Overlapping free trade agreements : a gravity analysis of South Africa - EU - SADC hub-and-spoke

OVERLAPPING FREE TRADE

AGREEMENTS

A Gravity Analysis of the South Africa – EU – SADC

Hub-and-Spoke

Master’s Thesis

Author: Anna Abate Bessomo

Student Number: 11385251

E-Mail: abatebea@tcd.ie

Supervisor: Dr. Boe Thio

Second Reader: Dr. Dirk Veestraeten

Program: Msc Economics – International Economics and Globalisation

Date: 13/07/2017

2

Statement of Originality

This document is written by Student Anna Abate Bessomo, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

3

Abstract

This thesis studies how the formation of the overlapping free trade agreements (hub-and-spoke free trade network) between South Africa and the EU and between South Africa and the Southern African Development Community, formed in 2000, has affected trade flows. The three-good, three-county model of comparative advantage developed by Deltas et al. (2012), predicts that a hub-and-spoke free trade network increases trade flows beyond the pure trade liberalisation effect of free trade agreements – the hub-and-spoke effect, and affects the sectorial composition of trade as the hub, i.e. South Africa, can engage in indirect arbitrage to exploit differences in comparative advantage. The results indicate that the Free Trade Agreements increased trade between 20% and 30%. There is no evidence for an aggregated hub-and-spoke effect, but the overlapping Free Trade Agreements seem to have affected the sectorial composition of trade, even if not in the patterns predicted by the model.

Keywords: Hub-and-spoke free trade agreements; overlapping free trade agreements; indirect

4

Table of Contents

Abbreviations ... 5

1.

Introduction ... 6

2.

Theory ... 7

2.1.

FTA Effects ... 7

2.2.

HAS Effects ... 8

2.3.

North-South and South-South FTAs ... 9

2.4.

The Model ... 9

3.

Overview over the countries’ trade policies ... 12

3.1.

South Africa ... 12

3.2.

Southern African Customs Union ... 12

3.3.

Southern African Development Community ... 13

3.4.

European Union ... 17

3.5.

The Free Trade Agreements ... 18

4.

Empirical Analysis ... 19

4.1.

Methodology ... 19

4.2.

Data ... 26

4.3.

Estimation Results ... 29

5.

Conclusion ... 37

5.1.

Limitations and Critique ... 38

5.2.

Further Fields for Study ... 39

6.

Bibliography ... 41

7.

Appendix ... 46

7.1.

Sectors by CA ... 46

7.2.

Estimation Results ... 49

5

Abbreviations

CA Comparative Advantage

CET Common external tariff

COMESA Common Market for Eastern and Southern Africa

CU Customs Union

DRC Democratic Republic of the Congo

EAC Eastern African Community

EBA Everything but Arms

EEC European Economic Community

EFTA European Free Trade Association

EM Emerging Market Economy

EPA Economic Partnership Agreement

EU European Union

FDI Foreign Direct Investment

FE Fixed effects estimator

FTA Free trade agreement

GATT General Agreement on Tariffs and Trade

GSP Generalised System of Preferences

HAS Hub-and-spoke free trade network

HS Harmonized Commodity Description and Coding System

IMF International Monetary Fund

IV Instrumental variable estimator

LDC Least Developed Country

MFN Most favoured nation

OEC The Observatory for Economic Complexity

OLS Ordinary Least Squares

PPML Poisson Pseudo-Maximum Likelihood

PTA Preferential Trade Agreement

RE Random effects estimator

ROO Rules of Origin

ROW Rest of world

RTA Regional Trading Agreement

SACU Southern African Customs Union

SADC Southern African Development Community

SSA Sub-Saharan Africa

TDCA Trade, Development and Co-operation Agreement

TFP Total Factor Productivity

UAE United Arab Emirates

UK United Kingdom

UNCTAD United Nations Conference on Trade and Development

USA United States of America

WTO World Trade Organisation

ZA South Africa

6

1.

Introduction

The number of Free Trade Agreements (FTAs) has increased significantly since the early 1990s. This has raised new issues: e.g. hub-and-spoke free trade networks (HAS), where the hub has entered FTAs with all its spokes, while the spokes did not form FTAs between themselves. This allows the hub preferential access to all spokes’ markets while each spoke only benefits from preferential access to the hub’s market. The emergence of HAS has opened many research questions regarding trade flows, indirect arbitrage, welfare effects and knowledge spill-overs.

First, some definition should be clarified: Following WTO definitions, FTAs are one type of Regional Trading Agreements (RTA). FTAs are bilateral or multilateral, reciprocal agreements to abolish tariffs while maintaining full control over commercial policies towards the rest of the world (ROW). They are therefore not as far-reaching as customs unions (CU), the other type of RTA, which additionally also establish a common external tariff (Gandolfo, 1998). RTAs differ from Preferential Trade Agreements (PTA) in their reciprocity. PTAs, under WTO definitions, are unilateral trade preferences, such as the Generalised System of Preferences (GSP). These terms are sometimes used interchangeably in the literature, but for this paper, I will stick to these definitions. In addition to the RTAs defined by the WTO, the EU started implementing Economic Partnership Agreements (EPA). These are trade and development agreements aimed at African, Caribbean and Pacific countries. In addition to trade liberalisation, these agreements also include support mechanisms for sustainable development and poverty reduction.

Generally, the effect of FTAs on trade flows is divided into trade creation and trade diversion. Trade creation refers to new trade following the reduction in tariffs, leading to a shift in production towards the most efficient producer implying a more efficient allocation of resources and is considered welfare enhancing. Trade diversion occurs if the elimination of tariffs induces a country to import from a FTA-partner instead of from a more efficient country outside the FTA. In this case, the most efficient producer is no longer competitive because of his tariff-disadvantage. This leads to a less efficient allocation of resources (Gandolfo, 1998). In the case of a FTA that also experiences trade diversion, the net welfare change is ambiguous (Krueger, 1999).

FTAs incorporate rules of origin (ROO). In a HAS setting, ROO prevent the hub from engaging in direct arbitrage: importing goods tariff-free from one spoke and re-exporting them tariff free to the other. But ROO do not prohibit indirect arbitrage: the hub can increase tariff free imports from one spoke to satisfy a larger share of domestic demand and export the freed-up domestic capacities to the other spoke. The possibility of indirect arbitrage is the reason why HAS can increase trade beyond the FTA liberalisation effect, the HAS effect (Deltas et al. 2012).

Empirically, this paper will focus on South Africa (ZA) as the hub, the EU and the Southern African Development Community (SADC) as the spokes.

The EU and ZA formed an FTA covering goods which came into force on 01/01/2000. ZA and the SADC entered an FTA covering goods on 01/09/2000.

The research question to answer is:

How did the formation of the ZA-EU and ZA-SADC FTAs affect trade flows and did it give rise to a HAS effect and indirect arbitrage at the sector level?

I will approach the question broadly following the methodology proposed by Deltas et al. (2012), by estimating a gravity equation to model trade flows. The estimation will be based on import-export data spanning 1994 to 2014. A welfare analysis is beyond the scope of this paper but identifying the effect of HAS on trade flows is an import first step towards a more general analysis.

7

The countries included in the analysis were chosen based on three reasons: Firstly, as a practical motivation, ZA and the SADC are part of few other FTAs which makes the identification of any effect easier. Additionally, this paper presents two interesting nuances: Firstly, the development configuration – an emerging country is the hub linking advanced to developing countries. Secondly, the two spokes are in themselves transnational communities. Therefore, this paper presents an extension to the vast literature on the trade effects of FTAs and to the emerging literature on HAS effects. As the number of FTAs keeps increasing, a better understanding of the particularities of HAS networks is likely to become increasingly important, as more and more countries might find themselves in HAS, intentionally or unintentionally. HAS concerns might therefore also gain importance in policy discussions.

This paper will proceed as follows: Chapter 2 summarises the literature and introduces the model. Chapter 3 discusses the countries’ trade policies. Chapter 4 comprise the empirical analysis including the methodology, the data and the estimation results. Chapter 5 concludes.

2. Theory

This chapter presents an overview of the literature on FTAs and on HAS effects. It also entails a summary of the theoretical model developed by Deltas et al. (2012). Together with the overview of the countries’ trade policies in chapter 3, this motivates the specific research question answered in the later chapters of this thesis.

2.1. FTA Effects

The overall effect of FTAs on trade flows has received much attention in the literature. Traditionally, the effect is estimated by a gravity equation:

𝑙𝑛(𝑇𝑖𝑗𝑡) = 𝛽0+ 𝛽1ln(𝑌𝑖𝑡) + 𝛽2ln(𝑌𝑗𝑡) + 𝛽3𝐹𝑇𝐴𝑖𝑗𝑡+ 𝜷𝑿 + 𝜀𝑖𝑗𝑡

Where Tij is the country pair’s trade, Yi(j) indicates the GDP of country i(j) and FTAij is a dummy variable

which takes the value of one if the two countries have formed an FTA. Additionally, X, is a vector of

other control variables, e.g. dummy variables for a common language or a shared border; and εij is the

error term. The coefficient on FTAij, β3, is typically the main coefficient of interest, the effect size. It

measures the effect of the FTA on trade flows. Different studies find greatly differing estimates for the effect size (even for the same FTA) but also use widely varying methods (Cipollina and Salvatici, 2010). Therefore, I will focus on two meta-studies in this theory overview:

Cipollina and Salvatici (2010) gather 1827 point estimates for RTA effect sizes in their sample. The majority is positive. They test the null hypothesis that the point estimates are jointly equal to zero. The null is robustly and strongly rejected implying a positive effect size. Overall, the authors’ most reliable estimate of the size effect is a trade increase of 40% due to FTA liberalisation. Comparing the effect of RTAs over time, the more recent agreements have a larger effect which is also consistent with the evolution from “shallow” to “deep” RTAs that reduce trade costs through additional domestic reforms. The World Bank (2005) finds that, generally, FTAs create more trade than they divert: of the 19 studies included, nine found net trade creation, ten net trade diversion. Trade diversion is less likely if countries joining an FTA also lower their tariffs towards the ROW. Generally, FTAs have a higher chance of success if market fundamentals (e.g. strong competition, access to cheap inputs) are well-developed. For many developing countries, this implies the need for behind-the-border reforms accompanying or preceding trade liberalisation. Whether an FTA generates economic gains may also depend on the restrictiveness of the ROO: the more complicated the rules, the higher the compliance costs, the higher the needed net trade gain to offset these costs. Restrictive ROO for final goods also

8

divert trade because they incentivise countries to import intermediary goods from FTA partners (World Bank, 2005).

2.2. HAS Effects

Most empirical studies use two different approaches: general equilibrium models and gravity equations.

2.2.1. Gravity Equations

Looking at the Israel-EU-USA HAS, with Israel being the hub, Deltas et al. (2012) find evidence for indirect arbitrage at the sectoral level using import-export data. If FTAs are formed sequentially, trade between the future hub and its first FTA-partner increases (usually) after the hub signs the second FTA. De Benedictis et al. (2005) use a panel of export data with the EU15 as the hub and Central- and Eastern-European nations as the periphery. They find that FTAs between spokes prevent HAS relations from occurring and increase intra-periphery trade. Hur et al. (2010) conclude that HAS have a positive effect on trade beyond the direct trade liberalising effect of FTAs and that the hub can export more, incentivising other countries to seek hub-status by entering more FTAs. However, using a panel of 175 countries, Lee et al. (2008) show that trade creation approaches zero as more and more countries join overlapping FTAs. Therefore, more HAS do not necessarily increase global trade.

2.2.2. General Equilibrium Analysis

Generally, hub-status implies a disproportional welfare increase relative to the spokes after trade liberalisation (Chong and Hur, 2007; Yildiz, 2012 and Das and Andriamananjara, 2006). Chong and Hur (2007) also find, that, for small open economies, being a hub is preferable to being in a free trade zone with both spokes while the spokes lose in welfare terms (even though these losses might be small relative to their GDP). Yildiz (2012) extends these findings by noting that, in the presence of oligopoly and asymmetric costs, being a hub is even preferable to global free trade and a HAS could even deliver higher global welfare than multilateral free trade, if the hub is efficient enough compared to the spokes. Intuitively, this result is based on the preferential treatment of the hub’s exports in all its foreign markets. This reallocates a higher share of the world’s output to the low-cost producer. Das and Andriamananjara (2006) add that HAS also affect direct and indirect technological spillovers.

2.2.3. Additional HAS Studies

HAS may also bias FDI flows and industry location in favour of the hub, allowing the hub, all else equal, to set higher corporate taxes (Darby et al. 2014). HAS integration beyond a critical level could even increase industry-relocation to the hub suddenly and widen wage divergences between the hub and the spokes and between the individual spokes (Puga and Venables, 1997). Cao (2015) examines the effect of HAS on FTA creation and finds that for any country-pair the probability of forming a FTA is larger if HAS is not a concern or if the countries try to secure hub-status. HAS concerns and hub-status-seeking have higher partial effects on the probability of FTA formation than other determinants (e.g. distance or common language).

Baldwin (2009) discusses the political economy of HAS concerning East Asia. He notes that HAS hurt the entire region by making it less attractive to FDI inflows. Also, HAS might be divisive politically as it favours the hub, economically and politically, at the spokes’ expense. On the other hand, HAS may be a better path toward multilateral free trade than CU because the creation of new FTAs by the (future) hub may be an option even when the enlargement of the CU is politically infeasible (Mukunoki and Tachi, 2006).

9

2.3. North-South and South-South FTAs

One particularity of this analysis is the development configuration. Here, one spoke, the EU is an industrialised ‘northern’ country, while the other spoke, the SADC is a developing ‘southern’ country. ZA, an emerging economy, is somewhere in between.

Large northern countries receive few direct benefits from forming FTAs with smaller southern countries. Developing countries wanting to form an FTA with a developed country should first reduce their MFN tariff levels to prevent negative terms-of-trade shocks. Generally, ROO are more restrictive (implying more trade diversion) in North-South than in South-South FTAs. South-South FTAs are normally less successful than others, because of the small size of the markets and little difference in comparative advantage. On average North-South agreements have a shorter negotiation and implementation period than South-South FTAs making them less susceptible to lobbying (World Bank 2005).

Busse argues that ZA benefits from the FTA with the EU regarding its speed of integration into the world economy and that the FTA may help to “lock in economic reforms, which is particularly important for transition and emerging market economies” (Busse, 2000, p.153). Additional to the political benefits, HAS leads to real income convergence between the hub and the higher-income spoke, while it leads to divergence between the hub and the lower-income spoke. The scope for trade creation within a HAS is maximised when the spokes have very different income levels (Umemoto, 2003).

2.4. The Model

Deltas et al. (2012) develop a three-country, three-good specific factor model to make theoretical predictions about the effects of HAS. The source of international trade in this model are differences in factor endowments implying a specific pattern of comparative advantage (CA).

2.4.1. Set-up

The model assumes a perfectly competitive economy with three countries producing three goods. Each country is populated by a homogeneous number of agents with the same preferences, each supplying one unit of labour. Labour can move freely across sectors (implying wage equalisation across industries). Each country is also endowed with three sector specific inputs. Each sector specific input is abundant in a different country. The three goods are produced under a Cobb-Douglas production function using labour and the sector specific input. Neither labour nor the sector specific inputs can move across borders and technology is the same in each country. Tariffs take on iceberg form, i.e. a tariff of τ implies that 1 + τ units of the good must be shipped for one unit to reach the destination. Each country has a CA in the good whose sector specific input is abundant in the country.

Assuming three countries, South (S), Middle (M) and North (N), populated by the mass of agents, γS_,

γM _{and y}N _{respectively, and each endowed with three sector specific inputs, K}

1, K2 and K3, so that: {𝐾₁𝑁_{, 𝐾} 2𝑁, 𝐾3𝑁} = 𝛾𝑁(1 + 𝛼, 1,1) {𝐾1𝑆, 𝐾2𝑆, 𝐾3𝑆} = 𝛾𝑆(1, 1 + 𝛼, 1) {𝐾₁𝑀_{, 𝐾} 2𝑀, 𝐾3𝑀} = 𝛾𝑀(1, 1,1 + 𝛼)

Where α captures the strength of the CA: e.g. country N has, relative to country S and M, a CA of α in

good 1 (which uses the sector specific input K1). A value of 1 for α would imply that country N’s relative

10

Each country exports the good which uses its abundant sector specific input, ergo:

N good 1 good 2 and good 3.

Country S exports good 2 and imports good 1 and good 3.

M good 3 good 1 and good 2.

From the perspective of country M, countries N and S are competing importers, while their exports are complements. The equivalent holds for the other countries.

2.4.2. Symmetric countries

Under the assumption of symmetric countries (γS_=γM_=yN_{) trade must be balanced between all}

country-pairs, i.e. the total value of imports must equal the total value of exports for bilateral trade. All existing tariffs are assumed to be non-prohibitive.

Then, equilibrium conditions under a bilateral FTA (between N and M), a multilateral FTA (including S, N, and M) and a HAS (with M being the hub), can be derived algebraically. The equilibrium conditions imply a vector of total consumption, production, labour demand, wages, prices and tariffs so that markets clear and trade is balanced. From the equilibrium conditions under the different trade scenarios, the following proposition follows:

The first result, trade reduction between the original FTA partners in response to a move to multilateral free trade, builds on the fact that trade flows do not depend on absolute but on relative trade barriers. When the three countries move from a bilateral FTA towards multilateral free trade, the trade barriers between N and M increase relative to the trade barriers between the two countries and S. Trade between the original FTA partners decreases. The second result stems from the fact that, under asymmetric trade liberalisation, the hub (M in this scenario) can engage in indirect arbitrage if the two spokes have a CA in different goods. Trade between the original FTA partners increases and excess trade is created. Consequently, trade liberalisation to a third country has an ambiguous effect on bilateral trade flows, an ambiguity that goes beyond the effects of relative trade barriers: when relative trade barriers increase between two countries (countries N and M here), trade between them may either increase or decrease, depending on the trade configuration relative to the third country. HAS effects only occur if the exports of the two spokes are complements, i.e. if the two spokes have CAs in different goods. If their exports were substitutes, i.e. the same good, excess trade would disappear. Even though the HAS effect in this model seems to depend on the perfect substitutability between goods of the same variety from different countries, this need not be the case: in a more complex Krugman-style model, with CA at the sector level and monopolistic competition at the sub-sector level,

Proposition I. Assume three countries N, S, and M; non-prohibitive tariffs and three trade configurations: a) a bilateral FTA between N and M; b) a multilateral FTA involving N, M, and S and c) a HAS where M is the hub and ROO are binding. If γS _{= γ}M _{= y}N _{and τ > 0, then}

1.

A move from the bilateral FTA between N and M towards a multilateral FTA will

decrease trade between N and M as a share of GDP in M – i.e. symmetric trade

liberalisation reduces trade between the original partners

2.

A move from a bilateral FTA between N and M towards a HAS with M being the

hub will increase trade between N and M as a share of GDP in M – i.e. asymmetric

trade liberalisation increases trade between the original partners

11

the HAS effect would still emerge. Then, the existence of HAS effects depends on the varieties within the same sector being closer substitutes than the varieties across different sectors (Deltas et al., 2012).

2.4.3. Asymmetric countries

The presence of size asymmetries complicates the analysis because country-pair trade no longer needs to be balanced. The authors apply numerical analysis under a broad parameter space to retain as much generality as possible. The important parameters are the relative size differences (i.e. the relative difference between γS_{, γ}M _{and y}N_{) and the strength of the CA (i.e. α).}

In general, the numerical analysis supports the results made in proposition I. Point I.1. is confirmed: a move from a bilateral FTA to multilateral free trade decreases trade between the original FTA partners. Point I.2. is confirmed for most parameter combinations: a move from a bilateral FTA to a HAS increases trade between the original FTA partners, unless both the CA is very strong and the relative size of the spokes is very large.

The HAS effect disappears for high values of CA and large spokes because two effects occur simultaneously when the bilateral FTA is replaced with the HAS: from the perspective of the hub, firstly, the relative trade barrier to the original FTA partner increases, decreasing the hub’s incentive to trade. In contrast, the opportunity to engage in indirect arbitrage raises the hub’s incentive to trade with its original FTA partner. If the hub is relatively small compared to the spokes, its export good is relatively scarce. This implies a high relative price of the good and, consequentially, a low relative price for the spokes’ export goods. Therefore, the value of the excess trade created by indirect arbitrage is low. Then, the first, negative effect is more likely to dominate the weaker, positive effect. Accordingly, the HAS effect may be overturned. Since proposition I.1. always holds, it is also true that, even for values of relative size and CA that preclude the emergence of HAS effects, the trade decrease between the original FTA partners will be less, when the trade configuration changes to a HAS, than when it changes to multilateral free trade (Deltas et al., 2012).

2.4.4. The ZA-EU-SADC HAS

It should be noted that, in this case, HAS effects may not materialise because one spoke, the EU, is relatively large compared to ZA, with some countries such as Germany, France or Italy dwarfing ZA in economic size. The other spoke, the SADC, is relatively small compared to ZA, with all individual countries being considerably smaller than ZA in (economic) size. The magnitude of the relative size differences can be seen in figure 4.1. Whether these size differences matter also depend on the differences in CA and no conclusion can be drawn before the empirical analysis.

12

3.

Overview over the countries’ trade policies

This chapter aims to give a concise overview of the countries’ economic situations and trade policies. While the differences between the advanced economies of the EU and the developing and emerging countries of the SADC might seem obvious, the individual members of the SADC still exhibit considerable heterogeneity: from Zimbabwe, a low income country with a strained socio-political climate and a recent history of hyperinflation, to the Seychelles, on the brink of becoming a high income country, the different SADC members differ in their development stage, size, economic specialisation, trade patterns, institutional framework and general openness to trade. Even though general openness to trade depends on multiple factors, I will note each country’s simple average applied MFN tariff rate as an indicator. This rate is imposed on all trade between WTO members in the absence of RTAs. While a full discussion of all differences would exceed the scope of this thesis, a short impression should be given.

3.1. South Africa

From 1986 to 1991, ZA was subject to sanctions on trade and finance through which the country’s most important trade partners (USA, EU, Japan) tried to put pressure on the Apartheid regime. The sanctions were lifted in 1991 after the government implemented reforms which cumulated in the 1994 constitution putting an official end to the Apartheid era. The effectiveness and long-term consequences of the sanctions remain a topic of discussion (cp. Keller, 1993 and Hefti and Staehelin-Witt, 2013).

ZA is classified as an upper-middle income country and is the second largest economy in Sub-Saharan Africa (SSA) after Nigeria. ZA is considered an emerging market (EM) and has been a GATT member since 1948. ZA has seen an increase in real GDP per capita of 35% since 1994, even though the median EM’s real GDP per capita has increased by double that rate. Living standards rose significantly after democratisation. Nonetheless, ZA has amongst the highest current account deficit, unemployment and inequality rates in the emerging world. The corporate sector is globally competitive and the financial system is sophisticated leading to a globally integrated economy. ZA is a regional trade and financial hub (IMF, 2016a and WTO, 2015a).

As is part of the Southern African Customs Union (SACU), imports to ZA are subject to the common external tariff (CET).

Imports and exports are rather diversified. Manufacturing and mining represent the main share of merchandise exports. Agricultural exports have been steadily increasing. More than 50% of imports are manufactured products. Imports remain important as they supply ZA’s manufacturing sector with raw materials and high-quality equipment (WTO, 2015a).

The destination and source countries for ZA’s trade are quite diversified. Southern African exports mainly go to other African countries (most notably other SACU members), the EU, USA and China. Most imports are sourced from the EU, USA and China (WTO, 2015a). Intra-African trade has increased in importance, with the import share coming from other SSA countries more than doubling over the last decade. SSA now absorbs around 30% of ZA’s exports (IMF, 2016a).

3.2. Southern African Customs Union

The SACU is made up of Botswana, Lesotho, Namibia, ZA and Swaziland. It is the oldest CU worldwide, first established in 1910. The SACU administers a CET whose revenue is combined into a common revenue pool and shared amongst the members following a revenue sharing formula (Elago, 2016).

13

SACU countries rely heavily on ZA for their common customs receipt, as ZA accounts for 85% of all imports (IMF, 2016). For all countries except ZA, custom revenue constitutes a significant share of the national budget (Elago, 2016).

Even excluding ZA, the SACU countries display considerable heterogeneity. The main country characteristics for each SACU country are summarised in table 3.1.

In general, the SACU countries still face challenges in the realms of poverty, unemployment, inequality and HIV/ AIDS prevalence: e.g. the HIV/AIDS prevalence in Swaziland of 31% is amongst the highest in the world and life expectancy has fallen to 49 years. Lesotho and Namibia still struggle with food insecurities. Additionally, Botswana’s significant reliance on commodity exports makes its economy vulnerable to international price fluctuations. On the positive side, Botswana is ranked as the second most competitive country in SSA (behind ZA) and the SSA country with the lowest levels of corruption and Namibia hosts one of the most developed financial systems in SSA (WTO, 2015a).

In 2015, the CET of the SACU implied a simple average applied MFN rate of 7.6% (WTO et al., 2016). The SACU signed a FTA with the European Free Trade Association (EFTA), i.e. Iceland, Norway, Switzerland and Lichtenstein, in 2008 which ended in 2015 (WTO, 2017).

Table 3.1.: SACU country characteristics (WTO, 2015a)

Classification Main exports Main imports Main export markets Main import markets Accession to GATT or WTO Botswana Upper-middle income Diamonds and meat; shift to apparel (preferential treatment in the US) Europe (especially UK) and SACU; Asia and America growing in importance South Africa 1987 Lesotho Low-income; Least developed country (LDC) Textiles and clothing SACU (especially ZA), USA, EU ZA, Asia, Europe 1988 Namibia Upper-middle income Minerals (mostly diamonds) and manufacturing ZA and EU ZA (but a sizeable share of re-exports) 1992 Swaziland Lower-middle income Sugar and sugar-based products; manufacturing

Foodstuff ZA and EU SACU 1993

3.3. Southern African Development Community

In addition to the five countries that make up the SACU, the SADC also consists of Angola, the Democratic Republic of the Congo (DCR), Madagascar, Malawi, Mauritius, Mozambique, the Seychelles, Tanzania, Zambia and Zimbabwe. Their main characteristics are summarised in table 3.2.

14

Generally, the SADC unites countries from the entire spectrum of developing countries. On the one hand are the Seychelles and Mauritius – upper-middle income countries with high human development and social indicators and relatively high living standards. The Seychelles, despite a balance of payments crisis in 2008-2009, now have a GDP per capita putting it close to high income countries (IMF, 2015). Mauritius has attracted considerable FDI inflows (spurred by increased competitiveness) and has seen one of the strongest growth performances in Africa (WTO, 2014). On the other hand, there are countries such as Madagascar, which experienced four socio-political crises in 20 years and where the share of people living on less than US$2 per day has increased to over 90% and malnutrition remains a serious problem (WTO, 2015c). Malawi and Zimbabwe also belong to the poorest countries on Earth, both being heavily dependent on fluctuating international donor support. Malawi’s situation is aggravated by a high and increasing population growth rate, high cost of doing business and the mainly informal labour market (WTO, 2016b). The challenges facing Zimbabwe are almost too many to summarise. Amongst the main ones are: high unemployment; exhausted international reserves and unsustainable external debt of which a significant share is in arrears; a strained socio-political climate; insecure property rights; waning international support and widespread state intervention and ownership, causing the supply of key goods and services to be costly and inefficient. Until the establishment of a multi-currency system in 2009, making the US dollar the de facto legal tender to re-establish some macroeconomic stability, Zimbabwe experienced a period of hyperinflation (WTO, 2011). Somewhere in between these two extremes fall countries such as Tanzania and Zambia, which both achieved solid economic growth over the last decade. Despite being a LDC with a poverty rate exceeding 30%, macroeconomic stability and structural reforms have transformed Tanzania into a more market-orientated and outward-looking economy (WTO, 2012). Zambia has even been re-classified as a lower-middle income country in 2014, mainly due to considerable improvements in life expectancy and education (WTO, 2016c). The Zambian, Mozambican and Angolan economies’ dependencies on commodity exports; copper, aluminium and oil respectively, leave these countries vulnerable to international price fluctuations (WTO, 2016c; IMF, 2016b; WTO, 2008).

It should be noted, that the descriptions of the countries are only short summaries and that their economic situations are a lot more complicated and nuanced than this analysis allows to depict. Angola, for example, had a poverty rate of 37% in 2009 and is amongst the ten countries with the highest child mortality rates. Yet, it also is the second largest oil producer in SSA (after Nigeria) and its banking sector is ranked third in SSA (WTO, 2015b). Even Zimbabwe, with all its challenges, still has, compared to other African countries, a relatively well-diversified and large manufacturing sector (WTO, 2011).

It should be noted, that not all members of the SADC signed the corresponding FTA: Angola, the DCR and Madagascar chose not to sign or to ratify the agreement. The Seychelles only joined the FTA in 2015, i.e. after the time period analysed.

On 10/06/2016, the EU signed an EPA with the SADC EPA group: Botswana, Lesotho, Mozambique, Namibia, ZA and Swaziland. Angola can choose to join the agreement in the future (European Commission, 2016a). Malawi, Mauritius, the Seychelles and Zimbabwe signed a preliminary EPA with the EU, which came into force in 2012: the EU – Eastern and Southern Africa States Interim EPA (WTO, 2017). Additionally, some SADC countries are also members of the Common Market for Eastern and Southern Africa (COMESA) (summarised in table 3.3.). Mozambique, Tanzania and Zimbabwe are also members of the Global System of Trade Preferences among Developing Countries (WTO, 2017). In addition, Mauritius also formed a FTA with Turkey and a PTA with Pakistan (WTO, 2017).

15

Table 3.2.: SADC country characteristics

Classification Main exports Main imports Main export markets Main import markets MFN rate 20151 GATT or WTO2 Angola (WTO, 2015b) Upper-middle income; LDC

Oil (products) and diamonds

China, EU, USA, and India

Portugal, China, Rep. of Korea and Brazil

11.4% 1994

DCR

(WTO, 2016a)

LDC Mining (cobalt, copper,

diamonds, gold and petroleum)

Foodstuff, chemical products, transport

equipment and machinery

China, Zambia, the EU and the Middle East

EU, ZA, Zambia and China 10.9% 1997 Madagascar (WTO, 2015c) LDC Clothing, mining,

shrimps and crabs, agriculture and services

Decreasingly EU and US; increasingly Middle

East and Asia, mainly China; ZA and

Mauritius

EU, United Arab Emirates (UAE) and

China

11.7% 1963

Malawi

(WTO, 2016b)

LDC Tobacco, tea, sugar and

uranium

Manufacturing Africa and EU ZA, Mozambique, India,

the EU and China

12.7% 1964 Mauritius (WTO, 2014) Upper-middle income Services, clothing, textiles and sugar

Oil and foodstuff EU EU, China and India 1.0% 1970

Mozambique

(IMF, 2016b; WTO, 2008)

LDC Aluminium, electricity

and agricultural goods (cashews, cotton, refined sugar, tobacco

and fishery products)

Alumina (which is refined and re-exported), oil, food and

chemicals

EU and ZA Australia and ZA 10.1% 1992

1 _{MFN rates taken from WTO et al. (2016).} 2 _{Year the country joined the GATT or WTO}

16

Classification Main exports Main imports Main export markets Main import markets MFN rate 2015 GATT or WTO Seychelles (IMF, 2015; OEC; 2017) Upper-middle income

Fish products, refined petroleum, cement, and animal meal and

pellets

Refined petroleum, fish products, ships, boats

and paper labels

UAE, the EU (particularly France and

the UK), Japan and Mauritius

UAE, Spain, France, ZA and Mauritius 2.9% 2015 Tanzania (WTO, 2012) Upper-middle income; LDC agriculture (traditionally coffee,

tobacco, tea) and mining

machinery, transport equipment, chemicals,

oil and food

East African Community (EAC)

common market (mainly Kenya) and other African nations

(mostly ZA), Switzerland, China and

Japan India and EU 12.9% (EAC CET) 1961 Zambia (WTO, 2016c) Lower-middle income

Copper machinery, transport

equipment, oil and automotive parts

Europe, particularly Switzerland

Other African nations (especially ZA and DRC), EU, Middle East

and China

13.6% 1982

Zimbabwe

(WTO, 2011)

17

In addition to the EPAs, most SADC countries already benefit from the EU GSP or the Everything But Arms (EBA) scheme. The GSP program started in 1971, EBA came into being in 2001. Under these schemes, most of the countries’ exports receive preferential, duty free and quota-free access to the EU market. The SADC countries that benefit from the EU GSP are: Botswana, Namibia and Swaziland. From EBA are benefiting: Angola, DCR, Lesotho, Madagascar, Malawi, Mozambique, Tanzania and Zambia (European Commission, 2014 and 2016b). The composition of the main RTAs is summarised in table 3.3.

Table 3.3: Summary SADC

Country Member SACU Member SADC

FTA Member COMESA GSP/ EBA beneficiary Angola (✓) ✓ Botswana ✓ ✓ ✓ DRC ✓ ✓ Lesotho ✓ ✓ ✓ Madagascar ✓ ✓ Malawi ✓ ✓ ✓ Mauritius ✓ ✓ Mozambique ✓ ✓ Namibia ✓ ✓ ✓ Seychelles ✓ ✓ South Africa ✓ ✓ Swaziland ✓ ✓ ✓ ✓ Tanzania ✓ ✓ Zambia ✓ ✓ ✓ Zimbabwe ✓ ✓

3.4. European Union

The EU is a political and economic union founded in 1958 and made up of 28 countries. The member countries are summarised in table 3.2. The EU goes beyond a CU: it is a single market where goods, services, capital and people can move freely. As of 2016, 19 out of the 28 countries also use a common currency, the Euro. Except from the UK, who voted to leave the EU in 2016, and Denmark, all other EU countries are expected to join the Eurozone eventually (European Union, 2017). All members have varying degrees of competence in different sectors affecting trade flows. The EU remains one of the largest economies and trading entities worldwide and is highly integrated into the world economy. The largest share of EU merchandise imports and exports is comprised of manufactured goods. The main EU export sectors are transport equipment, machinery and chemicals. In addition to manufactures, the EU mainly imports fuel. The largest EU export markets are the USA, followed by Switzerland and Asia, particularly China and Japan. Most EU imports arrive from Asia, especially China, Russia, USA, Switzerland and Norway (WTO, 2015d). The simple average applied MFN rate was 4.3% in 2015 (WTO et al., 2016).

In addition to the FTAs with ZA, the SADC EPA and Eastern and Southern Africa States Interim EPA mentioned above, the EU partakes in 35 other RTAs, illustrating its integration into the world economy. The EU is also in the process of negotiating, ratifying and implementing more RTAs (WTO, 2017).

18

Table 3.4: EU member countries (European Union, 2017)

Country Accession Date

Belgium 01/01/1958 France Germany Italy Luxembourg Netherlands Denmark 01/01/1973 Ireland United Kingdom Greece 01/01/1981 Portugal 01/01/1986 Spain Austria 01/01/1995 Finland Sweden Cyprus 01/05/2004 Czech Republic Estonia Hungary Latvia Lithuania Malta Poland Slovakia Slovenia Bulgaria 01/01/2007 Romania Croatia 01/07/2013

3.5. The Free Trade Agreements

3.5.1. The Trade, Development and Co-operation Agreement

The Trade, Development and Co-operation Agreement (TDCA) is the FTA governing the trade relations between the EU and ZA. It was signed in 1999 and came into force in the beginning of 2000. The implementation period ended in 2012. The TDCA determines the elimination of tariffs on most industrial and agricultural products. The speed with which tariffs are phased out depends on the country and sector, where the maximum liberalisation period is set to 10 years for the EU and to 12 years for ZA.

The TDCA also includes provision to further trade in services, FDI, economic and development co-operation. It also contains an outline for financial and technical assistance in the form of grants and loans extended to ZA. Furthermore, the TDCA also describes the ROO (Agreement on Trade, Development and Cooperation between the European Community and its Member States, of the one part, and the Republic of South Africa, of the other part, 1999).

The TDCA covers 90% of the bilateral trade. Since the beginning of the agreement, trade in goods between the EU and ZA increased by over 120%. FDI has increased by a factor of five. After its

19

ratification, the TDCA will be replaced by the EU – SADC EPA signed in 2016 (European Commission, 2017).

3.5.2. Protocol on Trade in the Southern African Development Community

The Protocol on Trade was signed in 1996 and came into force on 01/09/2000. It was amended in 2000, 2007 and 2008. Its implementation was complete in 2015. The Protocol aims at liberalising intra-regional trade in goods and services. It sets out the elimination of tariff and non-tariff barriers to trade within eight years, allowing for longer implementation periods for critical sectors. The agreement also limits export duties and quantitative import restrictions.

Additionally, the Protocol intends to harmonise and to co-ordinate customs procedures and trade policies based on international standards. The agreement also includes a commitment to co-operation in the realms of industrial development, finance and investment. As usual with a FTA, ROO are included (Protocol on Trade in the Southern African Development Community (SADC) Region 1999).

Figure 3.1. illustrates ZA’s international trading environment. Nested circles represent sub-units that are part of the larger unit. Arrows symbolise RTAs. Even though ZA is embedded in the SACU, the HAS concept still applies, since the SACU is only a CU and not a single market (as opposed to the EU which is a CU and a single market). Therefore, ROO still apply and only ZA is a contracting partner of the TDCA.

Figure 3.1.: ZA’s trading environment

4.

Empirical Analysis

This chapter describes the empirical specification, estimation techniques, the data and the empirical results.

4.1. Methodology

The estimation method generally follows Deltas et al. (2012) by setting up a gravity equation, adopting a differences-in-differences treatment effect approach.

The time period for this analysis is 1994 to 2014. Starting in 1994 allows to ignore any possible effect the sanctions on ZA might have had on trade (particularly with the EU) and allows to only include trade after the completion of the Uruguay Round of WTO trade negotiations. Since the end of the Uruguay round led to a significant reduction in multilateral trade barriers, starting afterwards removes the necessity to control for the change in the world trading climate. Only country-pairs involving ZA are included in the analysis to avoid possible confounding effects. This implies that the year fixed effects now control for the secular change in ZA’s trade and not the average secular change between any

20

country pairs. Additionally, the elasticity of GDP now describes the response of ZA’s trade to changes in GDP, instead of the global average. This also implies that country and country-pair fixed effects coincide. (Deltas et al., 2012).

4.1.1. Aggregate Analysis

The first step in the empirical estimation is to determine the overall effect of the FTAs on the aggregate level. Since the two spokes are heterogeneous transnational entities, I will extend Deltas et al.’s framework by adding, for EU and SADC countries, interaction terms between country(-pair) fixed effects and the FTA dummies to allow trade liberalisation to have had different effects on different country-pairs’ trade.

The log-linearized gravity equation to be estimated is therefore equation 4.1.

Equation 4.1. log(𝑇𝑟𝑎𝑑𝑒𝑗𝑡) = 𝛽0+ 𝛽𝐸𝑈𝐸𝑈𝑗𝑡+ 𝛽𝑆𝐴𝐷𝐶𝑆𝐴𝐷𝐶𝑗𝑡 + 𝛽𝐸𝐹𝑇𝐴𝐸𝐹𝑇𝐴𝑗𝑡+ 𝛼2ln(𝐺𝐷𝑃𝑍𝐴𝑡𝐺𝐷𝑃𝑗𝑡) + ∑ 𝜙𝑡 𝑡 𝑦𝑡+ ∑ 𝜓𝑗 𝑗 𝑐𝑗 + ∑ 𝛾𝑗 𝑗∈𝐸𝑈 𝑐𝑗𝐸𝑈𝑗𝑡+ ∑ 𝛾𝑗 𝑗∈𝑆𝐴𝐷𝐶 𝑐𝑗𝑆𝐴𝐷𝐶𝑗𝑡+ 𝜀𝑗𝑡

Where, Tradejt is the bilateral trade between ZA and partner country j in year t. EUjt is a dummy variable

that equals one if country j is part of the ZA-EU FTA in year t. SADCjt is a dummy variable equal to one

if country j is part of the ZA-SADC FTA in year t. EFTAjt is a dummy variable equal to one if country j is

part ZA-EFTA FTA in year t. GDPZA(j)t is the nominal GDP of ZA (country j) in year t. Additionally, cj are

country-pair fixed effects, yt indicate the year fixed effects and εjt is the random error term (Deltas et

al., 2012).

Because the implementation period of the FTAs stretched over eight to twelve years, the FTAs will be allowed to have had different effects over time. The estimation period of 20 years is divided into three periods: 1994-1999, 2000-2007, 2008-2014. The first period, 1994-1999, is the benchmark before the implementation of the FTAs. The second regression equation is therefore:

Equation 4.2. log(𝑇𝑟𝑎𝑑𝑒𝑗𝑡) = 𝛽0+ 𝛽𝐸𝑈1𝐼𝑡𝜖[2000]𝐸𝑈𝑗𝑡+ 𝛽𝐸𝑈2𝐼𝑡𝜖[2008]𝐸𝑈𝑗𝑡+ 𝛽𝑆𝐴𝐷𝐶1𝐼𝑡𝜖[2000]𝑆𝐴𝐷𝐶𝑗𝑡 + 𝛽𝑆𝐴𝐷𝐶2𝐼𝑡𝜖[2008]𝑆𝐴𝐷𝐶𝑗𝑡+ 𝛽𝐸𝐹𝑇𝐴𝐸𝐹𝑇𝐴𝑗𝑡+ 𝛼2ln(𝐺𝐷𝑃𝑍𝐴𝑡𝐺𝐷𝑃𝑗𝑡) + ∑ 𝜙𝑡 𝑡 𝑦𝑡 + ∑ 𝜓𝑗 𝑗 𝑐𝑗+ ∑ 𝛾𝑗 𝑗∈𝐸𝑈 𝑐𝑗𝐸𝑈𝑗𝑡+ ∑ 𝛾𝑗 𝑗∈𝑆𝐴𝐷𝐶 𝑐𝑗𝑆𝐴𝐷𝐶𝑗𝑡+ 𝜀𝑗𝑡

Where the Is are indicator variables that take the value of one if the observation falls within the indexed period, e.g. Itϵ[2000] = 1 if 2000 ≤ t ≤ 2007 and Itϵ[2008] = 1 if 2008 ≤ t ≤ 2014.

4.1.2. Aggregate Analysis – HAS effect

The second step in the estimation is to try to identify a possible HAS effect.

Deltas et al. (2012) rely on the time elapsed between the formation of the two FTAs in their example to estimate the HAS effect: the Israel-EU FTA was signed 10 years before the Israel-USA FTA. They compare the coefficient on the Israel-EU FTA dummy before and after the formation of the Israel-USA FTA and find, consistent with theory, that the coefficient on the Israel-EU FTA dummy increases after the formation of the Israel-USA FTA. Since ZA signed both FTAs in the same year, this identification

21

approach is not feasible. I will follow Hur et al.’s (2010) approach by relying on an additional dummy variable to identify a possible HAS effect. This leads to a gravity equation as specified by equation 4.3.

Equation 4.3. log(𝑇𝑟𝑎𝑑𝑒𝑗𝑡) = 𝛽0+ 𝛽𝐸𝑈𝐸𝑈𝑗𝑡+ 𝛽𝑆𝐴𝐷𝐶𝑆𝐴𝐷𝐶𝑗𝑡+ 𝛽𝐻𝐴𝑆𝐻𝐴𝑆𝑗𝑡 + 𝛽𝐸𝐹𝑇𝐴𝐸𝐹𝑇𝐴𝑗𝑡+ 𝛼2ln(𝐺𝐷𝑃𝑍𝐴𝑡𝐺𝐷𝑃𝑗𝑡) + ∑ 𝜙𝑡 𝑡 𝑦𝑡+ ∑ 𝜓𝑗 𝑗 𝑐𝑗 + ∑ 𝛾𝑗 𝑗∈𝐸𝑈 𝑐𝑗𝐸𝑈𝑗𝑡+ ∑ 𝛾𝑗 𝑗∈𝑆𝐴𝐷𝐶 𝑐𝑗𝑆𝐴𝐷𝐶𝑗𝑡+ 𝜀𝑗𝑡

The HASjt dummy takes the value of one if country j is part of the EU or the SADC FTA in year t. It

measures the HAS effect, while the coefficients on the EUjt and SADCjt dummies measure the effect of

each FTA in isolation.

As above, the estimation will be repeated allowing the FTAs to have had different effects over two time periods. Equation 4.4. log(𝑇𝑟𝑎𝑑𝑒𝑗𝑡) = 𝛽0+ 𝛽𝐸𝑈1𝐼𝑡𝜖[2000]𝐸𝑈𝑗𝑡+ 𝛽𝐸𝑈2𝐼𝑡𝜖[2008]𝐸𝑈𝑗𝑡+ 𝛽𝑆𝐴𝐷𝐶1𝐼𝑡𝜖[2000]𝑆𝐴𝐷𝐶𝑗𝑡 + 𝛽𝑆𝐴𝐷𝐶2𝐼𝑡𝜖[2008]𝑆𝐴𝐷𝐶𝑗𝑡+ 𝛽𝐻𝐴𝑆1𝐼𝑡𝜖[2000]𝐻𝐴𝑆𝑗𝑡+ 𝛽𝐻𝐴𝑆2𝐼𝑡𝜖[2008]𝐻𝐴𝑆𝑗𝑡 + 𝛽𝐸𝐹𝑇𝐴𝐸𝐹𝑇𝐴𝑗𝑡+ 𝛼2ln(𝐺𝐷𝑃𝑍𝐴𝑡𝐺𝐷𝑃𝑗𝑡) + ∑ 𝜙𝑡 𝑡 𝑦𝑡+ ∑ 𝜓𝑗 𝑗 𝑐𝑗 + ∑ 𝛾𝑗 𝑗∈𝐸𝑈 𝑐𝑗𝐸𝑈𝑗𝑡+ ∑ 𝛾𝑗 𝑗∈𝑆𝐴𝐷𝐶 𝑐𝑗𝑆𝐴𝐷𝐶𝑗𝑡+ 𝜀𝑗𝑡

Where all variables are defined as before.

4.1.3. Sectorial Analysis – Indirect Arbitrage

The third estimation step is trying to identify indirect arbitrage on the sector level. The methodology is again taken from Deltas et al. (2012).

Firstly, the sectors in which each spoke has a CA must be identified. Empirically, the sectors in which a spoke has a CA will be specified by looking at net sector-exports before the signing of the FTAs whereby trade flows will be normalised to allow for the possibility of unbalanced trade. Not taking account of trade imbalances would bias the identification of CAs in favour of the surplus country. Since, possibly, both spokes have a CA relative to ZA, the concept must be narrowed down further: in the context of indirect arbitrage, a spoke is only said to have a CA relative to ZA if the other spoke does not. This is compatible with the model introduced in section 2.4., since indirect arbitrage is only expected to occur if the two spokes’ exports are complements, not substitutes. Therefore, sectors are identified in which one spoke has a CA relative to ZA and to the other spoke.

Let XPjkt denote ZA’s exports to partner j in sector k in year t, and MPjkt ZA’s imports from partner j in

sector k in year t. The EU has a CA in sector k if ∑𝑡 ∈(1994−1999)𝑀𝑃𝐸𝑈𝑘𝑡 ∑ ∑𝑘 𝑡 ∈(1994−1999)𝑀𝑃𝐸𝑈𝑘𝑡 > ∑𝑡 ∈(1994−1999)𝑋𝑃𝐸𝑈𝑘𝑡 ∑ ∑𝑘 𝑡 ∈(1994−1999)𝑋𝑃𝐸𝑈𝑘𝑡 And, simultaneously, ∑𝑡 ∈(1994−1999)𝑀𝑃𝑆𝐴𝐷𝐶𝑘𝑡 ∑ ∑𝑘 𝑡 ∈(1994−1999)𝑀𝑃𝑆𝐴𝐷𝐶𝑘𝑡 < ∑𝑡 ∈(1994−1999)𝑋𝑃𝑆𝐴𝐷𝐶𝑘𝑡 ∑ ∑𝑘 𝑡 ∈(1994−1999)𝑋𝑃𝑆𝐴𝐷𝐶𝑘𝑡

22

The first condition implies that, for the EU to have a CA in sector k, EU net exports must be positive after trade imbalances have been corrected. The second condition implies, that SADC’s normalised net exports must, at the same time, be negative.

The sectors in which the SADC and ZA have a CA can be determined analogously.

With these conditions, three dummy variables can be defined: AdvEUk, AdvSADCk and AdvZAk which

take the value of one if the EU, the SADC or ZA, respectively, have the CA in sector k. Since the dummy variables are time-invariant, CA is assumed to be static during the time analysed.

Secondly, the correct dependent variable needs to be defined; the Partner Ratio. Since the goal is to see if the formation of the HAS affected trade flows between different regions and in different sectors, the trade data needs to be aggregated by region and sector: define three regions indexed by j where j=EU, SADC, ROW and define four types of sectors, k, depending on who has the CA: k=EU, SADC, ZA, MISC. Therefore, k=EU indicates that the EU has a CA in sector k; k=SADC and k=ZA imply that the SADC or ZA respectively enjoy the CA, while k=MISC are all other sectors. As for the identification of the CA above, the dependent variable needs to be normalised to correct for imbalanced trade by the aggregate trade between the countries. Since this normalisation considers the direction of trade, the dependent variable also differentiates between imports and exports. The dependent variable takes the form:

𝑃𝑎𝑟𝑡𝑛𝑒𝑟𝑅𝑎𝑡𝑖𝑜𝑗𝑘𝑑𝑡=

𝑇𝑟𝑎𝑑𝑒𝑗𝑘𝑑𝑡

∑ 𝑇𝑟𝑎𝑑𝑒𝑘 𝑗𝑘𝑑𝑡

Where d indicates the direction of trade: d = ex indicates ZA’s exports while d = im are imports from ZA’s perspective. Trade measures the value of directional, bilateral trade flows between ZA and region j in sector k in year t.

Finally, the regression equation can be specified as:

Equation 4.5.

𝑃𝑎𝑟𝑡𝑛𝑒𝑟𝑅𝑎𝑡𝑖𝑜𝑗𝑘𝑑𝑡

= 𝜆𝐴𝑑𝑣𝐸𝑈_𝐸𝑈𝑥+ 𝜆𝐴𝑑𝑣𝐸𝑈_𝐸𝑈𝑚+ 𝜆𝐴𝑑𝑣𝑆𝐴𝐷𝐶_𝑆𝐴𝐷𝐶𝑥+ 𝜆𝐴𝑑𝑣𝑆𝐴𝐷𝐶_𝑆𝐴𝐷𝐶𝑚+ 𝜆𝐴𝑑𝑣𝐸𝑈_𝑆𝐴𝐷𝐶𝑥

+ 𝜆𝐴𝑑𝑣𝐸𝑈_𝑆𝐴𝐷𝐶𝑚+ 𝜆𝐴𝑑𝑣𝑆𝐴𝐷𝐶_𝐸𝑈𝑥+ 𝜆𝐴𝑑𝑣𝑆𝐴𝐷𝐶_𝐸𝑈𝑚+ 𝜳𝟏𝒋𝒅𝒕+ 𝚿𝟐𝒋𝒅𝒕+ 𝚽𝒌𝒅+ 𝑣𝑗𝑘𝑑𝑡

To identify indirect arbitrage, the analysis relies on eight fixed effects expressed as dummy variables:

λAdvEu_EUx takes the value of one, if the EU has a CA and the trade flow is an export to the EU and the

HAS is in place, i.e. t ϵ {2000, 2014}. λAdvEu_EUm indicates that the EU enjoys the CA and the observation

corresponds to imports to ZA from the EU while the HAS is in effect. The other six fixed effects are defined symmetrically. The coefficients on the λs measure HAS effects in one sector and trade direction. For notational simplicity, the coefficients are left implicit and will be referred to by their respective λs.

To account for possible systemic factors, further fixed effects are included: Ψ1jdt is a vector of four

fixed effects for partner country and trade direction combinations prior to the signing of the FTAs. This

allows for different intercepts for partner-directions. Similarly, Ψ2jdt is defined as a vector of

partner-direction dummies following the FTA formation. Φkd is a vector of fixed effects controlling for

23

The theory predicts four testable relations:

4.1.4. Estimation Methods

Even though gravity equations are successful empirically, for a long time, micro-foundations were missing which made the approach susceptible to estimation biases (Baldwin and Taglioni, 2006; Anderson and van Wincoop, 2003). A full discussion of their theoretical basis is beyond the scope of this thesis. Instead, I will highlight some common empirical mistakes to be avoided. One additional empirical issue to be dealt with in gravity equations are zero trade flows. This sub-chapter will also discuss different estimators under the aspect of their performance when the sample includes (many) zero trade observations. The three estimators discussed are: Ordinary Least Squares (OLS), the Heckman correction and Poisson Pseudo-Maximum Likelihood (PPML). Additionally, recent literature has advanced many other possibly appealing estimators, including semi- and non-parametric estimation approaches, a discussion of which would exceed the scope of this thesis (cp. Martin and Pham, 2015).

4.1.4.1. Common Empirical Mistakes

Baldwin and Taglioni (2006) identify three types of empirical mistakes: gold-medal mistakes are caused by omitted variables that are correlated with the trade cost terms and bias the results. This problem is caused by the endogeneity of the FTA-formation decision. This implies an instrumental variable (IV) estimation approach but, since finding a suitable instrument can be problematic, country and country-pair dummies are a partial solution (De Benedictus and Taglioni, 2011). Silver-medal mistakes are linked to different measurements of trade flows (Baldwin and Taglioni, 2006). If trade is unbalanced and averages of bilateral trade are used, the appropriate specification of the dependent variable is the sum of the logs – not the log of the sums. If possible, import-export data should be used, since the trade direction is valuable information in the estimation process (De Benedictis and Taglioni, 2011).

Bronze-medal mistakes are related to the deflation of nominal trade values by the U.S. price index

which may create biases caused by spurious correlations, since inflation exhibits a global trend (Baldwin and Taglioni, 2006). Since gravity equations are modified expenditure equations, trade and GDP should not be deflated by a price index on theoretical grounds, regardless of the empirical issues related to deflation, e.g. the lack of appropriate deflators (De Benedictis and Taglioni, 2011). Therefore, in this thesis, all trade flows and GDP levels are in nominal terms and expressed in US$ as the common numeraire.

4.1.4.2. Ordinary Least Squares

Since gravity equations are usually estimated in log-log specification, OLS drops all zero trade flows since the log of zero is undefined. Therefore, it restricts the estimation to the sub-sample of non-zero

Proposition II: The effects of indirect arbitrage

1. ZA increased its net exports to the SADC in sectors in which the EU has a CA

𝜆

_{𝐴𝑑𝑣𝐸𝑈_𝑆𝐴𝐷𝐶𝑥}

− 𝜆

_{𝐴𝑑𝑣𝐸𝑈_𝑆𝐴𝐷𝐶𝑚}

> 0

2. ZA increased its net imports from the EU in sectors in which the EU has a CA

𝜆

_{𝐴𝑑𝑣𝐸𝑈_𝐸𝑈𝑥}

− 𝜆

_{𝐴𝑑𝑣𝐸𝑈_𝐸𝑈𝑚}

< 0

3. ZA increased its net exports to the EU in sectors in which the SADC has a CA

𝜆

𝐴𝑑𝑣𝑆𝐴𝐷𝐶_𝐸𝑈𝑥

− 𝜆

𝐴𝑑𝑣𝑆𝐴𝐷𝐶_𝐸𝑈𝑚

> 0

4. ZA increased its net imports from the SADC in sectors in which the SADC has a CA

𝜆

_{𝐴𝑑𝑣𝑆𝐴𝐷𝐶_𝑆𝐴𝐷𝐶𝑥}

− 𝜆

_{𝐴𝑑𝑣𝑆𝐴𝐷𝐶_𝑆𝐴𝐷𝐶𝑚}

< 0

24

observations.3 _{This may introduce an estimation bias, namely selection bias, because zero trade flows}

are non-random. They indicate the absence of trade suggesting trade barriers too high for a given level of demand and supply (De Benedictus and Taglioni, 2011). The selection bias is fundamentally an omitted variable bias with the probability of being selected (i.e. the probability of observing positive trade) as the omitted variable. This probability is certainly correlated with some of the regressors, since the probability of positive trade depends on the trade costs (and their proxies normally included in gravity equations, e.g. FTA dummy variables or distance) (Shepherd, 2012).

Another way of looking at endogeneity bias stems from the distribution of the log-error term under heteroscedasticity. Consider the standard gravity equation as introduced in chapter 2.1. (for simplicity without the vector of additional control variables):

𝑙𝑛(𝑇𝑖𝑗𝑡) = 𝛽0+ 𝛽1ln(𝑌𝑖𝑡) + 𝛽2ln(𝑌𝑗𝑡) + 𝛽3𝐹𝑇𝐴𝑖𝑗𝑡+ 𝜀𝑖𝑗𝑡

This gravity equation is the log-linearized form of the actual model equation: 𝑇𝑖𝑗𝑡 = 𝛽0𝑌𝑖𝑡𝛽1𝑌𝑗𝑡

𝛽2_𝐹𝑇𝐴

𝑖𝑗𝑡 𝛽3_𝜃

𝑖𝑗𝑡

Where 𝜀𝑖𝑗𝑡 = ln (𝜃𝑖𝑗𝑡). Since 𝜃𝑖𝑗𝑡 appears in log-form in the log-linearized equation, the expected value

of 𝜀𝑖𝑗𝑡 = ln (𝜃𝑖𝑗𝑡) depends on the higher moments of 𝜃𝑖𝑗𝑡, including its variance. If 𝜃𝑖𝑗𝑡 is

heteroscedastic, which is probable, then the expected value of εijt depends on one or more

independent variables since it depends on the variance of 𝜃𝑖𝑗𝑡. This violates the Conditional Mean Zero

or Exogeneity assumption needed for consistent and unbiased OLS estimation, i.e. 𝐸(𝜀𝑖𝑗𝑡|𝑌𝑖𝑡, 𝑌𝑗𝑡, 𝐹𝑇𝐴𝑖𝑗𝑡) ≠ 0. This type of heteroscedasticity cannot be controlled for by fixed effects

since it does not only affect the estimated standard errors but also the parameter estimates (Shepherd, 2012). Using simulation results, Santos Silva and Tenreyro (2006, 2011) show that heteroscedasticity is quantitively and qualitatively important and the bias introduced by heteroscedasticity can be severe, even when fixed effects are included. Generally, OLS estimates for exporter and importer fixed effects are downward biased for large countries and upward biased for small countries, implying that OLS overestimates trade for large markets and underestimates it for small ones (Fally, 2015).

Despite its many econometric issues, OLS is still often considered the benchmark estimation method.

Fixed Effects Estimation

To take advantage of the panel structure of the data, in addition to (pooled) OLS, a fixed effect (FE) estimation will be carried out. FE estimation controls for unobserved, time-invariant, entity-specific heterogeneity that may be correlated with the other regressors.

Numerically, FE estimation and pooled OLS with a full set of country- and time-dummy variables should yield the same results, despite their computationally different approaches (Torres-Reyna, 2007).

4.1.4.3. Heckman Correction

The Heckman correction (Heckman, 1979), is a two-stage estimation that splits the usual gravity model into two equations: an outcome equation, which identifies that relation between trade flows and a set of explanatory variables, and a selection equation, which relates the unobservable probability of observing positive trade to the same set of explanatory variables and, preferably, one extra explanatory variable not included in the outcome equation, i.e. a variable that affects the probability bilateral trade occurs but not the value of trade: the exclusion restriction. The first step is to estimate

3 _{Some empirical studies have tried to solve this issue by adding a small constant to trade flows, making all}

observations strictly positive and allowing for a log transformation of the entire sample. But, Santos Silva and Tenreyro (2006) show that this will lead to an estimation bias that may be substantial.

25

the probability that a certain observation will be in the estimation sample using a probit model. The

probit results are then used to estimate the inverse Mills ratio, 𝜆̂𝑖𝑗𝑡, which, if included in the outcome

equation, corrects the selection bias. Generally, in empirical work, instead of estimating the Heckman model in two stages, a maximum likelihood estimation is used to estimate the outcome and selection equation simultaneously to avoid econometric complications (Shepherd, 2012; Bushway et al., 2007).

The Exclusion Restriction

One empirical issue related to the Heckman procedure is the exclusion restriction. The exclusion restriction is not strictly necessary. However, in the absence of a valid exclusion restriction, the model is identifiable solely by the non-linearity of the inverse Mills ratio (Busway et al., 2007). But, since the inverse Mills ratio approximates a linear function over wide, mid-level ranges for the independent variables, this increases multicollinearity concerns, reduces robustness and inflates standard errors (Puhani, 2000). I will follow Helpman et al. (2008) in setting up a variable measuring religious similarities. They argue that the “probability that two randomly drawn persons, one from each country, share the same religion raises export volumes” (Helpman et al., 2008, p.459) and that a common language and religion affect the fixed costs of trade, and therefore only the trade decision, not the trade volume, and are valid exclusion restrictions. The religious similarity variable is also used by Gauto (2012).

4.1.4.4. The Poisson Pseudo-Maximum Likelihood Estimator

The PPML estimator was introduced to the gravity literature by Santos Silva and Tenreyro (2006). The intuitive idea behind the PPML relies on estimating the model in its original, non-log linearized form. Therefore, PPML estimates the level-form of the model introduced in chapter 4.1.4.2.:

𝑇𝑖𝑗𝑡 = 𝑒𝛽0𝑌𝑖𝑡𝛽1𝑌𝑗𝑡 𝛽₂

𝐹𝑇𝐴_𝑖𝑗𝑡𝛽3_𝜃

𝑖𝑗𝑡

Normally, the Poisson model is used for count data where the depend variable can only take on positive, integer values, but it is increasingly used for other kinds of data. Assuming the gravity equation includes the correct independent variables, the PPML estimator yields consistent estimates without the data needing to be distributed as Poisson or even as count data. The only distributional assumption imposed is that the conditional variance of Tijt is proportional to its conditional mean

(Santos Silva and Tenreyro, 2006).

Zero-Inflation Poisson

Simulations have shown that, in large samples, the Zero-inflated Poisson (ZIP) estimator leads to a smaller bias. Empirically, these estimators, the PPML and ZIP, might lead to only slightly different or even completely identical estimation results (Martin and Pham, 2015). The ZIP model is usually used for data that follows a Poisson distribution but exhibits an excess number of zeros. ZIP should only be applied to large samples, since it is even harder to fit than standard logistic models since it is not known which zeros are due to the logit and which due to the Poisson model, and even standard logistic regression maximum likelihood estimators have an infinite bias in finite samples (Lambert, 1992).

4.1.5. Pre- and Post-Estimation Testing

Before and after the regression analysis, multiple tests will be carried out.

Pre-estimation, unit root tests will be applied to judge the degree of autocorrelation in GDP and in trade. The augmented Dickey-Fuller test and the Phillips-Perron unit root tests will be used with different specification (Dickey and Fuller, 1979; Phillips and Perron, 1988). Additionally, the Hausman test and the test of overidentifying restrictions (orthogonality conditions) will be employed to choose between FE and random effects (RE) (Hausman, 1978; Arellano, 1993).