older than 1984 are dropped and missing values are interpolated linearly. The problem is that for some variables such as industry and agriculture share of GDP roughly 16 percent of the observations are still missing and that in the case of tertiary school enrollment also 12 percent of the observations are missing. After omitting the observations for all countries, where the needed variables are not fully present from 1984 to 2011, the dataset contains only a remainder of 56 countries with almost no OECD countries in it. This dataset is subsequently labelled with a.
In addition, Table 10 in the Appendix gives a detailed overview of the history of EU enlargement and is taken into account to assure that none of the countries in the donor pool is affected neither by a customs union nor EU membership. However, countries such as Bulgaria and Romania, which have been joining the EU in 2007 are not omitted, because the main focus lays on the effects of a customs union on Turkish GDP per capita levels after five (2001) and ten years (2006). This is resulting in a strongly balanced panel with no missing observations and 56 country observations spanning fully from 1984 to 2011 (synth a).
Nevertheless, the donor pool is far from being ideal, because Abadie et al. (2015) argue that countries need to be chosen, which are similar to the country of interest, that is, Turkey. For this reason, all countries that are not located in greater Europe, Middle East (including Pakistan and India) or North Africa, are dropped and subsequently the dataset contains only a remainder of 16 countries (synth b).
3.3.2 Dataset for Synth A-C*
To come up with a longer time period and thus, a greater period prior to the enforcement of the customs union between Turkey and the EU, we compile a second dataset from different sources. The investment, agriculture and manufacturing share of GDP are all taken from the UN National Accounts Main Aggregates Database (UN Statistics Division, 2015). Population and GDP per capita in 2014 US$7are sourced from the Total Economy Database (The Conference Board, 2016). The education data on secondary and tertiary enrollment rates is obtained from Barro and Lee (2013). Since the latter is only available on a five-year basis, the missing values are generated following a linear trend. This is resulting in the dataset being used to estimate synth a* containing 56 countries with their observations spanning fully over 41 years from 1970 to 2010. Due to the fact that the education data is only available until 2010, thus all years greater than 2010 have been omitted.
However, in a subsequent step we limit the dataset to countries laying within greater Europe, Middle East (including Pakistan and India) and North Africa leading to 22 country observations eventually (synth b*).
Lastly – and mainly due to robustness checks – two additional variables are added
to the dataset of synth a*, namely life expectancy at birth (World Bank, 2016b) and a democracy index (Pemstein et al., 2010), which are then used to estimate synth c*. synth c* is discussed in detail only in section 3.5. Table 12 in the Appendix displays all the variables and the corresponding sources that have been used for estimating synth a-c*.
3.4
Results
The results for both methodological approaches are discussed below. The regression re-sults are considered as first evidence, whereas synthetic control groups are considered to have advantageous characteristics, because they are addressing endogeneity, omitted vari-ables, measurement errors and causality concerns (Campos et al., 2014). Consequently, we discuss the synthetic control groups more broadly and in detail to assess the robustness of the findings.
3.4.1 Regression Results
The results of the regressions for model 1-4 are reported in Table 2. Model 4 yields the highest r-squared and adjusted r-squared. Therefore, it is considered to explain most of the variation in Turkish GDP growth, when compared to the other models. However, in model 4 only foreign direct investment and total factor productivity are found to be significant for explaining the variation in Turkish GDP growth. The only difference between model 4 and model 3 is that in the former time fixed effects are included, turning to model 3 additionally capital formation is found to be significant.
Moreover, also model 2 does not yield any significant effects for the customs union dummy, which is splitting the sample in a prior to 1996 period (CU dummy = 0) and a period from 1996 onwards (CU dummy = 1). In this model, the initial income level, sec-ondary school enrollment and negatively inflation are found to be significant at explaining Turkish GDP growth.
Finally, only the most simplified model (model 1 ) is able to explain GDP growth sig-nificantly with capital formation, population growth and total factor productivity. Most importantly, also the customs union dummy is positive and significant at the 5 percent level. That means, that according to model 1 4.17 percent of the variation in Turkish GDP growth can be explained by the customs union. However, the r-squared (0.511) is relatively low and indicates that the model is explaining only have of the variation in GDP growth.
Table 2 – OLS regressions for Turkey with dependent variable GDP growth.
model 1 model 2 model 3 model 4 variables ln(GDP p.c.) 40.82*** (12.88) Capital formation 59.86*** 80.85** 50.46 (19.26) (36.15) (35.58) Agriculture share 22.89 124.3 (83.27) (86.35) Manufacturing yhare -35.42 28.27 (41.53) (45.28) Government consumption -66.28 (68.22) Secondary enrollment -0.920* -0.788 -0.304 (0.469) (0.551) (0.529) Tertiary enrollment -0.934 -1.315 -0.533 (0.738) (0.802) (0.811) ln(life expectancy) -29.09 104.1 39.59 (59.90) (86.11) (86.82) Population growth 465.9** (211.4) Democracy index -0.994 (2.046) FDI -2.750** -2.486** (1.112) (1.016) Trade 0.159 0.190 0.133 (0.149) (0.128) (0.119) Inflation -0.0894*** -0.0557 -0.0516 (0.0286) (0.0329) (0.0318) TFP 38.82*** 36.63* 55.72*** (10.31) (20.13) (19.73) CU dummy 4.170** 1.869 -1.961 0.286 (1.758) (3.161) (3.197) (3.081) Constant -55.08*** -222.7 -451.5 -229.8 (10.41) (207.8) (349.4) (351.1)
Time fixed effects no no no yes Observations 41 41 37 37 R-squared 0.511 0.462 0.687 0.772 R-squared adjusted 0.457 0.305 0.549 0.628
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
3.4.2 Synthetic Control Group Results
Table 3 – Actual and synth a/synth b with average variable values for Turkey.
actual synth a synth b
Capital formation 23.51 23.23 19.06 Population growth 1.86 1.85 1.84 Agriculture share of GDP 17.91 17.85 17.31 Industry share of GDP 31.58 31.67 31.16 Secondary enrollment 49.25 50.25 50.28 Tertiary enrollment 12.50 13.13 13.13 GDP p.c.(1990) 7793.08 7783.35 7786.65 GDP p.c.(1995) 7959.05 8385.82 8385.72 RMSPE 398.69 500.46 N 53 16 T 28 28 ∆ 5 years -7.24% -11.64% ∆ 10 years 7.74% 4.36%
Notes: See subsection Data to see, how synth a and synth b are compiled or refer to Table 11 in the Appendix for a detailed explanation of the variables.
Both datasets can be used to construct synthetic controls with differences in the match-ing variables smaller than one percentage point (for non-GDP per capita variables; Table 3). The different datasets yield quantitatively similar results where they only differ in the five- and ten-year change in GDP per capita, where synth b yields smaller changes. A similar picture emerges using synth a* through synth c* (see Table 4). After five years of forming a customs union (in 2001) the differences of the GDP per capita levels between actual and synthetic Turkey range from -11.64 percent (synth b) to -1.53 per-cent (synth c*). Contrarily, after ten years (in 2006) they range from 3.65 (synth a*) percent to 17.93 percent (synth c*).
yields higher predictive power when compared to b, in which countries are limited to geographically proximate ones. Hence, we can neither conclude that limiting the donor pool to similar countries is necessary nor that it yields better results. Campos et al. (2014) argue similarly and confirm that this limitation might not be important, when estimating synthetic controls for EU membership. Additionally, all countries and their weights are reported in Figure 14.
Table 4 – Actual and synth a-c* with average variable values for Turkey.
actual* synth a* synth b* synth c*
Capital formation 19.71 19.74 23.98 19.75 Population growth 2.23 2.17 2.21 2.23 Agriculture share of GDP 17.19 17.19 18.09 17.21 Industry share of GDP 23.35 23.38 16.48 19.79 Secondary enrollment 12.47 19.73 21.71 22.45 Tertiary enrollment 6.58 6.60 8.15 6.60 ln(Life expectancy) 4.09 4.10 Democracy Index 0.30 0.30 GDP p.c.(1970) 6095.00 6139.02 6251.66 6190.17 GDP p.c.(1995) 11434.10 11460.49 11463.78 11442.54 RMSPE 346.37 393.8647 334.46 N 53 22 52 T 41 41 41 ∆ 5 years -7.81% -10.07% -1.53% ∆ 10 years 3.65% 4.48% 17.93%
Notes: synth a-c* are compiled from a different dataset. See subsection Data for more information or refer to Table 12 in the Appendix for a detailed explanation of the variables.
Turning to synth a* in particular, the average synthetic values do not deviate much from their actual counter parts as can be seen in Table 4. That means that the synthetic controls have good explanatory power in terms of explaining GDP per capita before the customs union and thus, are also expected to have good predictive power in the period from 1996 onwards. Taking into account specifically the second column describing the synthetic control being created from dataset a*, the investment share of GDP (or the so-called capital formation) deviates only by 0.02 percentage points from the actual value and the actual agricultural share is even matched exactly by synth a*. On the other hand, the largest deviation from actual Turkey can be found in the case of secondary school enrollment by 7.26 percentage points. GDP per capita in 1970 deviates by 45 US$ and in 1990 only by roughly 26 US$. In general, GDP per capita levels seem to be over estimated in the three synthetic control groups (synth a-c*).
Figure 3 – Comparing growth rates between synth c* and actual Turkey. 2.7 2.43 2.82 1.4 0 1 2 3
average GDP p.c. growth (in percent)
1986−1996 1996−2006
actual synthetic actual synthetic
Notes: A t-test was employed to test, whether the difference in mean GDP levels prior and from 1996 onwards for actual Turkey is different from synthetic Turkey. Thus, it is tested if H0 : mean(∆) = 0 against H1 : mean(∆) < 0,
H2 : mean(∆) 6= 0 and H3 : mean(∆) > 0. In this regard, it is defined
that mean(∆) = mean(synth. GDP p.c. − actual GDP p.c.). In the period prior to 1996, evidence is found that the mean(∆) = 0, whereas H1−3 can be
rejected. Contrarily, from 1996 onwards it is found that mean(∆) 6= 0 and that mean(∆) > 0 are significant at the one percent level.
overall effects of the customs union are significant and positive, when the average growth rates of actual Turkey and synth a* are taken into account. Figure 3 shows the average growth rates for actual Turkey and synth a* and distinguishes between the ten year period before and after the customs union was formed. It can be concluded that not only growth of actual Turkey between 1996 and 2006 was on average by 1.42 percentage points higher, but also that the mean difference of actual and synthetic Turkey post 1995 is found to be significantly different from and larger than zero.
In addition, Figure 4 plots the GDP per capita development in natural logarithms for Turkey and its corresponding synthetic control group for all four data specifications. The dashed vertical line labels the year, from which the customs union was formed onwards or in other words the cut-off year. In the top left-hand panel synth a can be found, whereas the top right-hand panel belongs to b, bottom left to a* and bottom right to b*. All four graphs explain the stylized facts as described above.
Figure 4 – Synthetic control groups for Turkey. 8.8 9 9.2 9.4 9.6 ln(GDP per capita) 1985 1990 1995 2000 2005 2010 Year SYNTH A 8.8 9 9.2 9.4 9.6 ln(GDP per capita) 1985 1990 1995 2000 2005 2010 Year SYNTH B 8.8 9 9.2 9.4 9.6 9.8 ln(GDP per capita) 1970 1980 1990 2000 2010 Year SYNTH A* 8.8 9 9.2 9.4 9.6 9.8 ln(GDP per capita) 1970 1980 1990 2000 2010 Year SYNTH B*
actual Turkey synthetic Turkey
Notes: In this figure the solid line refers to Y1t in equation (2), whereas the dashed line is referring to
PJ +1
j=2w∗jYjtin the same equation (see section 3.2 Methodology: Synthetic Control Groups).
be attributed most likely to the burst of the Dotcom bubble in 2001.
After all, considering synth a*, almost half of the Turkish average GDP growth from 1996-2006 can be explained due to the formation of a customs union. However, there might be two problems with this finding. First, the accuracy of the synthetic control group abates with increasing time, because it builds upon the GDP development of other countries before the enforcement of the customs union in order to replicate Turkey’s trend. As time goes by, there could be other countries rebuilding Turkey’s trend more precisely and the prediction quality mitigates in the long-run. Secondly, the robustness of the results still needs to be assessed.
3.5
Robustness and Limitations
robustness of the results above.8
However, to test the model’s precision even further, we add two more variables to assign more importance to human capital, when explaining GDP per capita. Following Barro (1996) life expectancy and a democracy index are added as explanatory variables. The resulting synthetic Turkey (synth c*) has even a lower RMSPE (334.46). Figure 5 plots the resulting GDP per capita values in natural logarithms. The differences – when compared to Figure 4 – can barely be obtained and confirm that the results are robust.
Figure 5 – Robustness check with broader set of variables: synth c*.
8.8 9 9.2 9.4 9.6 9.8 ln(GDP per capita) 1970 1980 1990 2000 2010 Year
actual Turkey synthetic Turkey
Another problem with the synthetic control method in the case of Turkey forming a customs union with the EU, is the fact that economic agents could anticipate these effects and change their behavior prior to 1996. For example, one could argue that investment increases earlier due to the anticipated liberalization of trade, whereas one assumption – as introduced earlier – states that there needs to be a clear cut-off point in time (Abadie et al., 2015). Consistent with this line of reasoning Campos et al. (2014) find clear anticipation effects in the case of EU membership. In order to test for anticipation effects the cut-off value defining the enforcement date of the customs union is altered to 1993 and results in Figure 6 with an RMSPE of 327.14. The figure does not display a different scenario, when compared to the results above, because the positive effects that the customs union would have had occur from 1996 onwards. However, the difference of synth a* and synthetic Turkey in Figure 6 can be explained by the methodology itself. The synthetic line in Figure 6 is steeper, because the dip in actual GDP per capita levels in 1994 is not taken
8The statistical software package STATA and the synth-module with the option allopt is used. According
into account for the optimization routine anymore. Consequently, the alteration leads to a different and steeper synthetic control graph. After all, it can be concluded that no clear anticipation effects can be found.
Figure 6 – synth a* with cut-off value set to 1993.
8.8 9 9.2 9.4 9.6 9.8 ln(GDP per capita) 1970 1980 1990 2000 2010 Year
actual Turkey synthetic Turkey
The main downside while applying the method of synthetic controls is that it does not allow to evaluate the significance as it is usually tested in econometrics. This can be either attributed to the fact that in comparative case studies the number of time periods or the number of countries in the donor pool is too small. On the contrary, this approach pays respect to possible endogeneity and omitted variable problems (Campos et al., 2014).
There are different approaches to overcome the missing possibility of hypotheses test-ing. Campos et al. (2014) use a difference-in-difference estimator for the actual and syn-thetic GDP series of each country and subsequently test, whether the difference between both series is significantly different from zero.
On the other hand, Abadie and Gardeazabal (2003) use so-called placebo studies, namely countries from the donor pool being not affected by forming a customs union with the EU to test, whether their synthetic counterpart matches the actual GDP per capita levels throughout the whole time period.
being not affected by an EU customs union – do not show any gap between the synthetic estimates and the actual GDP per capita values. In this line of reasoning, Turkey should be the exception with the largest differences between actual and synthetic GDP per capita from 1996 onwards.
Figure 7 – Placebo tests showing the gaps between synthetic and actual GDP per capita.
−2
−1
0
1
difference actual and synthetic GDP p.c. (in percent)
1970 1980 1990 2000 2010
Year
Figure 7 shows the estimated gaps between synthetic and actual GDP per capita for all countries from the donor pool (light grey lines) and the estimated gap for Turkey represented by the black solid line. It is obvious that Turkey is – abstracting from one positive outlier – the exception with a positive trend from 1996 onwards. These results also allow the assessment of the statistical significance. If a country would be picked at random, the probability of finding an effect as large as the one for Turkey or even larger, should be equal or smaller than five percent. Comparing the findings from Figure 7 to a statistical significance test at a common level (α = 0.05), the customs union with the EU has a significant impact on Turkish GDP per capita, because out of 53 countries only one shows an effect as large (or even larger) as Turkey.
Although we find the results to be robust and significant, there are some limitations to the estimates presented here. First, due to the fact that the synthetic countries are based on similarities of Turkey and countries in the donor pool before 1996 the underlying characteristics may change in the long-run and will thus mitigate the predictive power of the model. For example, one can argue that Turkey’s GDP per capita levels are matched perfectly by a set of countries before 1996 based on certain characteristics, but it could be that all the similarities have vanished 15 years later leading to biased results.
Turkey are reported in Table 14 in the Appendix. For example, synth c* is among others created from Brazil and Japan. If an exogenous shock has had a negative impact on either Brazil’s or Japan’s GDP per capita levels, but not on Turkey, it will lead to a downward bias for the synthetic GDP per capita levels. Ultimately, this would lead to a larger gap between actual and synthetic GDP per capita and thus give the impression that the effects from a customs union are larger than they actually are. We argue that synth a corrects for this bias, because it includes a lot more different countries and is thus more diversified when it comes to country specific effects. Nevertheless, the impact of forming a customs union is still positive. In this case (synth a), the growth rate which can be attributed to the customs union from 1996 to 2006 is on average 0.53 percent accounting for roughly 17 percent of overall GDP growth.
4
Gains from EU Membership
Since our objective is to compare the actual GDP per capita gains from an EU customs union for Turkey to the hypothetical gains from EU membership, a counterfactual for Turkey needs to be created to test for Hypothesis 2.
In this section, we use the estimates by Campos et al. (2014), who also used a synthetic control approach similar to the one in section 3, to predict potential gains of an EU membership for Turkey. The objective in this section is to make these gains applicable to the case of Turkey. In particular, this approach is used in order to make the effect of EU membership and the effects from a customs union on Turkish GDP comparable, because both estimates are based on synthetic controls and yield the differences in GDP per capita levels after five and ten years. In this regard, we also want to assure that even if the methodology produces slightly biased results, the biases will be found in both results, which makes them comparable eventually.
4.1
Methodology: Turkey as an EU Member
Table 5 shows the additional GDP per capita gains due to EU accession after five and ten years for 17 countries having already joined the EU as full-fledged member (see Campos et al. (2014) for the detailed derivation). They used the same methodology to derive the results as it has been applied in this paper to the case of Turkey forming a customs union with the EU and a similar dataset being based on the same variables.
confirmed by the other country observations in Table 5. Therefore, one could argue that this pattern suggests that Turkey as a relatively open and developing country would have gained substantially from EU membership. However, these results need to be applied to Turkey while controlling for country and time specific effects in order to come up with convincing results.
Table 5 – Estimated GDP per capita gains from EU accession.*
country year +5 years +10 years
Denmark 1973 10.29 14.30 United Kingdom 1973 4.82 8.59 Ireland 1973 5.24 9.40 Greece 1981 −11.59 −17.34 Portugal 1986 11.73 16.54 Spain 1986 9.35 13.66 Austria 1995 4.47 6.36 Finland 1995 2.19 4.02 Sweden 1995 0.82 2.35 Czech Republic 2004 2.11 5.62 Estonia 2004 16.34 24.15 Hungary 2004 8.73 12.30 Latvia 2004 18.02 31.69 Lithuania 2004 17.35 28.08 Poland 2004 8.67 5.93 Slowak Republic 2004 1.32 0.30 Slovenia 2004 6.33 10.35
Source: Campos et al. (2014)
*Measured as deviation between actual and synthetic
GDP per capita in percent and referred to as gap.
The country specific gains in terms of GDP per capita or in other words the gap between the synthetic and actual GDP per capita levels for EU members after their ac-cession, enters the regression as dependent variable. Subsequently, the already introduced channels through which integration affects GDP growth are used as regressors to explain this gap. While controlling for country and time specific effects, an equation is estimated in such a way that it can be applied to the case of Turkey. For this purpose, the following equation is taken into account as a starting point:
YYY = XXX β + DDD γ + ε,
in which YYY is a 34 × 1 vector of the gaps between synthetic and actual GDP per capita as estimated by Campos et al. (2014), DDD is a matrix with several dummy variables and X
X
exports as percentage of GDP) as a proxy for trade effects, total factor productivity as a proxy for single market effects, foreign direct investment as a proxy for capital flows, population growth as a proxy for migration and the control of corruption index as a proxy for the quality of institutions. Lastly, GDP per capita is used to capture the fact that relatively rich countries benefit less from EU accession due to the fact that their growth potential is lower and because of possible costs arising from transfers to poorer countries within the EU. Table 12 in the Appendix gives a detailed overview about the data sources for the variables used, whereas Table 1 summarizes the channels through which economic integration affects growth in theory.
4.2
Empirical Results
The descriptive statistics for all regressors and the dependent variable are given in Table 6. The differences in the gains from EU membership (gap) have already been discussed above in the previous section. The GDP per capita differences in the sample are substantial and range from 18, 300 US$ to 41, 184 US$, whereas the average country observation has a GDP per capita about 27, 036 US$. A lot of variation is also present in the countries’ trade as percentage of GDP ranging from 56 percent to 180 percent of GDP, the quality of the institutions ranging from 0.12 to 2.59 and total factor productivity ranging from 0.56 to 1.03. A notable feature of the data is foreign direct investment (FDI), where also negative values exist with a net outflow of capital in Hungary, Latvia and Slovenia in 2009 in the wake of the preliminary financial crisis. Negative values can also be found for population growth, especially for the Baltic States, Denmark and Portugal reflecting the overall demographic change that takes place in Europe. Contrarily, population growth is not varying much with a standard deviation of about 0.6 percent.
Table 6 – Descriptive statistics for regression variables.
variable mean sd min p50 max N Gap 8.60 9.54 -17.34 8.63 31.69 34 GDP p.c. 27036.12 6422.8 18304.51 26020.66 41184.11 34 Population growth 0.09 0.60 -1.64 0.12 1.43 34 Trade of GDP 97.05 41.51 35.51 90.26 180.06 34 TFP 0.77 0.15 0.56 0.78 1.03 34 Control of corruption 1.15 0.85 0.12 1.04 2.59 34 FDI 7.33 × e9 1.46 × e10 −2.97 × e9 1.74 × e9 8.13 × e10 34
